Revert "fix(test suite): forward compat test is not using latest neon_local (#7637 )"

This reverts commit ea531d448e.
Revert "chore(neon_test_utils): restrict installation to superuser" (#7679 )
2026-07-14 17:40:37 +00:00 · 2024-05-09 13:22:06 -04:00 · 2024-05-09 15:15:19 +00:00 · 2024-05-09 16:01:16 +02:00 · 2024-05-09 12:33:09 +01:00 · 2024-05-09 10:07:59 +01:00
125 changed files with 5599 additions and 1778 deletions
--- a/.github/workflows/build_and_test.yml
+++ b/.github/workflows/build_and_test.yml
@@ -236,27 +236,6 @@ jobs:
          submodules: true
          fetch-depth: 1

-      - name: Check Postgres submodules revision
-        shell: bash -euo pipefail {0}
-        run: |
-          # This is a temporary solution to ensure that the Postgres submodules revision is correct (i.e. the updated intentionally).
-          # Eventually it will be replaced by a regression test https://github.com/neondatabase/neon/pull/4603
-
-          FAILED=false
-          for postgres in postgres-v14 postgres-v15 postgres-v16; do
-            expected=$(cat vendor/revisions.json | jq --raw-output '."'"${postgres}"'"')
-            actual=$(git rev-parse "HEAD:vendor/${postgres}")
-            if [ "${expected}" != "${actual}" ]; then
-              echo >&2 "Expected ${postgres} rev to be at '${expected}', but it is at '${actual}'"
-              FAILED=true
-            fi
-          done
-
-          if [ "${FAILED}" = "true" ]; then
-            echo >&2 "Please update vendor/revisions.json if these changes are intentional"
-            exit 1
-          fi
-
      - name: Set pg 14 revision for caching
        id: pg_v14_rev
        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v14) >> $GITHUB_OUTPUT
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -25,9 +25,9 @@ checksum = "f26201604c87b1e01bd3d98f8d5d9a8fcbb815e8cedb41ffccbeb4bf593a35fe"

 [[package]]
 name = "ahash"
-version = "0.8.9"
+version = "0.8.11"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d713b3834d76b85304d4d525563c1276e2e30dc97cc67bfb4585a4a29fc2c89f"
+checksum = "e89da841a80418a9b391ebaea17f5c112ffaaa96f621d2c285b5174da76b9011"
 dependencies = [
 "cfg-if",
 "const-random",
@@ -284,9 +284,9 @@ checksum = "d468802bab17cbc0cc575e9b053f41e72aa36bfa6b7f55e3529ffa43161b97fa"

 [[package]]
 name = "aws-config"
-version = "1.1.4"
+version = "1.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8b30c39ebe61f75d1b3785362b1586b41991873c9ab3e317a9181c246fb71d82"
+checksum = "baaa0be6ee7d90b775ae6ccb6d2ba182b91219ec2001f92338773a094246af1d"
 dependencies = [
 "aws-credential-types",
 "aws-runtime",
@@ -309,14 +309,15 @@ dependencies = [
 "time",
 "tokio",
 "tracing",
+ "url",
 "zeroize",
 ]

 [[package]]
 name = "aws-credential-types"
-version = "1.1.8"
+version = "1.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fa8587ae17c8e967e4b05a62d495be2fb7701bec52a97f7acfe8a29f938384c8"
+checksum = "e16838e6c9e12125face1c1eff1343c75e3ff540de98ff7ebd61874a89bcfeb9"
 dependencies = [
 "aws-smithy-async",
 "aws-smithy-runtime-api",
@@ -326,9 +327,9 @@ dependencies = [

 [[package]]
 name = "aws-runtime"
-version = "1.1.8"
+version = "1.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b13dc54b4b49f8288532334bba8f87386a40571c47c37b1304979b556dc613c8"
+checksum = "785da4a15e7b166b505fd577e4560c7a7cd8fbdf842eb1336cbcbf8944ce56f1"
 dependencies = [
 "aws-credential-types",
 "aws-sigv4",
@@ -373,10 +374,11 @@ dependencies = [

 [[package]]
 name = "aws-sdk-s3"
-version = "1.14.0"
+version = "1.26.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "951f7730f51a2155c711c85c79f337fbc02a577fa99d2a0a8059acfce5392113"
+checksum = "7bc5ce518d4b8d16e0408de7bdf1b3097cec61a7daa979750a208f8d9934386d"
 dependencies = [
+ "ahash",
 "aws-credential-types",
 "aws-runtime",
 "aws-sigv4",
@@ -391,20 +393,25 @@ dependencies = [
 "aws-smithy-xml",
 "aws-types",
 "bytes",
+ "fastrand 2.0.0",
+ "hex",
+ "hmac",
 "http 0.2.9",
 "http-body 0.4.5",
+ "lru",
 "once_cell",
 "percent-encoding",
 "regex-lite",
+ "sha2",
 "tracing",
 "url",
 ]

 [[package]]
 name = "aws-sdk-sso"
-version = "1.12.0"
+version = "1.22.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f486420a66caad72635bc2ce0ff6581646e0d32df02aa39dc983bfe794955a5b"
+checksum = "ca3d6c4cba4e009391b72b0fcf12aff04ea3c9c3aa2ecaafa330326a8bd7e601"
 dependencies = [
 "aws-credential-types",
 "aws-runtime",
@@ -424,9 +431,9 @@ dependencies = [

 [[package]]
 name = "aws-sdk-ssooidc"
-version = "1.12.0"
+version = "1.22.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "39ddccf01d82fce9b4a15c8ae8608211ee7db8ed13a70b514bbfe41df3d24841"
+checksum = "73400dc239d14f63d932f4ca7b55af5e9ef1f857f7d70655249ccc287adb2570"
 dependencies = [
 "aws-credential-types",
 "aws-runtime",
@@ -446,9 +453,9 @@ dependencies = [

 [[package]]
 name = "aws-sdk-sts"
-version = "1.12.0"
+version = "1.22.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1a591f8c7e6a621a501b2b5d2e88e1697fcb6274264523a6ad4d5959889a41ce"
+checksum = "10f8858308af76fba3e5ffcf1bb56af5471574d2bdfaf0159470c25bc2f760e5"
 dependencies = [
 "aws-credential-types",
 "aws-runtime",
@@ -469,9 +476,9 @@ dependencies = [

 [[package]]
 name = "aws-sigv4"
-version = "1.2.0"
+version = "1.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "11d6f29688a4be9895c0ba8bef861ad0c0dac5c15e9618b9b7a6c233990fc263"
+checksum = "58b56f1cbe6fd4d0c2573df72868f20ab1c125ca9c9dbce17927a463433a2e57"
 dependencies = [
 "aws-credential-types",
 "aws-smithy-eventstream",
@@ -498,9 +505,9 @@ dependencies = [

 [[package]]
 name = "aws-smithy-async"
-version = "1.1.8"
+version = "1.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d26ea8fa03025b2face2b3038a63525a10891e3d8829901d502e5384a0d8cd46"
+checksum = "62220bc6e97f946ddd51b5f1361f78996e704677afc518a4ff66b7a72ea1378c"
 dependencies = [
 "futures-util",
 "pin-project-lite",
@@ -509,9 +516,9 @@ dependencies = [

 [[package]]
 name = "aws-smithy-checksums"
-version = "0.60.4"
+version = "0.60.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "be2acd1b9c6ae5859999250ed5a62423aedc5cf69045b844432de15fa2f31f2b"
+checksum = "83fa43bc04a6b2441968faeab56e68da3812f978a670a5db32accbdcafddd12f"
 dependencies = [
 "aws-smithy-http",
 "aws-smithy-types",
@@ -541,9 +548,9 @@ dependencies = [

 [[package]]
 name = "aws-smithy-http"
-version = "0.60.7"
+version = "0.60.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3f10fa66956f01540051b0aa7ad54574640f748f9839e843442d99b970d3aff9"
+checksum = "4a7de001a1b9a25601016d8057ea16e31a45fdca3751304c8edf4ad72e706c08"
 dependencies = [
 "aws-smithy-eventstream",
 "aws-smithy-runtime-api",
@@ -581,9 +588,9 @@ dependencies = [

 [[package]]
 name = "aws-smithy-runtime"
-version = "1.1.8"
+version = "1.5.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ec81002d883e5a7fd2bb063d6fb51c4999eb55d404f4fff3dd878bf4733b9f01"
+checksum = "c9ac79e9f3a4d576f3cd4a470a0275b138d9e7b11b1cd514a6858ae0a79dd5bb"
 dependencies = [
 "aws-smithy-async",
 "aws-smithy-http",
@@ -594,6 +601,7 @@ dependencies = [
 "h2 0.3.26",
 "http 0.2.9",
 "http-body 0.4.5",
+ "http-body 1.0.0",
 "hyper 0.14.26",
 "hyper-rustls 0.24.0",
 "once_cell",
@@ -606,9 +614,9 @@ dependencies = [

 [[package]]
 name = "aws-smithy-runtime-api"
-version = "1.2.0"
+version = "1.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9acb931e0adaf5132de878f1398d83f8677f90ba70f01f65ff87f6d7244be1c5"
+checksum = "04ec42c2f5c0e7796a2848dde4d9f3bf8ce12ccbb3d5aa40c52fa0cdd61a1c47"
 dependencies = [
 "aws-smithy-async",
 "aws-smithy-types",
@@ -623,16 +631,19 @@ dependencies = [

 [[package]]
 name = "aws-smithy-types"
-version = "1.1.8"
+version = "1.1.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "abe14dceea1e70101d38fbf2a99e6a34159477c0fb95e68e05c66bd7ae4c3729"
+checksum = "baf98d97bba6ddaba180f1b1147e202d8fe04940403a95a3f826c790f931bbd1"
 dependencies = [
 "base64-simd",
 "bytes",
 "bytes-utils",
 "futures-core",
 "http 0.2.9",
+ "http 1.1.0",
 "http-body 0.4.5",
+ "http-body 1.0.0",
+ "http-body-util",
 "itoa",
 "num-integer",
 "pin-project-lite",
@@ -646,18 +657,18 @@ dependencies = [

 [[package]]
 name = "aws-smithy-xml"
-version = "0.60.7"
+version = "0.60.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "872c68cf019c0e4afc5de7753c4f7288ce4b71663212771bf5e4542eb9346ca9"
+checksum = "d123fbc2a4adc3c301652ba8e149bf4bc1d1725affb9784eb20c953ace06bf55"
 dependencies = [
 "xmlparser",
 ]

 [[package]]
 name = "aws-types"
-version = "1.1.8"
+version = "1.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0dbf2f3da841a8930f159163175cf6a3d16ddde517c1b0fba7aa776822800f40"
+checksum = "5a43b56df2c529fe44cb4d92bd64d0479883fb9608ff62daede4df5405381814"
 dependencies = [
 "aws-credential-types",
 "aws-smithy-async",
@@ -2935,6 +2946,15 @@ version = "0.4.20"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b5e6163cb8c49088c2c36f57875e58ccd8c87c7427f7fbd50ea6710b2f3f2e8f"

+[[package]]
+name = "lru"
+version = "0.12.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d3262e75e648fce39813cb56ac41f3c3e3f65217ebf3844d818d1f9398cfb0dc"
+dependencies = [
+ "hashbrown 0.14.0",
+]
+
 [[package]]
 name = "match_cfg"
 version = "0.1.0"
@@ -4372,6 +4392,7 @@ dependencies = [
 "hyper 1.2.0",
 "hyper-tungstenite",
 "hyper-util",
+ "indexmap 2.0.1",
 "ipnet",
 "itertools",
 "lasso",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -52,14 +52,14 @@ azure_storage_blobs = "0.19"
 flate2 = "1.0.26"
 async-stream = "0.3"
 async-trait = "0.1"
-aws-config = { version = "1.1.4", default-features = false, features=["rustls"] }
-aws-sdk-s3 = "1.14"
+aws-config = { version = "1.3", default-features = false, features=["rustls"] }
+aws-sdk-s3 = "1.26"
 aws-sdk-iam = "1.15.0"
-aws-smithy-async = { version = "1.1.4", default-features = false, features=["rt-tokio"] }
-aws-smithy-types = "1.1.4"
-aws-credential-types = "1.1.4"
-aws-sigv4 = { version = "1.2.0", features = ["sign-http"] }
-aws-types = "1.1.7"
+aws-smithy-async = { version = "1.2.1", default-features = false, features=["rt-tokio"] }
+aws-smithy-types = "1.1.9"
+aws-credential-types = "1.2.0"
+aws-sigv4 = { version = "1.2.1", features = ["sign-http"] }
+aws-types = "1.2.0"
 axum = { version = "0.6.20", features = ["ws"] }
 base64 = "0.13.0"
 bincode = "1.3"
@@ -99,6 +99,7 @@ humantime = "2.1"
 humantime-serde = "1.1.1"
 hyper = "0.14"
 hyper-tungstenite = "0.13.0"
+indexmap = "2"
 inotify = "0.10.2"
 ipnet = "2.9.0"
 itertools = "0.10"
--- a/11
+++ b/11
@@ -81,11 +81,14 @@ $(POSTGRES_INSTALL_DIR)/build/%/config.status:
 		echo "'git submodule update --init --recursive --depth 2 --progress .' in project root.\n"; \
 		exit 1; }
 	mkdir -p $(POSTGRES_INSTALL_DIR)/build/$*
-	(cd $(POSTGRES_INSTALL_DIR)/build/$* && \
-	env PATH="$(EXTRA_PATH_OVERRIDES):$$PATH" $(ROOT_PROJECT_DIR)/vendor/postgres-$*/configure \
+
+	VERSION=$*; \
+	EXTRA_VERSION=$$(cd $(ROOT_PROJECT_DIR)/vendor/postgres-$$VERSION && git rev-parse HEAD); \
+	(cd $(POSTGRES_INSTALL_DIR)/build/$$VERSION && \
+	env PATH="$(EXTRA_PATH_OVERRIDES):$$PATH" $(ROOT_PROJECT_DIR)/vendor/postgres-$$VERSION/configure \
 		CFLAGS='$(PG_CFLAGS)' \
-		$(PG_CONFIGURE_OPTS) \
-		--prefix=$(abspath $(POSTGRES_INSTALL_DIR))/$* > configure.log)
+		$(PG_CONFIGURE_OPTS) --with-extra-version=" ($$EXTRA_VERSION)" \
+		--prefix=$(abspath $(POSTGRES_INSTALL_DIR))/$$VERSION > configure.log)

 # nicer alias to run 'configure'
 # Note: I've been unable to use templates for this part of our configuration.
--- a/compute_tools/src/bin/compute_ctl.rs
+++ b/compute_tools/src/bin/compute_ctl.rs
@@ -51,6 +51,7 @@ use tracing::{error, info, warn};
 use url::Url;

 use compute_api::responses::ComputeStatus;
+use compute_api::spec::ComputeSpec;

 use compute_tools::compute::{
    forward_termination_signal, ComputeNode, ComputeState, ParsedSpec, PG_PID,
@@ -69,6 +70,34 @@ use compute_tools::swap::resize_swap;
 const BUILD_TAG_DEFAULT: &str = "latest";

 fn main() -> Result<()> {
+    let (build_tag, clap_args) = init()?;
+
+    let (pg_handle, start_pg_result) = {
+        // Enter startup tracing context
+        let _startup_context_guard = startup_context_from_env();
+
+        let cli_args = process_cli(&clap_args)?;
+
+        let cli_spec = try_spec_from_cli(&clap_args, &cli_args)?;
+
+        let wait_spec_result = wait_spec(build_tag, cli_args, cli_spec)?;
+
+        start_postgres(&clap_args, wait_spec_result)?
+
+        // Startup is finished, exit the startup tracing span
+    };
+
+    // PostgreSQL is now running, if startup was successful. Wait until it exits.
+    let wait_pg_result = wait_postgres(pg_handle)?;
+
+    let delay_exit = cleanup_after_postgres_exit(start_pg_result)?;
+
+    maybe_delay_exit(delay_exit);
+
+    deinit_and_exit(wait_pg_result);
+}
+
+fn init() -> Result<(String, clap::ArgMatches)> {
    init_tracing_and_logging(DEFAULT_LOG_LEVEL)?;

    let mut signals = Signals::new([SIGINT, SIGTERM, SIGQUIT])?;
@@ -83,9 +112,15 @@ fn main() -> Result<()> {
        .to_string();
    info!("build_tag: {build_tag}");

-    let matches = cli().get_matches();
-    let pgbin_default = String::from("postgres");
-    let pgbin = matches.get_one::<String>("pgbin").unwrap_or(&pgbin_default);
+    Ok((build_tag, cli().get_matches()))
+}
+
+fn process_cli(matches: &clap::ArgMatches) -> Result<ProcessCliResult> {
+    let pgbin_default = "postgres";
+    let pgbin = matches
+        .get_one::<String>("pgbin")
+        .map(|s| s.as_str())
+        .unwrap_or(pgbin_default);

    let ext_remote_storage = matches
        .get_one::<String>("remote-ext-config")
@@ -113,6 +148,30 @@ fn main() -> Result<()> {
    let spec_path = matches.get_one::<String>("spec-path");
    let resize_swap_on_bind = matches.get_flag("resize-swap-on-bind");

+    Ok(ProcessCliResult {
+        connstr,
+        pgdata,
+        pgbin,
+        ext_remote_storage,
+        http_port,
+        spec_json,
+        spec_path,
+        resize_swap_on_bind,
+    })
+}
+
+struct ProcessCliResult<'clap> {
+    connstr: &'clap str,
+    pgdata: &'clap str,
+    pgbin: &'clap str,
+    ext_remote_storage: Option<&'clap str>,
+    http_port: u16,
+    spec_json: Option<&'clap String>,
+    spec_path: Option<&'clap String>,
+    resize_swap_on_bind: bool,
+}
+
+fn startup_context_from_env() -> Option<opentelemetry::ContextGuard> {
    // Extract OpenTelemetry context for the startup actions from the
    // TRACEPARENT and TRACESTATE env variables, and attach it to the current
    // tracing context.
@@ -149,7 +208,7 @@ fn main() -> Result<()> {
    if let Ok(val) = std::env::var("TRACESTATE") {
        startup_tracing_carrier.insert("tracestate".to_string(), val);
    }
-    let startup_context_guard = if !startup_tracing_carrier.is_empty() {
+    if !startup_tracing_carrier.is_empty() {
        use opentelemetry::propagation::TextMapPropagator;
        use opentelemetry::sdk::propagation::TraceContextPropagator;
        let guard = TraceContextPropagator::new()
@@ -159,8 +218,17 @@ fn main() -> Result<()> {
        Some(guard)
    } else {
        None
-    };
+    }
+}

+fn try_spec_from_cli(
+    matches: &clap::ArgMatches,
+    ProcessCliResult {
+        spec_json,
+        spec_path,
+        ..
+    }: &ProcessCliResult,
+) -> Result<CliSpecParams> {
    let compute_id = matches.get_one::<String>("compute-id");
    let control_plane_uri = matches.get_one::<String>("control-plane-uri");

@@ -201,6 +269,34 @@ fn main() -> Result<()> {
        }
    };

+    Ok(CliSpecParams {
+        spec,
+        live_config_allowed,
+    })
+}
+
+struct CliSpecParams {
+    /// If a spec was provided via CLI or file, the [`ComputeSpec`]
+    spec: Option<ComputeSpec>,
+    live_config_allowed: bool,
+}
+
+fn wait_spec(
+    build_tag: String,
+    ProcessCliResult {
+        connstr,
+        pgdata,
+        pgbin,
+        ext_remote_storage,
+        resize_swap_on_bind,
+        http_port,
+        ..
+    }: ProcessCliResult,
+    CliSpecParams {
+        spec,
+        live_config_allowed,
+    }: CliSpecParams,
+) -> Result<WaitSpecResult> {
    let mut new_state = ComputeState::new();
    let spec_set;

@@ -239,8 +335,6 @@ fn main() -> Result<()> {
    let _http_handle =
        launch_http_server(http_port, &compute).expect("cannot launch http endpoint thread");

-    let extension_server_port: u16 = http_port;
-
    if !spec_set {
        // No spec provided, hang waiting for it.
        info!("no compute spec provided, waiting");
@@ -269,6 +363,29 @@ fn main() -> Result<()> {
        state.start_time = now;
    }

+    Ok(WaitSpecResult {
+        compute,
+        http_port,
+        resize_swap_on_bind,
+    })
+}
+
+struct WaitSpecResult {
+    compute: Arc<ComputeNode>,
+    // passed through from ProcessCliResult
+    http_port: u16,
+    resize_swap_on_bind: bool,
+}
+
+fn start_postgres(
+    // need to allow unused because `matches` is only used if target_os = "linux"
+    #[allow(unused_variables)] matches: &clap::ArgMatches,
+    WaitSpecResult {
+        compute,
+        http_port,
+        resize_swap_on_bind,
+    }: WaitSpecResult,
+) -> Result<(Option<PostgresHandle>, StartPostgresResult)> {
    // We got all we need, update the state.
    let mut state = compute.state.lock().unwrap();
    state.status = ComputeStatus::Init;
@@ -318,10 +435,10 @@ fn main() -> Result<()> {
        }
    }

+    let extension_server_port: u16 = http_port;
+
    // Start Postgres
    let mut pg = None;
-    let mut exit_code = None;
-
    if !prestartup_failed {
        pg = match compute.start_compute(extension_server_port) {
            Ok(pg) => Some(pg),
@@ -376,7 +493,7 @@ fn main() -> Result<()> {
            // This token is used internally by the monitor to clean up all threads
            let token = CancellationToken::new();

-            let vm_monitor = &rt.as_ref().map(|rt| {
+            let vm_monitor = rt.as_ref().map(|rt| {
                rt.spawn(vm_monitor::start(
                    Box::leak(Box::new(vm_monitor::Args {
                        cgroup: cgroup.cloned(),
@@ -389,12 +506,41 @@ fn main() -> Result<()> {
        }
    }

+    Ok((
+        pg,
+        StartPostgresResult {
+            delay_exit,
+            compute,
+            #[cfg(target_os = "linux")]
+            rt,
+            #[cfg(target_os = "linux")]
+            token,
+            #[cfg(target_os = "linux")]
+            vm_monitor,
+        },
+    ))
+}
+
+type PostgresHandle = (std::process::Child, std::thread::JoinHandle<()>);
+
+struct StartPostgresResult {
+    delay_exit: bool,
+    // passed through from WaitSpecResult
+    compute: Arc<ComputeNode>,
+
+    #[cfg(target_os = "linux")]
+    rt: Option<tokio::runtime::Runtime>,
+    #[cfg(target_os = "linux")]
+    token: tokio_util::sync::CancellationToken,
+    #[cfg(target_os = "linux")]
+    vm_monitor: Option<tokio::task::JoinHandle<Result<()>>>,
+}
+
+fn wait_postgres(pg: Option<PostgresHandle>) -> Result<WaitPostgresResult> {
    // Wait for the child Postgres process forever. In this state Ctrl+C will
    // propagate to Postgres and it will be shut down as well.
+    let mut exit_code = None;
    if let Some((mut pg, logs_handle)) = pg {
-        // Startup is finished, exit the startup tracing span
-        drop(startup_context_guard);
-
        let ecode = pg
            .wait()
            .expect("failed to start waiting on Postgres process");
@@ -409,6 +555,25 @@ fn main() -> Result<()> {
        exit_code = ecode.code()
    }

+    Ok(WaitPostgresResult { exit_code })
+}
+
+struct WaitPostgresResult {
+    exit_code: Option<i32>,
+}
+
+fn cleanup_after_postgres_exit(
+    StartPostgresResult {
+        mut delay_exit,
+        compute,
+        #[cfg(target_os = "linux")]
+        vm_monitor,
+        #[cfg(target_os = "linux")]
+        token,
+        #[cfg(target_os = "linux")]
+        rt,
+    }: StartPostgresResult,
+) -> Result<bool> {
    // Terminate the vm_monitor so it releases the file watcher on
    // /sys/fs/cgroup/neon-postgres.
    // Note: the vm-monitor only runs on linux because it requires cgroups.
@@ -450,13 +615,19 @@ fn main() -> Result<()> {
        error!("error while checking for core dumps: {err:?}");
    }

+    Ok(delay_exit)
+}
+
+fn maybe_delay_exit(delay_exit: bool) {
    // If launch failed, keep serving HTTP requests for a while, so the cloud
    // control plane can get the actual error.
    if delay_exit {
        info!("giving control plane 30s to collect the error before shutdown");
        thread::sleep(Duration::from_secs(30));
    }
+}

+fn deinit_and_exit(WaitPostgresResult { exit_code }: WaitPostgresResult) -> ! {
    // Shutdown trace pipeline gracefully, so that it has a chance to send any
    // pending traces before we exit. Shutting down OTEL tracing provider may
    // hang for quite some time, see, for example:
--- a/control_plane/src/bin/neon_local.rs
+++ b/control_plane/src/bin/neon_local.rs
@@ -9,8 +9,11 @@ use anyhow::{anyhow, bail, Context, Result};
 use clap::{value_parser, Arg, ArgAction, ArgMatches, Command, ValueEnum};
 use compute_api::spec::ComputeMode;
 use control_plane::endpoint::ComputeControlPlane;
-use control_plane::local_env::{InitForceMode, LocalEnv};
-use control_plane::pageserver::{PageServerNode, PAGESERVER_REMOTE_STORAGE_DIR};
+use control_plane::local_env::{
+    InitForceMode, LocalEnv, NeonBroker, NeonLocalInitConf, NeonLocalInitPageserverConf,
+    SafekeeperConf,
+};
+use control_plane::pageserver::PageServerNode;
 use control_plane::safekeeper::SafekeeperNode;
 use control_plane::storage_controller::StorageController;
 use control_plane::{broker, local_env};
@@ -52,44 +55,6 @@ const DEFAULT_PG_VERSION: &str = "15";

 const DEFAULT_PAGESERVER_CONTROL_PLANE_API: &str = "http://127.0.0.1:1234/upcall/v1/";

-fn default_conf(num_pageservers: u16) -> String {
-    let mut template = format!(
-        r#"
-# Default built-in configuration, defined in main.rs
-control_plane_api = '{DEFAULT_PAGESERVER_CONTROL_PLANE_API}'
-
-[broker]
-listen_addr = '{DEFAULT_BROKER_ADDR}'
-
-[[safekeepers]]
-id = {DEFAULT_SAFEKEEPER_ID}
-pg_port = {DEFAULT_SAFEKEEPER_PG_PORT}
-http_port = {DEFAULT_SAFEKEEPER_HTTP_PORT}
-
-"#,
-    );
-
-    for i in 0..num_pageservers {
-        let pageserver_id = NodeId(DEFAULT_PAGESERVER_ID.0 + i as u64);
-        let pg_port = DEFAULT_PAGESERVER_PG_PORT + i;
-        let http_port = DEFAULT_PAGESERVER_HTTP_PORT + i;
-
-        template += &format!(
-            r#"
-[[pageservers]]
-id = {pageserver_id}
-listen_pg_addr = '127.0.0.1:{pg_port}'
-listen_http_addr = '127.0.0.1:{http_port}'
-pg_auth_type = '{trust_auth}'
-http_auth_type = '{trust_auth}'
-"#,
-            trust_auth = AuthType::Trust,
-        )
-    }
-
-    template
-}
-
 ///
 /// Timelines tree element used as a value in the HashMap.
 ///
@@ -152,7 +117,7 @@ fn main() -> Result<()> {
    };

    match subcommand_result {
-        Ok(Some(updated_env)) => updated_env.persist_config(&updated_env.base_data_dir)?,
+        Ok(Some(updated_env)) => updated_env.persist_config()?,
        Ok(None) => (),
        Err(e) => {
            eprintln!("command failed: {e:?}");
@@ -341,55 +306,65 @@ fn parse_timeline_id(sub_match: &ArgMatches) -> anyhow::Result<Option<TimelineId
 }

 fn handle_init(init_match: &ArgMatches) -> anyhow::Result<LocalEnv> {
-    let num_pageservers = init_match
-        .get_one::<u16>("num-pageservers")
-        .expect("num-pageservers arg has a default");
-    // Create config file
-    let toml_file: String = if let Some(config_path) = init_match.get_one::<PathBuf>("config") {
+    let num_pageservers = init_match.get_one::<u16>("num-pageservers");
+
+    let force = init_match.get_one("force").expect("we set a default value");
+
+    // Create the in-memory `LocalEnv` that we'd normally load from disk in `load_config`.
+    let init_conf: NeonLocalInitConf = if let Some(config_path) =
+        init_match.get_one::<PathBuf>("config")
+    {
+        // User (likely the Python test suite) provided a description of the environment.
+        if num_pageservers.is_some() {
+            bail!("Cannot specify both --num-pageservers and --config, use key `pageservers` in the --config file instead");
+        }
        // load and parse the file
-        std::fs::read_to_string(config_path).with_context(|| {
+        let contents = std::fs::read_to_string(config_path).with_context(|| {
            format!(
                "Could not read configuration file '{}'",
                config_path.display()
            )
-        })?
+        })?;
+        toml_edit::de::from_str(&contents)?
    } else {
-        // Built-in default config
-        default_conf(*num_pageservers)
+        // User (likely interactive) did not provide a description of the environment, give them the default
+        NeonLocalInitConf {
+            control_plane_api: Some(Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap())),
+            broker: NeonBroker {
+                listen_addr: DEFAULT_BROKER_ADDR.parse().unwrap(),
+            },
+            safekeepers: vec![SafekeeperConf {
+                id: DEFAULT_SAFEKEEPER_ID,
+                pg_port: DEFAULT_SAFEKEEPER_PG_PORT,
+                http_port: DEFAULT_SAFEKEEPER_HTTP_PORT,
+                ..Default::default()
+            }],
+            pageservers: (0..num_pageservers.copied().unwrap_or(1))
+                .map(|i| {
+                    let pageserver_id = NodeId(DEFAULT_PAGESERVER_ID.0 + i as u64);
+                    let pg_port = DEFAULT_PAGESERVER_PG_PORT + i;
+                    let http_port = DEFAULT_PAGESERVER_HTTP_PORT + i;
+                    NeonLocalInitPageserverConf {
+                        id: pageserver_id,
+                        listen_pg_addr: format!("127.0.0.1:{pg_port}"),
+                        listen_http_addr: format!("127.0.0.1:{http_port}"),
+                        pg_auth_type: AuthType::Trust,
+                        http_auth_type: AuthType::Trust,
+                        other: Default::default(),
+                    }
+                })
+                .collect(),
+            pg_distrib_dir: None,
+            neon_distrib_dir: None,
+            default_tenant_id: TenantId::from_array(std::array::from_fn(|_| 0)),
+            storage_controller: None,
+            control_plane_compute_hook_api: None,
+        }
    };

-    let pageserver_config: toml_edit::Document =
-        if let Some(path) = init_match.get_one::<PathBuf>("pageserver-config") {
-            std::fs::read_to_string(path)?.parse()?
-        } else {
-            toml_edit::Document::new()
-        };
-
-    let pg_version = init_match
-        .get_one::<u32>("pg-version")
-        .copied()
-        .context("Failed to parse postgres version from the argument string")?;
-
-    let mut env =
-        LocalEnv::parse_config(&toml_file).context("Failed to create neon configuration")?;
-    let force = init_match.get_one("force").expect("we set a default value");
-    env.init(pg_version, force)
-        .context("Failed to initialize neon repository")?;
-
-    // Create remote storage location for default LocalFs remote storage
-    std::fs::create_dir_all(env.base_data_dir.join(PAGESERVER_REMOTE_STORAGE_DIR))?;
-
-    // Initialize pageserver, create initial tenant and timeline.
-    for ps_conf in &env.pageservers {
-        PageServerNode::from_env(&env, ps_conf)
-            .initialize(&pageserver_config)
-            .unwrap_or_else(|e| {
-                eprintln!("pageserver init failed: {e:?}");
-                exit(1);
-            });
-    }
-
-    Ok(env)
+    LocalEnv::init(init_conf, force)
+        .context("materialize initial neon_local environment on disk")?;
+    Ok(LocalEnv::load_config().expect("freshly written config should be loadable"))
 }

 /// The default pageserver is the one where CLI tenant/timeline operations are sent by default.
@@ -835,6 +810,8 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                .copied()
                .unwrap_or(false);

+            let allow_multiple = sub_args.get_flag("allow-multiple");
+
            let mode = match (lsn, hot_standby) {
                (Some(lsn), false) => ComputeMode::Static(lsn),
                (None, true) => ComputeMode::Replica,
@@ -852,7 +829,9 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                _ => {}
            }

-            cplane.check_conflicting_endpoints(mode, tenant_id, timeline_id)?;
+            if !allow_multiple {
+                cplane.check_conflicting_endpoints(mode, tenant_id, timeline_id)?;
+            }

            cplane.new_endpoint(
                &endpoint_id,
@@ -881,6 +860,8 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re

            let remote_ext_config = sub_args.get_one::<String>("remote-ext-config");

+            let allow_multiple = sub_args.get_flag("allow-multiple");
+
            // If --safekeepers argument is given, use only the listed safekeeper nodes.
            let safekeepers =
                if let Some(safekeepers_str) = sub_args.get_one::<String>("safekeepers") {
@@ -906,11 +887,13 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                .cloned()
                .unwrap_or_default();

-            cplane.check_conflicting_endpoints(
-                endpoint.mode,
-                endpoint.tenant_id,
-                endpoint.timeline_id,
-            )?;
+            if !allow_multiple {
+                cplane.check_conflicting_endpoints(
+                    endpoint.mode,
+                    endpoint.tenant_id,
+                    endpoint.timeline_id,
+                )?;
+            }

            let (pageservers, stripe_size) = if let Some(pageserver_id) = pageserver_id {
                let conf = env.get_pageserver_conf(pageserver_id).unwrap();
@@ -1410,9 +1393,7 @@ fn cli() -> Command {
    let num_pageservers_arg = Arg::new("num-pageservers")
        .value_parser(value_parser!(u16))
        .long("num-pageservers")
-        .help("How many pageservers to create (default 1)")
-        .required(false)
-        .default_value("1");
+        .help("How many pageservers to create (default 1)");

    let update_catalog = Arg::new("update-catalog")
        .value_parser(value_parser!(bool))
@@ -1426,6 +1407,12 @@ fn cli() -> Command {
        .help("If set, will create test user `user` and `neondb` database. Requires `update-catalog = true`")
        .required(false);

+    let allow_multiple = Arg::new("allow-multiple")
+        .help("Allow multiple primary endpoints running on the same branch. Shouldn't be used normally, but useful for tests.")
+        .long("allow-multiple")
+        .action(ArgAction::SetTrue)
+        .required(false);
+
    Command::new("Neon CLI")
        .arg_required_else_help(true)
        .version(GIT_VERSION)
@@ -1440,14 +1427,6 @@ fn cli() -> Command {
                        .value_parser(value_parser!(PathBuf))
                        .value_name("config")
                )
-                .arg(
-                    Arg::new("pageserver-config")
-                        .long("pageserver-config")
-                        .required(false)
-                        .value_parser(value_parser!(PathBuf))
-                        .value_name("pageserver-config")
-                        .help("Merge the provided pageserver config into the one generated by neon_local."),
-                )
                .arg(pg_version_arg.clone())
                .arg(force_arg)
        )
@@ -1588,6 +1567,7 @@ fn cli() -> Command {
                    .arg(pg_version_arg.clone())
                    .arg(hot_standby_arg.clone())
                    .arg(update_catalog)
+                    .arg(allow_multiple.clone())
                )
                .subcommand(Command::new("start")
                    .about("Start postgres.\n If the endpoint doesn't exist yet, it is created.")
@@ -1596,6 +1576,7 @@ fn cli() -> Command {
                    .arg(safekeepers_arg)
                    .arg(remote_ext_config_args)
                    .arg(create_test_user)
+                    .arg(allow_multiple.clone())
                )
                .subcommand(Command::new("reconfigure")
                            .about("Reconfigure the endpoint")
--- a/control_plane/src/local_env.rs
+++ b/control_plane/src/local_env.rs
@@ -3,7 +3,7 @@
 //! Now it also provides init method which acts like a stub for proper installation
 //! script which will use local paths.

-use anyhow::{bail, ensure, Context};
+use anyhow::{bail, Context};

 use clap::ValueEnum;
 use postgres_backend::AuthType;
@@ -23,6 +23,8 @@ use utils::{
    id::{NodeId, TenantId, TenantTimelineId, TimelineId},
 };

+use crate::pageserver::PageServerNode;
+use crate::pageserver::PAGESERVER_REMOTE_STORAGE_DIR;
 use crate::safekeeper::SafekeeperNode;

 pub const DEFAULT_PG_VERSION: u32 = 15;
@@ -34,7 +36,7 @@ pub const DEFAULT_PG_VERSION: u32 = 15;
 // to 'neon_local init --config=<path>' option. See control_plane/simple.conf for
 // an example.
 //
-#[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
+#[derive(PartialEq, Eq, Clone, Debug)]
 pub struct LocalEnv {
    // Base directory for all the nodes (the pageserver, safekeepers and
    // compute endpoints).
@@ -42,59 +44,99 @@ pub struct LocalEnv {
    // This is not stored in the config file. Rather, this is the path where the
    // config file itself is. It is read from the NEON_REPO_DIR env variable or
    // '.neon' if not given.
-    #[serde(skip)]
    pub base_data_dir: PathBuf,

    // Path to postgres distribution. It's expected that "bin", "include",
    // "lib", "share" from postgres distribution are there. If at some point
    // in time we will be able to run against vanilla postgres we may split that
    // to four separate paths and match OS-specific installation layout.
-    #[serde(default)]
    pub pg_distrib_dir: PathBuf,

    // Path to pageserver binary.
-    #[serde(default)]
    pub neon_distrib_dir: PathBuf,

    // Default tenant ID to use with the 'neon_local' command line utility, when
    // --tenant_id is not explicitly specified.
-    #[serde(default)]
    pub default_tenant_id: Option<TenantId>,

    // used to issue tokens during e.g pg start
-    #[serde(default)]
    pub private_key_path: PathBuf,

    pub broker: NeonBroker,

    // Configuration for the storage controller (1 per neon_local environment)
-    #[serde(default)]
    pub storage_controller: NeonStorageControllerConf,

    /// This Vec must always contain at least one pageserver
+    /// Populdated by [`Self::load_config`] from the individual `pageserver.toml`s.
+    /// NB: not used anymore except for informing users that they need to change their `.neon/config`.
    pub pageservers: Vec<PageServerConf>,

-    #[serde(default)]
    pub safekeepers: Vec<SafekeeperConf>,

    // Control plane upcall API for pageserver: if None, we will not run storage_controller  If set, this will
    // be propagated into each pageserver's configuration.
-    #[serde(default)]
    pub control_plane_api: Option<Url>,

    // Control plane upcall API for storage controller.  If set, this will be propagated into the
    // storage controller's configuration.
-    #[serde(default)]
    pub control_plane_compute_hook_api: Option<Url>,

    /// Keep human-readable aliases in memory (and persist them to config), to hide ZId hex strings from the user.
-    #[serde(default)]
    // A `HashMap<String, HashMap<TenantId, TimelineId>>` would be more appropriate here,
    // but deserialization into a generic toml object as `toml::Value::try_from` fails with an error.
    // https://toml.io/en/v1.0.0 does not contain a concept of "a table inside another table".
+    pub branch_name_mappings: HashMap<String, Vec<(TenantId, TimelineId)>>,
+}
+
+/// On-disk state stored in `.neon/config`.
+#[derive(PartialEq, Eq, Clone, Debug, Default, Serialize, Deserialize)]
+#[serde(default, deny_unknown_fields)]
+pub struct OnDiskConfig {
+    pub pg_distrib_dir: PathBuf,
+    pub neon_distrib_dir: PathBuf,
+    pub default_tenant_id: Option<TenantId>,
+    pub private_key_path: PathBuf,
+    pub broker: NeonBroker,
+    pub storage_controller: NeonStorageControllerConf,
+    #[serde(
+        skip_serializing,
+        deserialize_with = "fail_if_pageservers_field_specified"
+    )]
+    pub pageservers: Vec<PageServerConf>,
+    pub safekeepers: Vec<SafekeeperConf>,
+    pub control_plane_api: Option<Url>,
+    pub control_plane_compute_hook_api: Option<Url>,
    branch_name_mappings: HashMap<String, Vec<(TenantId, TimelineId)>>,
 }

+fn fail_if_pageservers_field_specified<'de, D>(_: D) -> Result<Vec<PageServerConf>, D::Error>
+where
+    D: serde::Deserializer<'de>,
+{
+    Err(serde::de::Error::custom(
+        "The 'pageservers' field is no longer used; pageserver.toml is now authoritative; \
+         Please remove the `pageservers` from your .neon/config.",
+    ))
+}
+
+/// The description of the neon_local env to be initialized by `neon_local init --config`.
+#[derive(Clone, Debug, Deserialize)]
+#[serde(deny_unknown_fields)]
+pub struct NeonLocalInitConf {
+    // TODO: do we need this? Seems unused
+    pub pg_distrib_dir: Option<PathBuf>,
+    // TODO: do we need this? Seems unused
+    pub neon_distrib_dir: Option<PathBuf>,
+    pub default_tenant_id: TenantId,
+    pub broker: NeonBroker,
+    pub storage_controller: Option<NeonStorageControllerConf>,
+    pub pageservers: Vec<NeonLocalInitPageserverConf>,
+    pub safekeepers: Vec<SafekeeperConf>,
+    pub control_plane_api: Option<Option<Url>>,
+    pub control_plane_compute_hook_api: Option<Option<Url>>,
+}
+
 /// Broker config for cluster internal communication.
 #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
 #[serde(default)]
@@ -141,24 +183,18 @@ impl NeonBroker {
    }
 }

+// neon_local needs to know this subset of pageserver configuration.
+// For legacy reasons, this information is duplicated from `pageserver.toml` into `.neon/config`.
+// It can get stale if `pageserver.toml` is changed.
+// TODO(christian): don't store this at all in `.neon/config`, always load it from `pageserver.toml`
 #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
 #[serde(default, deny_unknown_fields)]
 pub struct PageServerConf {
-    // node id
    pub id: NodeId,
-
-    // Pageserver connection settings
    pub listen_pg_addr: String,
    pub listen_http_addr: String,
-
-    // auth type used for the PG and HTTP ports
    pub pg_auth_type: AuthType,
    pub http_auth_type: AuthType,
-
-    pub(crate) virtual_file_io_engine: Option<String>,
-    pub(crate) get_vectored_impl: Option<String>,
-    pub(crate) get_impl: Option<String>,
-    pub(crate) validate_vectored_get: Option<bool>,
 }

 impl Default for PageServerConf {
@@ -169,10 +205,40 @@ impl Default for PageServerConf {
            listen_http_addr: String::new(),
            pg_auth_type: AuthType::Trust,
            http_auth_type: AuthType::Trust,
-            virtual_file_io_engine: None,
-            get_vectored_impl: None,
-            get_impl: None,
-            validate_vectored_get: None,
+        }
+    }
+}
+
+/// The toml that can be passed to `neon_local init --config`.
+/// This is a subset of the `pageserver.toml` configuration.
+// TODO(christian): use pageserver_api::config::ConfigToml (PR #7656)
+#[derive(Clone, Debug, serde::Deserialize, serde::Serialize)]
+pub struct NeonLocalInitPageserverConf {
+    pub id: NodeId,
+    pub listen_pg_addr: String,
+    pub listen_http_addr: String,
+    pub pg_auth_type: AuthType,
+    pub http_auth_type: AuthType,
+    #[serde(flatten)]
+    pub other: HashMap<String, toml::Value>,
+}
+
+impl From<&NeonLocalInitPageserverConf> for PageServerConf {
+    fn from(conf: &NeonLocalInitPageserverConf) -> Self {
+        let NeonLocalInitPageserverConf {
+            id,
+            listen_pg_addr,
+            listen_http_addr,
+            pg_auth_type,
+            http_auth_type,
+            other: _,
+        } = conf;
+        Self {
+            id: *id,
+            listen_pg_addr: listen_pg_addr.clone(),
+            listen_http_addr: listen_http_addr.clone(),
+            pg_auth_type: *pg_auth_type,
+            http_auth_type: *http_auth_type,
        }
    }
 }
@@ -360,44 +426,7 @@ impl LocalEnv {
            .collect()
    }

-    /// Create a LocalEnv from a config file.
-    ///
-    /// Unlike 'load_config', this function fills in any defaults that are missing
-    /// from the config file.
-    pub fn parse_config(toml: &str) -> anyhow::Result<Self> {
-        let mut env: LocalEnv = toml::from_str(toml)?;
-
-        // Find postgres binaries.
-        // Follow POSTGRES_DISTRIB_DIR if set, otherwise look in "pg_install".
-        // Note that later in the code we assume, that distrib dirs follow the same pattern
-        // for all postgres versions.
-        if env.pg_distrib_dir == Path::new("") {
-            if let Some(postgres_bin) = env::var_os("POSTGRES_DISTRIB_DIR") {
-                env.pg_distrib_dir = postgres_bin.into();
-            } else {
-                let cwd = env::current_dir()?;
-                env.pg_distrib_dir = cwd.join("pg_install")
-            }
-        }
-
-        // Find neon binaries.
-        if env.neon_distrib_dir == Path::new("") {
-            env::current_exe()?
-                .parent()
-                .unwrap()
-                .clone_into(&mut env.neon_distrib_dir);
-        }
-
-        if env.pageservers.is_empty() {
-            anyhow::bail!("Configuration must contain at least one pageserver");
-        }
-
-        env.base_data_dir = base_path();
-
-        Ok(env)
-    }
-
-    /// Locate and load config
+    ///  Construct `Self` from on-disk state.
    pub fn load_config() -> anyhow::Result<Self> {
        let repopath = base_path();

@@ -411,38 +440,129 @@ impl LocalEnv {
        // TODO: check that it looks like a neon repository

        // load and parse file
-        let config = fs::read_to_string(repopath.join("config"))?;
-        let mut env: LocalEnv = toml::from_str(config.as_str())?;
+        let config_file_contents = fs::read_to_string(repopath.join("config"))?;
+        let on_disk_config: OnDiskConfig = toml::from_str(config_file_contents.as_str())?;
+        let mut env = {
+            let OnDiskConfig {
+                pg_distrib_dir,
+                neon_distrib_dir,
+                default_tenant_id,
+                private_key_path,
+                broker,
+                storage_controller,
+                pageservers,
+                safekeepers,
+                control_plane_api,
+                control_plane_compute_hook_api,
+                branch_name_mappings,
+            } = on_disk_config;
+            LocalEnv {
+                base_data_dir: repopath.clone(),
+                pg_distrib_dir,
+                neon_distrib_dir,
+                default_tenant_id,
+                private_key_path,
+                broker,
+                storage_controller,
+                pageservers,
+                safekeepers,
+                control_plane_api,
+                control_plane_compute_hook_api,
+                branch_name_mappings,
+            }
+        };

-        env.base_data_dir = repopath;
+        // The source of truth for pageserver configuration is the pageserver.toml.
+        assert!(
+            env.pageservers.is_empty(),
+            "we ensure this during deserialization"
+        );
+        env.pageservers = {
+            let iter = std::fs::read_dir(&repopath).context("open dir")?;
+            let mut pageservers = Vec::new();
+            for res in iter {
+                let dentry = res?;
+                const PREFIX: &str = "pageserver_";
+                let dentry_name = dentry
+                    .file_name()
+                    .into_string()
+                    .ok()
+                    .with_context(|| format!("non-utf8 dentry: {:?}", dentry.path()))
+                    .unwrap();
+                if !dentry_name.starts_with(PREFIX) {
+                    continue;
+                }
+                if !dentry.file_type().context("determine file type")?.is_dir() {
+                    anyhow::bail!("expected a directory, got {:?}", dentry.path());
+                }
+                let id = dentry_name[PREFIX.len()..]
+                    .parse::<NodeId>()
+                    .with_context(|| format!("parse id from {:?}", dentry.path()))?;
+                // TODO(christian): use pageserver_api::config::ConfigToml (PR #7656)
+                #[derive(serde::Serialize, serde::Deserialize)]
+                // (allow unknown fields, unlike PageServerConf)
+                struct PageserverConfigTomlSubset {
+                    id: NodeId,
+                    listen_pg_addr: String,
+                    listen_http_addr: String,
+                    pg_auth_type: AuthType,
+                    http_auth_type: AuthType,
+                }
+                let config_toml_path = dentry.path().join("pageserver.toml");
+                let config_toml: PageserverConfigTomlSubset = toml_edit::de::from_str(
+                    &std::fs::read_to_string(&config_toml_path)
+                        .with_context(|| format!("read {:?}", config_toml_path))?,
+                )
+                .context("parse pageserver.toml")?;
+                let PageserverConfigTomlSubset {
+                    id: config_toml_id,
+                    listen_pg_addr,
+                    listen_http_addr,
+                    pg_auth_type,
+                    http_auth_type,
+                } = config_toml;
+                let conf = PageServerConf {
+                    id: {
+                        anyhow::ensure!(
+                            config_toml_id == id,
+                            "id mismatch: config_toml.id={config_toml_id} id={id}",
+                        );
+                        id
+                    },
+                    listen_pg_addr,
+                    listen_http_addr,
+                    pg_auth_type,
+                    http_auth_type,
+                };
+                pageservers.push(conf);
+            }
+            pageservers
+        };

        Ok(env)
    }

-    pub fn persist_config(&self, base_path: &Path) -> anyhow::Result<()> {
-        // Currently, the user first passes a config file with 'neon_local init --config=<path>'
-        // We read that in, in `create_config`, and fill any missing defaults. Then it's saved
-        // to .neon/config. TODO: We lose any formatting and comments along the way, which is
-        // a bit sad.
-        let mut conf_content = r#"# This file describes a local deployment of the page server
-# and safekeeeper node. It is read by the 'neon_local' command-line
-# utility.
-"#
-        .to_string();
-
-        // Convert the LocalEnv to a toml file.
-        //
-        // This could be as simple as this:
-        //
-        // conf_content += &toml::to_string_pretty(env)?;
-        //
-        // But it results in a "values must be emitted before tables". I'm not sure
-        // why, AFAICS the table, i.e. 'safekeepers: Vec<SafekeeperConf>' is last.
-        // Maybe rust reorders the fields to squeeze avoid padding or something?
-        // In any case, converting to toml::Value first, and serializing that, works.
-        // See https://github.com/alexcrichton/toml-rs/issues/142
-        conf_content += &toml::to_string_pretty(&toml::Value::try_from(self)?)?;
+    pub fn persist_config(&self) -> anyhow::Result<()> {
+        Self::persist_config_impl(
+            &self.base_data_dir,
+            &OnDiskConfig {
+                pg_distrib_dir: self.pg_distrib_dir.clone(),
+                neon_distrib_dir: self.neon_distrib_dir.clone(),
+                default_tenant_id: self.default_tenant_id,
+                private_key_path: self.private_key_path.clone(),
+                broker: self.broker.clone(),
+                storage_controller: self.storage_controller.clone(),
+                pageservers: vec![], // it's skip_serializing anyway
+                safekeepers: self.safekeepers.clone(),
+                control_plane_api: self.control_plane_api.clone(),
+                control_plane_compute_hook_api: self.control_plane_compute_hook_api.clone(),
+                branch_name_mappings: self.branch_name_mappings.clone(),
+            },
+        )
+    }

+    pub fn persist_config_impl(base_path: &Path, config: &OnDiskConfig) -> anyhow::Result<()> {
+        let conf_content = &toml::to_string_pretty(config)?;
        let target_config_path = base_path.join("config");
        fs::write(&target_config_path, conf_content).with_context(|| {
            format!(
@@ -467,17 +587,13 @@ impl LocalEnv {
        }
    }

-    //
-    // Initialize a new Neon repository
-    //
-    pub fn init(&mut self, pg_version: u32, force: &InitForceMode) -> anyhow::Result<()> {
-        // check if config already exists
-        let base_path = &self.base_data_dir;
-        ensure!(
-            base_path != Path::new(""),
-            "repository base path is missing"
-        );
+    /// Materialize the [`NeonLocalInitConf`] to disk. Called during [`neon_local init`].
+    pub fn init(conf: NeonLocalInitConf, force: &InitForceMode) -> anyhow::Result<()> {
+        let base_path = base_path();
+        assert_ne!(base_path, Path::new(""));
+        let base_path = &base_path;

+        // create base_path dir
        if base_path.exists() {
            match force {
                InitForceMode::MustNotExist => {
@@ -509,70 +625,96 @@ impl LocalEnv {
                }
            }
        }
-
-        if !self.pg_bin_dir(pg_version)?.join("postgres").exists() {
-            bail!(
-                "Can't find postgres binary at {}",
-                self.pg_bin_dir(pg_version)?.display()
-            );
-        }
-        for binary in ["pageserver", "safekeeper"] {
-            if !self.neon_distrib_dir.join(binary).exists() {
-                bail!(
-                    "Can't find binary '{binary}' in neon distrib dir '{}'",
-                    self.neon_distrib_dir.display()
-                );
-            }
-        }
-
        if !base_path.exists() {
            fs::create_dir(base_path)?;
        }

+        let NeonLocalInitConf {
+            pg_distrib_dir,
+            neon_distrib_dir,
+            default_tenant_id,
+            broker,
+            storage_controller,
+            pageservers,
+            safekeepers,
+            control_plane_api,
+            control_plane_compute_hook_api,
+        } = conf;
+
+        // Find postgres binaries.
+        // Follow POSTGRES_DISTRIB_DIR if set, otherwise look in "pg_install".
+        // Note that later in the code we assume, that distrib dirs follow the same pattern
+        // for all postgres versions.
+        let pg_distrib_dir = pg_distrib_dir.unwrap_or_else(|| {
+            if let Some(postgres_bin) = env::var_os("POSTGRES_DISTRIB_DIR") {
+                postgres_bin.into()
+            } else {
+                let cwd = env::current_dir().unwrap();
+                cwd.join("pg_install")
+            }
+        });
+
+        // Find neon binaries.
+        let neon_distrib_dir = neon_distrib_dir
+            .unwrap_or_else(|| env::current_exe().unwrap().parent().unwrap().to_owned());
+
        // Generate keypair for JWT.
        //
        // The keypair is only needed if authentication is enabled in any of the
        // components. For convenience, we generate the keypair even if authentication
        // is not enabled, so that you can easily enable it after the initialization
-        // step. However, if the key generation fails, we treat it as non-fatal if
-        // authentication was not enabled.
-        if self.private_key_path == PathBuf::new() {
-            match generate_auth_keys(
-                base_path.join("auth_private_key.pem").as_path(),
-                base_path.join("auth_public_key.pem").as_path(),
-            ) {
-                Ok(()) => {
-                    self.private_key_path = PathBuf::from("auth_private_key.pem");
-                }
-                Err(e) => {
-                    if !self.auth_keys_needed() {
-                        eprintln!("Could not generate keypair for JWT authentication: {e}");
-                        eprintln!("Continuing anyway because authentication was not enabled");
-                        self.private_key_path = PathBuf::from("auth_private_key.pem");
-                    } else {
-                        return Err(e);
-                    }
-                }
-            }
+        // step.
+        generate_auth_keys(
+            base_path.join("auth_private_key.pem").as_path(),
+            base_path.join("auth_public_key.pem").as_path(),
+        )
+        .context("generate auth keys")?;
+        let private_key_path = PathBuf::from("auth_private_key.pem");
+
+        // create the runtime type because the remaining initialization code below needs
+        // a LocalEnv instance op operation
+        // TODO: refactor to avoid this, LocalEnv should only be constructed from on-disk state
+        let env = LocalEnv {
+            base_data_dir: base_path.clone(),
+            pg_distrib_dir,
+            neon_distrib_dir,
+            default_tenant_id: Some(default_tenant_id),
+            private_key_path,
+            broker,
+            storage_controller: storage_controller.unwrap_or_default(),
+            pageservers: pageservers.iter().map(Into::into).collect(),
+            safekeepers,
+            control_plane_api: control_plane_api.unwrap_or_default(),
+            control_plane_compute_hook_api: control_plane_compute_hook_api.unwrap_or_default(),
+            branch_name_mappings: Default::default(),
+        };
+
+        // create endpoints dir
+        fs::create_dir_all(env.endpoints_path())?;
+
+        // create safekeeper dirs
+        for safekeeper in &env.safekeepers {
+            fs::create_dir_all(SafekeeperNode::datadir_path_by_id(&env, safekeeper.id))?;
        }

-        fs::create_dir_all(self.endpoints_path())?;
-
-        for safekeeper in &self.safekeepers {
-            fs::create_dir_all(SafekeeperNode::datadir_path_by_id(self, safekeeper.id))?;
+        // initialize pageserver state
+        for (i, ps) in pageservers.into_iter().enumerate() {
+            let runtime_ps = &env.pageservers[i];
+            assert_eq!(&PageServerConf::from(&ps), runtime_ps);
+            fs::create_dir(env.pageserver_data_dir(ps.id))?;
+            PageServerNode::from_env(&env, runtime_ps)
+                .initialize(ps)
+                .context("pageserver init failed")?;
        }

-        self.persist_config(base_path)
-    }
+        // setup remote remote location for default LocalFs remote storage
+        std::fs::create_dir_all(env.base_data_dir.join(PAGESERVER_REMOTE_STORAGE_DIR))?;

-    fn auth_keys_needed(&self) -> bool {
-        self.pageservers.iter().any(|ps| {
-            ps.pg_auth_type == AuthType::NeonJWT || ps.http_auth_type == AuthType::NeonJWT
-        }) || self.safekeepers.iter().any(|sk| sk.auth_enabled)
+        env.persist_config()
    }
 }

-fn base_path() -> PathBuf {
+pub fn base_path() -> PathBuf {
    match std::env::var_os("NEON_REPO_DIR") {
        Some(val) => PathBuf::from(val),
        None => PathBuf::from(".neon"),
@@ -615,31 +757,3 @@ fn generate_auth_keys(private_key_path: &Path, public_key_path: &Path) -> anyhow
    }
    Ok(())
 }
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn simple_conf_parsing() {
-        let simple_conf_toml = include_str!("../simple.conf");
-        let simple_conf_parse_result = LocalEnv::parse_config(simple_conf_toml);
-        assert!(
-            simple_conf_parse_result.is_ok(),
-            "failed to parse simple config {simple_conf_toml}, reason: {simple_conf_parse_result:?}"
-        );
-
-        let string_to_replace = "listen_addr = '127.0.0.1:50051'";
-        let spoiled_url_str = "listen_addr = '!@$XOXO%^&'";
-        let spoiled_url_toml = simple_conf_toml.replace(string_to_replace, spoiled_url_str);
-        assert!(
-            spoiled_url_toml.contains(spoiled_url_str),
-            "Failed to replace string {string_to_replace} in the toml file {simple_conf_toml}"
-        );
-        let spoiled_url_parse_result = LocalEnv::parse_config(&spoiled_url_toml);
-        assert!(
-            spoiled_url_parse_result.is_err(),
-            "expected toml with invalid Url {spoiled_url_toml} to fail the parsing, but got {spoiled_url_parse_result:?}"
-        );
-    }
-}
--- a/control_plane/src/pageserver.rs
+++ b/control_plane/src/pageserver.rs
@@ -10,14 +10,15 @@ use std::io;
 use std::io::Write;
 use std::num::NonZeroU64;
 use std::path::PathBuf;
-use std::process::Command;
+use std::str::FromStr;
 use std::time::Duration;

 use anyhow::{bail, Context};
 use camino::Utf8PathBuf;
 use futures::SinkExt;
 use pageserver_api::models::{
-    self, LocationConfig, ShardParameters, TenantHistorySize, TenantInfo, TimelineInfo,
+    self, AuxFilePolicy, LocationConfig, ShardParameters, TenantHistorySize, TenantInfo,
+    TimelineInfo,
 };
 use pageserver_api::shard::TenantShardId;
 use pageserver_client::mgmt_api;
@@ -29,7 +30,7 @@ use utils::{
    lsn::Lsn,
 };

-use crate::local_env::PageServerConf;
+use crate::local_env::{NeonLocalInitPageserverConf, PageServerConf};
 use crate::{background_process, local_env::LocalEnv};

 /// Directory within .neon which will be used by default for LocalFs remote storage.
@@ -73,71 +74,23 @@ impl PageServerNode {
        }
    }

-    /// Merge overrides provided by the user on the command line with our default overides derived from neon_local configuration.
-    ///
-    /// These all end up on the command line of the `pageserver` binary.
-    fn neon_local_overrides(&self, cli_overrides: &toml_edit::Document) -> Vec<String> {
+    fn pageserver_init_make_toml(
+        &self,
+        conf: NeonLocalInitPageserverConf,
+    ) -> anyhow::Result<toml_edit::Document> {
+        assert_eq!(&PageServerConf::from(&conf), &self.conf, "during neon_local init, we derive the runtime state of ps conf (self.conf) from the --config flag fully");
+
+        // TODO(christian): instead of what we do here, create a pageserver_api::config::ConfigToml (PR #7656)
+
        // FIXME: the paths should be shell-escaped to handle paths with spaces, quotas etc.
        let pg_distrib_dir_param = format!(
            "pg_distrib_dir='{}'",
            self.env.pg_distrib_dir_raw().display()
        );

-        let PageServerConf {
-            id,
-            listen_pg_addr,
-            listen_http_addr,
-            pg_auth_type,
-            http_auth_type,
-            virtual_file_io_engine,
-            get_vectored_impl,
-            get_impl,
-            validate_vectored_get,
-        } = &self.conf;
-
-        let id = format!("id={}", id);
-
-        let http_auth_type_param = format!("http_auth_type='{}'", http_auth_type);
-        let listen_http_addr_param = format!("listen_http_addr='{}'", listen_http_addr);
-
-        let pg_auth_type_param = format!("pg_auth_type='{}'", pg_auth_type);
-        let listen_pg_addr_param = format!("listen_pg_addr='{}'", listen_pg_addr);
-        let virtual_file_io_engine = if let Some(virtual_file_io_engine) = virtual_file_io_engine {
-            format!("virtual_file_io_engine='{virtual_file_io_engine}'")
-        } else {
-            String::new()
-        };
-        let get_vectored_impl = if let Some(get_vectored_impl) = get_vectored_impl {
-            format!("get_vectored_impl='{get_vectored_impl}'")
-        } else {
-            String::new()
-        };
-        let get_impl = if let Some(get_impl) = get_impl {
-            format!("get_impl='{get_impl}'")
-        } else {
-            String::new()
-        };
-        let validate_vectored_get = if let Some(validate_vectored_get) = validate_vectored_get {
-            format!("validate_vectored_get={validate_vectored_get}")
-        } else {
-            String::new()
-        };
-
        let broker_endpoint_param = format!("broker_endpoint='{}'", self.env.broker.client_url());

-        let mut overrides = vec![
-            id,
-            pg_distrib_dir_param,
-            http_auth_type_param,
-            pg_auth_type_param,
-            listen_http_addr_param,
-            listen_pg_addr_param,
-            broker_endpoint_param,
-            virtual_file_io_engine,
-            get_vectored_impl,
-            get_impl,
-            validate_vectored_get,
-        ];
+        let mut overrides = vec![pg_distrib_dir_param, broker_endpoint_param];

        if let Some(control_plane_api) = &self.env.control_plane_api {
            overrides.push(format!(
@@ -147,7 +100,7 @@ impl PageServerNode {

            // Storage controller uses the same auth as pageserver: if JWT is enabled
            // for us, we will also need it to talk to them.
-            if matches!(http_auth_type, AuthType::NeonJWT) {
+            if matches!(conf.http_auth_type, AuthType::NeonJWT) {
                let jwt_token = self
                    .env
                    .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
@@ -156,28 +109,40 @@ impl PageServerNode {
            }
        }

-        if !cli_overrides.contains_key("remote_storage") {
+        if !conf.other.contains_key("remote_storage") {
            overrides.push(format!(
                "remote_storage={{local_path='../{PAGESERVER_REMOTE_STORAGE_DIR}'}}"
            ));
        }

-        if *http_auth_type != AuthType::Trust || *pg_auth_type != AuthType::Trust {
+        if conf.http_auth_type != AuthType::Trust || conf.pg_auth_type != AuthType::Trust {
            // Keys are generated in the toplevel repo dir, pageservers' workdirs
            // are one level below that, so refer to keys with ../
            overrides.push("auth_validation_public_key_path='../auth_public_key.pem'".to_owned());
        }

        // Apply the user-provided overrides
-        overrides.push(cli_overrides.to_string());
+        overrides.push(
+            toml_edit::ser::to_string_pretty(&conf)
+                .expect("we deserialized this from toml earlier"),
+        );

-        overrides
+        // Turn `overrides` into a toml document.
+        // TODO: above code is legacy code, it should be refactored to use toml_edit directly.
+        let mut config_toml = toml_edit::Document::new();
+        for fragment_str in overrides {
+            let fragment = toml_edit::Document::from_str(&fragment_str)
+                .expect("all fragments in `overrides` are valid toml documents, this function controls that");
+            for (key, item) in fragment.iter() {
+                config_toml.insert(key, item.clone());
+            }
+        }
+        Ok(config_toml)
    }

    /// Initializes a pageserver node by creating its config with the overrides provided.
-    pub fn initialize(&self, config_overrides: &toml_edit::Document) -> anyhow::Result<()> {
-        // First, run `pageserver --init` and wait for it to write a config into FS and exit.
-        self.pageserver_init(config_overrides)
+    pub fn initialize(&self, conf: NeonLocalInitPageserverConf) -> anyhow::Result<()> {
+        self.pageserver_init(conf)
            .with_context(|| format!("Failed to run init for pageserver node {}", self.conf.id))
    }

@@ -197,7 +162,7 @@ impl PageServerNode {
        self.start_node().await
    }

-    fn pageserver_init(&self, config_overrides: &toml_edit::Document) -> anyhow::Result<()> {
+    fn pageserver_init(&self, conf: NeonLocalInitPageserverConf) -> anyhow::Result<()> {
        let datadir = self.repo_path();
        let node_id = self.conf.id;
        println!(
@@ -208,36 +173,20 @@ impl PageServerNode {
        );
        io::stdout().flush()?;

-        if !datadir.exists() {
-            std::fs::create_dir(&datadir)?;
-        }
-
-        let datadir_path_str = datadir.to_str().with_context(|| {
-            format!("Cannot start pageserver node {node_id} in path that has no string representation: {datadir:?}")
-        })?;
-
-        // `pageserver --init` merges the `--config-override`s into a built-in default config,
-        // then writes out the merged product to `pageserver.toml`.
-        // TODO: just write the full `pageserver.toml` and get rid of `--config-override`.
-        let mut args = vec!["--init", "--workdir", datadir_path_str];
-        let overrides = self.neon_local_overrides(config_overrides);
-        for piece in &overrides {
-            args.push("--config-override");
-            args.push(piece);
-        }
-        let init_output = Command::new(self.env.pageserver_bin())
-            .args(args)
-            .envs(self.pageserver_env_variables()?)
-            .output()
-            .with_context(|| format!("Failed to run pageserver init for node {node_id}"))?;
-
-        anyhow::ensure!(
-            init_output.status.success(),
-            "Pageserver init for node {} did not finish successfully, stdout: {}, stderr: {}",
-            node_id,
-            String::from_utf8_lossy(&init_output.stdout),
-            String::from_utf8_lossy(&init_output.stderr),
-        );
+        let config = self
+            .pageserver_init_make_toml(conf)
+            .context("make pageserver toml")?;
+        let config_file_path = datadir.join("pageserver.toml");
+        let mut config_file = std::fs::OpenOptions::new()
+            .create_new(true)
+            .write(true)
+            .open(&config_file_path)
+            .with_context(|| format!("open pageserver toml for write: {config_file_path:?}"))?;
+        config_file
+            .write_all(config.to_string().as_bytes())
+            .context("write pageserver toml")?;
+        drop(config_file);
+        // TODO: invoke a TBD config-check command to validate that pageserver will start with the written config

        // Write metadata file, used by pageserver on startup to register itself with
        // the storage controller
@@ -429,11 +378,11 @@ impl PageServerNode {
                .map(serde_json::from_str)
                .transpose()
                .context("parse `timeline_get_throttle` from json")?,
-            switch_to_aux_file_v2: settings
-                .remove("switch_to_aux_file_v2")
-                .map(|x| x.parse::<bool>())
+            switch_aux_file_policy: settings
+                .remove("switch_aux_file_policy")
+                .map(|x| x.parse::<AuxFilePolicy>())
                .transpose()
-                .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
+                .context("Failed to parse 'switch_aux_file_policy'")?,
        };
        if !settings.is_empty() {
            bail!("Unrecognized tenant settings: {settings:?}")
@@ -552,11 +501,11 @@ impl PageServerNode {
                    .map(serde_json::from_str)
                    .transpose()
                    .context("parse `timeline_get_throttle` from json")?,
-                switch_to_aux_file_v2: settings
-                    .remove("switch_to_aux_file_v2")
-                    .map(|x| x.parse::<bool>())
+                switch_aux_file_policy: settings
+                    .remove("switch_aux_file_policy")
+                    .map(|x| x.parse::<AuxFilePolicy>())
                    .transpose()
-                    .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
+                    .context("Failed to parse 'switch_aux_file_policy'")?,
            }
        };

--- a/libs/metrics/src/lib.rs
+++ b/libs/metrics/src/lib.rs
@@ -480,6 +480,15 @@ impl<A: CounterPairAssoc> CounterPairVec<A> {
        let id = self.vec.with_labels(labels);
        self.vec.remove_metric(id)
    }
+
+    pub fn sample(&self, labels: <A::LabelGroupSet as LabelGroupSet>::Group<'_>) -> u64 {
+        let id = self.vec.with_labels(labels);
+        let metric = self.vec.get_metric(id);
+
+        let inc = metric.inc.count.load(std::sync::atomic::Ordering::Relaxed);
+        let dec = metric.dec.count.load(std::sync::atomic::Ordering::Relaxed);
+        inc.saturating_sub(dec)
+    }
 }

 impl<T, A> ::measured::metric::group::MetricGroup<T> for CounterPairVec<A>
--- a/libs/pageserver_api/src/keyspace.rs
+++ b/libs/pageserver_api/src/keyspace.rs
@@ -240,7 +240,7 @@ impl<'a> ShardedRange<'a> {
    /// pages that would not actually be stored on this node.
    ///
    /// Don't use this function in code that works with physical entities like layer files.
-    fn raw_size(range: &Range<Key>) -> u32 {
+    pub fn raw_size(range: &Range<Key>) -> u32 {
        if is_contiguous_range(range) {
            contiguous_range_len(range)
        } else {
--- a/libs/pageserver_api/src/models.rs
+++ b/libs/pageserver_api/src/models.rs
@@ -1,3 +1,4 @@
+pub mod detach_ancestor;
 pub mod partitioning;
 pub mod utilization;

@@ -8,6 +9,7 @@ use std::{
    collections::HashMap,
    io::{BufRead, Read},
    num::{NonZeroU64, NonZeroUsize},
+    str::FromStr,
    time::{Duration, SystemTime},
 };

@@ -303,7 +305,31 @@ pub struct TenantConfig {
    pub lazy_slru_download: Option<bool>,
    pub timeline_get_throttle: Option<ThrottleConfig>,
    pub image_layer_creation_check_threshold: Option<u8>,
-    pub switch_to_aux_file_v2: Option<bool>,
+    pub switch_aux_file_policy: Option<AuxFilePolicy>,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+pub enum AuxFilePolicy {
+    V1,
+    V2,
+    CrossValidation,
+}
+
+impl FromStr for AuxFilePolicy {
+    type Err = anyhow::Error;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let s = s.to_lowercase();
+        if s == "v1" {
+            Ok(Self::V1)
+        } else if s == "v2" {
+            Ok(Self::V2)
+        } else if s == "crossvalidation" || s == "cross_validation" {
+            Ok(Self::CrossValidation)
+        } else {
+            anyhow::bail!("cannot parse {} to aux file policy", s)
+        }
+    }
 }

 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
--- a/libs/pageserver_api/src/models/detach_ancestor.rs
+++ b/libs/pageserver_api/src/models/detach_ancestor.rs
@@ -0,0 +1,6 @@
+use utils::id::TimelineId;
+
+#[derive(Default, serde::Serialize)]
+pub struct AncestorDetached {
+    pub reparented_timelines: Vec<TimelineId>,
+}
--- a/libs/postgres_ffi/src/xlog_utils.rs
+++ b/libs/postgres_ffi/src/xlog_utils.rs
@@ -331,7 +331,10 @@ impl CheckPoint {
    /// Returns 'true' if the XID was updated.
    pub fn update_next_xid(&mut self, xid: u32) -> bool {
        // nextXid should be greater than any XID in WAL, so increment provided XID and check for wraparround.
-        let mut new_xid = std::cmp::max(xid.wrapping_add(1), pg_constants::FIRST_NORMAL_TRANSACTION_ID);
+        let mut new_xid = std::cmp::max(
+            xid.wrapping_add(1),
+            pg_constants::FIRST_NORMAL_TRANSACTION_ID,
+        );
        // To reduce number of metadata checkpoints, we forward align XID on XID_CHECKPOINT_INTERVAL.
        // XID_CHECKPOINT_INTERVAL should not be larger than BLCKSZ*CLOG_XACTS_PER_BYTE
        new_xid =
@@ -367,8 +370,16 @@ pub fn generate_wal_segment(segno: u64, system_id: u64, lsn: Lsn) -> Result<Byte
    let seg_off = lsn.segment_offset(WAL_SEGMENT_SIZE);

    let first_page_only = seg_off < XLOG_BLCKSZ;
-    let (shdr_rem_len, infoflags) = if first_page_only {
-        (seg_off, pg_constants::XLP_FIRST_IS_CONTRECORD)
+    // If first records starts in the middle of the page, pretend in page header
+    // there is a fake record which ends where first real record starts. This
+    // makes pg_waldump etc happy.
+    let (shdr_rem_len, infoflags) = if first_page_only && seg_off > 0 {
+        assert!(seg_off >= XLOG_SIZE_OF_XLOG_LONG_PHD);
+        // xlp_rem_len doesn't include page header, hence the subtraction.
+        (
+            seg_off - XLOG_SIZE_OF_XLOG_LONG_PHD,
+            pg_constants::XLP_FIRST_IS_CONTRECORD,
+        )
    } else {
        (0, 0)
    };
@@ -397,20 +408,22 @@ pub fn generate_wal_segment(segno: u64, system_id: u64, lsn: Lsn) -> Result<Byte

    if !first_page_only {
        let block_offset = lsn.page_offset_in_segment(WAL_SEGMENT_SIZE) as usize;
+        // see comments above about XLP_FIRST_IS_CONTRECORD and xlp_rem_len.
+        let (xlp_rem_len, xlp_info) = if page_off > 0 {
+            assert!(page_off >= XLOG_SIZE_OF_XLOG_SHORT_PHD as u64);
+            (
+                (page_off - XLOG_SIZE_OF_XLOG_SHORT_PHD as u64) as u32,
+                pg_constants::XLP_FIRST_IS_CONTRECORD,
+            )
+        } else {
+            (0, 0)
+        };
        let header = XLogPageHeaderData {
            xlp_magic: XLOG_PAGE_MAGIC as u16,
-            xlp_info: if page_off >= pg_constants::SIZE_OF_PAGE_HEADER as u64 {
-                pg_constants::XLP_FIRST_IS_CONTRECORD
-            } else {
-                0
-            },
+            xlp_info,
            xlp_tli: PG_TLI,
            xlp_pageaddr: lsn.page_lsn().0,
-            xlp_rem_len: if page_off >= pg_constants::SIZE_OF_PAGE_HEADER as u64 {
-                page_off as u32
-            } else {
-                0u32
-            },
+            xlp_rem_len,
            ..Default::default() // Put 0 in padding fields.
        };
        let hdr_bytes = header.encode()?;
--- a/libs/remote_storage/src/s3_bucket.rs
+++ b/libs/remote_storage/src/s3_bucket.rs
@@ -27,7 +27,7 @@ use aws_config::{
 };
 use aws_credential_types::provider::SharedCredentialsProvider;
 use aws_sdk_s3::{
-    config::{AsyncSleep, Builder, IdentityCache, Region, SharedAsyncSleep},
+    config::{AsyncSleep, IdentityCache, Region, SharedAsyncSleep},
    error::SdkError,
    operation::get_object::GetObjectError,
    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion, StorageClass},
@@ -75,13 +75,13 @@ struct GetObjectRequest {
 }
 impl S3Bucket {
    /// Creates the S3 storage, errors if incorrect AWS S3 configuration provided.
-    pub fn new(aws_config: &S3Config, timeout: Duration) -> anyhow::Result<Self> {
+    pub fn new(remote_storage_config: &S3Config, timeout: Duration) -> anyhow::Result<Self> {
        tracing::debug!(
            "Creating s3 remote storage for S3 bucket {}",
-            aws_config.bucket_name
+            remote_storage_config.bucket_name
        );

-        let region = Some(Region::new(aws_config.bucket_region.clone()));
+        let region = Some(Region::new(remote_storage_config.bucket_region.clone()));

        let provider_conf = ProviderConfig::without_region().with_region(region.clone());

@@ -113,6 +113,38 @@ impl S3Bucket {
        // AWS SDK requires us to specify how the RetryConfig should sleep when it wants to back off
        let sleep_impl: Arc<dyn AsyncSleep> = Arc::new(TokioSleep::new());

+        let sdk_config_loader: aws_config::ConfigLoader = aws_config::defaults(
+            #[allow(deprecated)] /* TODO: https://github.com/neondatabase/neon/issues/7665 */
+            BehaviorVersion::v2023_11_09(),
+        )
+        .region(region)
+        .identity_cache(IdentityCache::lazy().build())
+        .credentials_provider(SharedCredentialsProvider::new(credentials_provider))
+        .sleep_impl(SharedAsyncSleep::from(sleep_impl));
+
+        let sdk_config: aws_config::SdkConfig = std::thread::scope(|s| {
+            s.spawn(|| {
+                // TODO: make this function async.
+                tokio::runtime::Builder::new_current_thread()
+                    .enable_all()
+                    .build()
+                    .unwrap()
+                    .block_on(sdk_config_loader.load())
+            })
+            .join()
+            .unwrap()
+        });
+
+        let mut s3_config_builder = aws_sdk_s3::config::Builder::from(&sdk_config);
+
+        // Technically, the `remote_storage_config.endpoint` field only applies to S3 interactions.
+        // (In case we ever re-use the `sdk_config` for more than just the S3 client in the future)
+        if let Some(custom_endpoint) = remote_storage_config.endpoint.clone() {
+            s3_config_builder = s3_config_builder
+                .endpoint_url(custom_endpoint)
+                .force_path_style(true);
+        }
+
        // We do our own retries (see [`backoff::retry`]).  However, for the AWS SDK to enable rate limiting in response to throttling
        // responses (e.g. 429 on too many ListObjectsv2 requests), we must provide a retry config.  We set it to use at most one
        // attempt, and enable 'Adaptive' mode, which causes rate limiting to be enabled.
@@ -120,42 +152,36 @@ impl S3Bucket {
        retry_config
            .set_max_attempts(Some(1))
            .set_mode(Some(RetryMode::Adaptive));
+        s3_config_builder = s3_config_builder.retry_config(retry_config.build());

-        let mut config_builder = Builder::default()
-            .behavior_version(BehaviorVersion::v2023_11_09())
-            .region(region)
-            .identity_cache(IdentityCache::lazy().build())
-            .credentials_provider(SharedCredentialsProvider::new(credentials_provider))
-            .retry_config(retry_config.build())
-            .sleep_impl(SharedAsyncSleep::from(sleep_impl));
+        let s3_config = s3_config_builder.build();
+        let client = aws_sdk_s3::Client::from_conf(s3_config);

-        if let Some(custom_endpoint) = aws_config.endpoint.clone() {
-            config_builder = config_builder
-                .endpoint_url(custom_endpoint)
-                .force_path_style(true);
-        }
+        let prefix_in_bucket = remote_storage_config
+            .prefix_in_bucket
+            .as_deref()
+            .map(|prefix| {
+                let mut prefix = prefix;
+                while prefix.starts_with(REMOTE_STORAGE_PREFIX_SEPARATOR) {
+                    prefix = &prefix[1..]
+                }

-        let client = Client::from_conf(config_builder.build());
+                let mut prefix = prefix.to_string();
+                while prefix.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR) {
+                    prefix.pop();
+                }
+                prefix
+            });

-        let prefix_in_bucket = aws_config.prefix_in_bucket.as_deref().map(|prefix| {
-            let mut prefix = prefix;
-            while prefix.starts_with(REMOTE_STORAGE_PREFIX_SEPARATOR) {
-                prefix = &prefix[1..]
-            }
-
-            let mut prefix = prefix.to_string();
-            while prefix.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR) {
-                prefix.pop();
-            }
-            prefix
-        });
        Ok(Self {
            client,
-            bucket_name: aws_config.bucket_name.clone(),
-            max_keys_per_list_response: aws_config.max_keys_per_list_response,
+            bucket_name: remote_storage_config.bucket_name.clone(),
+            max_keys_per_list_response: remote_storage_config.max_keys_per_list_response,
            prefix_in_bucket,
-            concurrency_limiter: ConcurrencyLimiter::new(aws_config.concurrency_limit.get()),
-            upload_storage_class: aws_config.upload_storage_class.clone(),
+            concurrency_limiter: ConcurrencyLimiter::new(
+                remote_storage_config.concurrency_limit.get(),
+            ),
+            upload_storage_class: remote_storage_config.upload_storage_class.clone(),
            timeout,
        })
    }
--- a/libs/walproposer/src/api_bindings.rs
+++ b/libs/walproposer/src/api_bindings.rs
@@ -50,6 +50,14 @@ extern "C" fn get_flush_rec_ptr(wp: *mut WalProposer) -> XLogRecPtr {
    }
 }

+extern "C" fn update_donor(wp: *mut WalProposer, donor: *mut Safekeeper, donor_lsn: XLogRecPtr) {
+    unsafe {
+        let callback_data = (*(*wp).config).callback_data;
+        let api = callback_data as *mut Box<dyn ApiImpl>;
+        (*api).update_donor(&mut (*donor), donor_lsn)
+    }
+}
+
 extern "C" fn get_current_timestamp(wp: *mut WalProposer) -> TimestampTz {
    unsafe {
        let callback_data = (*(*wp).config).callback_data;
@@ -391,6 +399,7 @@ pub(crate) fn create_api() -> walproposer_api {
        get_shmem_state: Some(get_shmem_state),
        start_streaming: Some(start_streaming),
        get_flush_rec_ptr: Some(get_flush_rec_ptr),
+        update_donor: Some(update_donor),
        get_current_timestamp: Some(get_current_timestamp),
        conn_error_message: Some(conn_error_message),
        conn_status: Some(conn_status),
@@ -421,6 +430,32 @@ pub(crate) fn create_api() -> walproposer_api {
    }
 }

+pub fn empty_shmem() -> crate::bindings::WalproposerShmemState {
+    let empty_feedback = crate::bindings::PageserverFeedback {
+        present: false,
+        currentClusterSize: 0,
+        last_received_lsn: 0,
+        disk_consistent_lsn: 0,
+        remote_consistent_lsn: 0,
+        replytime: 0,
+        shard_number: 0,
+    };
+
+    crate::bindings::WalproposerShmemState {
+        propEpochStartLsn: crate::bindings::pg_atomic_uint64 { value: 0 },
+        donor_name: [0; 64],
+        donor_conninfo: [0; 1024],
+        donor_lsn: 0,
+        mutex: 0,
+        mineLastElectedTerm: crate::bindings::pg_atomic_uint64 { value: 0 },
+        backpressureThrottlingTime: crate::bindings::pg_atomic_uint64 { value: 0 },
+        currentClusterSize: crate::bindings::pg_atomic_uint64 { value: 0 },
+        shard_ps_feedback: [empty_feedback; 128],
+        num_shards: 0,
+        min_ps_feedback: empty_feedback,
+    }
+}
+
 impl std::fmt::Display for Level {
    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
        write!(f, "{:?}", self)
--- a/libs/walproposer/src/walproposer.rs
+++ b/libs/walproposer/src/walproposer.rs
@@ -1,8 +1,5 @@
 use std::ffi::CString;

-use postgres_ffi::WAL_SEGMENT_SIZE;
-use utils::{id::TenantTimelineId, lsn::Lsn};
-
 use crate::{
    api_bindings::{create_api, take_vec_u8, Level},
    bindings::{
@@ -10,6 +7,8 @@ use crate::{
        WalProposerCreate, WalProposerFree, WalProposerPoll, WalProposerStart,
    },
 };
+use postgres_ffi::WAL_SEGMENT_SIZE;
+use utils::{id::TenantTimelineId, lsn::Lsn};

 /// Rust high-level wrapper for C walproposer API. Many methods are not required
 /// for simple cases, hence todo!() in default implementations.
@@ -28,6 +27,10 @@ pub trait ApiImpl {
        todo!()
    }

+    fn update_donor(&self, _donor: &mut Safekeeper, _donor_lsn: u64) {
+        todo!()
+    }
+
    fn get_current_timestamp(&self) -> i64 {
        todo!()
    }
@@ -274,6 +277,7 @@ mod tests {
        sync::{atomic::AtomicUsize, mpsc::sync_channel},
    };

+    use std::cell::UnsafeCell;
    use utils::id::TenantTimelineId;

    use crate::{api_bindings::Level, bindings::NeonWALReadResult, walproposer::Wrapper};
@@ -297,6 +301,8 @@ mod tests {
        replies_ptr: AtomicUsize,
        // channel to send LSN to the main thread
        sync_channel: std::sync::mpsc::SyncSender<u64>,
+        // Shmem state, used for storing donor info
+        shmem: UnsafeCell<crate::bindings::WalproposerShmemState>,
    }

    impl MockImpl {
@@ -327,11 +333,22 @@ mod tests {
    }

    impl ApiImpl for MockImpl {
+        fn get_shmem_state(&self) -> *mut crate::bindings::WalproposerShmemState {
+            self.shmem.get()
+        }
+
        fn get_current_timestamp(&self) -> i64 {
            println!("get_current_timestamp");
            0
        }

+        fn update_donor(&self, donor: &mut crate::bindings::Safekeeper, donor_lsn: u64) {
+            let mut shmem = unsafe { *self.get_shmem_state() };
+            shmem.propEpochStartLsn.value = donor_lsn;
+            shmem.donor_conninfo = donor.conninfo;
+            shmem.donor_lsn = donor_lsn;
+        }
+
        fn conn_status(
            &self,
            _: &mut crate::bindings::Safekeeper,
@@ -507,6 +524,7 @@ mod tests {
            ],
            replies_ptr: AtomicUsize::new(0),
            sync_channel: sender,
+            shmem: UnsafeCell::new(crate::api_bindings::empty_shmem()),
        });
        let config = crate::walproposer::Config {
            ttid,
--- a/pageserver/benches/bench_layer_map.rs
+++ b/pageserver/benches/bench_layer_map.rs
@@ -1,7 +1,7 @@
 use pageserver::keyspace::{KeyPartitioning, KeySpace};
 use pageserver::repository::Key;
 use pageserver::tenant::layer_map::LayerMap;
-use pageserver::tenant::storage_layer::LayerFileName;
+use pageserver::tenant::storage_layer::LayerName;
 use pageserver::tenant::storage_layer::PersistentLayerDesc;
 use pageserver_api::shard::TenantShardId;
 use rand::prelude::{SeedableRng, SliceRandom, StdRng};
@@ -28,7 +28,7 @@ fn build_layer_map(filename_dump: PathBuf) -> LayerMap {
    let mut updates = layer_map.batch_update();
    for fname in filenames {
        let fname = fname.unwrap();
-        let fname = LayerFileName::from_str(&fname).unwrap();
+        let fname = LayerName::from_str(&fname).unwrap();
        let layer = PersistentLayerDesc::from(fname);

        let lsn_range = layer.get_lsn_range();
--- a/pageserver/compaction/src/compact_tiered.rs
+++ b/pageserver/compaction/src/compact_tiered.rs
@@ -24,7 +24,9 @@ use tracing::{debug, info};
 use std::collections::{HashSet, VecDeque};
 use std::ops::Range;

-use crate::helpers::{accum_key_values, keyspace_total_size, merge_delta_keys, overlaps_with};
+use crate::helpers::{
+    accum_key_values, keyspace_total_size, merge_delta_keys_buffered, overlaps_with,
+};
 use crate::interface::*;
 use utils::lsn::Lsn;

@@ -535,7 +537,10 @@ where
            }
        }
        // Open stream
-        let key_value_stream = std::pin::pin!(merge_delta_keys::<E>(deltas.as_slice(), ctx));
+        let key_value_stream =
+            std::pin::pin!(merge_delta_keys_buffered::<E>(deltas.as_slice(), ctx)
+                .await?
+                .map(Result::<_, anyhow::Error>::Ok));
        let mut new_jobs = Vec::new();

        // Slide a window through the keyspace
--- a/pageserver/compaction/src/helpers.rs
+++ b/pageserver/compaction/src/helpers.rs
@@ -9,10 +9,12 @@ use pageserver_api::shard::ShardIdentity;
 use pin_project_lite::pin_project;
 use std::collections::BinaryHeap;
 use std::collections::VecDeque;
+use std::fmt::Display;
 use std::future::Future;
 use std::ops::{DerefMut, Range};
 use std::pin::Pin;
 use std::task::{ready, Poll};
+use utils::lsn::Lsn;

 pub fn keyspace_total_size<K>(
    keyspace: &CompactionKeySpace<K>,
@@ -108,17 +110,40 @@ pub fn merge_delta_keys<'a, E: CompactionJobExecutor>(
    }
 }

+pub async fn merge_delta_keys_buffered<'a, E: CompactionJobExecutor + 'a>(
+    layers: &'a [E::DeltaLayer],
+    ctx: &'a E::RequestContext,
+) -> anyhow::Result<impl Stream<Item = <E::DeltaLayer as CompactionDeltaLayer<E>>::DeltaEntry<'a>>>
+{
+    let mut keys = Vec::new();
+    for l in layers {
+        // Boxing and casting to LoadFuture is required to obtain the right Sync bound.
+        // If we do l.load_keys(ctx).await? directly, there is a compilation error.
+        let load_future: LoadFuture<'a, _> = Box::pin(l.load_keys(ctx));
+        keys.extend(load_future.await?.into_iter());
+    }
+    keys.sort_by_key(|k| (k.key(), k.lsn()));
+    let stream = futures::stream::iter(keys.into_iter());
+    Ok(stream)
+}
+
 enum LazyLoadLayer<'a, E: CompactionJobExecutor> {
    Loaded(VecDeque<<E::DeltaLayer as CompactionDeltaLayer<E>>::DeltaEntry<'a>>),
    Unloaded(&'a E::DeltaLayer),
 }
 impl<'a, E: CompactionJobExecutor> LazyLoadLayer<'a, E> {
-    fn key(&self) -> E::Key {
+    fn min_key(&self) -> E::Key {
        match self {
            Self::Loaded(entries) => entries.front().unwrap().key(),
            Self::Unloaded(dl) => dl.key_range().start,
        }
    }
+    fn min_lsn(&self) -> Lsn {
+        match self {
+            Self::Loaded(entries) => entries.front().unwrap().lsn(),
+            Self::Unloaded(dl) => dl.lsn_range().start,
+        }
+    }
 }
 impl<'a, E: CompactionJobExecutor> PartialOrd for LazyLoadLayer<'a, E> {
    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
@@ -128,12 +153,12 @@ impl<'a, E: CompactionJobExecutor> PartialOrd for LazyLoadLayer<'a, E> {
 impl<'a, E: CompactionJobExecutor> Ord for LazyLoadLayer<'a, E> {
    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
        // reverse order so that we get a min-heap
-        other.key().cmp(&self.key())
+        (other.min_key(), other.min_lsn()).cmp(&(self.min_key(), self.min_lsn()))
    }
 }
 impl<'a, E: CompactionJobExecutor> PartialEq for LazyLoadLayer<'a, E> {
    fn eq(&self, other: &Self) -> bool {
-        self.key().eq(&other.key())
+        self.cmp(other) == std::cmp::Ordering::Equal
    }
 }
 impl<'a, E: CompactionJobExecutor> Eq for LazyLoadLayer<'a, E> {}
@@ -214,7 +239,7 @@ pub struct KeySize<K> {

 pub fn accum_key_values<'a, I, K, D, E>(input: I) -> impl Stream<Item = Result<KeySize<K>, E>>
 where
-    K: Eq,
+    K: Eq + PartialOrd + Display + Copy,
    I: Stream<Item = Result<D, E>>,
    D: CompactionDeltaEntry<'a, K>,
 {
@@ -229,12 +254,15 @@ where
                num_values: 1,
                size: first.size(),
            };
+            let mut last_key = accum.key;
            while let Some(this) = input.next().await {
                let this = this?;
                if this.key() == accum.key {
                    accum.size += this.size();
                    accum.num_values += 1;
                } else {
+                    assert!(last_key <= accum.key, "last_key={last_key} <= accum.key={}", accum.key);
+                    last_key = accum.key;
                    yield accum;
                    accum = KeySize {
                        key: this.key(),
@@ -243,6 +271,7 @@ where
                    };
                }
            }
+            assert!(last_key <= accum.key, "last_key={last_key} <= accum.key={}", accum.key);
            yield accum;
        }
    }
--- a/pageserver/compaction/tests/tests.rs
+++ b/pageserver/compaction/tests/tests.rs
@@ -1,5 +1,20 @@
+use once_cell::sync::OnceCell;
 use pageserver_compaction::interface::CompactionLayer;
 use pageserver_compaction::simulator::MockTimeline;
+use utils::logging;
+
+static LOG_HANDLE: OnceCell<()> = OnceCell::new();
+
+pub(crate) fn setup_logging() {
+    LOG_HANDLE.get_or_init(|| {
+        logging::init(
+            logging::LogFormat::Test,
+            logging::TracingErrorLayerEnablement::EnableWithRustLogFilter,
+            logging::Output::Stdout,
+        )
+        .expect("Failed to init test logging")
+    });
+}

 /// Test the extreme case that there are so many updates for a single key that
 /// even if we produce an extremely narrow delta layer, spanning just that one
@@ -11,13 +26,14 @@ use pageserver_compaction::simulator::MockTimeline;
 #[ignore]
 #[tokio::test]
 async fn test_many_updates_for_single_key() {
+    setup_logging();
    let mut executor = MockTimeline::new();
-    executor.target_file_size = 10_000_000; // 10 MB
+    executor.target_file_size = 1_000_000; // 1 MB

-    // Ingest 100 MB of updates to a single key.
+    // Ingest 10 MB of updates to a single key.
    for _ in 1..1000 {
        executor.ingest_uniform(100, 10, &(0..100_000)).unwrap();
-        executor.ingest_uniform(10_000, 10, &(0..1)).unwrap();
+        executor.ingest_uniform(1000, 10, &(0..1)).unwrap();
        executor.compact().await.unwrap();
    }

@@ -33,3 +49,26 @@ async fn test_many_updates_for_single_key() {
        }
    }
 }
+
+#[tokio::test]
+async fn test_simple_updates() {
+    setup_logging();
+    let mut executor = MockTimeline::new();
+    executor.target_file_size = 500_000; // 500 KB
+
+    // Ingest some traffic.
+    for _ in 1..400 {
+        executor.ingest_uniform(100, 500, &(0..100_000)).unwrap();
+    }
+
+    for l in executor.live_layers.iter() {
+        println!("layer {}: {}", l.short_id(), l.file_size());
+    }
+
+    println!("Running compaction...");
+    executor.compact().await.unwrap();
+
+    for l in executor.live_layers.iter() {
+        println!("layer {}: {}", l.short_id(), l.file_size());
+    }
+}
--- a/pageserver/ctl/src/index_part.rs
+++ b/pageserver/ctl/src/index_part.rs
@@ -3,7 +3,7 @@ use std::collections::HashMap;
 use anyhow::Context;
 use camino::Utf8PathBuf;
 use pageserver::tenant::remote_timeline_client::index::IndexLayerMetadata;
-use pageserver::tenant::storage_layer::LayerFileName;
+use pageserver::tenant::storage_layer::LayerName;
 use pageserver::tenant::{metadata::TimelineMetadata, IndexPart};
 use utils::lsn::Lsn;

@@ -19,7 +19,7 @@ pub(crate) async fn main(cmd: &IndexPartCmd) -> anyhow::Result<()> {
            let des: IndexPart = IndexPart::from_s3_bytes(&bytes).context("deserialize")?;
            #[derive(serde::Serialize)]
            struct Output<'a> {
-                layer_metadata: &'a HashMap<LayerFileName, IndexLayerMetadata>,
+                layer_metadata: &'a HashMap<LayerName, IndexLayerMetadata>,
                disk_consistent_lsn: Lsn,
                timeline_metadata: &'a TimelineMetadata,
            }
--- a/pageserver/src/aux_file.rs
+++ b/pageserver/src/aux_file.rs
@@ -1,3 +1,4 @@
+use bytes::{Buf, BufMut, Bytes};
 use pageserver_api::key::{Key, AUX_KEY_PREFIX, METADATA_KEY_SIZE};
 use tracing::warn;

@@ -61,6 +62,84 @@ pub fn encode_aux_file_key(path: &str) -> Key {
    }
 }

+const AUX_FILE_ENCODING_VERSION: u8 = 0x01;
+
+pub fn decode_file_value(val: &[u8]) -> anyhow::Result<Vec<(&str, &[u8])>> {
+    let mut ptr = val;
+    if ptr.is_empty() {
+        // empty value = no files
+        return Ok(Vec::new());
+    }
+    assert_eq!(
+        ptr.get_u8(),
+        AUX_FILE_ENCODING_VERSION,
+        "unsupported aux file value"
+    );
+    let mut files = vec![];
+    while ptr.has_remaining() {
+        let key_len = ptr.get_u32() as usize;
+        let key = &ptr[..key_len];
+        ptr.advance(key_len);
+        let val_len = ptr.get_u32() as usize;
+        let content = &ptr[..val_len];
+        ptr.advance(val_len);
+
+        let path = std::str::from_utf8(key)?;
+        files.push((path, content));
+    }
+    Ok(files)
+}
+
+/// Decode an aux file key-value pair into a list of files. The returned `Bytes` contains reference
+/// to the original value slice. Be cautious about memory consumption.
+pub fn decode_file_value_bytes(val: &Bytes) -> anyhow::Result<Vec<(String, Bytes)>> {
+    let mut ptr = val.clone();
+    if ptr.is_empty() {
+        // empty value = no files
+        return Ok(Vec::new());
+    }
+    assert_eq!(
+        ptr.get_u8(),
+        AUX_FILE_ENCODING_VERSION,
+        "unsupported aux file value"
+    );
+    let mut files = vec![];
+    while ptr.has_remaining() {
+        let key_len = ptr.get_u32() as usize;
+        let key = ptr.slice(..key_len);
+        ptr.advance(key_len);
+        let val_len = ptr.get_u32() as usize;
+        let content = ptr.slice(..val_len);
+        ptr.advance(val_len);
+
+        let path = std::str::from_utf8(&key)?.to_string();
+        files.push((path, content));
+    }
+    Ok(files)
+}
+
+pub fn encode_file_value(files: &[(&str, &[u8])]) -> anyhow::Result<Vec<u8>> {
+    if files.is_empty() {
+        // no files = empty value
+        return Ok(Vec::new());
+    }
+    let mut encoded = vec![];
+    encoded.put_u8(AUX_FILE_ENCODING_VERSION);
+    for (path, content) in files {
+        if path.len() > u32::MAX as usize {
+            anyhow::bail!("{} exceeds path size limit", path);
+        }
+        encoded.put_u32(path.len() as u32);
+        encoded.put_slice(path.as_bytes());
+        if content.len() > u32::MAX as usize {
+            anyhow::bail!("{} exceeds content size limit", path);
+        }
+        encoded.put_u32(content.len() as u32);
+        encoded.put_slice(content);
+    }
+    Ok(encoded)
+}
+
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -109,4 +188,21 @@ mod tests {
            encode_aux_file_key("other_file_not_supported").to_string()
        );
    }
+
+    #[test]
+    fn test_value_encoding() {
+        let files = vec![
+            ("pg_logical/1.file", "1111".as_bytes()),
+            ("pg_logical/2.file", "2222".as_bytes()),
+        ];
+        assert_eq!(
+            files,
+            decode_file_value(&encode_file_value(&files).unwrap()).unwrap()
+        );
+        let files = vec![];
+        assert_eq!(
+            files,
+            decode_file_value(&encode_file_value(&files).unwrap()).unwrap()
+        );
+    }
 }
--- a/pageserver/src/deletion_queue.rs
+++ b/pageserver/src/deletion_queue.rs
@@ -38,7 +38,7 @@ use deleter::DeleterMessage;
 use list_writer::ListWriterQueueMessage;
 use validator::ValidatorQueueMessage;

-use crate::{config::PageServerConf, tenant::storage_layer::LayerFileName};
+use crate::{config::PageServerConf, tenant::storage_layer::LayerName};

 // TODO: configurable for how long to wait before executing deletions

@@ -479,7 +479,7 @@ impl DeletionQueueClient {
        tenant_shard_id: TenantShardId,
        timeline_id: TimelineId,
        current_generation: Generation,
-        layers: Vec<(LayerFileName, LayerFileMetadata)>,
+        layers: Vec<(LayerName, LayerFileMetadata)>,
    ) -> Result<(), DeletionQueueError> {
        if current_generation.is_none() {
            debug!("Enqueuing deletions in legacy mode, skipping queue");
@@ -511,7 +511,7 @@ impl DeletionQueueClient {
        tenant_shard_id: TenantShardId,
        timeline_id: TimelineId,
        current_generation: Generation,
-        layers: Vec<(LayerFileName, LayerFileMetadata)>,
+        layers: Vec<(LayerName, LayerFileMetadata)>,
    ) -> Result<(), DeletionQueueError> {
        metrics::DELETION_QUEUE
            .keys_submitted
@@ -734,20 +734,20 @@ mod test {
    use crate::{
        control_plane_client::RetryForeverError,
        repository::Key,
-        tenant::{harness::TenantHarness, storage_layer::DeltaFileName},
+        tenant::{harness::TenantHarness, storage_layer::DeltaLayerName},
    };

    use super::*;
    pub const TIMELINE_ID: TimelineId =
        TimelineId::from_array(hex!("11223344556677881122334455667788"));

-    pub const EXAMPLE_LAYER_NAME: LayerFileName = LayerFileName::Delta(DeltaFileName {
+    pub const EXAMPLE_LAYER_NAME: LayerName = LayerName::Delta(DeltaLayerName {
        key_range: Key::from_i128(0x0)..Key::from_i128(0xFFFFFFFFFFFFFFFF),
        lsn_range: Lsn(0x00000000016B59D8)..Lsn(0x00000000016B5A51),
    });

    // When you need a second layer in a test.
-    pub const EXAMPLE_LAYER_NAME_ALT: LayerFileName = LayerFileName::Delta(DeltaFileName {
+    pub const EXAMPLE_LAYER_NAME_ALT: LayerName = LayerName::Delta(DeltaLayerName {
        key_range: Key::from_i128(0x0)..Key::from_i128(0xFFFFFFFFFFFFFFFF),
        lsn_range: Lsn(0x00000000016B5A51)..Lsn(0x00000000016B5A61),
    });
@@ -797,7 +797,7 @@ mod test {
        /// Returns remote layer file name, suitable for use in assert_remote_files
        fn write_remote_layer(
            &self,
-            file_name: LayerFileName,
+            file_name: LayerName,
            gen: Generation,
        ) -> anyhow::Result<String> {
            let tenant_shard_id = self.harness.tenant_shard_id;
@@ -952,7 +952,7 @@ mod test {
        let client = ctx.deletion_queue.new_client();
        client.recover(HashMap::new())?;

-        let layer_file_name_1: LayerFileName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
+        let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
        let tenant_shard_id = ctx.harness.tenant_shard_id;

        let content: Vec<u8> = "victim1 contents".into();
--- a/pageserver/src/deletion_queue/list_writer.rs
+++ b/pageserver/src/deletion_queue/list_writer.rs
@@ -34,7 +34,7 @@ use crate::deletion_queue::TEMP_SUFFIX;
 use crate::metrics;
 use crate::tenant::remote_timeline_client::remote_layer_path;
 use crate::tenant::remote_timeline_client::LayerFileMetadata;
-use crate::tenant::storage_layer::LayerFileName;
+use crate::tenant::storage_layer::LayerName;
 use crate::virtual_file::on_fatal_io_error;
 use crate::virtual_file::MaybeFatalIo;

@@ -59,7 +59,7 @@ pub(super) struct DeletionOp {
    // `layers` and `objects` are both just lists of objects.  `layers` is used if you do not
    // have a config object handy to project it to a remote key, and need the consuming worker
    // to do it for you.
-    pub(super) layers: Vec<(LayerFileName, LayerFileMetadata)>,
+    pub(super) layers: Vec<(LayerName, LayerFileMetadata)>,
    pub(super) objects: Vec<RemotePath>,

    /// The _current_ generation of the Tenant shard attachment in which we are enqueuing
--- a/pageserver/src/disk_usage_eviction_task.rs
+++ b/pageserver/src/disk_usage_eviction_task.rs
@@ -64,7 +64,7 @@ use crate::{
        mgr::TenantManager,
        remote_timeline_client::LayerFileMetadata,
        secondary::SecondaryTenant,
-        storage_layer::{AsLayerDesc, EvictionError, Layer, LayerFileName},
+        storage_layer::{AsLayerDesc, EvictionError, Layer, LayerName},
    },
 };

@@ -540,7 +540,12 @@ pub(crate) async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
                    js.spawn(async move {
                        layer
                            .secondary_tenant
-                            .evict_layer(tenant_manager.get_conf(), layer.timeline_id, layer.name)
+                            .evict_layer(
+                                tenant_manager.get_conf(),
+                                layer.timeline_id,
+                                layer.name,
+                                layer.metadata,
+                            )
                            .await;
                        Ok(file_size)
                    });
@@ -599,7 +604,7 @@ pub(crate) async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
 pub(crate) struct EvictionSecondaryLayer {
    pub(crate) secondary_tenant: Arc<SecondaryTenant>,
    pub(crate) timeline_id: TimelineId,
-    pub(crate) name: LayerFileName,
+    pub(crate) name: LayerName,
    pub(crate) metadata: LayerFileMetadata,
 }

@@ -632,9 +637,9 @@ impl EvictionLayer {
        }
    }

-    pub(crate) fn get_name(&self) -> LayerFileName {
+    pub(crate) fn get_name(&self) -> LayerName {
        match self {
-            Self::Attached(l) => l.layer_desc().filename(),
+            Self::Attached(l) => l.layer_desc().layer_name(),
            Self::Secondary(sl) => sl.name.clone(),
        }
    }
--- a/pageserver/src/http/routes.rs
+++ b/pageserver/src/http/routes.rs
@@ -63,6 +63,7 @@ use crate::tenant::remote_timeline_client::list_remote_timelines;
 use crate::tenant::secondary::SecondaryController;
 use crate::tenant::size::ModelInputs;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
+use crate::tenant::storage_layer::LayerName;
 use crate::tenant::timeline::CompactFlags;
 use crate::tenant::timeline::Timeline;
 use crate::tenant::SpawnMode;
@@ -1228,13 +1229,15 @@ async fn layer_download_handler(
    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
    let layer_file_name = get_request_param(&request, "layer_file_name")?;
    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let layer_name = LayerName::from_str(layer_file_name)
+        .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
    let state = get_state(&request);

    let timeline =
        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
            .await?;
    let downloaded = timeline
-        .download_layer(layer_file_name)
+        .download_layer(&layer_name)
        .await
        .map_err(ApiError::InternalServerError)?;

@@ -1258,11 +1261,14 @@ async fn evict_timeline_layer_handler(
    let layer_file_name = get_request_param(&request, "layer_file_name")?;
    let state = get_state(&request);

+    let layer_name = LayerName::from_str(layer_file_name)
+        .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
+
    let timeline =
        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
            .await?;
    let evicted = timeline
-        .evict_layer(layer_file_name)
+        .evict_layer(&layer_name)
        .await
        .map_err(ApiError::InternalServerError)?;

@@ -1827,6 +1833,75 @@ async fn timeline_download_remote_layers_handler_get(
    json_response(StatusCode::OK, info)
 }

+async fn timeline_detach_ancestor_handler(
+    request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    use crate::tenant::timeline::detach_ancestor::Options;
+    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
+    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+
+    let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
+
+    async move {
+        let mut options = Options::default();
+
+        let rewrite_concurrency =
+            parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
+        let copy_concurrency =
+            parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
+
+        [
+            (&mut options.rewrite_concurrency, rewrite_concurrency),
+            (&mut options.copy_concurrency, copy_concurrency),
+        ]
+        .into_iter()
+        .filter_map(|(target, val)| val.map(|val| (target, val)))
+        .for_each(|(target, val)| *target = val);
+
+        let state = get_state(&request);
+
+        let tenant = state
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
+
+        tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
+
+        let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
+        let ctx = &ctx;
+
+        let timeline = tenant
+            .get_timeline(timeline_id, true)
+            .map_err(|e| ApiError::NotFound(e.into()))?;
+
+        let (_guard, prepared) = timeline
+            .prepare_to_detach_from_ancestor(&tenant, options, ctx)
+            .await
+            .map_err(|e| ApiError::InternalServerError(e.into()))?;
+
+        let res = state
+            .tenant_manager
+            .complete_detaching_timeline_ancestor(tenant_shard_id, timeline_id, prepared, ctx)
+            .await;
+
+        match res {
+            Ok(reparented_timelines) => {
+                let resp = pageserver_api::models::detach_ancestor::AncestorDetached {
+                    reparented_timelines,
+                };
+
+                json_response(StatusCode::OK, resp)
+            }
+            Err(e) => Err(ApiError::InternalServerError(
+                e.context("timeline detach completion"),
+            )),
+        }
+    }
+    .instrument(span)
+    .await
+}
+
 async fn deletion_queue_flush(
    r: Request<Body>,
    cancel: CancellationToken,
@@ -2515,6 +2590,10 @@ pub fn make_router(
            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
            |r| api_handler(r, timeline_download_remote_layers_handler_get),
        )
+        .put(
+            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
+            |r| api_handler(r, timeline_detach_ancestor_handler),
+        )
        .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
            api_handler(r, timeline_delete_handler)
        })
--- a/pageserver/src/metrics.rs
+++ b/pageserver/src/metrics.rs
@@ -1512,29 +1512,80 @@ static REMOTE_TIMELINE_CLIENT_BYTES_FINISHED_COUNTER: Lazy<IntCounterVec> = Lazy
 });

 pub(crate) struct TenantManagerMetrics {
-    pub(crate) tenant_slots: UIntGauge,
+    tenant_slots_attached: UIntGauge,
+    tenant_slots_secondary: UIntGauge,
+    tenant_slots_inprogress: UIntGauge,
    pub(crate) tenant_slot_writes: IntCounter,
    pub(crate) unexpected_errors: IntCounter,
 }

+impl TenantManagerMetrics {
+    /// Helpers for tracking slots.  Note that these do not track the lifetime of TenantSlot objects
+    /// exactly: they track the lifetime of the slots _in the tenant map_.
+    pub(crate) fn slot_inserted(&self, slot: &TenantSlot) {
+        match slot {
+            TenantSlot::Attached(_) => {
+                self.tenant_slots_attached.inc();
+            }
+            TenantSlot::Secondary(_) => {
+                self.tenant_slots_secondary.inc();
+            }
+            TenantSlot::InProgress(_) => {
+                self.tenant_slots_inprogress.inc();
+            }
+        }
+    }
+
+    pub(crate) fn slot_removed(&self, slot: &TenantSlot) {
+        match slot {
+            TenantSlot::Attached(_) => {
+                self.tenant_slots_attached.dec();
+            }
+            TenantSlot::Secondary(_) => {
+                self.tenant_slots_secondary.dec();
+            }
+            TenantSlot::InProgress(_) => {
+                self.tenant_slots_inprogress.dec();
+            }
+        }
+    }
+
+    #[cfg(all(debug_assertions, not(test)))]
+    pub(crate) fn slots_total(&self) -> u64 {
+        self.tenant_slots_attached.get()
+            + self.tenant_slots_secondary.get()
+            + self.tenant_slots_inprogress.get()
+    }
+}
+
 pub(crate) static TENANT_MANAGER: Lazy<TenantManagerMetrics> = Lazy::new(|| {
-    TenantManagerMetrics {
-    tenant_slots: register_uint_gauge!(
+    let tenant_slots = register_uint_gauge_vec!(
        "pageserver_tenant_manager_slots",
        "How many slots currently exist, including all attached, secondary and in-progress operations",
+        &["mode"]
    )
-    .expect("failed to define a metric"),
-    tenant_slot_writes: register_int_counter!(
-        "pageserver_tenant_manager_slot_writes",
-        "Writes to a tenant slot, including all of create/attach/detach/delete"
-    )
-    .expect("failed to define a metric"),
-    unexpected_errors: register_int_counter!(
-        "pageserver_tenant_manager_unexpected_errors_total",
-        "Number of unexpected conditions encountered: nonzero value indicates a non-fatal bug."
-    )
-    .expect("failed to define a metric"),
-}
+    .expect("failed to define a metric");
+    TenantManagerMetrics {
+        tenant_slots_attached: tenant_slots
+            .get_metric_with_label_values(&["attached"])
+            .unwrap(),
+        tenant_slots_secondary: tenant_slots
+            .get_metric_with_label_values(&["secondary"])
+            .unwrap(),
+        tenant_slots_inprogress: tenant_slots
+            .get_metric_with_label_values(&["inprogress"])
+            .unwrap(),
+        tenant_slot_writes: register_int_counter!(
+            "pageserver_tenant_manager_slot_writes",
+            "Writes to a tenant slot, including all of create/attach/detach/delete"
+        )
+        .expect("failed to define a metric"),
+        unexpected_errors: register_int_counter!(
+            "pageserver_tenant_manager_unexpected_errors_total",
+            "Number of unexpected conditions encountered: nonzero value indicates a non-fatal bug."
+        )
+        .expect("failed to define a metric"),
+    }
 });

 pub(crate) struct DeletionQueueMetrics {
@@ -2275,6 +2326,7 @@ use std::time::{Duration, Instant};

 use crate::context::{PageContentKind, RequestContext};
 use crate::task_mgr::TaskKind;
+use crate::tenant::mgr::TenantSlot;

 /// Maintain a per timeline gauge in addition to the global gauge.
 struct PerTimelineRemotePhysicalSizeGauge {
@@ -2877,6 +2929,8 @@ pub fn preinitialize_metrics() {
        &WALRECEIVER_CANDIDATES_REMOVED,
        &tokio_epoll_uring::THREAD_LOCAL_LAUNCH_FAILURES,
        &tokio_epoll_uring::THREAD_LOCAL_LAUNCH_SUCCESSES,
+        &REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
+        &REMOTE_ONDEMAND_DOWNLOADED_BYTES,
    ]
    .into_iter()
    .for_each(|c| {
--- a/pageserver/src/pgdatadir_mapping.rs
+++ b/pageserver/src/pgdatadir_mapping.rs
@@ -10,9 +10,9 @@ use super::tenant::{PageReconstructError, Timeline};
 use crate::context::RequestContext;
 use crate::keyspace::{KeySpace, KeySpaceAccum};
 use crate::metrics::WAL_INGEST;
-use crate::repository::*;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id;
 use crate::walrecord::NeonWalRecord;
+use crate::{aux_file, repository::*};
 use anyhow::{ensure, Context};
 use bytes::{Buf, Bytes, BytesMut};
 use enum_map::Enum;
@@ -24,6 +24,7 @@ use pageserver_api::key::{
    AUX_FILES_KEY, CHECKPOINT_KEY, CONTROLFILE_KEY, DBDIR_KEY, TWOPHASEDIR_KEY,
 };
 use pageserver_api::keyspace::SparseKeySpace;
+use pageserver_api::models::AuxFilePolicy;
 use pageserver_api::reltag::{BlockNumber, RelTag, SlruKind};
 use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
 use postgres_ffi::BLCKSZ;
@@ -670,7 +671,7 @@ impl Timeline {
        self.get(CHECKPOINT_KEY, lsn, ctx).await
    }

-    pub(crate) async fn list_aux_files(
+    async fn list_aux_files_v1(
        &self,
        lsn: Lsn,
        ctx: &RequestContext,
@@ -688,6 +689,63 @@ impl Timeline {
        }
    }

+    async fn list_aux_files_v2(
+        &self,
+        lsn: Lsn,
+        ctx: &RequestContext,
+    ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
+        let kv = self
+            .scan(KeySpace::single(Key::metadata_aux_key_range()), lsn, ctx)
+            .await
+            .context("scan")?;
+        let mut result = HashMap::new();
+        for (_, v) in kv {
+            let v = v.context("get value")?;
+            let v = aux_file::decode_file_value_bytes(&v).context("value decode")?;
+            for (fname, content) in v {
+                result.insert(fname, content);
+            }
+        }
+        Ok(result)
+    }
+
+    pub(crate) async fn list_aux_files(
+        &self,
+        lsn: Lsn,
+        ctx: &RequestContext,
+    ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
+        match self.get_switch_aux_file_policy() {
+            AuxFilePolicy::V1 => self.list_aux_files_v1(lsn, ctx).await,
+            AuxFilePolicy::V2 => self.list_aux_files_v2(lsn, ctx).await,
+            AuxFilePolicy::CrossValidation => {
+                let v1_result = self.list_aux_files_v1(lsn, ctx).await;
+                let v2_result = self.list_aux_files_v2(lsn, ctx).await;
+                match (v1_result, v2_result) {
+                    (Ok(v1), Ok(v2)) => {
+                        if v1 != v2 {
+                            tracing::error!(
+                                "unmatched aux file v1 v2 result:\nv1 {v1:?}\nv2 {v2:?}"
+                            );
+                            return Err(PageReconstructError::Other(anyhow::anyhow!(
+                                "unmatched aux file v1 v2 result"
+                            )));
+                        }
+                        Ok(v1)
+                    }
+                    (Ok(_), Err(v2)) => {
+                        tracing::error!("aux file v1 returns Ok while aux file v2 returns an err");
+                        Err(v2)
+                    }
+                    (Err(v1), Ok(_)) => {
+                        tracing::error!("aux file v2 returns Ok while aux file v1 returns an err");
+                        Err(v1)
+                    }
+                    (Err(_), Err(v2)) => Err(v2),
+                }
+            }
+        }
+    }
+
    /// Does the same as get_current_logical_size but counted on demand.
    /// Used to initialize the logical size tracking on startup.
    ///
@@ -1389,6 +1447,9 @@ impl<'a> DatadirModification<'a> {
    }

    pub fn init_aux_dir(&mut self) -> anyhow::Result<()> {
+        if let AuxFilePolicy::V2 = self.tline.get_switch_aux_file_policy() {
+            return Ok(());
+        }
        let buf = AuxFilesDirectory::ser(&AuxFilesDirectory {
            files: HashMap::new(),
        })?;
@@ -1404,89 +1465,121 @@ impl<'a> DatadirModification<'a> {
        content: &[u8],
        ctx: &RequestContext,
    ) -> anyhow::Result<()> {
-        let file_path = path.to_string();
-        let content = if content.is_empty() {
-            None
-        } else {
-            Some(Bytes::copy_from_slice(content))
-        };
-
-        let n_files;
-        let mut aux_files = self.tline.aux_files.lock().await;
-        if let Some(mut dir) = aux_files.dir.take() {
-            // We already updated aux files in `self`: emit a delta and update our latest value.
-            dir.upsert(file_path.clone(), content.clone());
-            n_files = dir.files.len();
-            if aux_files.n_deltas == MAX_AUX_FILE_DELTAS {
-                self.put(
-                    AUX_FILES_KEY,
-                    Value::Image(Bytes::from(
-                        AuxFilesDirectory::ser(&dir).context("serialize")?,
-                    )),
-                );
-                aux_files.n_deltas = 0;
+        let policy = self.tline.get_switch_aux_file_policy();
+        if let AuxFilePolicy::V2 | AuxFilePolicy::CrossValidation = policy {
+            let key = aux_file::encode_aux_file_key(path);
+            // retrieve the key from the engine
+            let old_val = match self.get(key, ctx).await {
+                Ok(val) => Some(val),
+                Err(PageReconstructError::MissingKey(_)) => None,
+                Err(e) => return Err(e.into()),
+            };
+            let files = if let Some(ref old_val) = old_val {
+                aux_file::decode_file_value(old_val)?
            } else {
-                self.put(
-                    AUX_FILES_KEY,
-                    Value::WalRecord(NeonWalRecord::AuxFile { file_path, content }),
-                );
-                aux_files.n_deltas += 1;
-            }
-            aux_files.dir = Some(dir);
-        } else {
-            // Check if the AUX_FILES_KEY is initialized
-            match self.get(AUX_FILES_KEY, ctx).await {
-                Ok(dir_bytes) => {
-                    let mut dir = AuxFilesDirectory::des(&dir_bytes)?;
-                    // Key is already set, we may append a delta
-                    self.put(
-                        AUX_FILES_KEY,
-                        Value::WalRecord(NeonWalRecord::AuxFile {
-                            file_path: file_path.clone(),
-                            content: content.clone(),
-                        }),
-                    );
-                    dir.upsert(file_path, content);
-                    n_files = dir.files.len();
-                    aux_files.dir = Some(dir);
-                }
-                Err(
-                    e @ (PageReconstructError::AncestorStopping(_)
-                    | PageReconstructError::Cancelled
-                    | PageReconstructError::AncestorLsnTimeout(_)),
-                ) => {
-                    // Important that we do not interpret a shutdown error as "not found" and thereby
-                    // reset the map.
-                    return Err(e.into());
-                }
-                // Note: we added missing key error variant in https://github.com/neondatabase/neon/pull/7393 but
-                // the original code assumes all other errors are missing keys. Therefore, we keep the code path
-                // the same for now, though in theory, we should only match the `MissingKey` variant.
-                Err(
-                    PageReconstructError::Other(_)
-                    | PageReconstructError::WalRedo(_)
-                    | PageReconstructError::MissingKey { .. },
-                ) => {
-                    // Key is missing, we must insert an image as the basis for subsequent deltas.
+                Vec::new()
+            };
+            let new_files = if content.is_empty() {
+                files
+                    .into_iter()
+                    .filter(|(p, _)| &path != p)
+                    .collect::<Vec<_>>()
+            } else {
+                files
+                    .into_iter()
+                    .filter(|(p, _)| &path != p)
+                    .chain(std::iter::once((path, content)))
+                    .collect::<Vec<_>>()
+            };
+            let new_val = aux_file::encode_file_value(&new_files)?;
+            self.put(key, Value::Image(new_val.into()));
+        }

-                    let mut dir = AuxFilesDirectory {
-                        files: HashMap::new(),
-                    };
-                    dir.upsert(file_path, content);
+        if let AuxFilePolicy::V1 | AuxFilePolicy::CrossValidation = policy {
+            let file_path = path.to_string();
+            let content = if content.is_empty() {
+                None
+            } else {
+                Some(Bytes::copy_from_slice(content))
+            };
+
+            let n_files;
+            let mut aux_files = self.tline.aux_files.lock().await;
+            if let Some(mut dir) = aux_files.dir.take() {
+                // We already updated aux files in `self`: emit a delta and update our latest value.
+                dir.upsert(file_path.clone(), content.clone());
+                n_files = dir.files.len();
+                if aux_files.n_deltas == MAX_AUX_FILE_DELTAS {
                    self.put(
                        AUX_FILES_KEY,
                        Value::Image(Bytes::from(
                            AuxFilesDirectory::ser(&dir).context("serialize")?,
                        )),
                    );
-                    n_files = 1;
-                    aux_files.dir = Some(dir);
+                    aux_files.n_deltas = 0;
+                } else {
+                    self.put(
+                        AUX_FILES_KEY,
+                        Value::WalRecord(NeonWalRecord::AuxFile { file_path, content }),
+                    );
+                    aux_files.n_deltas += 1;
+                }
+                aux_files.dir = Some(dir);
+            } else {
+                // Check if the AUX_FILES_KEY is initialized
+                match self.get(AUX_FILES_KEY, ctx).await {
+                    Ok(dir_bytes) => {
+                        let mut dir = AuxFilesDirectory::des(&dir_bytes)?;
+                        // Key is already set, we may append a delta
+                        self.put(
+                            AUX_FILES_KEY,
+                            Value::WalRecord(NeonWalRecord::AuxFile {
+                                file_path: file_path.clone(),
+                                content: content.clone(),
+                            }),
+                        );
+                        dir.upsert(file_path, content);
+                        n_files = dir.files.len();
+                        aux_files.dir = Some(dir);
+                    }
+                    Err(
+                        e @ (PageReconstructError::AncestorStopping(_)
+                        | PageReconstructError::Cancelled
+                        | PageReconstructError::AncestorLsnTimeout(_)),
+                    ) => {
+                        // Important that we do not interpret a shutdown error as "not found" and thereby
+                        // reset the map.
+                        return Err(e.into());
+                    }
+                    // Note: we added missing key error variant in https://github.com/neondatabase/neon/pull/7393 but
+                    // the original code assumes all other errors are missing keys. Therefore, we keep the code path
+                    // the same for now, though in theory, we should only match the `MissingKey` variant.
+                    Err(
+                        PageReconstructError::Other(_)
+                        | PageReconstructError::WalRedo(_)
+                        | PageReconstructError::MissingKey { .. },
+                    ) => {
+                        // Key is missing, we must insert an image as the basis for subsequent deltas.
+
+                        let mut dir = AuxFilesDirectory {
+                            files: HashMap::new(),
+                        };
+                        dir.upsert(file_path, content);
+                        self.put(
+                            AUX_FILES_KEY,
+                            Value::Image(Bytes::from(
+                                AuxFilesDirectory::ser(&dir).context("serialize")?,
+                            )),
+                        );
+                        n_files = 1;
+                        aux_files.dir = Some(dir);
+                    }
                }
            }
-        }

-        self.pending_directory_entries
-            .push((DirectoryKind::AuxFiles, n_files));
+            self.pending_directory_entries
+                .push((DirectoryKind::AuxFiles, n_files));
+        }

        Ok(())
    }
--- a/pageserver/src/repository.rs
+++ b/pageserver/src/repository.rs
@@ -33,7 +33,6 @@ impl Value {
    }
 }

-#[cfg(test)]
 #[derive(Debug, PartialEq)]
 pub(crate) enum InvalidInput {
    TooShortValue,
@@ -42,10 +41,8 @@ pub(crate) enum InvalidInput {

 /// We could have a ValueRef where everything is `serde(borrow)`. Before implementing that, lets
 /// use this type for querying if a slice looks some particular way.
-#[cfg(test)]
 pub(crate) struct ValueBytes;

-#[cfg(test)]
 impl ValueBytes {
    pub(crate) fn will_init(raw: &[u8]) -> Result<bool, InvalidInput> {
        if raw.len() < 12 {
--- a/pageserver/src/task_mgr.rs
+++ b/pageserver/src/task_mgr.rs
@@ -319,6 +319,9 @@ pub enum TaskKind {
    // Eviction. One per timeline.
    Eviction,

+    // Ingest housekeeping (flushing ephemeral layers on time threshold or disk pressure)
+    IngestHousekeeping,
+
    /// See [`crate::disk_usage_eviction_task`].
    DiskUsageEviction,

@@ -367,6 +370,8 @@ pub enum TaskKind {

    #[cfg(test)]
    UnitTest,
+
+    DetachAncestor,
 }

 #[derive(Default)]
--- a/pageserver/src/tenant.rs
+++ b/pageserver/src/tenant.rs
@@ -322,6 +322,9 @@ pub struct Tenant {
    /// All [`Tenant::timelines`] of a given [`Tenant`] instance share the same [`throttle::Throttle`] instance.
    pub(crate) timeline_get_throttle:
        Arc<throttle::Throttle<&'static crate::metrics::tenant_throttling::TimelineGet>>,
+
+    /// An ongoing timeline detach must be checked during attempts to GC or compact a timeline.
+    ongoing_timeline_detach: std::sync::Mutex<Option<(TimelineId, utils::completion::Barrier)>>,
 }

 impl std::fmt::Debug for Tenant {
@@ -1676,6 +1679,34 @@ impl Tenant {
        Ok(())
    }

+    // Call through to all timelines to freeze ephemeral layers if needed.  Usually
+    // this happens during ingest: this background housekeeping is for freezing layers
+    // that are open but haven't been written to for some time.
+    async fn ingest_housekeeping(&self) {
+        // Scan through the hashmap and collect a list of all the timelines,
+        // while holding the lock. Then drop the lock and actually perform the
+        // compactions.  We don't want to block everything else while the
+        // compaction runs.
+        let timelines = {
+            self.timelines
+                .lock()
+                .unwrap()
+                .values()
+                .filter_map(|timeline| {
+                    if timeline.is_active() {
+                        Some(timeline.clone())
+                    } else {
+                        None
+                    }
+                })
+                .collect::<Vec<_>>()
+        };
+
+        for timeline in &timelines {
+            timeline.maybe_freeze_ephemeral_layer().await;
+        }
+    }
+
    pub fn current_state(&self) -> TenantState {
        self.state.borrow().clone()
    }
@@ -2529,6 +2560,7 @@ impl Tenant {
                &crate::metrics::tenant_throttling::TIMELINE_GET,
            )),
            tenant_conf: Arc::new(ArcSwap::from_pointee(attached_conf)),
+            ongoing_timeline_detach: std::sync::Mutex::default(),
        }
    }

@@ -3726,7 +3758,7 @@ pub(crate) mod harness {
                image_layer_creation_check_threshold: Some(
                    tenant_conf.image_layer_creation_check_threshold,
                ),
-                switch_to_aux_file_v2: Some(tenant_conf.switch_to_aux_file_v2),
+                switch_aux_file_policy: Some(tenant_conf.switch_aux_file_policy),
            }
        }
    }
--- a/pageserver/src/tenant/config.rs
+++ b/pageserver/src/tenant/config.rs
@@ -9,6 +9,7 @@
 //! may lead to a data loss.
 //!
 use anyhow::bail;
+use pageserver_api::models::AuxFilePolicy;
 use pageserver_api::models::CompactionAlgorithm;
 use pageserver_api::models::EvictionPolicy;
 use pageserver_api::models::{self, ThrottleConfig};
@@ -370,9 +371,9 @@ pub struct TenantConf {
    // Expresed in multiples of checkpoint distance.
    pub image_layer_creation_check_threshold: u8,

-    /// Switch to aux file v2. Switching this flag requires the user has not written any aux file into
+    /// Switch to a new aux file policy. Switching this flag requires the user has not written any aux file into
    /// the storage before, and this flag cannot be switched back. Otherwise there will be data corruptions.
-    pub switch_to_aux_file_v2: bool,
+    pub switch_aux_file_policy: AuxFilePolicy,
 }

 /// Same as TenantConf, but this struct preserves the information about
@@ -471,7 +472,7 @@ pub struct TenantConfOpt {

    #[serde(skip_serializing_if = "Option::is_none")]
    #[serde(default)]
-    pub switch_to_aux_file_v2: Option<bool>,
+    pub switch_aux_file_policy: Option<AuxFilePolicy>,
 }

 impl TenantConfOpt {
@@ -529,9 +530,9 @@ impl TenantConfOpt {
            image_layer_creation_check_threshold: self
                .image_layer_creation_check_threshold
                .unwrap_or(global_conf.image_layer_creation_check_threshold),
-            switch_to_aux_file_v2: self
-                .switch_to_aux_file_v2
-                .unwrap_or(global_conf.switch_to_aux_file_v2),
+            switch_aux_file_policy: self
+                .switch_aux_file_policy
+                .unwrap_or(global_conf.switch_aux_file_policy),
        }
    }
 }
@@ -573,7 +574,7 @@ impl Default for TenantConf {
            lazy_slru_download: false,
            timeline_get_throttle: crate::tenant::throttle::Config::disabled(),
            image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
-            switch_to_aux_file_v2: false,
+            switch_aux_file_policy: AuxFilePolicy::V1,
        }
    }
 }
@@ -648,7 +649,7 @@ impl From<TenantConfOpt> for models::TenantConfig {
            lazy_slru_download: value.lazy_slru_download,
            timeline_get_throttle: value.timeline_get_throttle.map(ThrottleConfig::from),
            image_layer_creation_check_threshold: value.image_layer_creation_check_threshold,
-            switch_to_aux_file_v2: value.switch_to_aux_file_v2,
+            switch_aux_file_policy: value.switch_aux_file_policy,
        }
    }
 }
--- a/pageserver/src/tenant/delete.rs
+++ b/pageserver/src/tenant/delete.rs
@@ -585,9 +585,20 @@ impl DeleteTenantFlow {

                    // FIXME: we should not be modifying this from outside of mgr.rs.
                    // This will go away when we simplify deletion (https://github.com/neondatabase/neon/issues/5080)
-                    crate::metrics::TENANT_MANAGER
-                        .tenant_slots
-                        .set(locked.len() as u64);
+
+                    // Update stats
+                    match &removed {
+                        TenantsMapRemoveResult::Occupied(slot) => {
+                            crate::metrics::TENANT_MANAGER.slot_removed(slot);
+                        }
+                        TenantsMapRemoveResult::InProgress(barrier) => {
+                            crate::metrics::TENANT_MANAGER
+                                .slot_removed(&TenantSlot::InProgress(barrier.clone()));
+                        }
+                        TenantsMapRemoveResult::Vacant => {
+                            // Nothing changed in map, no metric update
+                        }
+                    }

                    match removed {
                        TenantsMapRemoveResult::Occupied(TenantSlot::Attached(tenant)) => {
--- a/pageserver/src/tenant/metadata.rs
+++ b/pageserver/src/tenant/metadata.rs
@@ -207,6 +207,24 @@ impl TimelineMetadata {
        self.body.ancestor_lsn
    }

+    /// When reparenting, the `ancestor_lsn` does not change.
+    pub fn reparent(&mut self, timeline: &TimelineId) {
+        assert!(self.body.ancestor_timeline.is_some());
+        // no assertion for redoing this: it's fine, we may have to repeat this multiple times over
+        self.body.ancestor_timeline = Some(*timeline);
+    }
+
+    pub fn detach_from_ancestor(&mut self, timeline: &TimelineId, ancestor_lsn: &Lsn) {
+        if let Some(ancestor) = self.body.ancestor_timeline {
+            assert_eq!(ancestor, *timeline);
+        }
+        if self.body.ancestor_lsn != Lsn(0) {
+            assert_eq!(self.body.ancestor_lsn, *ancestor_lsn);
+        }
+        self.body.ancestor_timeline = None;
+        self.body.ancestor_lsn = Lsn(0);
+    }
+
    pub fn latest_gc_cutoff_lsn(&self) -> Lsn {
        self.body.latest_gc_cutoff_lsn
    }
--- a/pageserver/src/tenant/mgr.rs
+++ b/pageserver/src/tenant/mgr.rs
@@ -56,6 +56,7 @@ use utils::id::{TenantId, TimelineId};

 use super::delete::DeleteTenantError;
 use super::secondary::SecondaryTenant;
+use super::timeline::detach_ancestor::PreparedTimelineDetach;
 use super::TenantSharedResources;

 /// For a tenant that appears in TenantsMap, it may either be
@@ -246,6 +247,7 @@ impl TenantsMap {
        }
    }

+    #[cfg(all(debug_assertions, not(test)))]
    pub(crate) fn len(&self) -> usize {
        match self {
            TenantsMap::Initializing => 0,
@@ -746,6 +748,7 @@ pub async fn init_tenant_mgr(
            }
        };

+        METRICS.slot_inserted(&slot);
        tenants.insert(tenant_shard_id, slot);
    }

@@ -753,7 +756,7 @@ pub async fn init_tenant_mgr(

    let mut tenants_map = TENANTS.write().unwrap();
    assert!(matches!(&*tenants_map, &TenantsMap::Initializing));
-    METRICS.tenant_slots.set(tenants.len() as u64);
+
    *tenants_map = TenantsMap::Open(tenants);

    Ok(TenantManager {
@@ -824,6 +827,14 @@ fn tenant_spawn(
 async fn shutdown_all_tenants0(tenants: &std::sync::RwLock<TenantsMap>) {
    let mut join_set = JoinSet::new();

+    #[cfg(all(debug_assertions, not(test)))]
+    {
+        // Check that our metrics properly tracked the size of the tenants map.  This is a convenient location to check,
+        // as it happens implicitly at the end of tests etc.
+        let m = tenants.read().unwrap();
+        debug_assert_eq!(METRICS.slots_total(), m.len() as u64);
+    }
+
    // Atomically, 1. create the shutdown tasks and 2. prevent creation of new tenants.
    let (total_in_progress, total_attached) = {
        let mut m = tenants.write().unwrap();
@@ -1997,6 +2008,101 @@ impl TenantManager {
            })
            .collect())
    }
+
+    /// Completes an earlier prepared timeline detach ancestor.
+    pub(crate) async fn complete_detaching_timeline_ancestor(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        prepared: PreparedTimelineDetach,
+        ctx: &RequestContext,
+    ) -> Result<Vec<TimelineId>, anyhow::Error> {
+        struct RevertOnDropSlot(Option<SlotGuard>);
+
+        impl Drop for RevertOnDropSlot {
+            fn drop(&mut self) {
+                if let Some(taken) = self.0.take() {
+                    taken.revert();
+                }
+            }
+        }
+
+        impl RevertOnDropSlot {
+            fn into_inner(mut self) -> SlotGuard {
+                self.0.take().unwrap()
+            }
+        }
+
+        impl std::ops::Deref for RevertOnDropSlot {
+            type Target = SlotGuard;
+
+            fn deref(&self) -> &Self::Target {
+                self.0.as_ref().unwrap()
+            }
+        }
+
+        let slot_guard = tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::Any)?;
+        let slot_guard = RevertOnDropSlot(Some(slot_guard));
+
+        let tenant = {
+            let Some(old_slot) = slot_guard.get_old_value() else {
+                anyhow::bail!(
+                    "Tenant not found when trying to complete detaching timeline ancestor"
+                );
+            };
+
+            let Some(tenant) = old_slot.get_attached() else {
+                anyhow::bail!("Tenant is not in attached state");
+            };
+
+            if !tenant.is_active() {
+                anyhow::bail!("Tenant is not active");
+            }
+
+            tenant.clone()
+        };
+
+        let timeline = tenant.get_timeline(timeline_id, true)?;
+
+        let reparented = timeline
+            .complete_detaching_timeline_ancestor(&tenant, prepared, ctx)
+            .await?;
+
+        let mut slot_guard = slot_guard.into_inner();
+
+        let (_guard, progress) = utils::completion::channel();
+        match tenant.shutdown(progress, ShutdownMode::Hard).await {
+            Ok(()) => {
+                slot_guard.drop_old_value()?;
+            }
+            Err(_barrier) => {
+                slot_guard.revert();
+                // this really should not happen, at all, unless shutdown was already going?
+                anyhow::bail!("Cannot restart Tenant, already shutting down");
+            }
+        }
+
+        let tenant_path = self.conf.tenant_path(&tenant_shard_id);
+        let config = Tenant::load_tenant_config(self.conf, &tenant_shard_id)?;
+
+        let shard_identity = config.shard;
+        let tenant = tenant_spawn(
+            self.conf,
+            tenant_shard_id,
+            &tenant_path,
+            self.resources.clone(),
+            AttachedTenantConf::try_from(config)?,
+            shard_identity,
+            None,
+            self.tenants,
+            SpawnMode::Eager,
+            ctx,
+        )?;
+
+        slot_guard.upsert(TenantSlot::Attached(tenant))?;
+
+        Ok(reparented)
+    }
 }

 #[derive(Debug, thiserror::Error)]
@@ -2428,10 +2534,13 @@ impl SlotGuard {
                TenantsMap::Open(m) => m,
            };

+            METRICS.slot_inserted(&new_value);
+
            let replaced = m.insert(self.tenant_shard_id, new_value);
            self.upserted = true;
-
-            METRICS.tenant_slots.set(m.len() as u64);
+            if let Some(replaced) = replaced.as_ref() {
+                METRICS.slot_removed(replaced);
+            }

            replaced
        };
@@ -2541,9 +2650,13 @@ impl Drop for SlotGuard {
                }

                if self.old_value_is_shutdown() {
+                    METRICS.slot_removed(entry.get());
                    entry.remove();
                } else {
-                    entry.insert(self.old_value.take().unwrap());
+                    let inserting = self.old_value.take().unwrap();
+                    METRICS.slot_inserted(&inserting);
+                    let replaced = entry.insert(inserting);
+                    METRICS.slot_removed(&replaced);
                }
            }
            Entry::Vacant(_) => {
@@ -2554,8 +2667,6 @@ impl Drop for SlotGuard {
                );
            }
        }
-
-        METRICS.tenant_slots.set(m.len() as u64);
    }
 }

@@ -2635,7 +2746,9 @@ fn tenant_map_acquire_slot_impl(
            }
            _ => {
                let (completion, barrier) = utils::completion::channel();
-                v.insert(TenantSlot::InProgress(barrier));
+                let inserting = TenantSlot::InProgress(barrier);
+                METRICS.slot_inserted(&inserting);
+                v.insert(inserting);
                tracing::debug!("Vacant, inserted InProgress");
                Ok(SlotGuard::new(*tenant_shard_id, None, completion))
            }
@@ -2671,7 +2784,10 @@ fn tenant_map_acquire_slot_impl(
                _ => {
                    // Happy case: the slot was not in any state that violated our mode
                    let (completion, barrier) = utils::completion::channel();
-                    let old_value = o.insert(TenantSlot::InProgress(barrier));
+                    let in_progress = TenantSlot::InProgress(barrier);
+                    METRICS.slot_inserted(&in_progress);
+                    let old_value = o.insert(in_progress);
+                    METRICS.slot_removed(&old_value);
                    tracing::debug!("Occupied, replaced with InProgress");
                    Ok(SlotGuard::new(
                        *tenant_shard_id,
--- a/pageserver/src/tenant/remote_timeline_client.rs
+++ b/pageserver/src/tenant/remote_timeline_client.rs
@@ -240,7 +240,7 @@ use utils::id::{TenantId, TimelineId};
 use self::index::IndexPart;

 use super::metadata::MetadataUpdate;
-use super::storage_layer::{Layer, LayerFileName, ResidentLayer};
+use super::storage_layer::{Layer, LayerName, ResidentLayer};
 use super::upload_queue::SetDeletedFlagProgress;
 use super::Generation;

@@ -503,7 +503,7 @@ impl RemoteTimelineClient {
    /// On success, returns the size of the downloaded file.
    pub async fn download_layer_file(
        &self,
-        layer_file_name: &LayerFileName,
+        layer_file_name: &LayerName,
        layer_metadata: &LayerFileMetadata,
        cancel: &CancellationToken,
        ctx: &RequestContext,
@@ -570,7 +570,7 @@ impl RemoteTimelineClient {
        // ahead of what's _actually_ on the remote during index upload.
        upload_queue.latest_metadata = metadata.clone();

-        self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
+        self.schedule_index_upload(upload_queue);

        Ok(())
    }
@@ -591,7 +591,7 @@ impl RemoteTimelineClient {

        upload_queue.latest_metadata.apply(update);

-        self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
+        self.schedule_index_upload(upload_queue);

        Ok(())
    }
@@ -611,18 +611,14 @@ impl RemoteTimelineClient {
        let upload_queue = guard.initialized_mut()?;

        if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
-            self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
+            self.schedule_index_upload(upload_queue);
        }

        Ok(())
    }

    /// Launch an index-file upload operation in the background (internal function)
-    fn schedule_index_upload(
-        self: &Arc<Self>,
-        upload_queue: &mut UploadQueueInitialized,
-        metadata: TimelineMetadata,
-    ) {
+    fn schedule_index_upload(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
        let disk_consistent_lsn = upload_queue.latest_metadata.disk_consistent_lsn();

        info!(
@@ -631,11 +627,7 @@ impl RemoteTimelineClient {
            upload_queue.latest_files_changes_since_metadata_upload_scheduled,
        );

-        let index_part = IndexPart::new(
-            upload_queue.latest_files.clone(),
-            disk_consistent_lsn,
-            metadata,
-        );
+        let index_part = IndexPart::from(&*upload_queue);
        let op = UploadOp::UploadMetadata(index_part, disk_consistent_lsn);
        self.metric_begin(&op);
        upload_queue.queued_operations.push_back(op);
@@ -645,9 +637,61 @@ impl RemoteTimelineClient {
        self.launch_queued_tasks(upload_queue);
    }

+    pub(crate) async fn schedule_reparenting_and_wait(
+        self: &Arc<Self>,
+        new_parent: &TimelineId,
+    ) -> anyhow::Result<()> {
+        // FIXME: because of how Timeline::schedule_uploads works when called from layer flushing
+        // and reads the in-memory part we cannot do the detaching like this
+        let receiver = {
+            let mut guard = self.upload_queue.lock().unwrap();
+            let upload_queue = guard.initialized_mut()?;
+
+            upload_queue.latest_metadata.reparent(new_parent);
+
+            self.schedule_index_upload(upload_queue);
+
+            self.schedule_barrier0(upload_queue)
+        };
+
+        Self::wait_completion0(receiver).await
+    }
+
+    /// Schedules uploading a new version of `index_part.json` with the given layers added,
+    /// detaching from ancestor and waits for it to complete.
    ///
-    /// Launch an upload operation in the background.
-    ///
+    /// This is used with `Timeline::detach_ancestor` functionality.
+    pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
+        self: &Arc<Self>,
+        layers: &[Layer],
+        adopted: (TimelineId, Lsn),
+    ) -> anyhow::Result<()> {
+        let barrier = {
+            let mut guard = self.upload_queue.lock().unwrap();
+            let upload_queue = guard.initialized_mut()?;
+
+            upload_queue
+                .latest_metadata
+                .detach_from_ancestor(&adopted.0, &adopted.1);
+
+            for layer in layers {
+                upload_queue
+                    .latest_files
+                    .insert(layer.layer_desc().layer_name(), layer.metadata());
+            }
+
+            self.schedule_index_upload(upload_queue);
+
+            let barrier = self.schedule_barrier0(upload_queue);
+            self.launch_queued_tasks(upload_queue);
+            barrier
+        };
+
+        Self::wait_completion0(barrier).await
+    }
+
+    /// Launch an upload operation in the background; the file is added to be included in next
+    /// `index_part.json` upload.
    pub(crate) fn schedule_layer_file_upload(
        self: &Arc<Self>,
        layer: ResidentLayer,
@@ -669,13 +713,15 @@ impl RemoteTimelineClient {

        upload_queue
            .latest_files
-            .insert(layer.layer_desc().filename(), metadata.clone());
+            .insert(layer.layer_desc().layer_name(), metadata.clone());
        upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;

        info!(
-            "scheduled layer file upload {layer} gen={:?} shard={:?}",
-            metadata.generation, metadata.shard
+            gen=?metadata.generation,
+            shard=?metadata.shard,
+            "scheduled layer file upload {layer}",
        );
+
        let op = UploadOp::UploadLayer(layer, metadata);
        self.metric_begin(&op);
        upload_queue.queued_operations.push_back(op);
@@ -691,7 +737,7 @@ impl RemoteTimelineClient {
    /// successfully.
    pub fn schedule_layer_file_deletion(
        self: &Arc<Self>,
-        names: &[LayerFileName],
+        names: &[LayerName],
    ) -> anyhow::Result<()> {
        let mut guard = self.upload_queue.lock().unwrap();
        let upload_queue = guard.initialized_mut()?;
@@ -719,7 +765,7 @@ impl RemoteTimelineClient {
        // the layer files as "dangling". this is fine, at worst case we create work for the
        // scrubber.

-        let names = gc_layers.iter().map(|x| x.layer_desc().filename());
+        let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());

        self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);

@@ -734,14 +780,10 @@ impl RemoteTimelineClient {
        self: &Arc<Self>,
        upload_queue: &mut UploadQueueInitialized,
        names: I,
-    ) -> Vec<(LayerFileName, LayerFileMetadata)>
+    ) -> Vec<(LayerName, LayerFileMetadata)>
    where
-        I: IntoIterator<Item = LayerFileName>,
+        I: IntoIterator<Item = LayerName>,
    {
-        // Deleting layers doesn't affect the values stored in TimelineMetadata,
-        // so we don't need update it. Just serialize it.
-        let metadata = upload_queue.latest_metadata.clone();
-
        // Decorate our list of names with each name's metadata, dropping
        // names that are unexpectedly missing from our metadata.  This metadata
        // is later used when physically deleting layers, to construct key paths.
@@ -780,7 +822,7 @@ impl RemoteTimelineClient {
        // index_part update, because that needs to be uploaded before we can actually delete the
        // files.
        if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
-            self.schedule_index_upload(upload_queue, metadata);
+            self.schedule_index_upload(upload_queue);
        }

        with_metadata
@@ -790,7 +832,7 @@ impl RemoteTimelineClient {
    /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
    pub(crate) fn schedule_deletion_of_unlinked(
        self: &Arc<Self>,
-        layers: Vec<(LayerFileName, LayerFileMetadata)>,
+        layers: Vec<(LayerName, LayerFileMetadata)>,
    ) -> anyhow::Result<()> {
        let mut guard = self.upload_queue.lock().unwrap();
        let upload_queue = guard.initialized_mut()?;
@@ -803,7 +845,7 @@ impl RemoteTimelineClient {
    fn schedule_deletion_of_unlinked0(
        self: &Arc<Self>,
        upload_queue: &mut UploadQueueInitialized,
-        mut with_metadata: Vec<(LayerFileName, LayerFileMetadata)>,
+        mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
    ) {
        // Filter out any layers which were not created by this tenant shard.  These are
        // layers that originate from some ancestor shard after a split, and may still
@@ -872,7 +914,7 @@ impl RemoteTimelineClient {
            self.schedule_layer_file_upload0(upload_queue, layer.clone());
        }

-        let names = compacted_from.iter().map(|x| x.layer_desc().filename());
+        let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());

        self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
        self.launch_queued_tasks(upload_queue);
@@ -882,12 +924,18 @@ impl RemoteTimelineClient {

    /// Wait for all previously scheduled uploads/deletions to complete
    pub(crate) async fn wait_completion(self: &Arc<Self>) -> anyhow::Result<()> {
-        let mut receiver = {
+        let receiver = {
            let mut guard = self.upload_queue.lock().unwrap();
            let upload_queue = guard.initialized_mut()?;
            self.schedule_barrier0(upload_queue)
        };

+        Self::wait_completion0(receiver).await
+    }
+
+    async fn wait_completion0(
+        mut receiver: tokio::sync::watch::Receiver<()>,
+    ) -> anyhow::Result<()> {
        if receiver.changed().await.is_err() {
            anyhow::bail!("wait_completion aborted because upload queue was stopped");
        }
@@ -1003,8 +1051,7 @@ impl RemoteTimelineClient {
            let deleted_at = Utc::now().naive_utc();
            stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);

-            let mut index_part = IndexPart::try_from(&stopped.upload_queue_for_deletion)
-                .context("IndexPart serialize")?;
+            let mut index_part = IndexPart::from(&stopped.upload_queue_for_deletion);
            index_part.deleted_at = Some(deleted_at);
            index_part
        };
@@ -1085,6 +1132,93 @@ impl RemoteTimelineClient {
        Ok(())
    }

+    /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
+    ///
+    /// This is not normally needed.
+    pub(crate) async fn upload_layer_file(
+        self: &Arc<Self>,
+        uploaded: &ResidentLayer,
+        cancel: &CancellationToken,
+    ) -> anyhow::Result<()> {
+        let remote_path = remote_layer_path(
+            &self.tenant_shard_id.tenant_id,
+            &self.timeline_id,
+            self.tenant_shard_id.to_index(),
+            &uploaded.layer_desc().layer_name(),
+            uploaded.metadata().generation,
+        );
+
+        backoff::retry(
+            || async {
+                upload::upload_timeline_layer(
+                    &self.storage_impl,
+                    uploaded.local_path(),
+                    &remote_path,
+                    uploaded.metadata().file_size(),
+                    cancel,
+                )
+                .await
+            },
+            TimeoutOrCancel::caused_by_cancel,
+            FAILED_UPLOAD_WARN_THRESHOLD,
+            FAILED_REMOTE_OP_RETRIES,
+            "upload a layer without adding it to latest files",
+            cancel,
+        )
+        .await
+        .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
+        .and_then(|x| x)
+        .context("upload a layer without adding it to latest files")
+    }
+
+    /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
+    /// not added to be part of a future `index_part.json` upload.
+    pub(crate) async fn copy_timeline_layer(
+        self: &Arc<Self>,
+        adopted: &Layer,
+        adopted_as: &Layer,
+        cancel: &CancellationToken,
+    ) -> anyhow::Result<()> {
+        let source_remote_path = remote_layer_path(
+            &self.tenant_shard_id.tenant_id,
+            &adopted
+                .get_timeline_id()
+                .expect("Source timeline should be alive"),
+            self.tenant_shard_id.to_index(),
+            &adopted.layer_desc().layer_name(),
+            adopted.metadata().generation,
+        );
+
+        let target_remote_path = remote_layer_path(
+            &self.tenant_shard_id.tenant_id,
+            &self.timeline_id,
+            self.tenant_shard_id.to_index(),
+            &adopted_as.layer_desc().layer_name(),
+            adopted_as.metadata().generation,
+        );
+
+        backoff::retry(
+            || async {
+                upload::copy_timeline_layer(
+                    &self.storage_impl,
+                    &source_remote_path,
+                    &target_remote_path,
+                    cancel,
+                )
+                .await
+            },
+            TimeoutOrCancel::caused_by_cancel,
+            FAILED_UPLOAD_WARN_THRESHOLD,
+            FAILED_REMOTE_OP_RETRIES,
+            "copy timeline layer",
+            cancel,
+        )
+        .await
+        .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
+        .and_then(|x| x)
+        .context("remote copy timeline layer")
+    }
+
    async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
        match tokio::time::timeout(
            DELETION_QUEUE_FLUSH_TIMEOUT,
@@ -1256,7 +1390,7 @@ impl RemoteTimelineClient {
        while let Some(next_op) = upload_queue.queued_operations.front() {
            // Can we run this task now?
            let can_run_now = match next_op {
-                UploadOp::UploadLayer(_, _) => {
+                UploadOp::UploadLayer(..) => {
                    // Can always be scheduled.
                    true
                }
@@ -1383,13 +1517,25 @@ impl RemoteTimelineClient {

            let upload_result: anyhow::Result<()> = match &task.op {
                UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
-                    let path = layer.local_path();
+                    let local_path = layer.local_path();
+
+                    // We should only be uploading layers created by this `Tenant`'s lifetime, so
+                    // the metadata in the upload should always match our current generation.
+                    assert_eq!(layer_metadata.generation, self.generation);
+
+                    let remote_path = remote_layer_path(
+                        &self.tenant_shard_id.tenant_id,
+                        &self.timeline_id,
+                        layer_metadata.shard,
+                        &layer.layer_desc().layer_name(),
+                        layer_metadata.generation,
+                    );
+
                    upload::upload_timeline_layer(
-                        self.conf,
                        &self.storage_impl,
-                        path,
-                        layer_metadata,
-                        self.generation,
+                        local_path,
+                        &remote_path,
+                        layer_metadata.file_size(),
                        &self.cancel,
                    )
                    .measure_remote_op(
@@ -1750,14 +1896,14 @@ pub fn remote_layer_path(
    tenant_id: &TenantId,
    timeline_id: &TimelineId,
    shard: ShardIndex,
-    layer_file_name: &LayerFileName,
+    layer_file_name: &LayerName,
    generation: Generation,
 ) -> RemotePath {
    // Generation-aware key format
    let path = format!(
        "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
        shard.get_suffix(),
-        layer_file_name.file_name(),
+        layer_file_name,
        generation.get_suffix()
    );

@@ -1818,29 +1964,6 @@ pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
    }
 }

-/// Files on the remote storage are stored with paths, relative to the workdir.
-/// That path includes in itself both tenant and timeline ids, allowing to have a unique remote storage path.
-///
-/// Errors if the path provided does not start from pageserver's workdir.
-pub fn remote_path(
-    conf: &PageServerConf,
-    local_path: &Utf8Path,
-    generation: Generation,
-) -> anyhow::Result<RemotePath> {
-    let stripped = local_path
-        .strip_prefix(&conf.workdir)
-        .context("Failed to strip workdir prefix")?;
-
-    let suffixed = format!("{0}{1}", stripped, generation.get_suffix());
-
-    RemotePath::new(Utf8Path::new(&suffixed)).with_context(|| {
-        format!(
-            "to resolve remote part of path {:?} for base {:?}",
-            local_path, conf.workdir
-        )
-    })
-}
-
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -1848,6 +1971,7 @@ mod tests {
        context::RequestContext,
        tenant::{
            harness::{TenantHarness, TIMELINE_ID},
+            storage_layer::layer::local_layer_path,
            Tenant, Timeline,
        },
        DEFAULT_PG_VERSION,
@@ -1876,8 +2000,8 @@ mod tests {
        TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
    }

-    fn assert_file_list(a: &HashSet<LayerFileName>, b: &[&str]) {
-        let mut avec: Vec<String> = a.iter().map(|x| x.file_name()).collect();
+    fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
+        let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
        avec.sort();

        let mut bvec = b.to_vec();
@@ -2003,7 +2127,7 @@ mod tests {
            .layer_metadata
            .keys()
            .map(|f| f.to_owned())
-            .collect::<HashSet<LayerFileName>>();
+            .collect::<HashSet<LayerName>>();
        let initial_layer = {
            assert!(initial_layers.len() == 1);
            initial_layers.into_iter().next().unwrap()
@@ -2029,12 +2153,21 @@ mod tests {
            ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
        ]
        .into_iter()
-        .map(|(name, contents): (LayerFileName, Vec<u8>)| {
-            std::fs::write(timeline_path.join(name.file_name()), &contents).unwrap();
+        .map(|(name, contents): (LayerName, Vec<u8>)| {
+
+            let local_path = local_layer_path(
+                harness.conf,
+                &timeline.tenant_shard_id,
+                &timeline.timeline_id,
+                &name,
+                &generation,
+            );
+            std::fs::write(&local_path, &contents).unwrap();

            Layer::for_resident(
                harness.conf,
                &timeline,
+                local_path,
                name,
                LayerFileMetadata::new(contents.len() as u64, generation, shard),
            )
@@ -2101,9 +2234,9 @@ mod tests {
                .map(|f| f.to_owned())
                .collect(),
            &[
-                &initial_layer.file_name(),
-                &layers[0].layer_desc().filename().file_name(),
-                &layers[1].layer_desc().filename().file_name(),
+                &initial_layer.to_string(),
+                &layers[0].layer_desc().layer_name().to_string(),
+                &layers[1].layer_desc().layer_name().to_string(),
            ],
        );
        assert_eq!(index_part.metadata, metadata);
@@ -2117,7 +2250,7 @@ mod tests {
        // keep using schedule_layer_file_deletion because we don't have a way to wait for the
        // spawn_blocking started by the drop.
        client
-            .schedule_layer_file_deletion(&[layers[0].layer_desc().filename()])
+            .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
            .unwrap();
        {
            let mut guard = client.upload_queue.lock().unwrap();
@@ -2135,9 +2268,9 @@ mod tests {
        }
        assert_remote_files(
            &[
-                &initial_layer.file_name(),
-                &layers[0].layer_desc().filename().file_name(),
-                &layers[1].layer_desc().filename().file_name(),
+                &initial_layer.to_string(),
+                &layers[0].layer_desc().layer_name().to_string(),
+                &layers[1].layer_desc().layer_name().to_string(),
                "index_part.json",
            ],
            &remote_timeline_dir,
@@ -2150,9 +2283,9 @@ mod tests {

        assert_remote_files(
            &[
-                &initial_layer.file_name(),
-                &layers[1].layer_desc().filename().file_name(),
-                &layers[2].layer_desc().filename().file_name(),
+                &initial_layer.to_string(),
+                &layers[1].layer_desc().layer_name().to_string(),
+                &layers[2].layer_desc().layer_name().to_string(),
                "index_part.json",
            ],
            &remote_timeline_dir,
@@ -2171,19 +2304,22 @@ mod tests {
            ..
        } = TestSetup::new("metrics").await.unwrap();
        let client = timeline.remote_client.as_ref().unwrap();
-        let timeline_path = harness.timeline_path(&TIMELINE_ID);

-        let layer_file_name_1: LayerFileName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
+        let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
+        let local_path = local_layer_path(
+            harness.conf,
+            &timeline.tenant_shard_id,
+            &timeline.timeline_id,
+            &layer_file_name_1,
+            &harness.generation,
+        );
        let content_1 = dummy_contents("foo");
-        std::fs::write(
-            timeline_path.join(layer_file_name_1.file_name()),
-            &content_1,
-        )
-        .unwrap();
+        std::fs::write(&local_path, &content_1).unwrap();

        let layer_file_1 = Layer::for_resident(
            harness.conf,
            &timeline,
+            local_path,
            layer_file_name_1.clone(),
            LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
        );
@@ -2252,12 +2388,7 @@ mod tests {

    async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
        // An empty IndexPart, just sufficient to ensure deserialization will succeed
-        let example_metadata = TimelineMetadata::example();
-        let example_index_part = IndexPart::new(
-            HashMap::new(),
-            example_metadata.disk_consistent_lsn(),
-            example_metadata,
-        );
+        let example_index_part = IndexPart::example();

        let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();

--- a/pageserver/src/tenant/remote_timeline_client/download.rs
+++ b/pageserver/src/tenant/remote_timeline_client/download.rs
@@ -21,7 +21,8 @@ use crate::config::PageServerConf;
 use crate::context::RequestContext;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::tenant::remote_timeline_client::{remote_layer_path, remote_timelines_path};
-use crate::tenant::storage_layer::LayerFileName;
+use crate::tenant::storage_layer::layer::local_layer_path;
+use crate::tenant::storage_layer::LayerName;
 use crate::tenant::Generation;
 use crate::virtual_file::{on_fatal_io_error, MaybeFatalIo, VirtualFile};
 use crate::TEMP_FILE_SUFFIX;
@@ -47,7 +48,7 @@ pub async fn download_layer_file<'a>(
    storage: &'a GenericRemoteStorage,
    tenant_shard_id: TenantShardId,
    timeline_id: TimelineId,
-    layer_file_name: &'a LayerFileName,
+    layer_file_name: &'a LayerName,
    layer_metadata: &'a LayerFileMetadata,
    cancel: &CancellationToken,
    ctx: &RequestContext,
@@ -55,7 +56,13 @@ pub async fn download_layer_file<'a>(
    debug_assert_current_span_has_tenant_and_timeline_id();

    let timeline_path = conf.timeline_path(&tenant_shard_id, &timeline_id);
-    let local_path = timeline_path.join(layer_file_name.file_name());
+    let local_path = local_layer_path(
+        conf,
+        &tenant_shard_id,
+        &timeline_id,
+        layer_file_name,
+        &layer_metadata.generation,
+    );

    let remote_path = remote_layer_path(
        &tenant_shard_id.tenant_id,
--- a/pageserver/src/tenant/remote_timeline_client/index.rs
+++ b/pageserver/src/tenant/remote_timeline_client/index.rs
@@ -6,10 +6,9 @@ use std::collections::HashMap;

 use chrono::NaiveDateTime;
 use serde::{Deserialize, Serialize};
-use utils::bin_ser::SerializeError;

 use crate::tenant::metadata::TimelineMetadata;
-use crate::tenant::storage_layer::LayerFileName;
+use crate::tenant::storage_layer::LayerName;
 use crate::tenant::upload_queue::UploadQueueInitialized;
 use crate::tenant::Generation;
 use pageserver_api::shard::ShardIndex;
@@ -76,7 +75,7 @@ pub struct IndexPart {
    ///
    /// Older versions of `IndexPart` will not have this property or have only a part of metadata
    /// that latest version stores.
-    pub layer_metadata: HashMap<LayerFileName, IndexLayerMetadata>,
+    pub layer_metadata: HashMap<LayerName, IndexLayerMetadata>,

    // 'disk_consistent_lsn' is a copy of the 'disk_consistent_lsn' in the metadata.
    // It's duplicated for convenience when reading the serialized structure, but is
@@ -104,15 +103,14 @@ impl IndexPart {

    pub const FILE_NAME: &'static str = "index_part.json";

-    pub fn new(
-        layers_and_metadata: HashMap<LayerFileName, LayerFileMetadata>,
+    fn new(
+        layers_and_metadata: &HashMap<LayerName, LayerFileMetadata>,
        disk_consistent_lsn: Lsn,
        metadata: TimelineMetadata,
    ) -> Self {
-        // Transform LayerFileMetadata into IndexLayerMetadata
        let layer_metadata = layers_and_metadata
-            .into_iter()
-            .map(|(k, v)| (k, IndexLayerMetadata::from(v)))
+            .iter()
+            .map(|(k, v)| (k.to_owned(), IndexLayerMetadata::from(v)))
            .collect();

        Self {
@@ -141,20 +139,24 @@ impl IndexPart {
    pub fn to_s3_bytes(&self) -> serde_json::Result<Vec<u8>> {
        serde_json::to_vec(self)
    }
+
+    #[cfg(test)]
+    pub(crate) fn example() -> Self {
+        let example_metadata = TimelineMetadata::example();
+        Self::new(
+            &HashMap::new(),
+            example_metadata.disk_consistent_lsn(),
+            example_metadata,
+        )
+    }
 }

-impl TryFrom<&UploadQueueInitialized> for IndexPart {
-    type Error = SerializeError;
+impl From<&UploadQueueInitialized> for IndexPart {
+    fn from(uq: &UploadQueueInitialized) -> Self {
+        let disk_consistent_lsn = uq.latest_metadata.disk_consistent_lsn();
+        let metadata = uq.latest_metadata.clone();

-    fn try_from(upload_queue: &UploadQueueInitialized) -> Result<Self, Self::Error> {
-        let disk_consistent_lsn = upload_queue.latest_metadata.disk_consistent_lsn();
-        let metadata = upload_queue.latest_metadata.clone();
-
-        Ok(Self::new(
-            upload_queue.latest_files.clone(),
-            disk_consistent_lsn,
-            metadata,
-        ))
+        Self::new(&uq.latest_files, disk_consistent_lsn, metadata)
    }
 }

@@ -172,8 +174,8 @@ pub struct IndexLayerMetadata {
    pub shard: ShardIndex,
 }

-impl From<LayerFileMetadata> for IndexLayerMetadata {
-    fn from(other: LayerFileMetadata) -> Self {
+impl From<&LayerFileMetadata> for IndexLayerMetadata {
+    fn from(other: &LayerFileMetadata) -> Self {
        IndexLayerMetadata {
            file_size: other.file_size,
            generation: other.generation,
--- a/pageserver/src/tenant/remote_timeline_client/upload.rs
+++ b/pageserver/src/tenant/remote_timeline_client/upload.rs
@@ -12,18 +12,13 @@ use tokio_util::sync::CancellationToken;
 use utils::backoff;

 use super::Generation;
-use crate::{
-    config::PageServerConf,
-    tenant::remote_timeline_client::{
-        index::IndexPart, remote_index_path, remote_initdb_archive_path,
-        remote_initdb_preserved_archive_path, remote_path,
-    },
+use crate::tenant::remote_timeline_client::{
+    index::IndexPart, remote_index_path, remote_initdb_archive_path,
+    remote_initdb_preserved_archive_path,
 };
-use remote_storage::{GenericRemoteStorage, TimeTravelError};
+use remote_storage::{GenericRemoteStorage, RemotePath, TimeTravelError};
 use utils::id::{TenantId, TimelineId};

-use super::index::LayerFileMetadata;
-
 use tracing::info;

 /// Serializes and uploads the given index part data to the remote storage.
@@ -65,11 +60,10 @@ pub(crate) async fn upload_index_part<'a>(
 ///
 /// On an error, bumps the retries count and reschedules the entire task.
 pub(super) async fn upload_timeline_layer<'a>(
-    conf: &'static PageServerConf,
    storage: &'a GenericRemoteStorage,
-    source_path: &'a Utf8Path,
-    known_metadata: &'a LayerFileMetadata,
-    generation: Generation,
+    local_path: &'a Utf8Path,
+    remote_path: &'a RemotePath,
+    metadata_size: u64,
    cancel: &CancellationToken,
 ) -> anyhow::Result<()> {
    fail_point!("before-upload-layer", |_| {
@@ -78,8 +72,7 @@ pub(super) async fn upload_timeline_layer<'a>(

    pausable_failpoint!("before-upload-layer-pausable");

-    let storage_path = remote_path(conf, source_path, generation)?;
-    let source_file_res = fs::File::open(&source_path).await;
+    let source_file_res = fs::File::open(&local_path).await;
    let source_file = match source_file_res {
        Ok(source_file) => source_file,
        Err(e) if e.kind() == ErrorKind::NotFound => {
@@ -90,34 +83,49 @@ pub(super) async fn upload_timeline_layer<'a>(
            // it has been written to disk yet.
            //
            // This is tested against `test_compaction_delete_before_upload`
-            info!(path = %source_path, "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
+            info!(path = %local_path, "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
            return Ok(());
        }
-        Err(e) => {
-            Err(e).with_context(|| format!("open a source file for layer {source_path:?}"))?
-        }
+        Err(e) => Err(e).with_context(|| format!("open a source file for layer {local_path:?}"))?,
    };

    let fs_size = source_file
        .metadata()
        .await
-        .with_context(|| format!("get the source file metadata for layer {source_path:?}"))?
+        .with_context(|| format!("get the source file metadata for layer {local_path:?}"))?
        .len();

-    let metadata_size = known_metadata.file_size();
    if metadata_size != fs_size {
-        bail!("File {source_path:?} has its current FS size {fs_size} diferent from initially determined {metadata_size}");
+        bail!("File {local_path:?} has its current FS size {fs_size} diferent from initially determined {metadata_size}");
    }

    let fs_size = usize::try_from(fs_size)
-        .with_context(|| format!("convert {source_path:?} size {fs_size} usize"))?;
+        .with_context(|| format!("convert {local_path:?} size {fs_size} usize"))?;

    let reader = tokio_util::io::ReaderStream::with_capacity(source_file, super::BUFFER_SIZE);

    storage
-        .upload(reader, fs_size, &storage_path, None, cancel)
+        .upload(reader, fs_size, remote_path, None, cancel)
        .await
-        .with_context(|| format!("upload layer from local path '{source_path}'"))
+        .with_context(|| format!("upload layer from local path '{local_path}'"))
+}
+
+pub(super) async fn copy_timeline_layer(
+    storage: &GenericRemoteStorage,
+    source_path: &RemotePath,
+    target_path: &RemotePath,
+    cancel: &CancellationToken,
+) -> anyhow::Result<()> {
+    fail_point!("before-copy-layer", |_| {
+        bail!("failpoint before-copy-layer")
+    });
+
+    pausable_failpoint!("before-copy-layer-pausable");
+
+    storage
+        .copy_object(source_path, target_path, cancel)
+        .await
+        .with_context(|| format!("copy layer {source_path} to {target_path}"))
 }

 /// Uploads the given `initdb` data to the remote storage.
--- a/pageserver/src/tenant/secondary.rs
+++ b/pageserver/src/tenant/secondary.rs
@@ -21,8 +21,9 @@ use self::{
 use super::{
    config::{SecondaryLocationConfig, TenantConfOpt},
    mgr::TenantManager,
+    remote_timeline_client::LayerFileMetadata,
    span::debug_assert_current_span_has_tenant_id,
-    storage_layer::LayerFileName,
+    storage_layer::{layer::local_layer_path, LayerName},
 };

 use pageserver_api::{
@@ -181,7 +182,8 @@ impl SecondaryTenant {
        self: &Arc<Self>,
        conf: &PageServerConf,
        timeline_id: TimelineId,
-        name: LayerFileName,
+        name: LayerName,
+        metadata: LayerFileMetadata,
    ) {
        debug_assert_current_span_has_tenant_id();

@@ -195,9 +197,13 @@ impl SecondaryTenant {

        let now = SystemTime::now();

-        let path = conf
-            .timeline_path(&self.tenant_shard_id, &timeline_id)
-            .join(name.file_name());
+        let local_path = local_layer_path(
+            conf,
+            &self.tenant_shard_id,
+            &timeline_id,
+            &name,
+            &metadata.generation,
+        );

        let this = self.clone();

@@ -208,7 +214,7 @@ impl SecondaryTenant {
            // it, the secondary downloader could have seen an updated heatmap that
            // resulted in a layer being deleted.
            // Other local I/O errors are process-fatal: these should never happen.
-            let deleted = std::fs::remove_file(path);
+            let deleted = std::fs::remove_file(local_path);

            let not_found = deleted
                .as_ref()
--- a/pageserver/src/tenant/secondary/downloader.rs
+++ b/pageserver/src/tenant/secondary/downloader.rs
@@ -22,7 +22,7 @@ use crate::{
            FAILED_REMOTE_OP_RETRIES,
        },
        span::debug_assert_current_span_has_tenant_id,
-        storage_layer::LayerFileName,
+        storage_layer::{layer::local_layer_path, LayerName},
        tasks::{warn_when_period_overrun, BackgroundLoopKind},
    },
    virtual_file::{on_fatal_io_error, MaybeFatalIo, VirtualFile},
@@ -111,7 +111,7 @@ impl OnDiskState {
        _conf: &'static PageServerConf,
        _tenant_shard_id: &TenantShardId,
        _imeline_id: &TimelineId,
-        _ame: LayerFileName,
+        _ame: LayerName,
        metadata: LayerFileMetadata,
        access_time: SystemTime,
    ) -> Self {
@@ -124,10 +124,10 @@ impl OnDiskState {

 #[derive(Debug, Clone, Default)]
 pub(super) struct SecondaryDetailTimeline {
-    pub(super) on_disk_layers: HashMap<LayerFileName, OnDiskState>,
+    pub(super) on_disk_layers: HashMap<LayerName, OnDiskState>,

    /// We remember when layers were evicted, to prevent re-downloading them.
-    pub(super) evicted_at: HashMap<LayerFileName, SystemTime>,
+    pub(super) evicted_at: HashMap<LayerName, SystemTime>,
 }

 /// This state is written by the secondary downloader, it is opaque
@@ -621,12 +621,12 @@ impl<'a> TenantDownloader<'a> {
                let layers_in_heatmap = heatmap_timeline
                    .layers
                    .iter()
-                    .map(|l| &l.name)
+                    .map(|l| (&l.name, l.metadata.generation))
                    .collect::<HashSet<_>>();
                let layers_on_disk = timeline_state
                    .on_disk_layers
                    .iter()
-                    .map(|l| l.0)
+                    .map(|l| (l.0, l.1.metadata.generation))
                    .collect::<HashSet<_>>();

                let mut layer_count = layers_on_disk.len();
@@ -637,16 +637,24 @@ impl<'a> TenantDownloader<'a> {
                    .sum();

                // Remove on-disk layers that are no longer present in heatmap
-                for layer in layers_on_disk.difference(&layers_in_heatmap) {
+                for (layer_file_name, generation) in layers_on_disk.difference(&layers_in_heatmap) {
                    layer_count -= 1;
                    layer_byte_count -= timeline_state
                        .on_disk_layers
-                        .get(layer)
+                        .get(layer_file_name)
                        .unwrap()
                        .metadata
                        .file_size();

-                    delete_layers.push((*timeline_id, (*layer).clone()));
+                    let local_path = local_layer_path(
+                        self.conf,
+                        self.secondary_state.get_tenant_shard_id(),
+                        timeline_id,
+                        layer_file_name,
+                        generation,
+                    );
+
+                    delete_layers.push((*timeline_id, (*layer_file_name).clone(), local_path));
                }

                progress.bytes_downloaded += layer_byte_count;
@@ -661,11 +669,7 @@ impl<'a> TenantDownloader<'a> {
        }

        // Execute accumulated deletions
-        for (timeline_id, layer_name) in delete_layers {
-            let timeline_path = self
-                .conf
-                .timeline_path(self.secondary_state.get_tenant_shard_id(), &timeline_id);
-            let local_path = timeline_path.join(layer_name.to_string());
+        for (timeline_id, layer_name, local_path) in delete_layers {
            tracing::info!(timeline_id=%timeline_id, "Removing secondary local layer {layer_name} because it's absent in heatmap",);

            tokio::fs::remove_file(&local_path)
@@ -754,9 +758,6 @@ impl<'a> TenantDownloader<'a> {
    ) -> Result<(), UpdateError> {
        debug_assert_current_span_has_tenant_and_timeline_id();
        let tenant_shard_id = self.secondary_state.get_tenant_shard_id();
-        let timeline_path = self
-            .conf
-            .timeline_path(tenant_shard_id, &timeline.timeline_id);

        // Accumulate updates to the state
        let mut touched = Vec::new();
@@ -806,10 +807,14 @@ impl<'a> TenantDownloader<'a> {
                if cfg!(debug_assertions) {
                    // Debug for https://github.com/neondatabase/neon/issues/6966: check that the files we think
                    // are already present on disk are really there.
-                    let local_path = self
-                        .conf
-                        .timeline_path(tenant_shard_id, &timeline.timeline_id)
-                        .join(layer.name.file_name());
+                    let local_path = local_layer_path(
+                        self.conf,
+                        tenant_shard_id,
+                        &timeline.timeline_id,
+                        &layer.name,
+                        &layer.metadata.generation,
+                    );
+
                    match tokio::fs::metadata(&local_path).await {
                        Ok(meta) => {
                            tracing::debug!(
@@ -903,7 +908,13 @@ impl<'a> TenantDownloader<'a> {
            };

            if downloaded_bytes != layer.metadata.file_size {
-                let local_path = timeline_path.join(layer.name.to_string());
+                let local_path = local_layer_path(
+                    self.conf,
+                    tenant_shard_id,
+                    &timeline.timeline_id,
+                    &layer.name,
+                    &layer.metadata.generation,
+                );

                tracing::warn!(
                    "Downloaded layer {} with unexpected size {} != {}.  Removing download.",
@@ -986,7 +997,7 @@ async fn init_timeline_state(

    // As we iterate through layers found on disk, we will look up their metadata from this map.
    // Layers not present in metadata will be discarded.
-    let heatmap_metadata: HashMap<&LayerFileName, &HeatMapLayer> =
+    let heatmap_metadata: HashMap<&LayerName, &HeatMapLayer> =
        heatmap.layers.iter().map(|l| (&l.name, l)).collect();

    while let Some(dentry) = dir
@@ -1023,7 +1034,7 @@ async fn init_timeline_state(
            continue;
        }

-        match LayerFileName::from_str(file_name) {
+        match LayerName::from_str(file_name) {
            Ok(name) => {
                let remote_meta = heatmap_metadata.get(&name);
                match remote_meta {
--- a/pageserver/src/tenant/secondary/heatmap.rs
+++ b/pageserver/src/tenant/secondary/heatmap.rs
@@ -1,8 +1,6 @@
 use std::time::SystemTime;

-use crate::tenant::{
-    remote_timeline_client::index::IndexLayerMetadata, storage_layer::LayerFileName,
-};
+use crate::tenant::{remote_timeline_client::index::IndexLayerMetadata, storage_layer::LayerName};

 use serde::{Deserialize, Serialize};
 use serde_with::{serde_as, DisplayFromStr, TimestampSeconds};
@@ -31,7 +29,7 @@ pub(crate) struct HeatMapTimeline {
 #[serde_as]
 #[derive(Serialize, Deserialize)]
 pub(crate) struct HeatMapLayer {
-    pub(super) name: LayerFileName,
+    pub(super) name: LayerName,
    pub(super) metadata: IndexLayerMetadata,

    #[serde_as(as = "TimestampSeconds<i64>")]
@@ -42,7 +40,7 @@ pub(crate) struct HeatMapLayer {

 impl HeatMapLayer {
    pub(crate) fn new(
-        name: LayerFileName,
+        name: LayerName,
        metadata: IndexLayerMetadata,
        access_time: SystemTime,
    ) -> Self {
--- a/pageserver/src/tenant/storage_layer.rs
+++ b/pageserver/src/tenant/storage_layer.rs
@@ -1,11 +1,11 @@
 //! Common traits and structs for layers

 pub mod delta_layer;
-mod filename;
 pub mod image_layer;
 pub(crate) mod inmemory_layer;
 pub(crate) mod layer;
 mod layer_desc;
+mod layer_name;

 use crate::context::{AccessStatsBehavior, RequestContext};
 use crate::repository::Value;
@@ -34,10 +34,10 @@ use utils::rate_limit::RateLimit;
 use utils::{id::TimelineId, lsn::Lsn};

 pub use delta_layer::{DeltaLayer, DeltaLayerWriter, ValueRef};
-pub use filename::{DeltaFileName, ImageFileName, LayerFileName};
 pub use image_layer::{ImageLayer, ImageLayerWriter};
 pub use inmemory_layer::InMemoryLayer;
 pub use layer_desc::{PersistentLayerDesc, PersistentLayerKey};
+pub use layer_name::{DeltaLayerName, ImageLayerName, LayerName};

 pub(crate) use layer::{EvictionError, Layer, ResidentLayer};

@@ -646,8 +646,8 @@ pub mod tests {

    use super::*;

-    impl From<DeltaFileName> for PersistentLayerDesc {
-        fn from(value: DeltaFileName) -> Self {
+    impl From<DeltaLayerName> for PersistentLayerDesc {
+        fn from(value: DeltaLayerName) -> Self {
            PersistentLayerDesc::new_delta(
                TenantShardId::from([0; 18]),
                TimelineId::from_array([0; 16]),
@@ -658,8 +658,8 @@ pub mod tests {
        }
    }

-    impl From<ImageFileName> for PersistentLayerDesc {
-        fn from(value: ImageFileName) -> Self {
+    impl From<ImageLayerName> for PersistentLayerDesc {
+        fn from(value: ImageLayerName) -> Self {
            PersistentLayerDesc::new_img(
                TenantShardId::from([0; 18]),
                TimelineId::from_array([0; 16]),
@@ -670,11 +670,11 @@ pub mod tests {
        }
    }

-    impl From<LayerFileName> for PersistentLayerDesc {
-        fn from(value: LayerFileName) -> Self {
+    impl From<LayerName> for PersistentLayerDesc {
+        fn from(value: LayerName) -> Self {
            match value {
-                LayerFileName::Delta(d) => Self::from(d),
-                LayerFileName::Image(i) => Self::from(i),
+                LayerName::Delta(d) => Self::from(d),
+                LayerName::Image(i) => Self::from(i),
            }
        }
    }
--- a/pageserver/src/tenant/storage_layer/delta_layer.rs
+++ b/pageserver/src/tenant/storage_layer/delta_layer.rs
@@ -57,6 +57,7 @@ use std::fs::File;
 use std::io::SeekFrom;
 use std::ops::Range;
 use std::os::unix::fs::FileExt;
+use std::str::FromStr;
 use std::sync::Arc;
 use tokio::sync::OnceCell;
 use tracing::*;
@@ -68,7 +69,8 @@ use utils::{
 };

 use super::{
-    AsLayerDesc, LayerAccessStats, PersistentLayerDesc, ResidentLayer, ValuesReconstructState,
+    AsLayerDesc, LayerAccessStats, LayerName, PersistentLayerDesc, ResidentLayer,
+    ValuesReconstructState,
 };

 ///
@@ -309,13 +311,13 @@ impl DeltaLayer {
            .and_then(|res| res)?;

        // not production code
-        let actual_filename = path.file_name().unwrap().to_owned();
-        let expected_filename = self.layer_desc().filename().file_name();
+        let actual_layer_name = LayerName::from_str(path.file_name().unwrap()).unwrap();
+        let expected_layer_name = self.layer_desc().layer_name();

-        if actual_filename != expected_filename {
+        if actual_layer_name != expected_layer_name {
            println!("warning: filename does not match what is expected from in-file summary");
-            println!("actual: {:?}", actual_filename);
-            println!("expected: {:?}", expected_filename);
+            println!("actual: {:?}", actual_layer_name.to_string());
+            println!("expected: {:?}", expected_layer_name.to_string());
        }

        Ok(Arc::new(loaded))
@@ -1139,15 +1141,15 @@ impl DeltaLayerInner {
        Ok(all_keys)
    }

-    /// Using the given writer, write out a truncated version, where LSNs higher than the
-    /// truncate_at are missing.
-    #[cfg(test)]
+    /// Using the given writer, write out a version which has the earlier Lsns than `until`.
+    ///
+    /// Return the amount of key value records pushed to the writer.
    pub(super) async fn copy_prefix(
        &self,
        writer: &mut DeltaLayerWriter,
-        truncate_at: Lsn,
+        until: Lsn,
        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
+    ) -> anyhow::Result<usize> {
        use crate::tenant::vectored_blob_io::{
            BlobMeta, VectoredReadBuilder, VectoredReadExtended,
        };
@@ -1211,6 +1213,8 @@ impl DeltaLayerInner {
        // FIXME: buffering of DeltaLayerWriter
        let mut per_blob_copy = Vec::new();

+        let mut records = 0;
+
        while let Some(item) = stream.try_next().await? {
            tracing::debug!(?item, "popped");
            let offset = item
@@ -1229,7 +1233,7 @@ impl DeltaLayerInner {

            prev = Option::from(item);

-            let actionable = actionable.filter(|x| x.0.lsn < truncate_at);
+            let actionable = actionable.filter(|x| x.0.lsn < until);

            let builder = if let Some((meta, offsets)) = actionable {
                // extend or create a new builder
@@ -1297,7 +1301,7 @@ impl DeltaLayerInner {
                    let will_init = crate::repository::ValueBytes::will_init(data)
                        .inspect_err(|_e| {
                            #[cfg(feature = "testing")]
-                            tracing::error!(data=?utils::Hex(data), err=?_e, "failed to parse will_init out of serialized value");
+                            tracing::error!(data=?utils::Hex(data), err=?_e, %key, %lsn, "failed to parse will_init out of serialized value");
                        })
                        .unwrap_or(false);

@@ -1314,7 +1318,10 @@ impl DeltaLayerInner {
                        )
                        .await;
                    per_blob_copy = tmp;
+
                    res?;
+
+                    records += 1;
                }

                buffer = Some(res.buf);
@@ -1326,7 +1333,7 @@ impl DeltaLayerInner {
            "with the sentinel above loop should had handled all"
        );

-        Ok(())
+        Ok(records)
    }

    pub(super) async fn dump(&self, ctx: &RequestContext) -> anyhow::Result<()> {
@@ -1399,7 +1406,6 @@ impl DeltaLayerInner {
        Ok(())
    }

-    #[cfg(test)]
    fn stream_index_forwards<'a, R>(
        &'a self,
        reader: &'a DiskBtreeReader<R, DELTA_KEY_SIZE>,
--- a/pageserver/src/tenant/storage_layer/image_layer.rs
+++ b/pageserver/src/tenant/storage_layer/image_layer.rs
@@ -54,6 +54,7 @@ use std::fs::File;
 use std::io::SeekFrom;
 use std::ops::Range;
 use std::os::unix::prelude::FileExt;
+use std::str::FromStr;
 use std::sync::Arc;
 use tokio::sync::OnceCell;
 use tokio_stream::StreamExt;
@@ -65,8 +66,10 @@ use utils::{
    lsn::Lsn,
 };

-use super::filename::ImageFileName;
-use super::{AsLayerDesc, Layer, PersistentLayerDesc, ResidentLayer, ValuesReconstructState};
+use super::layer_name::ImageLayerName;
+use super::{
+    AsLayerDesc, Layer, LayerName, PersistentLayerDesc, ResidentLayer, ValuesReconstructState,
+};

 ///
 /// Header stored in the beginning of the file
@@ -231,7 +234,7 @@ impl ImageLayer {
        conf: &PageServerConf,
        timeline_id: TimelineId,
        tenant_shard_id: TenantShardId,
-        fname: &ImageFileName,
+        fname: &ImageLayerName,
    ) -> Utf8PathBuf {
        let rand_string: String = rand::thread_rng()
            .sample_iter(&Alphanumeric)
@@ -267,13 +270,13 @@ impl ImageLayer {
            .and_then(|res| res)?;

        // not production code
-        let actual_filename = path.file_name().unwrap().to_owned();
-        let expected_filename = self.layer_desc().filename().file_name();
+        let actual_layer_name = LayerName::from_str(path.file_name().unwrap()).unwrap();
+        let expected_layer_name = self.layer_desc().layer_name();

-        if actual_filename != expected_filename {
+        if actual_layer_name != expected_layer_name {
            println!("warning: filename does not match what is expected from in-file summary");
-            println!("actual: {:?}", actual_filename);
-            println!("expected: {:?}", expected_filename);
+            println!("actual: {:?}", actual_layer_name.to_string());
+            println!("expected: {:?}", expected_layer_name.to_string());
        }

        Ok(loaded)
@@ -635,7 +638,7 @@ impl ImageLayerWriterInner {
            conf,
            timeline_id,
            tenant_shard_id,
-            &ImageFileName {
+            &ImageLayerName {
                key_range: key_range.clone(),
                lsn,
            },
--- a/pageserver/src/tenant/storage_layer/layer.rs
+++ b/pageserver/src/tenant/storage_layer/layer.rs
@@ -4,12 +4,13 @@ use pageserver_api::keyspace::KeySpace;
 use pageserver_api::models::{
    HistoricLayerInfo, LayerAccessKind, LayerResidenceEventReason, LayerResidenceStatus,
 };
-use pageserver_api::shard::ShardIndex;
+use pageserver_api::shard::{ShardIndex, TenantShardId};
 use std::ops::Range;
 use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
 use std::sync::{Arc, Weak};
 use std::time::{Duration, SystemTime};
 use tracing::Instrument;
+use utils::id::TimelineId;
 use utils::lsn::Lsn;
 use utils::sync::heavier_once_cell;

@@ -24,7 +25,7 @@ use crate::tenant::{remote_timeline_client::LayerFileMetadata, Timeline};
 use super::delta_layer::{self, DeltaEntry};
 use super::image_layer;
 use super::{
-    AsLayerDesc, LayerAccessStats, LayerAccessStatsReset, LayerFileName, PersistentLayerDesc,
+    AsLayerDesc, LayerAccessStats, LayerAccessStatsReset, LayerName, PersistentLayerDesc,
    ValueReconstructResult, ValueReconstructState, ValuesReconstructState,
 };

@@ -123,14 +124,42 @@ impl PartialEq for Layer {
    }
 }

+pub(crate) fn local_layer_path(
+    conf: &PageServerConf,
+    tenant_shard_id: &TenantShardId,
+    timeline_id: &TimelineId,
+    layer_file_name: &LayerName,
+    _generation: &Generation,
+) -> Utf8PathBuf {
+    let timeline_path = conf.timeline_path(tenant_shard_id, timeline_id);
+
+    timeline_path.join(layer_file_name.to_string())
+
+    // TODO: switch to enabling new-style layer paths after next release
+    // if generation.is_none() {
+    //     // Without a generation, we may only use legacy path style
+    //     timeline_path.join(layer_file_name.to_string())
+    // } else {
+    //     timeline_path.join(format!("{}-v1{}", layer_file_name, generation.get_suffix()))
+    // }
+}
+
 impl Layer {
    /// Creates a layer value for a file we know to not be resident.
    pub(crate) fn for_evicted(
        conf: &'static PageServerConf,
        timeline: &Arc<Timeline>,
-        file_name: LayerFileName,
+        file_name: LayerName,
        metadata: LayerFileMetadata,
    ) -> Self {
+        let local_path = local_layer_path(
+            conf,
+            &timeline.tenant_shard_id,
+            &timeline.timeline_id,
+            &file_name,
+            &metadata.generation,
+        );
+
        let desc = PersistentLayerDesc::from_filename(
            timeline.tenant_shard_id,
            timeline.timeline_id,
@@ -143,6 +172,7 @@ impl Layer {
        let owner = Layer(Arc::new(LayerInner::new(
            conf,
            timeline,
+            local_path,
            access_stats,
            desc,
            None,
@@ -159,7 +189,8 @@ impl Layer {
    pub(crate) fn for_resident(
        conf: &'static PageServerConf,
        timeline: &Arc<Timeline>,
-        file_name: LayerFileName,
+        local_path: Utf8PathBuf,
+        file_name: LayerName,
        metadata: LayerFileMetadata,
    ) -> ResidentLayer {
        let desc = PersistentLayerDesc::from_filename(
@@ -184,6 +215,7 @@ impl Layer {
            LayerInner::new(
                conf,
                timeline,
+                local_path,
                access_stats,
                desc,
                Some(inner),
@@ -225,9 +257,19 @@ impl Layer {
                LayerResidenceStatus::Resident,
                LayerResidenceEventReason::LayerCreate,
            );
+
+            let local_path = local_layer_path(
+                conf,
+                &timeline.tenant_shard_id,
+                &timeline.timeline_id,
+                &desc.layer_name(),
+                &timeline.generation,
+            );
+
            LayerInner::new(
                conf,
                timeline,
+                local_path,
                access_stats,
                desc,
                Some(inner),
@@ -410,6 +452,13 @@ impl Layer {
        self.0.metadata()
    }

+    pub(crate) fn get_timeline_id(&self) -> Option<TimelineId> {
+        self.0
+            .timeline
+            .upgrade()
+            .map(|timeline| timeline.timeline_id)
+    }
+
    /// Traditional debug dumping facility
    #[allow(unused)]
    pub(crate) async fn dump(&self, verbose: bool, ctx: &RequestContext) -> anyhow::Result<()> {
@@ -641,7 +690,7 @@ impl Drop for LayerInner {
        let span = tracing::info_span!(parent: None, "layer_delete", tenant_id = %self.layer_desc().tenant_shard_id.tenant_id, shard_id=%self.layer_desc().tenant_shard_id.shard_slug(), timeline_id = %self.layer_desc().timeline_id);

        let path = std::mem::take(&mut self.path);
-        let file_name = self.layer_desc().filename();
+        let file_name = self.layer_desc().layer_name();
        let file_size = self.layer_desc().file_size;
        let timeline = self.timeline.clone();
        let meta = self.metadata();
@@ -709,19 +758,17 @@ impl Drop for LayerInner {
 }

 impl LayerInner {
+    #[allow(clippy::too_many_arguments)]
    fn new(
        conf: &'static PageServerConf,
        timeline: &Arc<Timeline>,
+        local_path: Utf8PathBuf,
        access_stats: LayerAccessStats,
        desc: PersistentLayerDesc,
        downloaded: Option<Arc<DownloadedLayer>>,
        generation: Generation,
        shard: ShardIndex,
    ) -> Self {
-        let path = conf
-            .timeline_path(&timeline.tenant_shard_id, &timeline.timeline_id)
-            .join(desc.filename().to_string());
-
        let (inner, version, init_status) = if let Some(inner) = downloaded {
            let version = inner.version;
            let resident = ResidentOrWantedEvicted::Resident(inner);
@@ -736,8 +783,10 @@ impl LayerInner {

        LayerInner {
            conf,
-            debug_str: { format!("timelines/{}/{}", timeline.timeline_id, desc.filename()).into() },
-            path,
+            debug_str: {
+                format!("timelines/{}/{}", timeline.timeline_id, desc.layer_name()).into()
+            },
+            path: local_path,
            desc,
            timeline: Arc::downgrade(timeline),
            have_remote_client: timeline.remote_client.is_some(),
@@ -1074,7 +1123,7 @@ impl LayerInner {

        let result = client
            .download_layer_file(
-                &self.desc.filename(),
+                &self.desc.layer_name(),
                &self.metadata(),
                &timeline.cancel,
                ctx,
@@ -1211,7 +1260,7 @@ impl LayerInner {
    }

    fn info(&self, reset: LayerAccessStatsReset) -> HistoricLayerInfo {
-        let layer_file_name = self.desc.filename().file_name();
+        let layer_name = self.desc.layer_name().to_string();

        let resident = self
            .inner
@@ -1225,7 +1274,7 @@ impl LayerInner {
            let lsn_range = &self.desc.lsn_range;

            HistoricLayerInfo::Delta {
-                layer_file_name,
+                layer_file_name: layer_name,
                layer_file_size: self.desc.file_size,
                lsn_start: lsn_range.start,
                lsn_end: lsn_range.end,
@@ -1236,7 +1285,7 @@ impl LayerInner {
            let lsn = self.desc.image_layer_lsn();

            HistoricLayerInfo::Image {
-                layer_file_name,
+                layer_file_name: layer_name,
                layer_file_size: self.desc.file_size,
                lsn_start: lsn,
                remote: !resident,
@@ -1797,25 +1846,23 @@ impl ResidentLayer {
        }
    }

-    /// FIXME: truncate is bad name because we are not truncating anything, but copying the
-    /// filtered parts.
-    #[cfg(test)]
-    pub(super) async fn copy_delta_prefix(
+    /// Returns the amount of keys and values written to the writer.
+    pub(crate) async fn copy_delta_prefix(
        &self,
        writer: &mut super::delta_layer::DeltaLayerWriter,
-        truncate_at: Lsn,
+        until: Lsn,
        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
+    ) -> anyhow::Result<usize> {
        use LayerKind::*;

        let owner = &self.owner.0;

        match self.downloaded.get(owner, ctx).await? {
            Delta(ref d) => d
-                .copy_prefix(writer, truncate_at, ctx)
+                .copy_prefix(writer, until, ctx)
                .await
-                .with_context(|| format!("truncate {self}")),
-            Image(_) => anyhow::bail!(format!("cannot truncate image layer {self}")),
+                .with_context(|| format!("copy_delta_prefix until {until} of {self}")),
+            Image(_) => anyhow::bail!(format!("cannot copy_lsn_prefix of image layer {self}")),
        }
    }

--- a/pageserver/src/tenant/storage_layer/layer_desc.rs
+++ b/pageserver/src/tenant/storage_layer/layer_desc.rs
@@ -5,7 +5,7 @@ use utils::{id::TimelineId, lsn::Lsn};

 use crate::repository::Key;

-use super::{DeltaFileName, ImageFileName, LayerFileName};
+use super::{DeltaLayerName, ImageLayerName, LayerName};

 use serde::{Deserialize, Serialize};

@@ -51,7 +51,7 @@ impl PersistentLayerDesc {
    }

    pub fn short_id(&self) -> impl Display {
-        self.filename()
+        self.layer_name()
    }

    #[cfg(test)]
@@ -103,14 +103,14 @@ impl PersistentLayerDesc {
    pub fn from_filename(
        tenant_shard_id: TenantShardId,
        timeline_id: TimelineId,
-        filename: LayerFileName,
+        filename: LayerName,
        file_size: u64,
    ) -> Self {
        match filename {
-            LayerFileName::Image(i) => {
+            LayerName::Image(i) => {
                Self::new_img(tenant_shard_id, timeline_id, i.key_range, i.lsn, file_size)
            }
-            LayerFileName::Delta(d) => Self::new_delta(
+            LayerName::Delta(d) => Self::new_delta(
                tenant_shard_id,
                timeline_id,
                d.key_range,
@@ -132,34 +132,34 @@ impl PersistentLayerDesc {
        lsn..(lsn + 1)
    }

-    /// Get a delta file name for this layer.
+    /// Get a delta layer name for this layer.
    ///
    /// Panic: if this is not a delta layer.
-    pub fn delta_file_name(&self) -> DeltaFileName {
+    pub fn delta_layer_name(&self) -> DeltaLayerName {
        assert!(self.is_delta);
-        DeltaFileName {
+        DeltaLayerName {
            key_range: self.key_range.clone(),
            lsn_range: self.lsn_range.clone(),
        }
    }

-    /// Get a delta file name for this layer.
+    /// Get a image layer name for this layer.
    ///
    /// Panic: if this is not an image layer, or the lsn range is invalid
-    pub fn image_file_name(&self) -> ImageFileName {
+    pub fn image_layer_name(&self) -> ImageLayerName {
        assert!(!self.is_delta);
        assert!(self.lsn_range.start + 1 == self.lsn_range.end);
-        ImageFileName {
+        ImageLayerName {
            key_range: self.key_range.clone(),
            lsn: self.lsn_range.start,
        }
    }

-    pub fn filename(&self) -> LayerFileName {
+    pub fn layer_name(&self) -> LayerName {
        if self.is_delta {
-            self.delta_file_name().into()
+            self.delta_layer_name().into()
        } else {
-            self.image_file_name().into()
+            self.image_layer_name().into()
        }
    }

--- a/pageserver/src/tenant/storage_layer/layer_name.rs
+++ b/pageserver/src/tenant/storage_layer/layer_name.rs
@@ -2,40 +2,42 @@
 //! Helper functions for dealing with filenames of the image and delta layer files.
 //!
 use crate::repository::Key;
+use std::borrow::Cow;
 use std::cmp::Ordering;
 use std::fmt;
 use std::ops::Range;
 use std::str::FromStr;

+use regex::Regex;
 use utils::lsn::Lsn;

 use super::PersistentLayerDesc;

 // Note: Timeline::load_layer_map() relies on this sort order
 #[derive(PartialEq, Eq, Clone, Hash)]
-pub struct DeltaFileName {
+pub struct DeltaLayerName {
    pub key_range: Range<Key>,
    pub lsn_range: Range<Lsn>,
 }

-impl std::fmt::Debug for DeltaFileName {
+impl std::fmt::Debug for DeltaLayerName {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        use super::RangeDisplayDebug;

-        f.debug_struct("DeltaFileName")
+        f.debug_struct("DeltaLayerName")
            .field("key_range", &RangeDisplayDebug(&self.key_range))
            .field("lsn_range", &self.lsn_range)
            .finish()
    }
 }

-impl PartialOrd for DeltaFileName {
+impl PartialOrd for DeltaLayerName {
    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
        Some(self.cmp(other))
    }
 }

-impl Ord for DeltaFileName {
+impl Ord for DeltaLayerName {
    fn cmp(&self, other: &Self) -> Ordering {
        let mut cmp = self.key_range.start.cmp(&other.key_range.start);
        if cmp != Ordering::Equal {
@@ -55,16 +57,14 @@ impl Ord for DeltaFileName {
    }
 }

-/// Represents the filename of a DeltaLayer
+/// Represents the region of the LSN-Key space covered by a DeltaLayer
 ///
 /// ```text
 ///    <key start>-<key end>__<LSN start>-<LSN end>
 /// ```
-impl DeltaFileName {
-    ///
-    /// Parse a string as a delta file name. Returns None if the filename does not
-    /// match the expected pattern.
-    ///
+impl DeltaLayerName {
+    /// Parse the part of a delta layer's file name that represents the LayerName. Returns None
+    /// if the filename does not match the expected pattern.
    pub fn parse_str(fname: &str) -> Option<Self> {
        let mut parts = fname.split("__");
        let mut key_parts = parts.next()?.split('-');
@@ -74,10 +74,19 @@ impl DeltaFileName {
        let key_end_str = key_parts.next()?;
        let lsn_start_str = lsn_parts.next()?;
        let lsn_end_str = lsn_parts.next()?;
+
        if parts.next().is_some() || key_parts.next().is_some() || key_parts.next().is_some() {
            return None;
        }

+        if key_start_str.len() != 36
+            || key_end_str.len() != 36
+            || lsn_start_str.len() != 16
+            || lsn_end_str.len() != 16
+        {
+            return None;
+        }
+
        let key_start = Key::from_hex(key_start_str).ok()?;
        let key_end = Key::from_hex(key_end_str).ok()?;

@@ -94,14 +103,14 @@ impl DeltaFileName {
            // or panic?
        }

-        Some(DeltaFileName {
+        Some(DeltaLayerName {
            key_range: key_start..key_end,
            lsn_range: start_lsn..end_lsn,
        })
    }
 }

-impl fmt::Display for DeltaFileName {
+impl fmt::Display for DeltaLayerName {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(
            f,
@@ -115,29 +124,29 @@ impl fmt::Display for DeltaFileName {
 }

 #[derive(PartialEq, Eq, Clone, Hash)]
-pub struct ImageFileName {
+pub struct ImageLayerName {
    pub key_range: Range<Key>,
    pub lsn: Lsn,
 }

-impl std::fmt::Debug for ImageFileName {
+impl std::fmt::Debug for ImageLayerName {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        use super::RangeDisplayDebug;

-        f.debug_struct("ImageFileName")
+        f.debug_struct("ImageLayerName")
            .field("key_range", &RangeDisplayDebug(&self.key_range))
            .field("lsn", &self.lsn)
            .finish()
    }
 }

-impl PartialOrd for ImageFileName {
+impl PartialOrd for ImageLayerName {
    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
        Some(self.cmp(other))
    }
 }

-impl Ord for ImageFileName {
+impl Ord for ImageLayerName {
    fn cmp(&self, other: &Self) -> Ordering {
        let mut cmp = self.key_range.start.cmp(&other.key_range.start);
        if cmp != Ordering::Equal {
@@ -153,7 +162,7 @@ impl Ord for ImageFileName {
    }
 }

-impl ImageFileName {
+impl ImageLayerName {
    pub fn lsn_as_range(&self) -> Range<Lsn> {
        // Saves from having to copypaste this all over
        PersistentLayerDesc::image_layer_lsn_range(self.lsn)
@@ -161,16 +170,14 @@ impl ImageFileName {
 }

 ///
-/// Represents the filename of an ImageLayer
+/// Represents the part of the Key-LSN space covered by an ImageLayer
 ///
 /// ```text
 ///    <key start>-<key end>__<LSN>
 /// ```
-impl ImageFileName {
-    ///
-    /// Parse a string as an image file name. Returns None if the filename does not
-    /// match the expected pattern.
-    ///
+impl ImageLayerName {
+    /// Parse a string as then LayerName part of an image layer file name. Returns None if the
+    /// filename does not match the expected pattern.
    pub fn parse_str(fname: &str) -> Option<Self> {
        let mut parts = fname.split("__");
        let mut key_parts = parts.next()?.split('-');
@@ -182,19 +189,23 @@ impl ImageFileName {
            return None;
        }

+        if key_start_str.len() != 36 || key_end_str.len() != 36 || lsn_str.len() != 16 {
+            return None;
+        }
+
        let key_start = Key::from_hex(key_start_str).ok()?;
        let key_end = Key::from_hex(key_end_str).ok()?;

        let lsn = Lsn::from_hex(lsn_str).ok()?;

-        Some(ImageFileName {
+        Some(ImageLayerName {
            key_range: key_start..key_end,
            lsn,
        })
    }
 }

-impl fmt::Display for ImageFileName {
+impl fmt::Display for ImageLayerName {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(
            f,
@@ -205,21 +216,24 @@ impl fmt::Display for ImageFileName {
        )
    }
 }
+
+/// LayerName is the logical identity of a layer within a LayerMap at a moment in time.  The
+/// LayerName is not a unique filename, as the same LayerName may have multiple physical incarnations
+/// over time (e.g. across shard splits or compression). The physical filenames of layers in local
+/// storage and object names in remote storage consist of the LayerName plus some extra qualifiers
+/// that uniquely identify the physical incarnation of a layer (see [crate::tenant::remote_timeline_client::remote_layer_path])
+/// and [`crate::tenant::storage_layer::layer::local_layer_path`])
 #[derive(Debug, PartialEq, Eq, Hash, Clone)]
-pub enum LayerFileName {
-    Image(ImageFileName),
-    Delta(DeltaFileName),
+pub enum LayerName {
+    Image(ImageLayerName),
+    Delta(DeltaLayerName),
 }

-impl LayerFileName {
-    pub fn file_name(&self) -> String {
-        self.to_string()
-    }
-
+impl LayerName {
    /// Determines if this layer file is considered to be in future meaning we will discard these
    /// layers during timeline initialization from the given disk_consistent_lsn.
    pub(crate) fn is_in_future(&self, disk_consistent_lsn: Lsn) -> bool {
-        use LayerFileName::*;
+        use LayerName::*;
        match self {
            Image(file_name) if file_name.lsn > disk_consistent_lsn => true,
            Delta(file_name) if file_name.lsn_range.end > disk_consistent_lsn + 1 => true,
@@ -228,7 +242,7 @@ impl LayerFileName {
    }

    pub(crate) fn kind(&self) -> &'static str {
-        use LayerFileName::*;
+        use LayerName::*;
        match self {
            Delta(_) => "delta",
            Image(_) => "image",
@@ -236,7 +250,7 @@ impl LayerFileName {
    }
 }

-impl fmt::Display for LayerFileName {
+impl fmt::Display for LayerName {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::Image(fname) => write!(f, "{fname}"),
@@ -245,23 +259,36 @@ impl fmt::Display for LayerFileName {
    }
 }

-impl From<ImageFileName> for LayerFileName {
-    fn from(fname: ImageFileName) -> Self {
+impl From<ImageLayerName> for LayerName {
+    fn from(fname: ImageLayerName) -> Self {
        Self::Image(fname)
    }
 }
-impl From<DeltaFileName> for LayerFileName {
-    fn from(fname: DeltaFileName) -> Self {
+impl From<DeltaLayerName> for LayerName {
+    fn from(fname: DeltaLayerName) -> Self {
        Self::Delta(fname)
    }
 }

-impl FromStr for LayerFileName {
+impl FromStr for LayerName {
    type Err = String;

+    /// Conversion from either a physical layer filename, or the string-ization of
+    /// Self. When loading a physical layer filename, we drop any extra information
+    /// not needed to build Self.
    fn from_str(value: &str) -> Result<Self, Self::Err> {
-        let delta = DeltaFileName::parse_str(value);
-        let image = ImageFileName::parse_str(value);
+        let gen_suffix_regex = Regex::new("^(?<base>.+)(?<gen>-v1-[0-9a-f]{8})$").unwrap();
+        let file_name: Cow<str> = match gen_suffix_regex.captures(value) {
+            Some(captures) => captures
+                .name("base")
+                .expect("Non-optional group")
+                .as_str()
+                .into(),
+            None => value.into(),
+        };
+
+        let delta = DeltaLayerName::parse_str(&file_name);
+        let image = ImageLayerName::parse_str(&file_name);
        let ok = match (delta, image) {
            (None, None) => {
                return Err(format!(
@@ -276,7 +303,7 @@ impl FromStr for LayerFileName {
    }
 }

-impl serde::Serialize for LayerFileName {
+impl serde::Serialize for LayerName {
    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
    where
        S: serde::Serializer,
@@ -288,19 +315,19 @@ impl serde::Serialize for LayerFileName {
    }
 }

-impl<'de> serde::Deserialize<'de> for LayerFileName {
+impl<'de> serde::Deserialize<'de> for LayerName {
    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
    where
        D: serde::Deserializer<'de>,
    {
-        deserializer.deserialize_string(LayerFileNameVisitor)
+        deserializer.deserialize_string(LayerNameVisitor)
    }
 }

-struct LayerFileNameVisitor;
+struct LayerNameVisitor;

-impl<'de> serde::de::Visitor<'de> for LayerFileNameVisitor {
-    type Value = LayerFileName;
+impl<'de> serde::de::Visitor<'de> for LayerNameVisitor {
+    type Value = LayerName;

    fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
        write!(
@@ -315,3 +342,42 @@ impl<'de> serde::de::Visitor<'de> for LayerFileNameVisitor {
        v.parse().map_err(|e| E::custom(e))
    }
 }
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    #[test]
+    fn image_layer_parse() -> anyhow::Result<()> {
+        let expected = LayerName::Image(ImageLayerName {
+            key_range: Key::from_i128(0)
+                ..Key::from_hex("000000067F00000001000004DF0000000006").unwrap(),
+            lsn: Lsn::from_hex("00000000014FED58").unwrap(),
+        });
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-v1-00000001").map_err(|s| anyhow::anyhow!(s))?;
+        assert_eq!(parsed, expected,);
+
+        // Omitting generation suffix is valid
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58").map_err(|s| anyhow::anyhow!(s))?;
+        assert_eq!(parsed, expected,);
+
+        Ok(())
+    }
+
+    #[test]
+    fn delta_layer_parse() -> anyhow::Result<()> {
+        let expected = LayerName::Delta(DeltaLayerName {
+            key_range: Key::from_i128(0)
+                ..Key::from_hex("000000067F00000001000004DF0000000006").unwrap(),
+            lsn_range: Lsn::from_hex("00000000014FED58").unwrap()
+                ..Lsn::from_hex("000000000154C481").unwrap(),
+        });
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481-v1-00000001").map_err(|s| anyhow::anyhow!(s))?;
+        assert_eq!(parsed, expected);
+
+        // Omitting generation suffix is valid
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481").map_err(|s| anyhow::anyhow!(s))?;
+        assert_eq!(parsed, expected);
+
+        Ok(())
+    }
+}
--- a/pageserver/src/tenant/tasks.rs
+++ b/pageserver/src/tenant/tasks.rs
@@ -2,6 +2,7 @@
 //! such as compaction and GC

 use std::ops::ControlFlow;
+use std::str::FromStr;
 use std::sync::Arc;
 use std::time::{Duration, Instant};

@@ -9,9 +10,11 @@ use crate::context::{DownloadBehavior, RequestContext};
 use crate::metrics::TENANT_TASK_EVENTS;
 use crate::task_mgr;
 use crate::task_mgr::{TaskKind, BACKGROUND_RUNTIME};
+use crate::tenant::config::defaults::DEFAULT_COMPACTION_PERIOD;
 use crate::tenant::throttle::Stats;
 use crate::tenant::timeline::CompactionError;
 use crate::tenant::{Tenant, TenantState};
+use rand::Rng;
 use tokio_util::sync::CancellationToken;
 use tracing::*;
 use utils::{backoff, completion};
@@ -44,6 +47,7 @@ pub(crate) enum BackgroundLoopKind {
    Compaction,
    Gc,
    Eviction,
+    IngestHouseKeeping,
    ConsumptionMetricsCollectMetrics,
    ConsumptionMetricsSyntheticSizeWorker,
    InitialLogicalSizeCalculation,
@@ -132,6 +136,30 @@ pub fn start_background_loops(
            }
        },
    );
+
+    task_mgr::spawn(
+        BACKGROUND_RUNTIME.handle(),
+        TaskKind::IngestHousekeeping,
+        Some(tenant_shard_id),
+        None,
+        &format!("ingest housekeeping for tenant {tenant_shard_id}"),
+        false,
+        {
+            let tenant = Arc::clone(tenant);
+            let background_jobs_can_start = background_jobs_can_start.cloned();
+            async move {
+                let cancel = task_mgr::shutdown_token();
+                tokio::select! {
+                    _ = cancel.cancelled() => { return Ok(()) },
+                    _ = completion::Barrier::maybe_wait(background_jobs_can_start) => {}
+                };
+                ingest_housekeeping_loop(tenant, cancel)
+                    .instrument(info_span!("ingest_housekeeping_loop", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug()))
+                    .await;
+                Ok(())
+            }
+        },
+    );
 }

 ///
@@ -379,6 +407,61 @@ async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
    TENANT_TASK_EVENTS.with_label_values(&["stop"]).inc();
 }

+async fn ingest_housekeeping_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
+    TENANT_TASK_EVENTS.with_label_values(&["start"]).inc();
+    async {
+        loop {
+            tokio::select! {
+                _ = cancel.cancelled() => {
+                    return;
+                },
+                tenant_wait_result = wait_for_active_tenant(&tenant) => match tenant_wait_result {
+                    ControlFlow::Break(()) => return,
+                    ControlFlow::Continue(()) => (),
+                },
+            }
+
+            // We run ingest housekeeping with the same frequency as compaction: it is not worth
+            // having a distinct setting.  But we don't run it in the same task, because compaction
+            // blocks on acquiring the background job semaphore.
+            let period = tenant.get_compaction_period();
+
+            // If compaction period is set to zero (to disable it), then we will use a reasonable default
+            let period = if period == Duration::ZERO {
+                humantime::Duration::from_str(DEFAULT_COMPACTION_PERIOD)
+                    .unwrap()
+                    .into()
+            } else {
+                period
+            };
+
+            // Jitter the period by +/- 5%
+            let period =
+                rand::thread_rng().gen_range((period * (95)) / 100..(period * (105)) / 100);
+
+            // Always sleep first: we do not need to do ingest housekeeping early in the lifetime of
+            // a tenant, since it won't have started writing any ephemeral files yet.
+            if tokio::time::timeout(period, cancel.cancelled())
+                .await
+                .is_ok()
+            {
+                break;
+            }
+
+            let started_at = Instant::now();
+            tenant.ingest_housekeeping().await;
+
+            warn_when_period_overrun(
+                started_at.elapsed(),
+                period,
+                BackgroundLoopKind::IngestHouseKeeping,
+            );
+        }
+    }
+    .await;
+    TENANT_TASK_EVENTS.with_label_values(&["stop"]).inc();
+}
+
 async fn wait_for_active_tenant(tenant: &Arc<Tenant>) -> ControlFlow<()> {
    // if the tenant has a proper status already, no need to wait for anything
    if tenant.current_state() == TenantState::Active {
@@ -420,8 +503,6 @@ pub(crate) async fn random_init_delay(
    period: Duration,
    cancel: &CancellationToken,
 ) -> Result<(), Cancelled> {
-    use rand::Rng;
-
    if period == Duration::ZERO {
        return Ok(());
    }
--- a/pageserver/src/tenant/timeline.rs
+++ b/pageserver/src/tenant/timeline.rs
@@ -1,5 +1,6 @@
 mod compaction;
 pub mod delete;
+pub(crate) mod detach_ancestor;
 mod eviction_task;
 mod init;
 pub mod layer_manager;
@@ -22,8 +23,9 @@ use pageserver_api::{
    },
    keyspace::{KeySpaceAccum, SparseKeyPartitioning},
    models::{
-        CompactionAlgorithm, DownloadRemoteLayersTaskInfo, DownloadRemoteLayersTaskSpawnRequest,
-        EvictionPolicy, InMemoryLayerInfo, LayerMapInfo, TimelineState,
+        AuxFilePolicy, CompactionAlgorithm, DownloadRemoteLayersTaskInfo,
+        DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy, InMemoryLayerInfo, LayerMapInfo,
+        TimelineState,
    },
    reltag::BlockNumber,
    shard::{ShardIdentity, ShardNumber, TenantShardId},
@@ -58,6 +60,7 @@ use std::{
    ops::ControlFlow,
 };

+use crate::tenant::timeline::init::LocalLayerFileMetadata;
 use crate::tenant::{
    layer_map::{LayerMap, SearchResult},
    metadata::TimelineMetadata,
@@ -72,7 +75,7 @@ use crate::{
    disk_usage_eviction_task::finite_f32,
    tenant::storage_layer::{
        AsLayerDesc, DeltaLayerWriter, EvictionError, ImageLayerWriter, InMemoryLayer, Layer,
-        LayerAccessStatsReset, LayerFileName, ResidentLayer, ValueReconstructResult,
+        LayerAccessStatsReset, LayerName, ResidentLayer, ValueReconstructResult,
        ValueReconstructState, ValuesReconstructState,
    },
 };
@@ -862,9 +865,13 @@ impl Timeline {
                // Initialise the reconstruct state for the key with the cache
                // entry returned above.
                let mut reconstruct_state = ValuesReconstructState::new();
-                let mut key_state = VectoredValueReconstructState::default();
-                key_state.img = cached_page_img;
-                reconstruct_state.keys.insert(key, Ok(key_state));
+
+                // Only add the cached image to the reconstruct state when it exists.
+                if cached_page_img.is_some() {
+                    let mut key_state = VectoredValueReconstructState::default();
+                    key_state.img = cached_page_img;
+                    reconstruct_state.keys.insert(key, Ok(key_state));
+                }

                let vectored_res = self
                    .get_vectored_impl(keyspace.clone(), lsn, reconstruct_state, ctx)
@@ -1076,7 +1083,7 @@ impl Timeline {
        // We should generalize this into Keyspace::contains in the future.
        for range in &keyspace.ranges {
            if range.start.field1 < METADATA_KEY_BEGIN_PREFIX
-                || range.end.field1 >= METADATA_KEY_END_PREFIX
+                || range.end.field1 > METADATA_KEY_END_PREFIX
            {
                return Err(GetVectoredError::Other(anyhow::anyhow!(
                    "only metadata keyspace can be scanned"
@@ -1213,11 +1220,17 @@ impl Timeline {
        }
        reconstruct_timer.stop_and_record();

-        // Note that this is an approximation. Tracking the exact number of layers visited
-        // per key requires virtually unbounded memory usage and is inefficient
-        // (i.e. segment tree tracking each range queried from a layer)
-        crate::metrics::VEC_READ_NUM_LAYERS_VISITED
-            .observe(layers_visited as f64 / results.len() as f64);
+        // For aux file keys (v1 or v2) the vectored read path does not return an error
+        // when they're missing. Instead they are omitted from the resulting btree
+        // (this is a requirement, not a bug). Skip updating the metric in these cases
+        // to avoid infinite results.
+        if !results.is_empty() {
+            // Note that this is an approximation. Tracking the exact number of layers visited
+            // per key requires virtually unbounded memory usage and is inefficient
+            // (i.e. segment tree tracking each range queried from a layer)
+            crate::metrics::VEC_READ_NUM_LAYERS_VISITED
+                .observe(layers_visited as f64 / results.len() as f64);
+        }

        Ok(results)
    }
@@ -1494,15 +1507,21 @@ impl Timeline {
    /// Flush to disk all data that was written with the put_* functions
    #[instrument(skip(self), fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug(), timeline_id=%self.timeline_id))]
    pub(crate) async fn freeze_and_flush(&self) -> anyhow::Result<()> {
+        self.freeze_and_flush0().await
+    }
+
+    // This exists to provide a non-span creating version of `freeze_and_flush` we can call without
+    // polluting the span hierarchy.
+    pub(crate) async fn freeze_and_flush0(&self) -> anyhow::Result<()> {
        let to_lsn = self.freeze_inmem_layer(false).await;
        self.flush_frozen_layers_and_wait(to_lsn).await
    }

-    /// If there is no writer, and conditions for rolling the latest layer are met, then freeze it.
-    ///
-    /// This is for use in background housekeeping, to provide guarantees of layers closing eventually
-    /// even if there are no ongoing writes to drive that.
-    async fn maybe_freeze_ephemeral_layer(&self) {
+    // Check if an open ephemeral layer should be closed: this provides
+    // background enforcement of checkpoint interval if there is no active WAL receiver, to avoid keeping
+    // an ephemeral layer open forever when idle.  It also freezes layers if the global limit on
+    // ephemeral layer bytes has been breached.
+    pub(super) async fn maybe_freeze_ephemeral_layer(&self) {
        let Ok(_write_guard) = self.write_lock.try_lock() else {
            // If the write lock is held, there is an active wal receiver: rolling open layers
            // is their responsibility while they hold this lock.
@@ -1529,13 +1548,11 @@ impl Timeline {
                // we are a sharded tenant and have skipped some WAL
                let last_freeze_ts = *self.last_freeze_ts.read().unwrap();
                if last_freeze_ts.elapsed() >= self.get_checkpoint_timeout() {
-                    // This should be somewhat rare, so we log it at INFO level.
-                    //
-                    // We checked for checkpoint timeout so that a shard without any
-                    // data ingested (yet) doesn't write a remote index as soon as it
+                    // Only do this if have been layer-less longer than get_checkpoint_timeout, so that a shard
+                    // without any data ingested (yet) doesn't write a remote index as soon as it
                    // sees its LSN advance: we only do this if we've been layer-less
                    // for some time.
-                    tracing::info!(
+                    tracing::debug!(
                        "Advancing disk_consistent_lsn past WAL ingest gap {} -> {}",
                        disk_consistent_lsn,
                        last_record_lsn
@@ -1625,11 +1642,6 @@ impl Timeline {
            (guard, permit)
        };

-        // Prior to compaction, check if an open ephemeral layer should be closed: this provides
-        // background enforcement of checkpoint interval if there is no active WAL receiver, to avoid keeping
-        // an ephemeral layer open forever when idle.
-        self.maybe_freeze_ephemeral_layer().await;
-
        // this wait probably never needs any "long time spent" logging, because we already nag if
        // compaction task goes over it's period (20s) which is quite often in production.
        let (_guard, _permit) = tokio::select! {
@@ -1899,7 +1911,7 @@ impl Timeline {
    #[instrument(skip_all, fields(tenant_id = %self.tenant_shard_id.tenant_id, shard_id = %self.tenant_shard_id.shard_slug(), timeline_id = %self.timeline_id))]
    pub(crate) async fn download_layer(
        &self,
-        layer_file_name: &str,
+        layer_file_name: &LayerName,
    ) -> anyhow::Result<Option<bool>> {
        let Some(layer) = self.find_layer(layer_file_name).await else {
            return Ok(None);
@@ -1917,7 +1929,10 @@ impl Timeline {
    /// Evict just one layer.
    ///
    /// Returns `Ok(None)` in the case where the layer could not be found by its `layer_file_name`.
-    pub(crate) async fn evict_layer(&self, layer_file_name: &str) -> anyhow::Result<Option<bool>> {
+    pub(crate) async fn evict_layer(
+        &self,
+        layer_file_name: &LayerName,
+    ) -> anyhow::Result<Option<bool>> {
        let _gate = self
            .gate
            .enter()
@@ -1991,13 +2006,12 @@ const REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE: u64 = 10;

 // Private functions
 impl Timeline {
-    #[allow(dead_code)]
-    pub(crate) fn get_switch_to_aux_file_v2(&self) -> bool {
+    pub(crate) fn get_switch_aux_file_policy(&self) -> AuxFilePolicy {
        let tenant_conf = self.tenant_conf.load();
        tenant_conf
            .tenant_conf
-            .switch_to_aux_file_v2
-            .unwrap_or(self.conf.default_tenant_conf.switch_to_aux_file_v2)
+            .switch_aux_file_policy
+            .unwrap_or(self.conf.default_tenant_conf.switch_aux_file_policy)
    }

    pub(crate) fn get_lazy_slru_download(&self) -> bool {
@@ -2379,13 +2393,13 @@ impl Timeline {
        index_part: Option<IndexPart>,
    ) -> anyhow::Result<()> {
        use init::{Decision::*, Discovered, DismissedLayer};
-        use LayerFileName::*;
+        use LayerName::*;

        let mut guard = self.layers.write().await;

        let timer = self.metrics.load_layer_map_histo.start_timer();

-        // Scan timeline directory and create ImageFileName and DeltaFilename
+        // Scan timeline directory and create ImageLayerName and DeltaFilename
        // structs representing all files on disk
        let timeline_path = self
            .conf
@@ -2409,8 +2423,8 @@ impl Timeline {

                for discovered in discovered {
                    let (name, kind) = match discovered {
-                        Discovered::Layer(file_name, file_size) => {
-                            discovered_layers.push((file_name, file_size));
+                        Discovered::Layer(layer_file_name, local_path, file_size) => {
+                            discovered_layers.push((layer_file_name, local_path, file_size));
                            continue;
                        }
                        Discovered::Metadata => {
@@ -2460,31 +2474,30 @@ impl Timeline {
                        Ok(UseRemote { local, remote }) => {
                            // Remote is authoritative, but we may still choose to retain
                            // the local file if the contents appear to match
-                            if local.file_size() == remote.file_size() {
+                            if local.metadata.file_size() == remote.file_size() {
                                // Use the local file, but take the remote metadata so that we pick up
                                // the correct generation.
-                                UseLocal(remote)
+                                UseLocal(
+                                    LocalLayerFileMetadata {
+                                        metadata: remote,
+                                        local_path: local.local_path
+                                    }
+                                )
                            } else {
-                                path.push(name.file_name());
-                                init::cleanup_local_file_for_remote(&path, &local, &remote)?;
-                                path.pop();
+                                init::cleanup_local_file_for_remote(&local, &remote)?;
                                UseRemote { local, remote }
                            }
                        }
                        Ok(decision) => decision,
                        Err(DismissedLayer::Future { local }) => {
-                            if local.is_some() {
-                                path.push(name.file_name());
-                                init::cleanup_future_layer(&path, &name, disk_consistent_lsn)?;
-                                path.pop();
+                            if let Some(local) = local {
+                                init::cleanup_future_layer(&local.local_path, &name, disk_consistent_lsn)?;
                            }
                            needs_cleanup.push(name);
                            continue;
                        }
                        Err(DismissedLayer::LocalOnly(local)) => {
-                            path.push(name.file_name());
-                            init::cleanup_local_only_file(&path, &name, &local)?;
-                            path.pop();
+                            init::cleanup_local_only_file(&name, &local)?;
                            // this file never existed remotely, we will have to do rework
                            continue;
                        }
@@ -2498,9 +2511,9 @@ impl Timeline {
                    tracing::debug!(layer=%name, ?decision, "applied");

                    let layer = match decision {
-                        UseLocal(m) => {
-                            total_physical_size += m.file_size();
-                            Layer::for_resident(conf, &this, name, m).drop_eviction_guard()
+                        UseLocal(local) => {
+                            total_physical_size += local.metadata.file_size();
+                            Layer::for_resident(conf, &this, local.local_path, name, local.metadata).drop_eviction_guard()
                        }
                        Evicted(remote) | UseRemote { remote, .. } => {
                            Layer::for_evicted(conf, &this, name, remote)
@@ -2981,11 +2994,11 @@ impl Timeline {
        }
    }

-    async fn find_layer(&self, layer_file_name: &str) -> Option<Layer> {
+    async fn find_layer(&self, layer_name: &LayerName) -> Option<Layer> {
        let guard = self.layers.read().await;
        for historic_layer in guard.layer_map().iter_historic_layers() {
-            let historic_layer_name = historic_layer.filename().file_name();
-            if layer_file_name == historic_layer_name {
+            let historic_layer_name = historic_layer.layer_name();
+            if layer_name == &historic_layer_name {
                return Some(guard.get_from_desc(&historic_layer));
            }
        }
@@ -3014,8 +3027,8 @@ impl Timeline {
            let last_activity_ts = layer.access_stats().latest_activity_or_now();

            HeatMapLayer::new(
-                layer.layer_desc().filename(),
-                layer.metadata().into(),
+                layer.layer_desc().layer_name(),
+                (&layer.metadata()).into(),
                last_activity_ts,
            )
        });
@@ -3161,7 +3174,7 @@ impl Timeline {
            if let Some(open_layer) = &layers.open_layer {
                let start_lsn = open_layer.get_lsn_range().start;
                if cont_lsn > start_lsn {
-                    //info!("CHECKING for {} at {} on open layer {}", key, cont_lsn, open_layer.filename().display());
+                    //info!("CHECKING for {} at {} on open layer {}", key, cont_lsn, open_layer.layer_name().display());
                    // Get all the data needed to reconstruct the page version from this layer.
                    // But if we have an older cached page image, no need to go past that.
                    let lsn_floor = max(cached_lsn + 1, start_lsn);
@@ -3190,7 +3203,7 @@ impl Timeline {
            for frozen_layer in layers.frozen_layers.iter().rev() {
                let start_lsn = frozen_layer.get_lsn_range().start;
                if cont_lsn > start_lsn {
-                    //info!("CHECKING for {} at {} on frozen layer {}", key, cont_lsn, frozen_layer.filename().display());
+                    //info!("CHECKING for {} at {} on frozen layer {}", key, cont_lsn, frozen_layer.layer_name().display());
                    let lsn_floor = max(cached_lsn + 1, start_lsn);

                    let frozen_layer = frozen_layer.clone();
@@ -3517,7 +3530,7 @@ impl Timeline {
        Ok(ancestor)
    }

-    fn get_ancestor_timeline(&self) -> anyhow::Result<Arc<Timeline>> {
+    pub(crate) fn get_ancestor_timeline(&self) -> anyhow::Result<Arc<Timeline>> {
        let ancestor = self.ancestor_timeline.as_ref().with_context(|| {
            format!(
                "Ancestor is missing. Timeline id: {} Ancestor id {:?}",
@@ -4333,6 +4346,49 @@ impl Timeline {
            _ = self.cancel.cancelled() => {}
        )
    }
+
+    /// Detach this timeline from its ancestor by copying all of ancestors layers as this
+    /// Timelines layers up to the ancestor_lsn.
+    ///
+    /// Requires a timeline that:
+    /// - has an ancestor to detach from
+    /// - the ancestor does not have an ancestor -- follows from the original RFC limitations, not
+    /// a technical requirement
+    /// - has prev_lsn in remote storage (temporary restriction)
+    ///
+    /// After the operation has been started, it cannot be canceled. Upon restart it needs to be
+    /// polled again until completion.
+    ///
+    /// During the operation all timelines sharing the data with this timeline will be reparented
+    /// from our ancestor to be branches of this timeline.
+    pub(crate) async fn prepare_to_detach_from_ancestor(
+        self: &Arc<Timeline>,
+        tenant: &crate::tenant::Tenant,
+        options: detach_ancestor::Options,
+        ctx: &RequestContext,
+    ) -> Result<
+        (
+            completion::Completion,
+            detach_ancestor::PreparedTimelineDetach,
+        ),
+        detach_ancestor::Error,
+    > {
+        detach_ancestor::prepare(self, tenant, options, ctx).await
+    }
+
+    /// Completes the ancestor detach. This method is to be called while holding the
+    /// TenantManager's tenant slot, so during this method we cannot be deleted nor can any
+    /// timeline be deleted. After this method returns successfully, tenant must be reloaded.
+    ///
+    /// Pageserver receiving a SIGKILL during this operation is not supported (yet).
+    pub(crate) async fn complete_detaching_timeline_ancestor(
+        self: &Arc<Timeline>,
+        tenant: &crate::tenant::Tenant,
+        prepared: detach_ancestor::PreparedTimelineDetach,
+        ctx: &RequestContext,
+    ) -> Result<Vec<TimelineId>, anyhow::Error> {
+        detach_ancestor::complete(self, tenant, prepared, ctx).await
+    }
 }

 /// Top-level failure to compact.
@@ -4441,6 +4497,24 @@ impl Timeline {
        Ok(())
    }

+    async fn rewrite_layers(
+        self: &Arc<Self>,
+        replace_layers: Vec<(Layer, ResidentLayer)>,
+        drop_layers: Vec<Layer>,
+    ) -> anyhow::Result<()> {
+        let mut guard = self.layers.write().await;
+
+        guard.rewrite_layers(&replace_layers, &drop_layers, &self.metrics);
+
+        let upload_layers: Vec<_> = replace_layers.into_iter().map(|r| r.1).collect();
+
+        if let Some(remote_client) = self.remote_client.as_ref() {
+            remote_client.schedule_compaction_update(&drop_layers, &upload_layers)?;
+        }
+
+        Ok(())
+    }
+
    /// Schedules the uploads of the given image layers
    fn upload_new_image_layers(
        self: &Arc<Self>,
@@ -4599,6 +4673,8 @@ impl Timeline {
        retain_lsns: Vec<Lsn>,
        new_gc_cutoff: Lsn,
    ) -> anyhow::Result<GcResult> {
+        // FIXME: if there is an ongoing detach_from_ancestor, we should just skip gc
+
        let now = SystemTime::now();
        let mut result: GcResult = GcResult::default();

@@ -4652,7 +4728,7 @@ impl Timeline {
            if l.get_lsn_range().end > horizon_cutoff {
                debug!(
                    "keeping {} because it's newer than horizon_cutoff {}",
-                    l.filename(),
+                    l.layer_name(),
                    horizon_cutoff,
                );
                result.layers_needed_by_cutoff += 1;
@@ -4663,7 +4739,7 @@ impl Timeline {
            if l.get_lsn_range().end > pitr_cutoff {
                debug!(
                    "keeping {} because it's newer than pitr_cutoff {}",
-                    l.filename(),
+                    l.layer_name(),
                    pitr_cutoff,
                );
                result.layers_needed_by_pitr += 1;
@@ -4682,7 +4758,7 @@ impl Timeline {
                if &l.get_lsn_range().start <= retain_lsn {
                    debug!(
                        "keeping {} because it's still might be referenced by child branch forked at {} is_dropped: xx is_incremental: {}",
-                        l.filename(),
+                        l.layer_name(),
                        retain_lsn,
                        l.is_incremental(),
                    );
@@ -4713,7 +4789,7 @@ impl Timeline {
            if !layers
                .image_layer_exists(&l.get_key_range(), &(l.get_lsn_range().end..new_gc_cutoff))
            {
-                debug!("keeping {} because it is the latest layer", l.filename());
+                debug!("keeping {} because it is the latest layer", l.layer_name());
                result.layers_not_updated += 1;
                continue 'outer;
            }
@@ -4721,7 +4797,7 @@ impl Timeline {
            // We didn't find any reason to keep this file, so remove it.
            debug!(
                "garbage collecting {} is_dropped: xx is_incremental: {}",
-                l.filename(),
+                l.layer_name(),
                l.is_incremental(),
            );
            layers_to_remove.push(l);
--- a/pageserver/src/tenant/timeline/compaction.rs
+++ b/pageserver/src/tenant/timeline/compaction.rs
@@ -15,7 +15,8 @@ use anyhow::{anyhow, Context};
 use enumset::EnumSet;
 use fail::fail_point;
 use itertools::Itertools;
-use pageserver_api::shard::{ShardIdentity, TenantShardId};
+use pageserver_api::keyspace::ShardedRange;
+use pageserver_api::shard::{ShardCount, ShardIdentity, TenantShardId};
 use tokio_util::sync::CancellationToken;
 use tracing::{debug, info, info_span, trace, warn, Instrument};
 use utils::id::TimelineId;
@@ -93,7 +94,7 @@ impl Timeline {
        // Define partitioning schema if needed

        // FIXME: the match should only cover repartitioning, not the next steps
-        match self
+        let partition_count = match self
            .repartition(
                self.get_last_record_lsn(),
                self.get_compaction_target_size(),
@@ -146,6 +147,7 @@ impl Timeline {
                assert!(sparse_layers.is_empty());

                self.upload_new_image_layers(dense_layers)?;
+                dense_partitioning.parts.len()
            }
            Err(err) => {
                // no partitioning? This is normal, if the timeline was just created
@@ -157,9 +159,150 @@ impl Timeline {
                if !self.cancel.is_cancelled() {
                    tracing::error!("could not compact, repartitioning keyspace failed: {err:?}");
                }
+                1
            }
        };

+        if self.shard_identity.count >= ShardCount::new(2) {
+            // Limit the number of layer rewrites to the number of partitions: this means its
+            // runtime should be comparable to a full round of image layer creations, rather than
+            // being potentially much longer.
+            let rewrite_max = partition_count;
+
+            self.compact_shard_ancestors(rewrite_max, ctx).await?;
+        }
+
+        Ok(())
+    }
+
+    /// Check for layers that are elegible to be rewritten:
+    /// - Shard splitting: After a shard split, ancestor layers beyond pitr_interval, so that
+    ///   we don't indefinitely retain keys in this shard that aren't needed.
+    /// - For future use: layers beyond pitr_interval that are in formats we would
+    ///   rather not maintain compatibility with indefinitely.
+    ///
+    /// Note: this phase may read and write many gigabytes of data: use rewrite_max to bound
+    /// how much work it will try to do in each compaction pass.
+    async fn compact_shard_ancestors(
+        self: &Arc<Self>,
+        rewrite_max: usize,
+        _ctx: &RequestContext,
+    ) -> anyhow::Result<()> {
+        let mut drop_layers = Vec::new();
+        let layers_to_rewrite: Vec<Layer> = Vec::new();
+
+        // We will use the PITR cutoff as a condition for rewriting layers.
+        let pitr_cutoff = self.gc_info.read().unwrap().cutoffs.pitr;
+
+        let layers = self.layers.read().await;
+        for layer_desc in layers.layer_map().iter_historic_layers() {
+            let layer = layers.get_from_desc(&layer_desc);
+            if layer.metadata().shard.shard_count == self.shard_identity.count {
+                // This layer does not belong to a historic ancestor, no need to re-image it.
+                continue;
+            }
+
+            // This layer was created on an ancestor shard: check if it contains any data for this shard.
+            let sharded_range = ShardedRange::new(layer_desc.get_key_range(), &self.shard_identity);
+            let layer_local_page_count = sharded_range.page_count();
+            let layer_raw_page_count = ShardedRange::raw_size(&layer_desc.get_key_range());
+            if layer_local_page_count == 0 {
+                // This ancestral layer only covers keys that belong to other shards.
+                // We include the full metadata in the log: if we had some critical bug that caused
+                // us to incorrectly drop layers, this would simplify manually debugging + reinstating those layers.
+                info!(%layer, old_metadata=?layer.metadata(),
+                    "dropping layer after shard split, contains no keys for this shard.",
+                );
+
+                if cfg!(debug_assertions) {
+                    // Expensive, exhaustive check of keys in this layer: this guards against ShardedRange's calculations being
+                    // wrong.  If ShardedRange claims the local page count is zero, then no keys in this layer
+                    // should be !is_key_disposable()
+                    let range = layer_desc.get_key_range();
+                    let mut key = range.start;
+                    while key < range.end {
+                        debug_assert!(self.shard_identity.is_key_disposable(&key));
+                        key = key.next();
+                    }
+                }
+
+                drop_layers.push(layer);
+                continue;
+            } else if layer_local_page_count != u32::MAX
+                && layer_local_page_count == layer_raw_page_count
+            {
+                debug!(%layer,
+                    "layer is entirely shard local ({} keys), no need to filter it",
+                    layer_local_page_count
+                );
+                continue;
+            }
+
+            // Don't bother re-writing a layer unless it will at least halve its size
+            if layer_local_page_count != u32::MAX
+                && layer_local_page_count > layer_raw_page_count / 2
+            {
+                debug!(%layer,
+                    "layer is already mostly local ({}/{}), not rewriting",
+                    layer_local_page_count,
+                    layer_raw_page_count
+                );
+            }
+
+            // Don't bother re-writing a layer if it is within the PITR window: it will age-out eventually
+            // without incurring the I/O cost of a rewrite.
+            if layer_desc.get_lsn_range().end >= pitr_cutoff {
+                debug!(%layer, "Skipping rewrite of layer still in PITR window ({} >= {})",
+                    layer_desc.get_lsn_range().end, pitr_cutoff);
+                continue;
+            }
+
+            if layer_desc.is_delta() {
+                // We do not yet implement rewrite of delta layers
+                debug!(%layer, "Skipping rewrite of delta layer");
+                continue;
+            }
+
+            // Only rewrite layers if they would have different remote paths: either they belong to this
+            // shard but an old generation, or they belonged to another shard.  This also implicitly
+            // guarantees that the layer is persistent in remote storage (as only remote persistent
+            // layers are carried across shard splits, any local-only layer would be in the current generation)
+            if layer.metadata().generation == self.generation
+                && layer.metadata().shard.shard_count == self.shard_identity.count
+            {
+                debug!(%layer, "Skipping rewrite, is not from old generation");
+                continue;
+            }
+
+            if layers_to_rewrite.len() >= rewrite_max {
+                tracing::info!(%layer, "Will rewrite layer on a future compaction, already rewrote {}",
+                    layers_to_rewrite.len()
+                );
+                continue;
+            }
+
+            // Fall through: all our conditions for doing a rewrite passed.
+            // TODO: implement rewriting
+            tracing::debug!(%layer, "Would rewrite layer");
+        }
+
+        // Drop the layers read lock: we will acquire it for write in [`Self::rewrite_layers`]
+        drop(layers);
+
+        // TODO: collect layers to rewrite
+        let replace_layers = Vec::new();
+
+        // Update the LayerMap so that readers will use the new layers, and enqueue it for writing to remote storage
+        self.rewrite_layers(replace_layers, drop_layers).await?;
+
+        if let Some(remote_client) = self.remote_client.as_ref() {
+            // We wait for all uploads to complete before finishing this compaction stage.  This is not
+            // necessary for correctness, but it simplifies testing, and avoids proceeding with another
+            // Timeline's compaction while this timeline's uploads may be generating lots of disk I/O
+            // load.
+            remote_client.wait_completion().await?;
+        }
+
        Ok(())
    }

--- a/pageserver/src/tenant/timeline/delete.rs
+++ b/pageserver/src/tenant/timeline/delete.rs
@@ -422,6 +422,10 @@ impl DeleteTimelineFlow {
    pub(crate) fn is_finished(&self) -> bool {
        matches!(self, Self::Finished)
    }
+
+    pub(crate) fn is_not_started(&self) -> bool {
+        matches!(self, Self::NotStarted)
+    }
 }

 struct DeletionGuard(OwnedMutexGuard<DeleteTimelineFlow>);
--- a/pageserver/src/tenant/timeline/detach_ancestor.rs
+++ b/pageserver/src/tenant/timeline/detach_ancestor.rs
@@ -0,0 +1,550 @@
+use std::sync::Arc;
+
+use super::{layer_manager::LayerManager, Timeline};
+use crate::{
+    context::{DownloadBehavior, RequestContext},
+    task_mgr::TaskKind,
+    tenant::{
+        storage_layer::{AsLayerDesc as _, DeltaLayerWriter, Layer, ResidentLayer},
+        Tenant,
+    },
+    virtual_file::{MaybeFatalIo, VirtualFile},
+};
+use tokio_util::sync::CancellationToken;
+use tracing::Instrument;
+use utils::{completion, generation::Generation, id::TimelineId, lsn::Lsn};
+
+#[derive(Debug, thiserror::Error)]
+pub(crate) enum Error {
+    #[error("no ancestors")]
+    NoAncestor,
+    #[error("too many ancestors")]
+    TooManyAncestors,
+    #[error("shutting down, please retry later")]
+    ShuttingDown,
+    #[error("detached timeline must receive writes before the operation")]
+    DetachedTimelineNeedsWrites,
+    #[error("flushing failed")]
+    FlushAncestor(#[source] anyhow::Error),
+    #[error("layer download failed")]
+    RewrittenDeltaDownloadFailed(#[source] anyhow::Error),
+    #[error("copying LSN prefix locally failed")]
+    CopyDeltaPrefix(#[source] anyhow::Error),
+    #[error("upload rewritten layer")]
+    UploadRewritten(#[source] anyhow::Error),
+
+    #[error("ancestor is already being detached by: {}", .0)]
+    OtherTimelineDetachOngoing(TimelineId),
+
+    #[error("remote copying layer failed")]
+    CopyFailed(#[source] anyhow::Error),
+
+    #[error("unexpected error")]
+    Unexpected(#[source] anyhow::Error),
+}
+
+pub(crate) struct PreparedTimelineDetach {
+    layers: Vec<Layer>,
+}
+
+/// TODO: this should be part of PageserverConf because we cannot easily modify cplane arguments.
+#[derive(Debug)]
+pub(crate) struct Options {
+    pub(crate) rewrite_concurrency: std::num::NonZeroUsize,
+    pub(crate) copy_concurrency: std::num::NonZeroUsize,
+}
+
+impl Default for Options {
+    fn default() -> Self {
+        Self {
+            rewrite_concurrency: std::num::NonZeroUsize::new(2).unwrap(),
+            copy_concurrency: std::num::NonZeroUsize::new(10).unwrap(),
+        }
+    }
+}
+
+/// See [`Timeline::prepare_to_detach_from_ancestor`]
+pub(super) async fn prepare(
+    detached: &Arc<Timeline>,
+    tenant: &Tenant,
+    options: Options,
+    ctx: &RequestContext,
+) -> Result<(completion::Completion, PreparedTimelineDetach), Error> {
+    use Error::*;
+
+    if detached.remote_client.as_ref().is_none() {
+        unimplemented!("no new code for running without remote storage");
+    }
+
+    let Some((ancestor, ancestor_lsn)) = detached
+        .ancestor_timeline
+        .as_ref()
+        .map(|tl| (tl.clone(), detached.ancestor_lsn))
+    else {
+        return Err(NoAncestor);
+    };
+
+    if !ancestor_lsn.is_valid() {
+        return Err(NoAncestor);
+    }
+
+    if ancestor.ancestor_timeline.is_some() {
+        // non-technical requirement; we could flatten N ancestors just as easily but we chose
+        // not to
+        return Err(TooManyAncestors);
+    }
+
+    if detached.get_prev_record_lsn() == Lsn::INVALID
+        || detached.disk_consistent_lsn.load() == ancestor_lsn
+    {
+        // this is to avoid a problem that after detaching we would be unable to start up the
+        // compute because of "PREV_LSN: invalid".
+        return Err(DetachedTimelineNeedsWrites);
+    }
+
+    // before we acquire the gate, we must mark the ancestor as having a detach operation
+    // ongoing which will block other concurrent detach operations so we don't get to ackward
+    // situations where there would be two branches trying to reparent earlier branches.
+    let (guard, barrier) = completion::channel();
+
+    {
+        let mut guard = tenant.ongoing_timeline_detach.lock().unwrap();
+        if let Some((tl, other)) = guard.as_ref() {
+            if !other.is_ready() {
+                return Err(OtherTimelineDetachOngoing(*tl));
+            }
+        }
+        *guard = Some((detached.timeline_id, barrier));
+    }
+
+    let _gate_entered = detached.gate.enter().map_err(|_| ShuttingDown)?;
+
+    if ancestor_lsn >= ancestor.get_disk_consistent_lsn() {
+        let span =
+            tracing::info_span!("freeze_and_flush", ancestor_timeline_id=%ancestor.timeline_id);
+        async {
+            let started_at = std::time::Instant::now();
+            let freeze_and_flush = ancestor.freeze_and_flush0();
+            let mut freeze_and_flush = std::pin::pin!(freeze_and_flush);
+
+            let res =
+                tokio::time::timeout(std::time::Duration::from_secs(1), &mut freeze_and_flush)
+                    .await;
+
+            let res = match res {
+                Ok(res) => res,
+                Err(_elapsed) => {
+                    tracing::info!("freezing and flushing ancestor is still ongoing");
+                    freeze_and_flush.await
+                }
+            };
+
+            res.map_err(FlushAncestor)?;
+
+            // we do not need to wait for uploads to complete but we do need `struct Layer`,
+            // copying delta prefix is unsupported currently for `InMemoryLayer`.
+            tracing::info!(
+                elapsed_ms = started_at.elapsed().as_millis(),
+                "froze and flushed the ancestor"
+            );
+            Ok(())
+        }
+        .instrument(span)
+        .await?;
+    }
+
+    let end_lsn = ancestor_lsn + 1;
+
+    let (filtered_layers, straddling_branchpoint, rest_of_historic) = {
+        // we do not need to start from our layers, because they can only be layers that come
+        // *after* ancestor_lsn
+        let layers = tokio::select! {
+            guard = ancestor.layers.read() => guard,
+            _ = detached.cancel.cancelled() => {
+                return Err(ShuttingDown);
+            }
+            _ = ancestor.cancel.cancelled() => {
+                return Err(ShuttingDown);
+            }
+        };
+
+        // between retries, these can change if compaction or gc ran in between. this will mean
+        // we have to redo work.
+        partition_work(ancestor_lsn, &layers)
+    };
+
+    // TODO: layers are already sorted by something: use that to determine how much of remote
+    // copies are already done.
+    tracing::info!(filtered=%filtered_layers, to_rewrite = straddling_branchpoint.len(), historic=%rest_of_historic.len(), "collected layers");
+
+    // TODO: copying and lsn prefix copying could be done at the same time with a single fsync after
+    let mut new_layers: Vec<Layer> =
+        Vec::with_capacity(straddling_branchpoint.len() + rest_of_historic.len());
+
+    {
+        tracing::debug!(to_rewrite = %straddling_branchpoint.len(), "copying prefix of delta layers");
+
+        let mut tasks = tokio::task::JoinSet::new();
+
+        let mut wrote_any = false;
+
+        let limiter = Arc::new(tokio::sync::Semaphore::new(
+            options.rewrite_concurrency.get(),
+        ));
+
+        for layer in straddling_branchpoint {
+            let limiter = limiter.clone();
+            let timeline = detached.clone();
+            let ctx = ctx.detached_child(TaskKind::DetachAncestor, DownloadBehavior::Download);
+
+            tasks.spawn(async move {
+                let _permit = limiter.acquire().await;
+                let copied =
+                    upload_rewritten_layer(end_lsn, &layer, &timeline, &timeline.cancel, &ctx)
+                        .await?;
+                Ok(copied)
+            });
+        }
+
+        while let Some(res) = tasks.join_next().await {
+            match res {
+                Ok(Ok(Some(copied))) => {
+                    wrote_any = true;
+                    tracing::info!(layer=%copied, "rewrote and uploaded");
+                    new_layers.push(copied);
+                }
+                Ok(Ok(None)) => {}
+                Ok(Err(e)) => return Err(e),
+                Err(je) => return Err(Unexpected(je.into())),
+            }
+        }
+
+        // FIXME: the fsync should be mandatory, after both rewrites and copies
+        if wrote_any {
+            let timeline_dir = VirtualFile::open(
+                &detached
+                    .conf
+                    .timeline_path(&detached.tenant_shard_id, &detached.timeline_id),
+            )
+            .await
+            .fatal_err("VirtualFile::open for timeline dir fsync");
+            timeline_dir
+                .sync_all()
+                .await
+                .fatal_err("VirtualFile::sync_all timeline dir");
+        }
+    }
+
+    let mut tasks = tokio::task::JoinSet::new();
+    let limiter = Arc::new(tokio::sync::Semaphore::new(options.copy_concurrency.get()));
+
+    for adopted in rest_of_historic {
+        let limiter = limiter.clone();
+        let timeline = detached.clone();
+
+        tasks.spawn(
+            async move {
+                let _permit = limiter.acquire().await;
+                let owned =
+                    remote_copy(&adopted, &timeline, timeline.generation, &timeline.cancel).await?;
+                tracing::info!(layer=%owned, "remote copied");
+                Ok(owned)
+            }
+            .in_current_span(),
+        );
+    }
+
+    while let Some(res) = tasks.join_next().await {
+        match res {
+            Ok(Ok(owned)) => {
+                new_layers.push(owned);
+            }
+            Ok(Err(failed)) => {
+                return Err(failed);
+            }
+            Err(je) => return Err(Unexpected(je.into())),
+        }
+    }
+
+    // TODO: fsync directory again if we hardlinked something
+
+    let prepared = PreparedTimelineDetach { layers: new_layers };
+
+    Ok((guard, prepared))
+}
+
+fn partition_work(
+    ancestor_lsn: Lsn,
+    source_layermap: &LayerManager,
+) -> (usize, Vec<Layer>, Vec<Layer>) {
+    let mut straddling_branchpoint = vec![];
+    let mut rest_of_historic = vec![];
+
+    let mut later_by_lsn = 0;
+
+    for desc in source_layermap.layer_map().iter_historic_layers() {
+        // off by one chances here:
+        // - start is inclusive
+        // - end is exclusive
+        if desc.lsn_range.start > ancestor_lsn {
+            later_by_lsn += 1;
+            continue;
+        }
+
+        let target = if desc.lsn_range.start <= ancestor_lsn
+            && desc.lsn_range.end > ancestor_lsn
+            && desc.is_delta
+        {
+            // TODO: image layer at Lsn optimization
+            &mut straddling_branchpoint
+        } else {
+            &mut rest_of_historic
+        };
+
+        target.push(source_layermap.get_from_desc(&desc));
+    }
+
+    (later_by_lsn, straddling_branchpoint, rest_of_historic)
+}
+
+async fn upload_rewritten_layer(
+    end_lsn: Lsn,
+    layer: &Layer,
+    target: &Arc<Timeline>,
+    cancel: &CancellationToken,
+    ctx: &RequestContext,
+) -> Result<Option<Layer>, Error> {
+    use Error::UploadRewritten;
+    let copied = copy_lsn_prefix(end_lsn, layer, target, ctx).await?;
+
+    let Some(copied) = copied else {
+        return Ok(None);
+    };
+
+    // FIXME: better shuttingdown error
+    target
+        .remote_client
+        .as_ref()
+        .unwrap()
+        .upload_layer_file(&copied, cancel)
+        .await
+        .map_err(UploadRewritten)?;
+
+    Ok(Some(copied.into()))
+}
+
+async fn copy_lsn_prefix(
+    end_lsn: Lsn,
+    layer: &Layer,
+    target_timeline: &Arc<Timeline>,
+    ctx: &RequestContext,
+) -> Result<Option<ResidentLayer>, Error> {
+    use Error::{CopyDeltaPrefix, RewrittenDeltaDownloadFailed};
+
+    tracing::debug!(%layer, %end_lsn, "copying lsn prefix");
+
+    let mut writer = DeltaLayerWriter::new(
+        target_timeline.conf,
+        target_timeline.timeline_id,
+        target_timeline.tenant_shard_id,
+        layer.layer_desc().key_range.start,
+        layer.layer_desc().lsn_range.start..end_lsn,
+    )
+    .await
+    .map_err(CopyDeltaPrefix)?;
+
+    let resident = layer
+        .download_and_keep_resident()
+        .await
+        // likely shutdown
+        .map_err(RewrittenDeltaDownloadFailed)?;
+
+    let records = resident
+        .copy_delta_prefix(&mut writer, end_lsn, ctx)
+        .await
+        .map_err(CopyDeltaPrefix)?;
+
+    drop(resident);
+
+    tracing::debug!(%layer, records, "copied records");
+
+    if records == 0 {
+        drop(writer);
+        // TODO: we might want to store an empty marker in remote storage for this
+        // layer so that we will not needlessly walk `layer` on repeated attempts.
+        Ok(None)
+    } else {
+        // reuse the key instead of adding more holes between layers by using the real
+        // highest key in the layer.
+        let reused_highest_key = layer.layer_desc().key_range.end;
+        let copied = writer
+            .finish(reused_highest_key, target_timeline, ctx)
+            .await
+            .map_err(CopyDeltaPrefix)?;
+
+        tracing::debug!(%layer, %copied, "new layer produced");
+
+        Ok(Some(copied))
+    }
+}
+
+/// Creates a new Layer instance for the adopted layer, and ensures it is found from the remote
+/// storage on successful return without the adopted layer being added to `index_part.json`.
+async fn remote_copy(
+    adopted: &Layer,
+    adoptee: &Arc<Timeline>,
+    generation: Generation,
+    cancel: &CancellationToken,
+) -> Result<Layer, Error> {
+    use Error::CopyFailed;
+
+    // depending if Layer::keep_resident we could hardlink
+
+    let mut metadata = adopted.metadata();
+    debug_assert!(metadata.generation <= generation);
+    metadata.generation = generation;
+
+    let owned = crate::tenant::storage_layer::Layer::for_evicted(
+        adoptee.conf,
+        adoptee,
+        adopted.layer_desc().layer_name(),
+        metadata,
+    );
+
+    // FIXME: better shuttingdown error
+    adoptee
+        .remote_client
+        .as_ref()
+        .unwrap()
+        .copy_timeline_layer(adopted, &owned, cancel)
+        .await
+        .map(move |()| owned)
+        .map_err(CopyFailed)
+}
+
+/// See [`Timeline::complete_detaching_timeline_ancestor`].
+pub(super) async fn complete(
+    detached: &Arc<Timeline>,
+    tenant: &Tenant,
+    prepared: PreparedTimelineDetach,
+    _ctx: &RequestContext,
+) -> Result<Vec<TimelineId>, anyhow::Error> {
+    let rtc = detached
+        .remote_client
+        .as_ref()
+        .expect("has to have a remote timeline client for timeline ancestor detach");
+
+    let PreparedTimelineDetach { layers } = prepared;
+
+    let ancestor = detached
+        .get_ancestor_timeline()
+        .expect("must still have a ancestor");
+    let ancestor_lsn = detached.get_ancestor_lsn();
+
+    // publish the prepared layers before we reparent any of the timelines, so that on restart
+    // reparented timelines find layers. also do the actual detaching.
+    //
+    // if we crash after this operation, we will at least come up having detached a timeline, but
+    // we cannot go back and reparent the timelines which would had been reparented in normal
+    // execution.
+    //
+    // this is not perfect, but it avoids us a retry happening after a compaction or gc on restart
+    // which could give us a completely wrong layer combination.
+    rtc.schedule_adding_existing_layers_to_index_detach_and_wait(
+        &layers,
+        (ancestor.timeline_id, ancestor_lsn),
+    )
+    .await?;
+
+    let mut tasks = tokio::task::JoinSet::new();
+
+    // because we are now keeping the slot in progress, it is unlikely that there will be any
+    // timeline deletions during this time. if we raced one, then we'll just ignore it.
+    tenant
+        .timelines
+        .lock()
+        .unwrap()
+        .values()
+        .filter_map(|tl| {
+            if Arc::ptr_eq(tl, detached) {
+                return None;
+            }
+
+            if !tl.is_active() {
+                return None;
+            }
+
+            let tl_ancestor = tl.ancestor_timeline.as_ref()?;
+            let is_same = Arc::ptr_eq(&ancestor, tl_ancestor);
+            let is_earlier = tl.get_ancestor_lsn() <= ancestor_lsn;
+
+            let is_deleting = tl
+                .delete_progress
+                .try_lock()
+                .map(|flow| !flow.is_not_started())
+                .unwrap_or(true);
+
+            if is_same && is_earlier && !is_deleting {
+                Some(tl.clone())
+            } else {
+                None
+            }
+        })
+        .for_each(|timeline| {
+            // important in this scope: we are holding the Tenant::timelines lock
+            let span = tracing::info_span!("reparent", reparented=%timeline.timeline_id);
+            let new_parent = detached.timeline_id;
+
+            tasks.spawn(
+                async move {
+                    let res = timeline
+                        .remote_client
+                        .as_ref()
+                        .expect("reparented has to have remote client because detached has one")
+                        .schedule_reparenting_and_wait(&new_parent)
+                        .await;
+
+                    match res {
+                        Ok(()) => Some(timeline),
+                        Err(e) => {
+                            // with the use of tenant slot, we no longer expect these.
+                            tracing::warn!("reparenting failed: {e:#}");
+                            None
+                        }
+                    }
+                }
+                .instrument(span),
+            );
+        });
+
+    let reparenting_candidates = tasks.len();
+    let mut reparented = Vec::with_capacity(tasks.len());
+
+    while let Some(res) = tasks.join_next().await {
+        match res {
+            Ok(Some(timeline)) => {
+                tracing::info!(reparented=%timeline.timeline_id, "reparenting done");
+                reparented.push(timeline.timeline_id);
+            }
+            Ok(None) => {
+                // lets just ignore this for now. one or all reparented timelines could had
+                // started deletion, and that is fine.
+            }
+            Err(je) if je.is_cancelled() => unreachable!("not used"),
+            Err(je) if je.is_panic() => {
+                // ignore; it's better to continue with a single reparenting failing (or even
+                // all of them) in order to get to the goal state.
+                //
+                // these timelines will never be reparentable, but they can be always detached as
+                // separate tree roots.
+            }
+            Err(je) => tracing::error!("unexpected join error: {je:?}"),
+        }
+    }
+
+    if reparenting_candidates != reparented.len() {
+        tracing::info!("failed to reparent some candidates");
+    }
+
+    Ok(reparented)
+}
--- a/pageserver/src/tenant/timeline/init.rs
+++ b/pageserver/src/tenant/timeline/init.rs
@@ -6,13 +6,13 @@ use crate::{
            self,
            index::{IndexPart, LayerFileMetadata},
        },
-        storage_layer::LayerFileName,
+        storage_layer::LayerName,
        Generation,
    },
    METADATA_FILE_NAME,
 };
 use anyhow::Context;
-use camino::Utf8Path;
+use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::shard::ShardIndex;
 use std::{collections::HashMap, str::FromStr};
 use utils::lsn::Lsn;
@@ -20,7 +20,7 @@ use utils::lsn::Lsn;
 /// Identified files in the timeline directory.
 pub(super) enum Discovered {
    /// The only one we care about
-    Layer(LayerFileName, u64),
+    Layer(LayerName, Utf8PathBuf, u64),
    /// Old ephmeral files from previous launches, should be removed
    Ephemeral(String),
    /// Old temporary timeline files, unsure what these really are, should be removed
@@ -43,10 +43,10 @@ pub(super) fn scan_timeline_dir(path: &Utf8Path) -> anyhow::Result<Vec<Discovere
        let direntry = direntry?;
        let file_name = direntry.file_name().to_string();

-        let discovered = match LayerFileName::from_str(&file_name) {
+        let discovered = match LayerName::from_str(&file_name) {
            Ok(file_name) => {
                let file_size = direntry.metadata()?.len();
-                Discovered::Layer(file_name, file_size)
+                Discovered::Layer(file_name, direntry.path().to_owned(), file_size)
            }
            Err(_) => {
                if file_name == METADATA_FILE_NAME {
@@ -72,6 +72,28 @@ pub(super) fn scan_timeline_dir(path: &Utf8Path) -> anyhow::Result<Vec<Discovere
    Ok(ret)
 }

+/// Whereas `LayerFileMetadata` describes the metadata we would store in remote storage,
+/// this structure extends it with metadata describing the layer's presence in local storage.
+#[derive(Clone, Debug)]
+pub(super) struct LocalLayerFileMetadata {
+    pub(super) metadata: LayerFileMetadata,
+    pub(super) local_path: Utf8PathBuf,
+}
+
+impl LocalLayerFileMetadata {
+    pub fn new(
+        local_path: Utf8PathBuf,
+        file_size: u64,
+        generation: Generation,
+        shard: ShardIndex,
+    ) -> Self {
+        Self {
+            local_path,
+            metadata: LayerFileMetadata::new(file_size, generation, shard),
+        }
+    }
+}
+
 /// Decision on what to do with a layer file after considering its local and remote metadata.
 #[derive(Clone, Debug)]
 pub(super) enum Decision {
@@ -80,11 +102,11 @@ pub(super) enum Decision {
    /// The layer is present locally, but local metadata does not match remote; we must
    /// delete it and treat it as evicted.
    UseRemote {
-        local: LayerFileMetadata,
+        local: LocalLayerFileMetadata,
        remote: LayerFileMetadata,
    },
    /// The layer is present locally, and metadata matches.
-    UseLocal(LayerFileMetadata),
+    UseLocal(LocalLayerFileMetadata),
 }

 /// A layer needs to be left out of the layer map.
@@ -92,39 +114,42 @@ pub(super) enum Decision {
 pub(super) enum DismissedLayer {
    /// The related layer is is in future compared to disk_consistent_lsn, it must not be loaded.
    Future {
-        /// The local metadata. `None` if the layer is only known through [`IndexPart`].
-        local: Option<LayerFileMetadata>,
+        /// `None` if the layer is only known through [`IndexPart`].
+        local: Option<LocalLayerFileMetadata>,
    },
    /// The layer only exists locally.
    ///
    /// In order to make crash safe updates to layer map, we must dismiss layers which are only
    /// found locally or not yet included in the remote `index_part.json`.
-    LocalOnly(LayerFileMetadata),
+    LocalOnly(LocalLayerFileMetadata),
 }

 /// Merges local discoveries and remote [`IndexPart`] to a collection of decisions.
 pub(super) fn reconcile(
-    discovered: Vec<(LayerFileName, u64)>,
+    discovered: Vec<(LayerName, Utf8PathBuf, u64)>,
    index_part: Option<&IndexPart>,
    disk_consistent_lsn: Lsn,
    generation: Generation,
    shard: ShardIndex,
-) -> Vec<(LayerFileName, Result<Decision, DismissedLayer>)> {
+) -> Vec<(LayerName, Result<Decision, DismissedLayer>)> {
    use Decision::*;

-    // name => (local, remote)
-    type Collected = HashMap<LayerFileName, (Option<LayerFileMetadata>, Option<LayerFileMetadata>)>;
+    // name => (local_metadata, remote_metadata)
+    type Collected =
+        HashMap<LayerName, (Option<LocalLayerFileMetadata>, Option<LayerFileMetadata>)>;

    let mut discovered = discovered
        .into_iter()
-        .map(|(name, file_size)| {
+        .map(|(layer_name, local_path, file_size)| {
            (
-                name,
+                layer_name,
                // The generation and shard here will be corrected to match IndexPart in the merge below, unless
                // it is not in IndexPart, in which case using our current generation makes sense
                // because it will be uploaded in this generation.
                (
-                    Some(LayerFileMetadata::new(file_size, generation, shard)),
+                    Some(LocalLayerFileMetadata::new(
+                        local_path, file_size, generation, shard,
+                    )),
                    None,
                ),
            )
@@ -153,7 +178,7 @@ pub(super) fn reconcile(
                Err(DismissedLayer::Future { local })
            } else {
                match (local, remote) {
-                    (Some(local), Some(remote)) if local != remote => {
+                    (Some(local), Some(remote)) if local.metadata != remote => {
                        Ok(UseRemote { local, remote })
                    }
                    (Some(x), Some(_)) => Ok(UseLocal(x)),
@@ -177,12 +202,12 @@ pub(super) fn cleanup(path: &Utf8Path, kind: &str) -> anyhow::Result<()> {
 }

 pub(super) fn cleanup_local_file_for_remote(
-    path: &Utf8Path,
-    local: &LayerFileMetadata,
+    local: &LocalLayerFileMetadata,
    remote: &LayerFileMetadata,
 ) -> anyhow::Result<()> {
-    let local_size = local.file_size();
+    let local_size = local.metadata.file_size();
    let remote_size = remote.file_size();
+    let path = &local.local_path;

    let file_name = path.file_name().expect("must be file path");
    tracing::warn!("removing local file {file_name:?} because it has unexpected length {local_size}; length in remote index is {remote_size}");
@@ -199,7 +224,7 @@ pub(super) fn cleanup_local_file_for_remote(

 pub(super) fn cleanup_future_layer(
    path: &Utf8Path,
-    name: &LayerFileName,
+    name: &LayerName,
    disk_consistent_lsn: Lsn,
 ) -> anyhow::Result<()> {
    // future image layers are allowed to be produced always for not yet flushed to disk
@@ -211,12 +236,14 @@ pub(super) fn cleanup_future_layer(
 }

 pub(super) fn cleanup_local_only_file(
-    path: &Utf8Path,
-    name: &LayerFileName,
-    local: &LayerFileMetadata,
+    name: &LayerName,
+    local: &LocalLayerFileMetadata,
 ) -> anyhow::Result<()> {
    let kind = name.kind();
-    tracing::info!("found local-only {kind} layer {name}, metadata {local:?}");
-    std::fs::remove_file(path)?;
+    tracing::info!(
+        "found local-only {kind} layer {name}, metadata {:?}",
+        local.metadata
+    );
+    std::fs::remove_file(&local.local_path)?;
    Ok(())
 }
--- a/pageserver/src/tenant/timeline/layer_manager.rs
+++ b/pageserver/src/tenant/timeline/layer_manager.rs
@@ -205,6 +205,24 @@ impl LayerManager {
        updates.flush();
    }

+    /// Called when compaction is completed.
+    pub(crate) fn rewrite_layers(
+        &mut self,
+        rewrite_layers: &[(Layer, ResidentLayer)],
+        drop_layers: &[Layer],
+        _metrics: &TimelineMetrics,
+    ) {
+        let mut updates = self.layer_map.batch_update();
+
+        // TODO: implement rewrites (currently this code path only used for drops)
+        assert!(rewrite_layers.is_empty());
+
+        for l in drop_layers {
+            Self::delete_historic_layer(l, &mut updates, &mut self.layer_fmgr);
+        }
+        updates.flush();
+    }
+
    /// Called when garbage collect has selected the layers to be removed.
    pub(crate) fn finish_gc_timeline(&mut self, gc_layers: &[Layer]) {
        let mut updates = self.layer_map.batch_update();
@@ -276,7 +294,7 @@ impl<T: AsLayerDesc + Clone> LayerFileManager<T> {
        // A layer's descriptor is present in the LayerMap => the LayerFileManager contains a layer for the descriptor.
        self.0
            .get(&desc.key())
-            .with_context(|| format!("get layer from desc: {}", desc.filename()))
+            .with_context(|| format!("get layer from desc: {}", desc.layer_name()))
            .expect("not found")
            .clone()
    }
--- a/pageserver/src/tenant/upload_queue.rs
+++ b/pageserver/src/tenant/upload_queue.rs
@@ -1,4 +1,4 @@
-use super::storage_layer::LayerFileName;
+use super::storage_layer::LayerName;
 use super::storage_layer::ResidentLayer;
 use crate::tenant::metadata::TimelineMetadata;
 use crate::tenant::remote_timeline_client::index::IndexPart;
@@ -45,7 +45,7 @@ pub(crate) struct UploadQueueInitialized {

    /// All layer files stored in the remote storage, taking into account all
    /// in-progress and queued operations
-    pub(crate) latest_files: HashMap<LayerFileName, LayerFileMetadata>,
+    pub(crate) latest_files: HashMap<LayerName, LayerFileMetadata>,

    /// How many file uploads or deletions been scheduled, since the
    /// last (scheduling of) metadata index upload?
@@ -89,7 +89,7 @@ pub(crate) struct UploadQueueInitialized {
    /// Putting this behind a testing feature to catch problems in tests, but assuming we could have a
    /// bug causing leaks, then it's better to not leave this enabled for production builds.
    #[cfg(feature = "testing")]
-    pub(crate) dangling_files: HashMap<LayerFileName, Generation>,
+    pub(crate) dangling_files: HashMap<LayerName, Generation>,

    /// Set to true when we have inserted the `UploadOp::Shutdown` into the `inprogress_tasks`.
    pub(crate) shutting_down: bool,
@@ -281,7 +281,7 @@ pub(crate) struct UploadTask {
 /// for timeline deletion, which skips this queue and goes directly to DeletionQueue.
 #[derive(Debug)]
 pub(crate) struct Delete {
-    pub(crate) layers: Vec<(LayerFileName, LayerFileMetadata)>,
+    pub(crate) layers: Vec<(LayerName, LayerFileMetadata)>,
 }

 #[derive(Debug)]
--- a/pgxn/neon/Makefile
+++ b/pgxn/neon/Makefile
@@ -14,7 +14,8 @@ OBJS = \
 	relsize_cache.o \
 	walproposer.o \
 	walproposer_pg.o \
-	control_plane_connector.o
+	control_plane_connector.o \
+	walsender_hooks.o

 PG_CPPFLAGS = -I$(libpq_srcdir)
 SHLIB_LINK_INTERNAL = $(libpq)
--- a/pgxn/neon/libpagestore.c
+++ b/pgxn/neon/libpagestore.c
@@ -49,7 +49,7 @@ char	   *neon_auth_token;
 int			readahead_buffer_size = 128;
 int			flush_every_n_requests = 8;

-int         neon_protocol_version = 1;
+int         neon_protocol_version = 2;

 static int	n_reconnect_attempts = 0;
 static int	max_reconnect_attempts = 60;
@@ -860,7 +860,7 @@ pg_init_libpagestore(void)
 							"Version of compute<->page server protocol",
 							NULL,
 							&neon_protocol_version,
-							1, /* default to old protocol for now */
+							2, /* use protocol version 2 */
 							1, /* min */
 							2, /* max */
 							PGC_SU_BACKEND,
--- a/pgxn/neon/neon.c
+++ b/pgxn/neon/neon.c
@@ -34,6 +34,7 @@
 #include "walproposer.h"
 #include "pagestore_client.h"
 #include "control_plane_connector.h"
+#include "walsender_hooks.h"

 PG_MODULE_MAGIC;
 void		_PG_init(void);
@@ -265,7 +266,6 @@ LogicalSlotsMonitorMain(Datum main_arg)
 	}
 }

-
 void
 _PG_init(void)
 {
@@ -279,6 +279,7 @@ _PG_init(void)

 	pg_init_libpagestore();
 	pg_init_walproposer();
+        WalSender_Custom_XLogReaderRoutines = NeonOnDemandXLogReaderRoutines;

 	InitLogicalReplicationMonitor();

--- a/pgxn/neon/neon_walreader.c
+++ b/pgxn/neon/neon_walreader.c
@@ -36,10 +36,7 @@

 static NeonWALReadResult NeonWALReadRemote(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size count, TimeLineID tli);
 static NeonWALReadResult NeonWALReaderReadMsg(NeonWALReader *state);
-static void NeonWALReaderResetRemote(NeonWALReader *state);
 static bool NeonWALReadLocal(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size count, TimeLineID tli);
-static bool neon_wal_segment_open(NeonWALReader *state, XLogSegNo nextSegNo, TimeLineID *tli_p);
-static void neon_wal_segment_close(NeonWALReader *state);
 static bool is_wal_segment_exists(XLogSegNo segno, int segsize,
 								  TimeLineID tli);

@@ -82,8 +79,9 @@ struct NeonWALReader
 	XLogRecPtr	req_lsn;
 	Size		req_len;
 	Size		req_progress;
-	WalProposer *wp;			/* we learn donor through walproposer */
+	char		donor_conninfo[MAXCONNINFO];
 	char		donor_name[64]; /* saved donor safekeeper name for logging */
+	XLogRecPtr	donor_lsn;
 	/* state of connection to safekeeper */
 	NeonWALReaderRemoteState rem_state;
 	WalProposerConn *wp_conn;
@@ -107,7 +105,7 @@ struct NeonWALReader

 /* palloc and initialize NeonWALReader */
 NeonWALReader *
-NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, WalProposer *wp, char *log_prefix)
+NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, char *log_prefix)
 {
 	NeonWALReader *reader;

@@ -123,8 +121,6 @@ NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, WalPropose
 	reader->seg.ws_tli = 0;
 	reader->segcxt.ws_segsize = wal_segment_size;

-	reader->wp = wp;
-
 	reader->rem_state = RS_NONE;

 	if (log_prefix)
@@ -204,21 +200,16 @@ NeonWALReadRemote(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size cou
 {
 	if (state->rem_state == RS_NONE)
 	{
-		XLogRecPtr	donor_lsn;
-
-		/* no connection yet; start one */
-		Safekeeper *donor = GetDonor(state->wp, &donor_lsn);
-
-		if (donor == NULL)
+		if (!NeonWALReaderUpdateDonor(state))
 		{
 			snprintf(state->err_msg, sizeof(state->err_msg),
 					 "failed to establish remote connection to fetch WAL: no donor available");
 			return NEON_WALREAD_ERROR;
+
 		}
-		snprintf(state->donor_name, sizeof(state->donor_name), "%s:%s", donor->host, donor->port);
-		nwr_log(LOG, "establishing connection to %s, flush_lsn %X/%X to fetch WAL",
-				state->donor_name, LSN_FORMAT_ARGS(donor_lsn));
-		state->wp_conn = libpqwp_connect_start(donor->conninfo);
+		/* no connection yet; start one */
+		nwr_log(LOG, "establishing connection to %s, lsn=%X/%X to fetch WAL", state->donor_name, LSN_FORMAT_ARGS(state->donor_lsn));
+		state->wp_conn = libpqwp_connect_start(state->donor_conninfo);
 		if (PQstatus(state->wp_conn->pg_conn) == CONNECTION_BAD)
 		{
 			snprintf(state->err_msg, sizeof(state->err_msg),
@@ -251,10 +242,22 @@ NeonWALReadRemote(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size cou
 				{
 					/* connection successfully established */
 					char		start_repl_query[128];
+					term_t		term = pg_atomic_read_u64(&GetWalpropShmemState()->mineLastElectedTerm);

+					/*
+					 * Set elected walproposer's term to pull only data from
+					 * its history. Note: for logical walsender it means we
+					 * might stream WAL not yet committed by safekeepers. It
+					 * would be cleaner to fix this.
+					 *
+					 * mineLastElectedTerm shouldn't be 0 at this point
+					 * because we checked above that donor exists and it
+					 * appears only after successfull election.
+					 */
+					Assert(term > 0);
 					snprintf(start_repl_query, sizeof(start_repl_query),
 							 "START_REPLICATION PHYSICAL %X/%X (term='" UINT64_FORMAT "')",
-							 LSN_FORMAT_ARGS(startptr), state->wp->propTerm);
+							 LSN_FORMAT_ARGS(startptr), term);
 					nwr_log(LOG, "connection to %s to fetch WAL succeeded, running %s",
 							state->donor_name, start_repl_query);
 					if (!libpqwp_send_query(state->wp_conn, start_repl_query))
@@ -404,6 +407,10 @@ NeonWALReadRemote(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size cou
 			state->req_lsn = InvalidXLogRecPtr;
 			state->req_len = 0;
 			state->req_progress = 0;
+
+			/* Update the current segment info. */
+			state->seg.ws_tli = tli;
+
 			return NEON_WALREAD_SUCCESS;
 		}
 	}
@@ -526,7 +533,7 @@ err:
 }

 /* reset remote connection and request in progress */
-static void
+void
 NeonWALReaderResetRemote(NeonWALReader *state)
 {
 	state->req_lsn = InvalidXLogRecPtr;
@@ -691,13 +698,25 @@ NeonWALReadLocal(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size coun
 	return true;
 }

+XLogRecPtr
+NeonWALReaderGetRemLsn(NeonWALReader *state)
+{
+	return state->rem_lsn;
+}
+
+const WALOpenSegment *
+NeonWALReaderGetSegment(NeonWALReader *state)
+{
+	return &state->seg;
+}
+
 /*
 * Copy of vanilla wal_segment_open, but returns false in case of error instead
 * of ERROR, with errno set.
 *
 * XLogReaderRoutine->segment_open callback for local pg_wal files
 */
-static bool
+bool
 neon_wal_segment_open(NeonWALReader *state, XLogSegNo nextSegNo,
 					  TimeLineID *tli_p)
 {
@@ -724,7 +743,7 @@ is_wal_segment_exists(XLogSegNo segno, int segsize, TimeLineID tli)
 }

 /* copy of vanilla wal_segment_close with NeonWALReader */
-static void
+void
 neon_wal_segment_close(NeonWALReader *state)
 {
 	if (state->seg.ws_file >= 0)
@@ -740,3 +759,19 @@ NeonWALReaderErrMsg(NeonWALReader *state)
 {
 	return state->err_msg;
 }
+
+/*
+ * Returns true if there is a donor, and false otherwise
+ */
+bool
+NeonWALReaderUpdateDonor(NeonWALReader *state)
+{
+	WalproposerShmemState *wps = GetWalpropShmemState();
+
+	SpinLockAcquire(&wps->mutex);
+	memcpy(state->donor_name, wps->donor_name, sizeof(state->donor_name));
+	memcpy(state->donor_conninfo, wps->donor_conninfo, sizeof(state->donor_conninfo));
+	state->donor_lsn = wps->donor_lsn;
+	SpinLockRelease(&wps->mutex);
+	return state->donor_name[0] != '\0';
+}
--- a/pgxn/neon/neon_walreader.h
+++ b/pgxn/neon/neon_walreader.h
@@ -19,12 +19,19 @@ typedef enum
 	NEON_WALREAD_ERROR,
 } NeonWALReadResult;

-extern NeonWALReader *NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, WalProposer *wp, char *log_prefix);
+extern NeonWALReader *NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, char *log_prefix);
 extern void NeonWALReaderFree(NeonWALReader *state);
+extern void NeonWALReaderResetRemote(NeonWALReader *state);
 extern NeonWALReadResult NeonWALRead(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size count, TimeLineID tli);
 extern pgsocket NeonWALReaderSocket(NeonWALReader *state);
 extern uint32 NeonWALReaderEvents(NeonWALReader *state);
 extern bool NeonWALReaderIsRemConnEstablished(NeonWALReader *state);
 extern char *NeonWALReaderErrMsg(NeonWALReader *state);
+extern XLogRecPtr NeonWALReaderGetRemLsn(NeonWALReader *state);
+extern const WALOpenSegment *NeonWALReaderGetSegment(NeonWALReader *state);
+extern bool neon_wal_segment_open(NeonWALReader *state, XLogSegNo nextSegNo, TimeLineID *tli_p);
+extern void neon_wal_segment_close(NeonWALReader *state);
+extern bool NeonWALReaderUpdateDonor(NeonWALReader *state);
+

 #endif							/* __NEON_WALREADER_H__ */
--- a/pgxn/neon/walproposer.c
+++ b/pgxn/neon/walproposer.c
@@ -80,7 +80,7 @@ static int	CompareLsn(const void *a, const void *b);
 static char *FormatSafekeeperState(Safekeeper *sk);
 static void AssertEventsOkForState(uint32 events, Safekeeper *sk);
 static char *FormatEvents(WalProposer *wp, uint32 events);
-
+static void UpdateDonorShmem(WalProposer *wp);

 WalProposer *
 WalProposerCreate(WalProposerConfig *config, walproposer_api api)
@@ -922,7 +922,8 @@ static void
 DetermineEpochStartLsn(WalProposer *wp)
 {
 	TermHistory *dth;
-	int          n_ready = 0;
+	int			n_ready = 0;
+	WalproposerShmemState *walprop_shared;

 	wp->propEpochStartLsn = InvalidXLogRecPtr;
 	wp->donorEpoch = 0;
@@ -964,16 +965,18 @@ DetermineEpochStartLsn(WalProposer *wp)
 	if (n_ready < wp->quorum)
 	{
 		/*
-		 * This is a rare case that can be triggered if safekeeper has voted and disconnected.
-		 * In this case, its state will not be SS_IDLE and its vote cannot be used, because
-		 * we clean up `voteResponse` in `ShutdownConnection`.
+		 * This is a rare case that can be triggered if safekeeper has voted
+		 * and disconnected. In this case, its state will not be SS_IDLE and
+		 * its vote cannot be used, because we clean up `voteResponse` in
+		 * `ShutdownConnection`.
 		 */
 		wp_log(FATAL, "missing majority of votes, collected %d, expected %d, got %d", wp->n_votes, wp->quorum, n_ready);
 	}

 	/*
-	 * If propEpochStartLsn is 0, it means flushLsn is 0 everywhere, we are bootstrapping
-	 * and nothing was committed yet. Start streaming then from the basebackup LSN.
+	 * If propEpochStartLsn is 0, it means flushLsn is 0 everywhere, we are
+	 * bootstrapping and nothing was committed yet. Start streaming then from
+	 * the basebackup LSN.
 	 */
 	if (wp->propEpochStartLsn == InvalidXLogRecPtr && !wp->config->syncSafekeepers)
 	{
@@ -984,11 +987,12 @@ DetermineEpochStartLsn(WalProposer *wp)
 		}
 		wp_log(LOG, "bumped epochStartLsn to the first record %X/%X", LSN_FORMAT_ARGS(wp->propEpochStartLsn));
 	}
+	pg_atomic_write_u64(&wp->api.get_shmem_state(wp)->propEpochStartLsn, wp->propEpochStartLsn);

 	/*
-	 * Safekeepers are setting truncateLsn after timelineStartLsn is known, so it
-	 * should never be zero at this point, if we know timelineStartLsn.
-	 * 
+	 * Safekeepers are setting truncateLsn after timelineStartLsn is known, so
+	 * it should never be zero at this point, if we know timelineStartLsn.
+	 *
 	 * timelineStartLsn can be zero only on the first syncSafekeepers run.
 	 */
 	Assert((wp->truncateLsn != InvalidXLogRecPtr) ||
@@ -1022,10 +1026,9 @@ DetermineEpochStartLsn(WalProposer *wp)
 	 * since which we are going to write according to the consensus. If not,
 	 * we must bail out, as clog and other non rel data is inconsistent.
 	 */
+	walprop_shared = wp->api.get_shmem_state(wp);
 	if (!wp->config->syncSafekeepers)
 	{
-		WalproposerShmemState *walprop_shared = wp->api.get_shmem_state(wp);
-
 		/*
 		 * Basebackup LSN always points to the beginning of the record (not
 		 * the page), as StartupXLOG most probably wants it this way.
@@ -1040,7 +1043,7 @@ DetermineEpochStartLsn(WalProposer *wp)
 			 * compute (who could generate WAL) is ok.
 			 */
 			if (!((dth->n_entries >= 1) && (dth->entries[dth->n_entries - 1].term ==
-											walprop_shared->mineLastElectedTerm)))
+											pg_atomic_read_u64(&walprop_shared->mineLastElectedTerm))))
 			{
 				/*
 				 * Panic to restart PG as we need to retake basebackup.
@@ -1054,8 +1057,8 @@ DetermineEpochStartLsn(WalProposer *wp)
 					   LSN_FORMAT_ARGS(wp->api.get_redo_start_lsn(wp)));
 			}
 		}
-		walprop_shared->mineLastElectedTerm = wp->propTerm;
 	}
+	pg_atomic_write_u64(&walprop_shared->mineLastElectedTerm, wp->propTerm);
 }

 /*
@@ -1105,9 +1108,13 @@ SendProposerElected(Safekeeper *sk)
 	{
 		/* safekeeper is empty or no common point, start from the beginning */
 		sk->startStreamingAt = wp->propTermHistory.entries[0].lsn;
-		wp_log(LOG, "no common point with sk %s:%s, streaming since first term at %X/%X, timelineStartLsn=%X/%X, termHistory.n_entries=%u" ,
-		 	 sk->host, sk->port, LSN_FORMAT_ARGS(sk->startStreamingAt), LSN_FORMAT_ARGS(wp->timelineStartLsn), wp->propTermHistory.n_entries);
-		/* wp->timelineStartLsn == InvalidXLogRecPtr can be only when timeline is created manually (test_s3_wal_replay) */
+		wp_log(LOG, "no common point with sk %s:%s, streaming since first term at %X/%X, timelineStartLsn=%X/%X, termHistory.n_entries=%u",
+			   sk->host, sk->port, LSN_FORMAT_ARGS(sk->startStreamingAt), LSN_FORMAT_ARGS(wp->timelineStartLsn), wp->propTermHistory.n_entries);
+
+		/*
+		 * wp->timelineStartLsn == InvalidXLogRecPtr can be only when timeline
+		 * is created manually (test_s3_wal_replay)
+		 */
 		Assert(sk->startStreamingAt == wp->timelineStartLsn || wp->timelineStartLsn == InvalidXLogRecPtr);
 	}
 	else
@@ -1177,6 +1184,12 @@ StartStreaming(Safekeeper *sk)
 	sk->active_state = SS_ACTIVE_SEND;
 	sk->streamingAt = sk->startStreamingAt;

+	/*
+	 * Donors can only be in SS_ACTIVE state, so we potentially update the
+	 * donor when we switch one to SS_ACTIVE.
+	 */
+	UpdateDonorShmem(sk->wp);
+
 	/* event set will be updated inside SendMessageToNode */
 	SendMessageToNode(sk);
 }
@@ -1568,17 +1581,17 @@ GetAcknowledgedByQuorumWALPosition(WalProposer *wp)
 * none if it doesn't exist. donor_lsn is set to end position of the donor to
 * the best of our knowledge.
 */
-Safekeeper *
-GetDonor(WalProposer *wp, XLogRecPtr *donor_lsn)
+static void
+UpdateDonorShmem(WalProposer *wp)
 {
 	Safekeeper *donor = NULL;
 	int			i;
-	*donor_lsn = InvalidXLogRecPtr;
+	XLogRecPtr	donor_lsn = InvalidXLogRecPtr;

 	if (wp->n_votes < wp->quorum)
 	{
-		wp_log(WARNING, "GetDonor called before elections are won");
-		return NULL;
+		wp_log(WARNING, "UpdateDonorShmem called before elections are won");
+		return;
 	}

 	/*
@@ -1589,7 +1602,7 @@ GetDonor(WalProposer *wp, XLogRecPtr *donor_lsn)
 	if (wp->safekeeper[wp->donor].state >= SS_IDLE)
 	{
 		donor = &wp->safekeeper[wp->donor];
-		*donor_lsn = wp->propEpochStartLsn;
+		donor_lsn = wp->propEpochStartLsn;
 	}

 	/*
@@ -1601,13 +1614,19 @@ GetDonor(WalProposer *wp, XLogRecPtr *donor_lsn)
 	{
 		Safekeeper *sk = &wp->safekeeper[i];

-		if (sk->state == SS_ACTIVE && sk->appendResponse.flushLsn > *donor_lsn)
+		if (sk->state == SS_ACTIVE && sk->appendResponse.flushLsn > donor_lsn)
 		{
 			donor = sk;
-			*donor_lsn = sk->appendResponse.flushLsn;
+			donor_lsn = sk->appendResponse.flushLsn;
 		}
 	}
-	return donor;
+
+	if (donor == NULL)
+	{
+		wp_log(WARNING, "UpdateDonorShmem didn't find a suitable donor, skipping");
+		return;
+	}
+	wp->api.update_donor(wp, donor, donor_lsn);
 }

 /*
@@ -1617,7 +1636,7 @@ static void
 HandleSafekeeperResponse(WalProposer *wp, Safekeeper *sk)
 {
 	XLogRecPtr	candidateTruncateLsn;
-	XLogRecPtr  newCommitLsn;
+	XLogRecPtr	newCommitLsn;

 	newCommitLsn = GetAcknowledgedByQuorumWALPosition(wp);
 	if (newCommitLsn > wp->commitLsn)
@@ -1627,7 +1646,7 @@ HandleSafekeeperResponse(WalProposer *wp, Safekeeper *sk)
 		BroadcastAppendRequest(wp);
 	}

-	/* 
+	/*
 	 * Unlock syncrep waiters, update ps_feedback, CheckGracefulShutdown().
 	 * The last one will terminate the process if the shutdown is requested
 	 * and WAL is committed by the quorum. BroadcastAppendRequest() should be
--- a/pgxn/neon/walproposer.h
+++ b/pgxn/neon/walproposer.h
@@ -284,14 +284,19 @@ typedef struct PageserverFeedback

 typedef struct WalproposerShmemState
 {
+	pg_atomic_uint64 propEpochStartLsn;
+	char		donor_name[64];
+	char		donor_conninfo[MAXCONNINFO];
+	XLogRecPtr	donor_lsn;
+
 	slock_t		mutex;
-	term_t		mineLastElectedTerm;
+	pg_atomic_uint64 mineLastElectedTerm;
 	pg_atomic_uint64 backpressureThrottlingTime;
 	pg_atomic_uint64 currentClusterSize;

 	/* last feedback from each shard */
 	PageserverFeedback shard_ps_feedback[MAX_SHARDS];
-	int num_shards;
+	int			num_shards;

 	/* aggregated feedback with min LSNs across shards */
 	PageserverFeedback min_ps_feedback;
@@ -465,6 +470,9 @@ typedef struct walproposer_api
 	/* Get pointer to the latest available WAL. */
 	XLogRecPtr	(*get_flush_rec_ptr) (WalProposer *wp);

+	/* Update current donor info in WalProposer Shmem */
+	void		(*update_donor) (WalProposer *wp, Safekeeper *donor, XLogRecPtr donor_lsn);
+
 	/* Get current time. */
 	TimestampTz (*get_current_timestamp) (WalProposer *wp);

@@ -497,7 +505,7 @@ typedef struct walproposer_api
 	 *
 	 * On success, the data is placed in *buf. It is valid until the next call
 	 * to this function.
-	 * 
+	 *
 	 * Returns PG_ASYNC_READ_FAIL on closed connection.
 	 */
 	PGAsyncReadResult (*conn_async_read) (Safekeeper *sk, char **buf, int *amount);
@@ -545,13 +553,14 @@ typedef struct walproposer_api
 	 * Returns 0 if timeout is reached, 1 if some event happened. Updates
 	 * events mask to indicate events and sets sk to the safekeeper which has
 	 * an event.
-	 * 
+	 *
 	 * On timeout, events is set to WL_NO_EVENTS. On socket event, events is
 	 * set to WL_SOCKET_READABLE and/or WL_SOCKET_WRITEABLE. When socket is
 	 * closed, events is set to WL_SOCKET_READABLE.
-	 * 
-	 * WL_SOCKET_WRITEABLE is usually set only when we need to flush the buffer.
-	 * It can be returned only if caller asked for this event in the last *_event_set call.
+	 *
+	 * WL_SOCKET_WRITEABLE is usually set only when we need to flush the
+	 * buffer. It can be returned only if caller asked for this event in the
+	 * last *_event_set call.
 	 */
 	int			(*wait_event_set) (WalProposer *wp, long timeout, Safekeeper **sk, uint32 *events);

@@ -571,9 +580,9 @@ typedef struct walproposer_api
 	void		(*finish_sync_safekeepers) (WalProposer *wp, XLogRecPtr lsn);

 	/*
-	 * Called after every AppendResponse from the safekeeper. Used to propagate
-	 * backpressure feedback and to confirm WAL persistence (has been commited
-	 * on the quorum of safekeepers).
+	 * Called after every AppendResponse from the safekeeper. Used to
+	 * propagate backpressure feedback and to confirm WAL persistence (has
+	 * been commited on the quorum of safekeepers).
 	 */
 	void		(*process_safekeeper_feedback) (WalProposer *wp, Safekeeper *sk);

@@ -716,12 +725,14 @@ extern void WalProposerBroadcast(WalProposer *wp, XLogRecPtr startpos, XLogRecPt
 extern void WalProposerPoll(WalProposer *wp);
 extern void WalProposerFree(WalProposer *wp);

+extern WalproposerShmemState *GetWalpropShmemState();
+
 /*
 * WaitEventSet API doesn't allow to remove socket, so walproposer_pg uses it to
 * recreate set from scratch, hence the export.
 */
 extern void SafekeeperStateDesiredEvents(Safekeeper *sk, uint32 *sk_events, uint32 *nwr_events);
-extern Safekeeper *GetDonor(WalProposer *wp, XLogRecPtr *donor_lsn);
+extern TimeLineID walprop_pg_get_timeline_id(void);


 #define WPEVENT		1337		/* special log level for walproposer internal
--- a/pgxn/neon/walproposer_pg.c
+++ b/pgxn/neon/walproposer_pg.c
@@ -85,7 +85,6 @@ static void walprop_pg_init_standalone_sync_safekeepers(void);
 static void walprop_pg_init_walsender(void);
 static void walprop_pg_init_bgworker(void);
 static TimestampTz walprop_pg_get_current_timestamp(WalProposer *wp);
-static TimeLineID walprop_pg_get_timeline_id(void);
 static void walprop_pg_load_libpqwalreceiver(void);

 static process_interrupts_callback_t PrevProcessInterruptsCallback;
@@ -94,6 +93,8 @@ static shmem_startup_hook_type prev_shmem_startup_hook_type;
 static shmem_request_hook_type prev_shmem_request_hook = NULL;
 static void walproposer_shmem_request(void);
 #endif
+static void WalproposerShmemInit_SyncSafekeeper(void);
+

 static void StartProposerReplication(WalProposer *wp, StartReplicationCmd *cmd);
 static void WalSndLoop(WalProposer *wp);
@@ -136,6 +137,7 @@ WalProposerSync(int argc, char *argv[])
 	WalProposer *wp;

 	init_walprop_config(true);
+	WalproposerShmemInit_SyncSafekeeper();
 	walprop_pg_init_standalone_sync_safekeepers();
 	walprop_pg_load_libpqwalreceiver();

@@ -281,6 +283,8 @@ WalproposerShmemInit(void)
 	{
 		memset(walprop_shared, 0, WalproposerShmemSize());
 		SpinLockInit(&walprop_shared->mutex);
+		pg_atomic_init_u64(&walprop_shared->propEpochStartLsn, 0);
+		pg_atomic_init_u64(&walprop_shared->mineLastElectedTerm, 0);
 		pg_atomic_init_u64(&walprop_shared->backpressureThrottlingTime, 0);
 		pg_atomic_init_u64(&walprop_shared->currentClusterSize, 0);
 	}
@@ -289,6 +293,17 @@ WalproposerShmemInit(void)
 	return found;
 }

+static void
+WalproposerShmemInit_SyncSafekeeper(void)
+{
+	walprop_shared = palloc(WalproposerShmemSize());
+	memset(walprop_shared, 0, WalproposerShmemSize());
+	SpinLockInit(&walprop_shared->mutex);
+	pg_atomic_init_u64(&walprop_shared->propEpochStartLsn, 0);
+	pg_atomic_init_u64(&walprop_shared->mineLastElectedTerm, 0);
+	pg_atomic_init_u64(&walprop_shared->backpressureThrottlingTime, 0);
+}
+
 #define BACK_PRESSURE_DELAY 10000L // 0.01 sec

 static bool
@@ -399,6 +414,13 @@ nwp_shmem_startup_hook(void)
 	WalproposerShmemInit();
 }

+WalproposerShmemState *
+GetWalpropShmemState()
+{
+	Assert(walprop_shared != NULL);
+	return walprop_shared;
+}
+
 static WalproposerShmemState *
 walprop_pg_get_shmem_state(WalProposer *wp)
 {
@@ -431,14 +453,15 @@ record_pageserver_feedback(PageserverFeedback *ps_feedback)
 	for (int i = 0; i < walprop_shared->num_shards; i++)
 	{
 		PageserverFeedback *feedback = &walprop_shared->shard_ps_feedback[i];
+
 		if (feedback->present)
 		{
 			if (min_feedback.last_received_lsn == InvalidXLogRecPtr || feedback->last_received_lsn < min_feedback.last_received_lsn)
 				min_feedback.last_received_lsn = feedback->last_received_lsn;
-			
+
 			if (min_feedback.disk_consistent_lsn == InvalidXLogRecPtr || feedback->disk_consistent_lsn < min_feedback.disk_consistent_lsn)
 				min_feedback.disk_consistent_lsn = feedback->disk_consistent_lsn;
-			
+
 			if (min_feedback.remote_consistent_lsn == InvalidXLogRecPtr || feedback->remote_consistent_lsn < min_feedback.remote_consistent_lsn)
 				min_feedback.remote_consistent_lsn = feedback->remote_consistent_lsn;
 		}
@@ -551,6 +574,7 @@ static void
 walprop_sigusr2(SIGNAL_ARGS)
 {
 	int			save_errno = errno;
+
 	got_SIGUSR2 = true;
 	SetLatch(MyLatch);
 	errno = save_errno;
@@ -598,7 +622,7 @@ walprop_pg_get_current_timestamp(WalProposer *wp)
 	return GetCurrentTimestamp();
 }

-static TimeLineID
+TimeLineID
 walprop_pg_get_timeline_id(void)
 {
 #if PG_VERSION_NUM >= 150000
@@ -617,6 +641,20 @@ walprop_pg_load_libpqwalreceiver(void)
 		wpg_log(ERROR, "libpqwalreceiver didn't initialize correctly");
 }

+static void
+walprop_pg_update_donor(WalProposer *wp, Safekeeper *donor, XLogRecPtr donor_lsn)
+{
+	WalproposerShmemState *wps = wp->api.get_shmem_state(wp);
+	char		donor_name[64];
+
+	pg_snprintf(donor_name, sizeof(donor_name), "%s:%s", donor->host, donor->port);
+	SpinLockAcquire(&wps->mutex);
+	memcpy(wps->donor_name, donor_name, sizeof(donor_name));
+	memcpy(wps->donor_conninfo, donor->conninfo, sizeof(donor->conninfo));
+	wps->donor_lsn = donor_lsn;
+	SpinLockRelease(&wps->mutex);
+}
+
 /* Helper function */
 static bool
 ensure_nonblocking_status(WalProposerConn *conn, bool is_nonblocking)
@@ -717,7 +755,6 @@ walprop_connect_start(Safekeeper *sk)
 {
 	Assert(sk->conn == NULL);
 	sk->conn = libpqwp_connect_start(sk->conninfo);
-
 }

 static WalProposerConnectPollStatusType
@@ -1091,7 +1128,7 @@ static void
 StartProposerReplication(WalProposer *wp, StartReplicationCmd *cmd)
 {
 	XLogRecPtr	FlushPtr;
-	 __attribute__((unused)) TimeLineID	currTLI;
+	__attribute__((unused)) TimeLineID currTLI;

 #if PG_VERSION_NUM < 150000
 	if (ThisTimeLineID == 0)
@@ -1295,116 +1332,13 @@ XLogBroadcastWalProposer(WalProposer *wp)
 	}
 }

-/* Download WAL before basebackup for logical walsenders from sk, if needed */
+/*
+  Used to download WAL before basebackup for logical walsenders from sk, no longer
+  needed because walsender always uses neon_walreader.
+ */
 static bool
 WalProposerRecovery(WalProposer *wp, Safekeeper *sk)
 {
-	char	   *err;
-	WalReceiverConn *wrconn;
-	WalRcvStreamOptions options;
-	char		conninfo[MAXCONNINFO];
-	TimeLineID	timeline;
-	XLogRecPtr	startpos;
-	XLogRecPtr	endpos;
-
-	startpos = GetLogRepRestartLSN(wp);
-	if (startpos == InvalidXLogRecPtr)
-		return true;			/* recovery not needed */
-	endpos = wp->propEpochStartLsn;
-
-	timeline = wp->greetRequest.timeline;
-
-	if (!neon_auth_token)
-	{
-		memcpy(conninfo, sk->conninfo, MAXCONNINFO);
-	}
-	else
-	{
-		int			written = 0;
-
-		written = snprintf((char *) conninfo, MAXCONNINFO, "password=%s %s", neon_auth_token, sk->conninfo);
-		if (written > MAXCONNINFO || written < 0)
-			wpg_log(FATAL, "could not append password to the safekeeper connection string");
-	}
-
-#if PG_MAJORVERSION_NUM < 16
-	wrconn = walrcv_connect(conninfo, false, "wal_proposer_recovery", &err);
-#else
-	wrconn = walrcv_connect(conninfo, false, false, "wal_proposer_recovery", &err);
-#endif
-
-	if (!wrconn)
-	{
-		ereport(WARNING,
-				(errmsg("could not connect to WAL acceptor %s:%s: %s",
-						sk->host, sk->port,
-						err)));
-		return false;
-	}
-	wpg_log(LOG,
-			"start recovery for logical replication from %s:%s starting from %X/%08X till %X/%08X timeline "
-			"%d",
-			sk->host, sk->port, (uint32) (startpos >> 32),
-			(uint32) startpos, (uint32) (endpos >> 32), (uint32) endpos, timeline);
-
-	options.logical = false;
-	options.startpoint = startpos;
-	options.slotname = NULL;
-	options.proto.physical.startpointTLI = timeline;
-
-	if (walrcv_startstreaming(wrconn, &options))
-	{
-		XLogRecPtr	rec_start_lsn;
-		XLogRecPtr	rec_end_lsn = 0;
-		int			len;
-		char	   *buf;
-		pgsocket	wait_fd = PGINVALID_SOCKET;
-
-		while ((len = walrcv_receive(wrconn, &buf, &wait_fd)) >= 0)
-		{
-			if (len == 0)
-			{
-				(void) WaitLatchOrSocket(
-										 MyLatch, WL_EXIT_ON_PM_DEATH | WL_SOCKET_READABLE, wait_fd,
-										 -1, WAIT_EVENT_WAL_RECEIVER_MAIN);
-			}
-			else
-			{
-				Assert(buf[0] == 'w' || buf[0] == 'k');
-				if (buf[0] == 'k')
-					continue;	/* keepalive */
-				memcpy(&rec_start_lsn, &buf[XLOG_HDR_START_POS],
-					   sizeof rec_start_lsn);
-				rec_start_lsn = pg_ntoh64(rec_start_lsn);
-				rec_end_lsn = rec_start_lsn + len - XLOG_HDR_SIZE;
-
-				/* write WAL to disk */
-				XLogWalPropWrite(sk->wp, &buf[XLOG_HDR_SIZE], len - XLOG_HDR_SIZE, rec_start_lsn);
-
-				ereport(DEBUG1,
-						(errmsg("Recover message %X/%X length %d",
-								LSN_FORMAT_ARGS(rec_start_lsn), len)));
-				if (rec_end_lsn >= endpos)
-					break;
-			}
-		}
-		ereport(LOG,
-				(errmsg("end of replication stream at %X/%X: %m",
-						LSN_FORMAT_ARGS(rec_end_lsn))));
-		walrcv_disconnect(wrconn);
-
-		/* failed to receive all WAL till endpos */
-		if (rec_end_lsn < endpos)
-			return false;
-	}
-	else
-	{
-		ereport(LOG,
-				(errmsg("primary server contains no more WAL on requested timeline %u LSN %X/%08X",
-						timeline, (uint32) (startpos >> 32), (uint32) startpos)));
-		return false;
-	}
-
 	return true;
 }

@@ -1545,7 +1479,7 @@ walprop_pg_wal_reader_allocate(Safekeeper *sk)

 	snprintf(log_prefix, sizeof(log_prefix), WP_LOG_PREFIX "sk %s:%s nwr: ", sk->host, sk->port);
 	Assert(!sk->xlogreader);
-	sk->xlogreader = NeonWALReaderAllocate(wal_segment_size, sk->wp->propEpochStartLsn, sk->wp, log_prefix);
+	sk->xlogreader = NeonWALReaderAllocate(wal_segment_size, sk->wp->propEpochStartLsn, log_prefix);
 	if (sk->xlogreader == NULL)
 		wpg_log(FATAL, "failed to allocate xlog reader");
 }
@@ -1960,8 +1894,8 @@ CombineHotStanbyFeedbacks(HotStandbyFeedback *hs, WalProposer *wp)
 static void
 walprop_pg_process_safekeeper_feedback(WalProposer *wp, Safekeeper *sk)
 {
-	HotStandbyFeedback	hsFeedback;
-	bool				needToAdvanceSlot = false;
+	HotStandbyFeedback hsFeedback;
+	bool		needToAdvanceSlot = false;

 	if (wp->config->syncSafekeepers)
 		return;
@@ -2095,22 +2029,25 @@ GetLogRepRestartLSN(WalProposer *wp)
 	return lrRestartLsn;
 }

-void SetNeonCurrentClusterSize(uint64 size)
+void
+SetNeonCurrentClusterSize(uint64 size)
 {
 	pg_atomic_write_u64(&walprop_shared->currentClusterSize, size);
 }

-uint64 GetNeonCurrentClusterSize(void)
+uint64
+GetNeonCurrentClusterSize(void)
 {
 	return pg_atomic_read_u64(&walprop_shared->currentClusterSize);
 }
-uint64 GetNeonCurrentClusterSize(void);
+uint64		GetNeonCurrentClusterSize(void);


 static const walproposer_api walprop_pg = {
 	.get_shmem_state = walprop_pg_get_shmem_state,
 	.start_streaming = walprop_pg_start_streaming,
 	.get_flush_rec_ptr = walprop_pg_get_flush_rec_ptr,
+	.update_donor = walprop_pg_update_donor,
 	.get_current_timestamp = walprop_pg_get_current_timestamp,
 	.conn_error_message = walprop_error_message,
 	.conn_status = walprop_status,
--- a/pgxn/neon/walsender_hooks.c
+++ b/pgxn/neon/walsender_hooks.c
@@ -0,0 +1,172 @@
+/*-------------------------------------------------------------------------
+ *
+ * walsender_hooks.c
+ *
+ * Implements XLogReaderRoutine in terms of NeonWALReader. Allows for
+ * fetching WAL from safekeepers, which normal xlogreader can't do.
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "walsender_hooks.h"
+#include "postgres.h"
+#include "fmgr.h"
+#include "access/xlogdefs.h"
+#include "replication/walsender.h"
+#include "access/xlog.h"
+#include "access/xlog_internal.h"
+#include "access/xlogreader.h"
+#include "miscadmin.h"
+#include "utils/wait_event.h"
+#include "utils/guc.h"
+#include "postmaster/interrupt.h"
+
+#include "neon_walreader.h"
+#include "walproposer.h"
+
+static NeonWALReader *wal_reader = NULL;
+extern XLogRecPtr WalSndWaitForWal(XLogRecPtr loc);
+extern bool GetDonorShmem(XLogRecPtr *donor_lsn);
+
+static XLogRecPtr
+NeonWALReadWaitForWAL(XLogRecPtr loc)
+{
+	while (!NeonWALReaderUpdateDonor(wal_reader))
+	{
+		pg_usleep(1000);
+		CHECK_FOR_INTERRUPTS();
+	}
+
+	return WalSndWaitForWal(loc);
+}
+
+static int
+NeonWALPageRead(
+				XLogReaderState *xlogreader,
+				XLogRecPtr targetPagePtr,
+				int reqLen,
+				XLogRecPtr targetRecPtr,
+				char *readBuf)
+{
+	XLogRecPtr	rem_lsn;
+
+	/* Wait for flush pointer to advance past our request */
+	XLogRecPtr	flushptr = NeonWALReadWaitForWAL(targetPagePtr + reqLen);
+	int			count;
+
+	if (flushptr < targetPagePtr + reqLen)
+		return -1;
+
+	/* Read at most XLOG_BLCKSZ bytes */
+	if (targetPagePtr + XLOG_BLCKSZ <= flushptr)
+		count = XLOG_BLCKSZ;
+	else
+		count = flushptr - targetPagePtr;
+
+	/*
+	 * Sometimes walsender requests non-monotonic sequences of WAL. If that's
+	 * the case, we have to reset streaming from remote at the correct
+	 * position. For example, walsender may try to verify the segment header
+	 * when trying to read in the middle of it.
+	 */
+	rem_lsn = NeonWALReaderGetRemLsn(wal_reader);
+	if (rem_lsn != InvalidXLogRecPtr && targetPagePtr != rem_lsn)
+	{
+		NeonWALReaderResetRemote(wal_reader);
+	}
+
+	for (;;)
+	{
+		NeonWALReadResult res = NeonWALRead(
+											wal_reader,
+											readBuf,
+											targetPagePtr,
+											count,
+											walprop_pg_get_timeline_id());
+
+		if (res == NEON_WALREAD_SUCCESS)
+		{
+			/*
+			 * Setting ws_tli is required by the XLogReaderRoutine, it is used
+			 * for segment name generation in error reports.
+			 *
+			 * ReadPageInternal updates ws_segno after calling cb on its own
+			 * and XLogReaderRoutine description doesn't require it, but
+			 * WALRead sets, let's follow it.
+			 */
+			xlogreader->seg.ws_tli = NeonWALReaderGetSegment(wal_reader)->ws_tli;
+			xlogreader->seg.ws_segno = NeonWALReaderGetSegment(wal_reader)->ws_segno;
+
+			/*
+			 * ws_file doesn't exist in case of remote read, and isn't used by
+			 * xlogreader except by WALRead on which we don't rely anyway.
+			 */
+			return count;
+		}
+		if (res == NEON_WALREAD_ERROR)
+		{
+			elog(ERROR, "[walsender] Failed to read WAL (req_lsn=%X/%X, len=%d): %s",
+				 LSN_FORMAT_ARGS(targetPagePtr),
+				 reqLen,
+				 NeonWALReaderErrMsg(wal_reader));
+			return -1;
+		}
+
+		/*
+		 * Res is WOULDBLOCK, so we wait on the socket, recreating event set
+		 * if necessary
+		 */
+		{
+
+			pgsocket	sock = NeonWALReaderSocket(wal_reader);
+			uint32_t	reader_events = NeonWALReaderEvents(wal_reader);
+			long		timeout_ms = 1000;
+
+			ResetLatch(MyLatch);
+			CHECK_FOR_INTERRUPTS();
+			if (ConfigReloadPending)
+			{
+				ConfigReloadPending = false;
+				ProcessConfigFile(PGC_SIGHUP);
+			}
+
+			WaitLatchOrSocket(
+							  MyLatch,
+							  WL_LATCH_SET | WL_EXIT_ON_PM_DEATH | reader_events,
+							  sock,
+							  timeout_ms,
+							  WAIT_EVENT_WAL_SENDER_MAIN);
+		}
+	}
+}
+
+static void
+NeonWALReadSegmentOpen(XLogReaderState *xlogreader, XLogSegNo nextSegNo, TimeLineID *tli_p)
+{
+	neon_wal_segment_open(wal_reader, nextSegNo, tli_p);
+	xlogreader->seg.ws_file = NeonWALReaderGetSegment(wal_reader)->ws_file;
+}
+
+static void
+NeonWALReadSegmentClose(XLogReaderState *xlogreader)
+{
+	neon_wal_segment_close(wal_reader);
+	xlogreader->seg.ws_file = NeonWALReaderGetSegment(wal_reader)->ws_file;
+}
+
+void
+NeonOnDemandXLogReaderRoutines(XLogReaderRoutine *xlr)
+{
+	if (!wal_reader)
+	{
+		XLogRecPtr	epochStartLsn = pg_atomic_read_u64(&GetWalpropShmemState()->propEpochStartLsn);
+
+		if (epochStartLsn == 0)
+		{
+			elog(ERROR, "Unable to start walsender when propEpochStartLsn is 0!");
+		}
+		wal_reader = NeonWALReaderAllocate(wal_segment_size, epochStartLsn, "[walsender] ");
+	}
+	xlr->page_read = NeonWALPageRead;
+	xlr->segment_open = NeonWALReadSegmentOpen;
+	xlr->segment_close = NeonWALReadSegmentClose;
+}
--- a/pgxn/neon/walsender_hooks.h
+++ b/pgxn/neon/walsender_hooks.h
@@ -0,0 +1,7 @@
+#ifndef __WALSENDER_HOOKS_H__
+#define __WALSENDER_HOOKS_H__
+
+struct XLogReaderRoutine;
+void		NeonOnDemandXLogReaderRoutines(struct XLogReaderRoutine *xlr);
+
+#endif
--- a/poetry.lock
+++ b/poetry.lock
@@ -158,6 +158,28 @@ files = [
 attrs = ">=16.0.0"
 pluggy = ">=0.4.0"

+[[package]]
+name = "annotated-types"
+version = "0.6.0"
+description = "Reusable constraint types to use with typing.Annotated"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "annotated_types-0.6.0-py3-none-any.whl", hash = "sha256:0641064de18ba7a25dee8f96403ebc39113d0cb953a01429249d5c7564666a43"},
+    {file = "annotated_types-0.6.0.tar.gz", hash = "sha256:563339e807e53ffd9c267e99fc6d9ea23eb8443c08f112651963e24e22f84a5d"},
+]
+
+[[package]]
+name = "antlr4-python3-runtime"
+version = "4.13.1"
+description = "ANTLR 4.13.1 runtime for Python 3"
+optional = false
+python-versions = "*"
+files = [
+    {file = "antlr4-python3-runtime-4.13.1.tar.gz", hash = "sha256:3cd282f5ea7cfb841537fe01f143350fdb1c0b1ce7981443a2fa8513fddb6d1a"},
+    {file = "antlr4_python3_runtime-4.13.1-py3-none-any.whl", hash = "sha256:78ec57aad12c97ac039ca27403ad61cb98aaec8a3f9bb8144f889aa0fa28b943"},
+]
+
 [[package]]
 name = "anyio"
 version = "4.3.0"
@@ -267,22 +289,23 @@ tests-no-zope = ["cloudpickle", "coverage[toml] (>=5.0.2)", "hypothesis", "mypy"

 [[package]]
 name = "aws-sam-translator"
-version = "1.48.0"
+version = "1.88.0"
 description = "AWS SAM Translator is a library that transform SAM templates into AWS CloudFormation templates"
 optional = false
-python-versions = ">=3.7, <=4.0, !=4.0"
+python-versions = "!=4.0,<=4.0,>=3.8"
 files = [
-    {file = "aws-sam-translator-1.48.0.tar.gz", hash = "sha256:7171037323dfa30f8f73e9bccb9210e4c384a585e087219a9518a5204f0a2c44"},
-    {file = "aws_sam_translator-1.48.0-py2-none-any.whl", hash = "sha256:be18dfa3dfe7ab291d281667c5f73ac62dbe6bfe86df7d122e4258b906b736f0"},
-    {file = "aws_sam_translator-1.48.0-py3-none-any.whl", hash = "sha256:ca4f8f9910d7713aeaba59346775bfb3198f6acb47c6704572f9bd3fc0fb5bf0"},
+    {file = "aws_sam_translator-1.88.0-py3-none-any.whl", hash = "sha256:aa93d498d8de3fb3d485c316155b1628144b823bbc176099a20de06df666fcac"},
+    {file = "aws_sam_translator-1.88.0.tar.gz", hash = "sha256:e77c65f3488566122277accd44a0f1ec018e37403e0d5fe25120d96e537e91a7"},
 ]

 [package.dependencies]
 boto3 = ">=1.19.5,<2.dev0"
-jsonschema = ">=3.2,<4.0"
+jsonschema = ">=3.2,<5"
+pydantic = ">=1.8,<3"
+typing-extensions = ">=4.4"

 [package.extras]
-dev = ["black (==20.8b1)", "boto3 (>=1.23,<2)", "click (>=7.1,<8.0)", "coverage (>=5.3,<6.0)", "dateparser (>=0.7,<1.0)", "docopt (>=0.6.2,<0.7.0)", "flake8 (>=3.8.4,<3.9.0)", "parameterized (>=0.7.4,<0.8.0)", "pylint (>=2.9.0,<2.10.0)", "pytest (>=6.2.5,<6.3.0)", "pytest-cov (>=2.10.1,<2.11.0)", "pytest-env (>=0.6.2,<0.7.0)", "pytest-xdist (>=2.5,<3.0)", "pyyaml (>=5.4,<6.0)", "requests (>=2.24.0,<2.25.0)", "tenacity (>=7.0.0,<7.1.0)", "tox (>=3.24,<4.0)"]
+dev = ["black (==24.3.0)", "boto3 (>=1.23,<2)", "boto3-stubs[appconfig,serverlessrepo] (>=1.19.5,<2.dev0)", "coverage (>=5.3,<8)", "dateparser (>=1.1,<2.0)", "mypy (>=1.3.0,<1.4.0)", "parameterized (>=0.7,<1.0)", "pytest (>=6.2,<8)", "pytest-cov (>=2.10,<5)", "pytest-env (>=0.6,<1)", "pytest-rerunfailures (>=9.1,<12)", "pytest-xdist (>=2.5,<4)", "pyyaml (>=6.0,<7.0)", "requests (>=2.28,<3.0)", "ruamel.yaml (==0.17.21)", "ruff (>=0.1.0,<0.2.0)", "tenacity (>=8.0,<9.0)", "types-PyYAML (>=6.0,<7.0)", "types-jsonschema (>=3.2,<4.0)"]

 [[package]]
 name = "aws-xray-sdk"
@@ -798,24 +821,26 @@ pycparser = "*"

 [[package]]
 name = "cfn-lint"
-version = "0.61.3"
+version = "0.87.1"
 description = "Checks CloudFormation templates for practices and behaviour that could potentially be improved"
 optional = false
-python-versions = ">=3.6, <=4.0, !=4.0"
+python-versions = "!=4.0,<=4.0,>=3.8"
 files = [
-    {file = "cfn-lint-0.61.3.tar.gz", hash = "sha256:3806e010d77901f5e935496df690c10e39676434a738fce1a1161cf9c7bd36a2"},
-    {file = "cfn_lint-0.61.3-py3-none-any.whl", hash = "sha256:8e9522fad0c7c98b31ecbdd4724f8d8a5787457cc0f71e62ae0d11104d6e52ab"},
+    {file = "cfn_lint-0.87.1-py3-none-any.whl", hash = "sha256:d450f450635fc223b6f66880ccac52a5fd1a52966fa1705f1ba52b88dfed3071"},
+    {file = "cfn_lint-0.87.1.tar.gz", hash = "sha256:b3ce9d3e5e0eadcea5d584c8ccaa00bf2a990a36a64d7ffd8683bc60b7e4f06f"},
 ]

 [package.dependencies]
-aws-sam-translator = ">=1.47.0"
+aws-sam-translator = ">=1.87.0"
 jschema-to-python = ">=1.2.3,<1.3.0"
 jsonpatch = "*"
-jsonschema = ">=3.0,<4.0"
+jsonschema = ">=3.0,<5"
 junit-xml = ">=1.9,<2.0"
-networkx = ">=2.4,<3.0"
+networkx = ">=2.4,<4"
 pyyaml = ">5.4"
+regex = ">=2021.7.1"
 sarif-om = ">=1.0.4,<1.1.0"
+sympy = ">=1.0.0"

 [[package]]
 name = "charset-normalizer"
@@ -931,24 +956,6 @@ websocket-client = ">=0.32.0"
 ssh = ["paramiko (>=2.4.2)"]
 tls = ["cryptography (>=1.3.4)", "idna (>=2.0.0)", "pyOpenSSL (>=17.5.0)"]

-[[package]]
-name = "ecdsa"
-version = "0.18.0"
-description = "ECDSA cryptographic signature library (pure python)"
-optional = false
-python-versions = ">=2.6, !=3.0.*, !=3.1.*, !=3.2.*"
-files = [
-    {file = "ecdsa-0.18.0-py2.py3-none-any.whl", hash = "sha256:80600258e7ed2f16b9aa1d7c295bd70194109ad5a30fdee0eaeefef1d4c559dd"},
-    {file = "ecdsa-0.18.0.tar.gz", hash = "sha256:190348041559e21b22a1d65cee485282ca11a6f81d503fddb84d5017e9ed1e49"},
-]
-
-[package.dependencies]
-six = ">=1.9.0"
-
-[package.extras]
-gmpy = ["gmpy"]
-gmpy2 = ["gmpy2"]
-
 [[package]]
 name = "exceptiongroup"
 version = "1.1.1"
@@ -1001,18 +1008,17 @@ dotenv = ["python-dotenv"]

 [[package]]
 name = "flask-cors"
-version = "3.0.10"
+version = "4.0.1"
 description = "A Flask extension adding a decorator for CORS support"
 optional = false
 python-versions = "*"
 files = [
-    {file = "Flask-Cors-3.0.10.tar.gz", hash = "sha256:b60839393f3b84a0f3746f6cdca56c1ad7426aa738b70d6c61375857823181de"},
-    {file = "Flask_Cors-3.0.10-py2.py3-none-any.whl", hash = "sha256:74efc975af1194fc7891ff5cd85b0f7478be4f7f59fe158102e91abb72bb4438"},
+    {file = "Flask_Cors-4.0.1-py2.py3-none-any.whl", hash = "sha256:f2a704e4458665580c074b714c4627dd5a306b333deb9074d0b1794dfa2fb677"},
+    {file = "flask_cors-4.0.1.tar.gz", hash = "sha256:eeb69b342142fdbf4766ad99357a7f3876a2ceb77689dc10ff912aac06c389e4"},
 ]

 [package.dependencies]
 Flask = ">=0.9"
-Six = "*"

 [[package]]
 name = "frozenlist"
@@ -1243,13 +1249,13 @@ files = [

 [[package]]
 name = "jinja2"
-version = "3.1.3"
+version = "3.1.4"
 description = "A very fast and expressive template engine."
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "Jinja2-3.1.3-py3-none-any.whl", hash = "sha256:7d6d50dd97d52cbc355597bd845fabfbac3f551e1f99619e39a35ce8c370b5fa"},
-    {file = "Jinja2-3.1.3.tar.gz", hash = "sha256:ac8bd6544d4bb2c9792bf3a159e80bba8fda7f07e81bc3aed565432d5925ba90"},
+    {file = "jinja2-3.1.4-py3-none-any.whl", hash = "sha256:bc5dd2abb727a5319567b7a813e6a2e7318c39f4f487cfe6c89c6f9c7d25197d"},
+    {file = "jinja2-3.1.4.tar.gz", hash = "sha256:4a3aee7acbbe7303aede8e9648d13b8bf88a429282aa6122a993f0ac800cb369"},
 ]

 [package.dependencies]
@@ -1269,6 +1275,23 @@ files = [
    {file = "jmespath-1.0.1.tar.gz", hash = "sha256:90261b206d6defd58fdd5e85f478bf633a2901798906be2ad389150c5c60edbe"},
 ]

+[[package]]
+name = "joserfc"
+version = "0.9.0"
+description = "The ultimate Python library for JOSE RFCs, including JWS, JWE, JWK, JWA, JWT"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "joserfc-0.9.0-py3-none-any.whl", hash = "sha256:4026bdbe2c196cd40574e916fa1e28874d99649412edaab0e373dec3077153fb"},
+    {file = "joserfc-0.9.0.tar.gz", hash = "sha256:eebca7f587b1761ce43a98ffd5327f2b600b9aa5bb0a77b947687f503ad43bc0"},
+]
+
+[package.dependencies]
+cryptography = "*"
+
+[package.extras]
+drafts = ["pycryptodome"]
+
 [[package]]
 name = "jschema-to-python"
 version = "1.2.3"
@@ -1310,6 +1333,20 @@ files = [
 [package.dependencies]
 jsonpointer = ">=1.9"

+[[package]]
+name = "jsonpath-ng"
+version = "1.6.1"
+description = "A final implementation of JSONPath for Python that aims to be standard compliant, including arithmetic and binary comparison operators and providing clear AST for metaprogramming."
+optional = false
+python-versions = "*"
+files = [
+    {file = "jsonpath-ng-1.6.1.tar.gz", hash = "sha256:086c37ba4917304850bd837aeab806670224d3f038fe2833ff593a672ef0a5fa"},
+    {file = "jsonpath_ng-1.6.1-py3-none-any.whl", hash = "sha256:8f22cd8273d7772eea9aaa84d922e0841aa36fdb8a2c6b7f6c3791a16a9bc0be"},
+]
+
+[package.dependencies]
+ply = "*"
+
 [[package]]
 name = "jsonpickle"
 version = "2.2.0"
@@ -1339,24 +1376,39 @@ files = [

 [[package]]
 name = "jsonschema"
-version = "3.2.0"
+version = "4.17.3"
 description = "An implementation of JSON Schema validation for Python"
 optional = false
-python-versions = "*"
+python-versions = ">=3.7"
 files = [
-    {file = "jsonschema-3.2.0-py2.py3-none-any.whl", hash = "sha256:4e5b3cf8216f577bee9ce139cbe72eca3ea4f292ec60928ff24758ce626cd163"},
-    {file = "jsonschema-3.2.0.tar.gz", hash = "sha256:c8a85b28d377cc7737e46e2d9f2b4f44ee3c0e1deac6bf46ddefc7187d30797a"},
+    {file = "jsonschema-4.17.3-py3-none-any.whl", hash = "sha256:a870ad254da1a8ca84b6a2905cac29d265f805acc57af304784962a2aa6508f6"},
+    {file = "jsonschema-4.17.3.tar.gz", hash = "sha256:0f864437ab8b6076ba6707453ef8f98a6a0d512a80e93f8abdb676f737ecb60d"},
 ]

 [package.dependencies]
 attrs = ">=17.4.0"
-pyrsistent = ">=0.14.0"
-setuptools = "*"
-six = ">=1.11.0"
+pyrsistent = ">=0.14.0,<0.17.0 || >0.17.0,<0.17.1 || >0.17.1,<0.17.2 || >0.17.2"

 [package.extras]
-format = ["idna", "jsonpointer (>1.13)", "rfc3987", "strict-rfc3339", "webcolors"]
-format-nongpl = ["idna", "jsonpointer (>1.13)", "rfc3339-validator", "rfc3986-validator (>0.1.0)", "webcolors"]
+format = ["fqdn", "idna", "isoduration", "jsonpointer (>1.13)", "rfc3339-validator", "rfc3987", "uri-template", "webcolors (>=1.11)"]
+format-nongpl = ["fqdn", "idna", "isoduration", "jsonpointer (>1.13)", "rfc3339-validator", "rfc3986-validator (>0.1.0)", "uri-template", "webcolors (>=1.11)"]
+
+[[package]]
+name = "jsonschema-spec"
+version = "0.1.6"
+description = "JSONSchema Spec with object-oriented paths"
+optional = false
+python-versions = ">=3.7.0,<4.0.0"
+files = [
+    {file = "jsonschema_spec-0.1.6-py3-none-any.whl", hash = "sha256:f2206d18c89d1824c1f775ba14ed039743b41a9167bd2c5bdb774b66b3ca0bbf"},
+    {file = "jsonschema_spec-0.1.6.tar.gz", hash = "sha256:90215863b56e212086641956b20127ccbf6d8a3a38343dad01d6a74d19482f76"},
+]
+
+[package.dependencies]
+jsonschema = ">=4.0.0,<4.18.0"
+pathable = ">=0.4.1,<0.5.0"
+PyYAML = ">=5.1"
+requests = ">=2.31.0,<3.0.0"

 [[package]]
 name = "junit-xml"
@@ -1372,6 +1424,52 @@ files = [
 [package.dependencies]
 six = "*"

+[[package]]
+name = "lazy-object-proxy"
+version = "1.10.0"
+description = "A fast and thorough lazy object proxy."
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "lazy-object-proxy-1.10.0.tar.gz", hash = "sha256:78247b6d45f43a52ef35c25b5581459e85117225408a4128a3daf8bf9648ac69"},
+    {file = "lazy_object_proxy-1.10.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:855e068b0358ab916454464a884779c7ffa312b8925c6f7401e952dcf3b89977"},
+    {file = "lazy_object_proxy-1.10.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7ab7004cf2e59f7c2e4345604a3e6ea0d92ac44e1c2375527d56492014e690c3"},
+    {file = "lazy_object_proxy-1.10.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:dc0d2fc424e54c70c4bc06787e4072c4f3b1aa2f897dfdc34ce1013cf3ceef05"},
+    {file = "lazy_object_proxy-1.10.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:e2adb09778797da09d2b5ebdbceebf7dd32e2c96f79da9052b2e87b6ea495895"},
+    {file = "lazy_object_proxy-1.10.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:b1f711e2c6dcd4edd372cf5dec5c5a30d23bba06ee012093267b3376c079ec83"},
+    {file = "lazy_object_proxy-1.10.0-cp310-cp310-win32.whl", hash = "sha256:76a095cfe6045c7d0ca77db9934e8f7b71b14645f0094ffcd842349ada5c5fb9"},
+    {file = "lazy_object_proxy-1.10.0-cp310-cp310-win_amd64.whl", hash = "sha256:b4f87d4ed9064b2628da63830986c3d2dca7501e6018347798313fcf028e2fd4"},
+    {file = "lazy_object_proxy-1.10.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:fec03caabbc6b59ea4a638bee5fce7117be8e99a4103d9d5ad77f15d6f81020c"},
+    {file = "lazy_object_proxy-1.10.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:02c83f957782cbbe8136bee26416686a6ae998c7b6191711a04da776dc9e47d4"},
+    {file = "lazy_object_proxy-1.10.0-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:009e6bb1f1935a62889ddc8541514b6a9e1fcf302667dcb049a0be5c8f613e56"},
+    {file = "lazy_object_proxy-1.10.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:75fc59fc450050b1b3c203c35020bc41bd2695ed692a392924c6ce180c6f1dc9"},
+    {file = "lazy_object_proxy-1.10.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:782e2c9b2aab1708ffb07d4bf377d12901d7a1d99e5e410d648d892f8967ab1f"},
+    {file = "lazy_object_proxy-1.10.0-cp311-cp311-win32.whl", hash = "sha256:edb45bb8278574710e68a6b021599a10ce730d156e5b254941754a9cc0b17d03"},
+    {file = "lazy_object_proxy-1.10.0-cp311-cp311-win_amd64.whl", hash = "sha256:e271058822765ad5e3bca7f05f2ace0de58a3f4e62045a8c90a0dfd2f8ad8cc6"},
+    {file = "lazy_object_proxy-1.10.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:e98c8af98d5707dcdecc9ab0863c0ea6e88545d42ca7c3feffb6b4d1e370c7ba"},
+    {file = "lazy_object_proxy-1.10.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:952c81d415b9b80ea261d2372d2a4a2332a3890c2b83e0535f263ddfe43f0d43"},
+    {file = "lazy_object_proxy-1.10.0-cp312-cp312-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:80b39d3a151309efc8cc48675918891b865bdf742a8616a337cb0090791a0de9"},
+    {file = "lazy_object_proxy-1.10.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:e221060b701e2aa2ea991542900dd13907a5c90fa80e199dbf5a03359019e7a3"},
+    {file = "lazy_object_proxy-1.10.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:92f09ff65ecff3108e56526f9e2481b8116c0b9e1425325e13245abfd79bdb1b"},
+    {file = "lazy_object_proxy-1.10.0-cp312-cp312-win32.whl", hash = "sha256:3ad54b9ddbe20ae9f7c1b29e52f123120772b06dbb18ec6be9101369d63a4074"},
+    {file = "lazy_object_proxy-1.10.0-cp312-cp312-win_amd64.whl", hash = "sha256:127a789c75151db6af398b8972178afe6bda7d6f68730c057fbbc2e96b08d282"},
+    {file = "lazy_object_proxy-1.10.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:9e4ed0518a14dd26092614412936920ad081a424bdcb54cc13349a8e2c6d106a"},
+    {file = "lazy_object_proxy-1.10.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5ad9e6ed739285919aa9661a5bbed0aaf410aa60231373c5579c6b4801bd883c"},
+    {file = "lazy_object_proxy-1.10.0-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2fc0a92c02fa1ca1e84fc60fa258458e5bf89d90a1ddaeb8ed9cc3147f417255"},
+    {file = "lazy_object_proxy-1.10.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:0aefc7591920bbd360d57ea03c995cebc204b424524a5bd78406f6e1b8b2a5d8"},
+    {file = "lazy_object_proxy-1.10.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:5faf03a7d8942bb4476e3b62fd0f4cf94eaf4618e304a19865abf89a35c0bbee"},
+    {file = "lazy_object_proxy-1.10.0-cp38-cp38-win32.whl", hash = "sha256:e333e2324307a7b5d86adfa835bb500ee70bfcd1447384a822e96495796b0ca4"},
+    {file = "lazy_object_proxy-1.10.0-cp38-cp38-win_amd64.whl", hash = "sha256:cb73507defd385b7705c599a94474b1d5222a508e502553ef94114a143ec6696"},
+    {file = "lazy_object_proxy-1.10.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:366c32fe5355ef5fc8a232c5436f4cc66e9d3e8967c01fb2e6302fd6627e3d94"},
+    {file = "lazy_object_proxy-1.10.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2297f08f08a2bb0d32a4265e98a006643cd7233fb7983032bd61ac7a02956b3b"},
+    {file = "lazy_object_proxy-1.10.0-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:18dd842b49456aaa9a7cf535b04ca4571a302ff72ed8740d06b5adcd41fe0757"},
+    {file = "lazy_object_proxy-1.10.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:217138197c170a2a74ca0e05bddcd5f1796c735c37d0eee33e43259b192aa424"},
+    {file = "lazy_object_proxy-1.10.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:9a3a87cf1e133e5b1994144c12ca4aa3d9698517fe1e2ca82977781b16955658"},
+    {file = "lazy_object_proxy-1.10.0-cp39-cp39-win32.whl", hash = "sha256:30b339b2a743c5288405aa79a69e706a06e02958eab31859f7f3c04980853b70"},
+    {file = "lazy_object_proxy-1.10.0-cp39-cp39-win_amd64.whl", hash = "sha256:a899b10e17743683b293a729d3a11f2f399e8a90c73b089e29f5d0fe3509f0dd"},
+    {file = "lazy_object_proxy-1.10.0-pp310.pp311.pp312.pp38.pp39-none-any.whl", hash = "sha256:80fa48bd89c8f2f456fc0765c11c23bf5af827febacd2f523ca5bc1893fcc09d"},
+]
+
 [[package]]
 name = "markupsafe"
 version = "2.1.1"
@@ -1423,64 +1521,80 @@ files = [

 [[package]]
 name = "moto"
-version = "4.1.2"
+version = "5.0.6"
 description = ""
 optional = false
-python-versions = ">=3.7"
+python-versions = ">=3.8"
 files = [
-    {file = "moto-4.1.2-py2.py3-none-any.whl", hash = "sha256:1b361ece638c74a657325378a259276f368aafce2f8be84f8143e69fa93ce8ec"},
-    {file = "moto-4.1.2.tar.gz", hash = "sha256:63431733d2a02c7bd652ad71ec1da442a0e0d580cbac5eeb50d440a2ce066eac"},
+    {file = "moto-5.0.6-py2.py3-none-any.whl", hash = "sha256:ca1e22831a741733b581ff2ef4d6ae2e1c6db1eab97af1b78b86ca2c6e88c609"},
+    {file = "moto-5.0.6.tar.gz", hash = "sha256:ad8b23f2b555ad694da8b2432a42b6d96beaaf67a4e7d932196a72193a2eee2c"},
 ]

 [package.dependencies]
+antlr4-python3-runtime = {version = "*", optional = true, markers = "extra == \"server\""}
 aws-xray-sdk = {version = ">=0.93,<0.96 || >0.96", optional = true, markers = "extra == \"server\""}
 boto3 = ">=1.9.201"
-botocore = ">=1.12.201"
+botocore = ">=1.14.0"
 cfn-lint = {version = ">=0.40.0", optional = true, markers = "extra == \"server\""}
 cryptography = ">=3.3.1"
-docker = {version = ">=2.5.1", optional = true, markers = "extra == \"server\""}
-ecdsa = {version = "!=0.15", optional = true, markers = "extra == \"server\""}
+docker = {version = ">=3.0.0", optional = true, markers = "extra == \"server\""}
 flask = {version = "<2.2.0 || >2.2.0,<2.2.1 || >2.2.1", optional = true, markers = "extra == \"server\""}
 flask-cors = {version = "*", optional = true, markers = "extra == \"server\""}
 graphql-core = {version = "*", optional = true, markers = "extra == \"server\""}
 Jinja2 = ">=2.10.1"
+joserfc = {version = ">=0.9.0", optional = true, markers = "extra == \"server\""}
 jsondiff = {version = ">=1.1.2", optional = true, markers = "extra == \"server\""}
-openapi-spec-validator = {version = ">=0.2.8", optional = true, markers = "extra == \"server\""}
+jsonpath-ng = {version = "*", optional = true, markers = "extra == \"server\""}
+openapi-spec-validator = {version = ">=0.5.0", optional = true, markers = "extra == \"server\""}
+py-partiql-parser = {version = "0.5.4", optional = true, markers = "extra == \"server\""}
 pyparsing = {version = ">=3.0.7", optional = true, markers = "extra == \"server\""}
 python-dateutil = ">=2.1,<3.0.0"
-python-jose = {version = ">=3.1.0,<4.0.0", extras = ["cryptography"], optional = true, markers = "extra == \"server\""}
 PyYAML = {version = ">=5.1", optional = true, markers = "extra == \"server\""}
 requests = ">=2.5"
-responses = ">=0.13.0"
+responses = ">=0.15.0"
 setuptools = {version = "*", optional = true, markers = "extra == \"server\""}
-sshpubkeys = {version = ">=3.1.0", optional = true, markers = "extra == \"server\""}
 werkzeug = ">=0.5,<2.2.0 || >2.2.0,<2.2.1 || >2.2.1"
 xmltodict = "*"

 [package.extras]
-all = ["PyYAML (>=5.1)", "aws-xray-sdk (>=0.93,!=0.96)", "cfn-lint (>=0.40.0)", "docker (>=2.5.1)", "ecdsa (!=0.15)", "graphql-core", "jsondiff (>=1.1.2)", "openapi-spec-validator (>=0.2.8)", "pyparsing (>=3.0.7)", "python-jose[cryptography] (>=3.1.0,<4.0.0)", "setuptools", "sshpubkeys (>=3.1.0)"]
-apigateway = ["PyYAML (>=5.1)", "ecdsa (!=0.15)", "openapi-spec-validator (>=0.2.8)", "python-jose[cryptography] (>=3.1.0,<4.0.0)"]
-apigatewayv2 = ["PyYAML (>=5.1)"]
+all = ["PyYAML (>=5.1)", "antlr4-python3-runtime", "aws-xray-sdk (>=0.93,!=0.96)", "cfn-lint (>=0.40.0)", "docker (>=3.0.0)", "graphql-core", "joserfc (>=0.9.0)", "jsondiff (>=1.1.2)", "jsonpath-ng", "multipart", "openapi-spec-validator (>=0.5.0)", "py-partiql-parser (==0.5.4)", "pyparsing (>=3.0.7)", "setuptools"]
+apigateway = ["PyYAML (>=5.1)", "joserfc (>=0.9.0)", "openapi-spec-validator (>=0.5.0)"]
+apigatewayv2 = ["PyYAML (>=5.1)", "openapi-spec-validator (>=0.5.0)"]
 appsync = ["graphql-core"]
-awslambda = ["docker (>=2.5.1)"]
-batch = ["docker (>=2.5.1)"]
-cloudformation = ["PyYAML (>=5.1)", "aws-xray-sdk (>=0.93,!=0.96)", "cfn-lint (>=0.40.0)", "docker (>=2.5.1)", "ecdsa (!=0.15)", "graphql-core", "jsondiff (>=1.1.2)", "openapi-spec-validator (>=0.2.8)", "pyparsing (>=3.0.7)", "python-jose[cryptography] (>=3.1.0,<4.0.0)", "setuptools", "sshpubkeys (>=3.1.0)"]
-cognitoidp = ["ecdsa (!=0.15)", "python-jose[cryptography] (>=3.1.0,<4.0.0)"]
-ds = ["sshpubkeys (>=3.1.0)"]
-dynamodb = ["docker (>=2.5.1)"]
-dynamodbstreams = ["docker (>=2.5.1)"]
-ebs = ["sshpubkeys (>=3.1.0)"]
-ec2 = ["sshpubkeys (>=3.1.0)"]
-efs = ["sshpubkeys (>=3.1.0)"]
-eks = ["sshpubkeys (>=3.1.0)"]
+awslambda = ["docker (>=3.0.0)"]
+batch = ["docker (>=3.0.0)"]
+cloudformation = ["PyYAML (>=5.1)", "aws-xray-sdk (>=0.93,!=0.96)", "cfn-lint (>=0.40.0)", "docker (>=3.0.0)", "graphql-core", "joserfc (>=0.9.0)", "jsondiff (>=1.1.2)", "openapi-spec-validator (>=0.5.0)", "py-partiql-parser (==0.5.4)", "pyparsing (>=3.0.7)", "setuptools"]
+cognitoidp = ["joserfc (>=0.9.0)"]
+dynamodb = ["docker (>=3.0.0)", "py-partiql-parser (==0.5.4)"]
+dynamodbstreams = ["docker (>=3.0.0)", "py-partiql-parser (==0.5.4)"]
 glue = ["pyparsing (>=3.0.7)"]
 iotdata = ["jsondiff (>=1.1.2)"]
-route53resolver = ["sshpubkeys (>=3.1.0)"]
-s3 = ["PyYAML (>=5.1)"]
-server = ["PyYAML (>=5.1)", "aws-xray-sdk (>=0.93,!=0.96)", "cfn-lint (>=0.40.0)", "docker (>=2.5.1)", "ecdsa (!=0.15)", "flask (!=2.2.0,!=2.2.1)", "flask-cors", "graphql-core", "jsondiff (>=1.1.2)", "openapi-spec-validator (>=0.2.8)", "pyparsing (>=3.0.7)", "python-jose[cryptography] (>=3.1.0,<4.0.0)", "setuptools", "sshpubkeys (>=3.1.0)"]
+proxy = ["PyYAML (>=5.1)", "antlr4-python3-runtime", "aws-xray-sdk (>=0.93,!=0.96)", "cfn-lint (>=0.40.0)", "docker (>=2.5.1)", "graphql-core", "joserfc (>=0.9.0)", "jsondiff (>=1.1.2)", "jsonpath-ng", "multipart", "openapi-spec-validator (>=0.5.0)", "py-partiql-parser (==0.5.4)", "pyparsing (>=3.0.7)", "setuptools"]
+resourcegroupstaggingapi = ["PyYAML (>=5.1)", "cfn-lint (>=0.40.0)", "docker (>=3.0.0)", "graphql-core", "joserfc (>=0.9.0)", "jsondiff (>=1.1.2)", "openapi-spec-validator (>=0.5.0)", "py-partiql-parser (==0.5.4)", "pyparsing (>=3.0.7)"]
+s3 = ["PyYAML (>=5.1)", "py-partiql-parser (==0.5.4)"]
+s3crc32c = ["PyYAML (>=5.1)", "crc32c", "py-partiql-parser (==0.5.4)"]
+server = ["PyYAML (>=5.1)", "antlr4-python3-runtime", "aws-xray-sdk (>=0.93,!=0.96)", "cfn-lint (>=0.40.0)", "docker (>=3.0.0)", "flask (!=2.2.0,!=2.2.1)", "flask-cors", "graphql-core", "joserfc (>=0.9.0)", "jsondiff (>=1.1.2)", "jsonpath-ng", "openapi-spec-validator (>=0.5.0)", "py-partiql-parser (==0.5.4)", "pyparsing (>=3.0.7)", "setuptools"]
 ssm = ["PyYAML (>=5.1)"]
+stepfunctions = ["antlr4-python3-runtime", "jsonpath-ng"]
 xray = ["aws-xray-sdk (>=0.93,!=0.96)", "setuptools"]

+[[package]]
+name = "mpmath"
+version = "1.3.0"
+description = "Python library for arbitrary-precision floating-point arithmetic"
+optional = false
+python-versions = "*"
+files = [
+    {file = "mpmath-1.3.0-py3-none-any.whl", hash = "sha256:a0b2b9fe80bbcd81a6647ff13108738cfb482d481d826cc0e02f5b35e5c88d2c"},
+    {file = "mpmath-1.3.0.tar.gz", hash = "sha256:7a28eb2a9774d00c7bc92411c19a89209d5da7c4c9a9e227be8330a23a25b91f"},
+]
+
+[package.extras]
+develop = ["codecov", "pycodestyle", "pytest (>=4.6)", "pytest-cov", "wheel"]
+docs = ["sphinx"]
+gmpy = ["gmpy2 (>=2.1.0a4)"]
+tests = ["pytest (>=4.6)"]
+
 [[package]]
 name = "multidict"
 version = "6.0.4"
@@ -1655,42 +1769,38 @@ test = ["codecov (>=2.1)", "pytest (>=7.1)", "pytest-cov (>=3.0)"]

 [[package]]
 name = "openapi-schema-validator"
-version = "0.2.3"
+version = "0.4.4"
 description = "OpenAPI schema validation for Python"
 optional = false
 python-versions = ">=3.7.0,<4.0.0"
 files = [
-    {file = "openapi-schema-validator-0.2.3.tar.gz", hash = "sha256:2c64907728c3ef78e23711c8840a423f0b241588c9ed929855e4b2d1bb0cf5f2"},
-    {file = "openapi_schema_validator-0.2.3-py3-none-any.whl", hash = "sha256:9bae709212a19222892cabcc60cafd903cbf4b220223f48583afa3c0e3cc6fc4"},
+    {file = "openapi_schema_validator-0.4.4-py3-none-any.whl", hash = "sha256:79f37f38ef9fd5206b924ed7a6f382cea7b649b3b56383c47f1906082b7b9015"},
+    {file = "openapi_schema_validator-0.4.4.tar.gz", hash = "sha256:c573e2be2c783abae56c5a1486ab716ca96e09d1c3eab56020d1dc680aa57bf8"},
 ]

 [package.dependencies]
-jsonschema = ">=3.0.0,<5.0.0"
+jsonschema = ">=4.0.0,<4.18.0"
+rfc3339-validator = "*"

 [package.extras]
-isodate = ["isodate"]
-rfc3339-validator = ["rfc3339-validator"]
-strict-rfc3339 = ["strict-rfc3339"]
+docs = ["sphinx (>=5.3.0,<6.0.0)", "sphinx-immaterial (>=0.11.0,<0.12.0)"]

 [[package]]
 name = "openapi-spec-validator"
-version = "0.4.0"
-description = "OpenAPI 2.0 (aka Swagger) and OpenAPI 3.0 spec validator"
+version = "0.5.7"
+description = "OpenAPI 2.0 (aka Swagger) and OpenAPI 3 spec validator"
 optional = false
 python-versions = ">=3.7.0,<4.0.0"
 files = [
-    {file = "openapi-spec-validator-0.4.0.tar.gz", hash = "sha256:97f258850afc97b048f7c2653855e0f88fa66ac103c2be5077c7960aca2ad49a"},
-    {file = "openapi_spec_validator-0.4.0-py3-none-any.whl", hash = "sha256:06900ac4d546a1df3642a779da0055be58869c598e3042a2fef067cfd99d04d0"},
+    {file = "openapi_spec_validator-0.5.7-py3-none-any.whl", hash = "sha256:8712d2879db7692974ef89c47a3ebfc79436442921ec3a826ac0ce80cde8c549"},
+    {file = "openapi_spec_validator-0.5.7.tar.gz", hash = "sha256:6c2d42180045a80fd6314de848b94310bdb0fa4949f4b099578b69f79d9fa5ac"},
 ]

 [package.dependencies]
-jsonschema = ">=3.2.0,<5.0.0"
-openapi-schema-validator = ">=0.2.0,<0.3.0"
-PyYAML = ">=5.1"
-setuptools = "*"
-
-[package.extras]
-requests = ["requests"]
+jsonschema = ">=4.0.0,<4.18.0"
+jsonschema-spec = ">=0.1.1,<0.2.0"
+lazy-object-proxy = ">=1.7.1,<2.0.0"
+openapi-schema-validator = ">=0.4.2,<0.5.0"

 [[package]]
 name = "packaging"
@@ -1703,6 +1813,17 @@ files = [
    {file = "packaging-23.0.tar.gz", hash = "sha256:b6ad297f8907de0fa2fe1ccbd26fdaf387f5f47c7275fedf8cce89f99446cf97"},
 ]

+[[package]]
+name = "pathable"
+version = "0.4.3"
+description = "Object-oriented paths"
+optional = false
+python-versions = ">=3.7.0,<4.0.0"
+files = [
+    {file = "pathable-0.4.3-py3-none-any.whl", hash = "sha256:cdd7b1f9d7d5c8b8d3315dbf5a86b2596053ae845f056f57d97c0eefff84da14"},
+    {file = "pathable-0.4.3.tar.gz", hash = "sha256:5c869d315be50776cc8a993f3af43e0c60dc01506b399643f919034ebf4cdcab"},
+]
+
 [[package]]
 name = "pbr"
 version = "5.9.0"
@@ -1729,6 +1850,17 @@ files = [
 dev = ["pre-commit", "tox"]
 testing = ["pytest", "pytest-benchmark"]

+[[package]]
+name = "ply"
+version = "3.11"
+description = "Python Lex & Yacc"
+optional = false
+python-versions = "*"
+files = [
+    {file = "ply-3.11-py2.py3-none-any.whl", hash = "sha256:096f9b8350b65ebd2fd1346b12452efe5b9607f7482813ffca50c22722a807ce"},
+    {file = "ply-3.11.tar.gz", hash = "sha256:00c7c1aaa88358b9c765b6d3000c6eec0ba42abca5351b095321aef446081da3"},
+]
+
 [[package]]
 name = "prometheus-client"
 version = "0.14.1"
@@ -1841,16 +1973,19 @@ files = [
 ]

 [[package]]
-name = "pyasn1"
-version = "0.4.8"
-description = "ASN.1 types and codecs"
+name = "py-partiql-parser"
+version = "0.5.4"
+description = "Pure Python PartiQL Parser"
 optional = false
 python-versions = "*"
 files = [
-    {file = "pyasn1-0.4.8-py2.py3-none-any.whl", hash = "sha256:39c7e2ec30515947ff4e87fb6f456dfc6e84857d34be479c9d4a4ba4bf46aa5d"},
-    {file = "pyasn1-0.4.8.tar.gz", hash = "sha256:aef77c9fb94a3ac588e87841208bdec464471d9871bd5050a287cc9a475cd0ba"},
+    {file = "py_partiql_parser-0.5.4-py2.py3-none-any.whl", hash = "sha256:3dc4295a47da9587681a96b35c6e151886fdbd0a4acbe0d97c4c68e5f689d315"},
+    {file = "py_partiql_parser-0.5.4.tar.gz", hash = "sha256:72e043919538fa63edae72fb59afc7e3fd93adbde656718a7d2b4666f23dd114"},
 ]

+[package.extras]
+dev = ["black (==22.6.0)", "flake8", "mypy", "pytest"]
+
 [[package]]
 name = "pycparser"
 version = "2.21"
@@ -1862,6 +1997,116 @@ files = [
    {file = "pycparser-2.21.tar.gz", hash = "sha256:e644fdec12f7872f86c58ff790da456218b10f863970249516d60a5eaca77206"},
 ]

+[[package]]
+name = "pydantic"
+version = "2.7.1"
+description = "Data validation using Python type hints"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "pydantic-2.7.1-py3-none-any.whl", hash = "sha256:e029badca45266732a9a79898a15ae2e8b14840b1eabbb25844be28f0b33f3d5"},
+    {file = "pydantic-2.7.1.tar.gz", hash = "sha256:e9dbb5eada8abe4d9ae5f46b9939aead650cd2b68f249bb3a8139dbe125803cc"},
+]
+
+[package.dependencies]
+annotated-types = ">=0.4.0"
+pydantic-core = "2.18.2"
+typing-extensions = ">=4.6.1"
+
+[package.extras]
+email = ["email-validator (>=2.0.0)"]
+
+[[package]]
+name = "pydantic-core"
+version = "2.18.2"
+description = "Core functionality for Pydantic validation and serialization"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "pydantic_core-2.18.2-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:9e08e867b306f525802df7cd16c44ff5ebbe747ff0ca6cf3fde7f36c05a59a81"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:f0a21cbaa69900cbe1a2e7cad2aa74ac3cf21b10c3efb0fa0b80305274c0e8a2"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:0680b1f1f11fda801397de52c36ce38ef1c1dc841a0927a94f226dea29c3ae3d"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:95b9d5e72481d3780ba3442eac863eae92ae43a5f3adb5b4d0a1de89d42bb250"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:c4fcf5cd9c4b655ad666ca332b9a081112cd7a58a8b5a6ca7a3104bc950f2038"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:9b5155ff768083cb1d62f3e143b49a8a3432e6789a3abee8acd005c3c7af1c74"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:553ef617b6836fc7e4df130bb851e32fe357ce36336d897fd6646d6058d980af"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:b89ed9eb7d616ef5714e5590e6cf7f23b02d0d539767d33561e3675d6f9e3857"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:75f7e9488238e920ab6204399ded280dc4c307d034f3924cd7f90a38b1829563"},
+    {file = "pydantic_core-2.18.2-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:ef26c9e94a8c04a1b2924149a9cb081836913818e55681722d7f29af88fe7b38"},
+    {file = "pydantic_core-2.18.2-cp310-none-win32.whl", hash = "sha256:182245ff6b0039e82b6bb585ed55a64d7c81c560715d1bad0cbad6dfa07b4027"},
+    {file = "pydantic_core-2.18.2-cp310-none-win_amd64.whl", hash = "sha256:e23ec367a948b6d812301afc1b13f8094ab7b2c280af66ef450efc357d2ae543"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:219da3f096d50a157f33645a1cf31c0ad1fe829a92181dd1311022f986e5fbe3"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:cc1cfd88a64e012b74e94cd00bbe0f9c6df57049c97f02bb07d39e9c852e19a4"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:05b7133a6e6aeb8df37d6f413f7705a37ab4031597f64ab56384c94d98fa0e90"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:224c421235f6102e8737032483f43c1a8cfb1d2f45740c44166219599358c2cd"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:b14d82cdb934e99dda6d9d60dc84a24379820176cc4a0d123f88df319ae9c150"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:2728b01246a3bba6de144f9e3115b532ee44bd6cf39795194fb75491824a1413"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:470b94480bb5ee929f5acba6995251ada5e059a5ef3e0dfc63cca287283ebfa6"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:997abc4df705d1295a42f95b4eec4950a37ad8ae46d913caeee117b6b198811c"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:75250dbc5290e3f1a0f4618db35e51a165186f9034eff158f3d490b3fed9f8a0"},
+    {file = "pydantic_core-2.18.2-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:4456f2dca97c425231d7315737d45239b2b51a50dc2b6f0c2bb181fce6207664"},
+    {file = "pydantic_core-2.18.2-cp311-none-win32.whl", hash = "sha256:269322dcc3d8bdb69f054681edff86276b2ff972447863cf34c8b860f5188e2e"},
+    {file = "pydantic_core-2.18.2-cp311-none-win_amd64.whl", hash = "sha256:800d60565aec896f25bc3cfa56d2277d52d5182af08162f7954f938c06dc4ee3"},
+    {file = "pydantic_core-2.18.2-cp311-none-win_arm64.whl", hash = "sha256:1404c69d6a676245199767ba4f633cce5f4ad4181f9d0ccb0577e1f66cf4c46d"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-macosx_10_12_x86_64.whl", hash = "sha256:fb2bd7be70c0fe4dfd32c951bc813d9fe6ebcbfdd15a07527796c8204bd36242"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:6132dd3bd52838acddca05a72aafb6eab6536aa145e923bb50f45e78b7251043"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d7d904828195733c183d20a54230c0df0eb46ec746ea1a666730787353e87182"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:c9bd70772c720142be1020eac55f8143a34ec9f82d75a8e7a07852023e46617f"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:2b8ed04b3582771764538f7ee7001b02e1170223cf9b75dff0bc698fadb00cf3"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:e6dac87ddb34aaec85f873d737e9d06a3555a1cc1a8e0c44b7f8d5daeb89d86f"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7ca4ae5a27ad7a4ee5170aebce1574b375de390bc01284f87b18d43a3984df72"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:886eec03591b7cf058467a70a87733b35f44707bd86cf64a615584fd72488b7c"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:ca7b0c1f1c983e064caa85f3792dd2fe3526b3505378874afa84baf662e12241"},
+    {file = "pydantic_core-2.18.2-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:4b4356d3538c3649337df4074e81b85f0616b79731fe22dd11b99499b2ebbdf3"},
+    {file = "pydantic_core-2.18.2-cp312-none-win32.whl", hash = "sha256:8b172601454f2d7701121bbec3425dd71efcb787a027edf49724c9cefc14c038"},
+    {file = "pydantic_core-2.18.2-cp312-none-win_amd64.whl", hash = "sha256:b1bd7e47b1558ea872bd16c8502c414f9e90dcf12f1395129d7bb42a09a95438"},
+    {file = "pydantic_core-2.18.2-cp312-none-win_arm64.whl", hash = "sha256:98758d627ff397e752bc339272c14c98199c613f922d4a384ddc07526c86a2ec"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-macosx_10_12_x86_64.whl", hash = "sha256:9fdad8e35f278b2c3eb77cbdc5c0a49dada440657bf738d6905ce106dc1de439"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:1d90c3265ae107f91a4f279f4d6f6f1d4907ac76c6868b27dc7fb33688cfb347"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:390193c770399861d8df9670fb0d1874f330c79caaca4642332df7c682bf6b91"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:82d5d4d78e4448683cb467897fe24e2b74bb7b973a541ea1dcfec1d3cbce39fb"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:4774f3184d2ef3e14e8693194f661dea5a4d6ca4e3dc8e39786d33a94865cefd"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:d4d938ec0adf5167cb335acb25a4ee69a8107e4984f8fbd2e897021d9e4ca21b"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e0e8b1be28239fc64a88a8189d1df7fad8be8c1ae47fcc33e43d4be15f99cc70"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:868649da93e5a3d5eacc2b5b3b9235c98ccdbfd443832f31e075f54419e1b96b"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:78363590ef93d5d226ba21a90a03ea89a20738ee5b7da83d771d283fd8a56761"},
+    {file = "pydantic_core-2.18.2-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:852e966fbd035a6468fc0a3496589b45e2208ec7ca95c26470a54daed82a0788"},
+    {file = "pydantic_core-2.18.2-cp38-none-win32.whl", hash = "sha256:6a46e22a707e7ad4484ac9ee9f290f9d501df45954184e23fc29408dfad61350"},
+    {file = "pydantic_core-2.18.2-cp38-none-win_amd64.whl", hash = "sha256:d91cb5ea8b11607cc757675051f61b3d93f15eca3cefb3e6c704a5d6e8440f4e"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-macosx_10_12_x86_64.whl", hash = "sha256:ae0a8a797a5e56c053610fa7be147993fe50960fa43609ff2a9552b0e07013e8"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:042473b6280246b1dbf530559246f6842b56119c2926d1e52b631bdc46075f2a"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1a388a77e629b9ec814c1b1e6b3b595fe521d2cdc625fcca26fbc2d44c816804"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:e25add29b8f3b233ae90ccef2d902d0ae0432eb0d45370fe315d1a5cf231004b"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f459a5ce8434614dfd39bbebf1041952ae01da6bed9855008cb33b875cb024c0"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:eff2de745698eb46eeb51193a9f41d67d834d50e424aef27df2fcdee1b153845"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a8309f67285bdfe65c372ea3722b7a5642680f3dba538566340a9d36e920b5f0"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:f93a8a2e3938ff656a7c1bc57193b1319960ac015b6e87d76c76bf14fe0244b4"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:22057013c8c1e272eb8d0eebc796701167d8377441ec894a8fed1af64a0bf399"},
+    {file = "pydantic_core-2.18.2-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:cfeecd1ac6cc1fb2692c3d5110781c965aabd4ec5d32799773ca7b1456ac636b"},
+    {file = "pydantic_core-2.18.2-cp39-none-win32.whl", hash = "sha256:0d69b4c2f6bb3e130dba60d34c0845ba31b69babdd3f78f7c0c8fae5021a253e"},
+    {file = "pydantic_core-2.18.2-cp39-none-win_amd64.whl", hash = "sha256:d9319e499827271b09b4e411905b24a426b8fb69464dfa1696258f53a3334641"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-macosx_10_12_x86_64.whl", hash = "sha256:a1874c6dd4113308bd0eb568418e6114b252afe44319ead2b4081e9b9521fe75"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-macosx_11_0_arm64.whl", hash = "sha256:ccdd111c03bfd3666bd2472b674c6899550e09e9f298954cfc896ab92b5b0e6d"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e18609ceaa6eed63753037fc06ebb16041d17d28199ae5aba0052c51449650a9"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6e5c584d357c4e2baf0ff7baf44f4994be121e16a2c88918a5817331fc7599d7"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:43f0f463cf89ace478de71a318b1b4f05ebc456a9b9300d027b4b57c1a2064fb"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-musllinux_1_1_aarch64.whl", hash = "sha256:e1b395e58b10b73b07b7cf740d728dd4ff9365ac46c18751bf8b3d8cca8f625a"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-musllinux_1_1_x86_64.whl", hash = "sha256:0098300eebb1c837271d3d1a2cd2911e7c11b396eac9661655ee524a7f10587b"},
+    {file = "pydantic_core-2.18.2-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:36789b70d613fbac0a25bb07ab3d9dba4d2e38af609c020cf4d888d165ee0bf3"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:3f9a801e7c8f1ef8718da265bba008fa121243dfe37c1cea17840b0944dfd72c"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-macosx_11_0_arm64.whl", hash = "sha256:3a6515ebc6e69d85502b4951d89131ca4e036078ea35533bb76327f8424531ce"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:20aca1e2298c56ececfd8ed159ae4dde2df0781988c97ef77d5c16ff4bd5b400"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:223ee893d77a310a0391dca6df00f70bbc2f36a71a895cecd9a0e762dc37b349"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:2334ce8c673ee93a1d6a65bd90327588387ba073c17e61bf19b4fd97d688d63c"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-musllinux_1_1_aarch64.whl", hash = "sha256:cbca948f2d14b09d20268cda7b0367723d79063f26c4ffc523af9042cad95592"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-musllinux_1_1_x86_64.whl", hash = "sha256:b3ef08e20ec49e02d5c6717a91bb5af9b20f1805583cb0adfe9ba2c6b505b5ae"},
+    {file = "pydantic_core-2.18.2-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:c6fdc8627910eed0c01aed6a390a252fe3ea6d472ee70fdde56273f198938374"},
+    {file = "pydantic_core-2.18.2.tar.gz", hash = "sha256:2e29d20810dfc3043ee13ac7d9e25105799817683348823f305ab3f349b9386e"},
+]
+
+[package.dependencies]
+typing-extensions = ">=4.6.0,<4.7.0 || >4.7.0"
+
 [[package]]
 name = "pyjwt"
 version = "2.4.0"
@@ -2116,28 +2361,6 @@ files = [
 [package.dependencies]
 six = ">=1.5"

-[[package]]
-name = "python-jose"
-version = "3.3.0"
-description = "JOSE implementation in Python"
-optional = false
-python-versions = "*"
-files = [
-    {file = "python-jose-3.3.0.tar.gz", hash = "sha256:55779b5e6ad599c6336191246e95eb2293a9ddebd555f796a65f838f07e5d78a"},
-    {file = "python_jose-3.3.0-py2.py3-none-any.whl", hash = "sha256:9b1376b023f8b298536eedd47ae1089bcdb848f1535ab30555cd92002d78923a"},
-]
-
-[package.dependencies]
-cryptography = {version = ">=3.4.0", optional = true, markers = "extra == \"cryptography\""}
-ecdsa = "!=0.15"
-pyasn1 = "*"
-rsa = "*"
-
-[package.extras]
-cryptography = ["cryptography (>=3.4.0)"]
-pycrypto = ["pyasn1", "pycrypto (>=2.6.0,<2.7.0)"]
-pycryptodome = ["pyasn1", "pycryptodome (>=3.3.1,<4.0.0)"]
-
 [[package]]
 name = "pywin32"
 version = "301"
@@ -2182,7 +2405,6 @@ files = [
    {file = "PyYAML-6.0.1-cp311-cp311-win_amd64.whl", hash = "sha256:bf07ee2fef7014951eeb99f56f39c9bb4af143d8aa3c21b1677805985307da34"},
    {file = "PyYAML-6.0.1-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:855fb52b0dc35af121542a76b9a84f8d1cd886ea97c84703eaa6d88e37a2ad28"},
    {file = "PyYAML-6.0.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:40df9b996c2b73138957fe23a16a4f0ba614f4c0efce1e9406a184b6d07fa3a9"},
-    {file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a08c6f0fe150303c1c6b71ebcd7213c2858041a7e01975da3a99aed1e7a378ef"},
    {file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6c22bec3fbe2524cde73d7ada88f6566758a8f7227bfbf93a408a9d86bcc12a0"},
    {file = "PyYAML-6.0.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:8d4e9c88387b0f5c7d5f281e55304de64cf7f9c0021a3525bd3b1c542da3b0e4"},
    {file = "PyYAML-6.0.1-cp312-cp312-win32.whl", hash = "sha256:d483d2cdf104e7c9fa60c544d92981f12ad66a457afae824d146093b8c294c54"},
@@ -2217,6 +2439,94 @@ files = [
    {file = "PyYAML-6.0.1.tar.gz", hash = "sha256:bfdf460b1736c775f2ba9f6a92bca30bc2095067b8a9d77876d1fad6cc3b4a43"},
 ]

+[[package]]
+name = "regex"
+version = "2024.4.28"
+description = "Alternative regular expression module, to replace re."
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "regex-2024.4.28-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:cd196d056b40af073d95a2879678585f0b74ad35190fac04ca67954c582c6b61"},
+    {file = "regex-2024.4.28-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:8bb381f777351bd534462f63e1c6afb10a7caa9fa2a421ae22c26e796fe31b1f"},
+    {file = "regex-2024.4.28-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:47af45b6153522733aa6e92543938e97a70ce0900649ba626cf5aad290b737b6"},
+    {file = "regex-2024.4.28-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:99d6a550425cc51c656331af0e2b1651e90eaaa23fb4acde577cf15068e2e20f"},
+    {file = "regex-2024.4.28-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:bf29304a8011feb58913c382902fde3395957a47645bf848eea695839aa101b7"},
+    {file = "regex-2024.4.28-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:92da587eee39a52c91aebea8b850e4e4f095fe5928d415cb7ed656b3460ae79a"},
+    {file = "regex-2024.4.28-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6277d426e2f31bdbacb377d17a7475e32b2d7d1f02faaecc48d8e370c6a3ff31"},
+    {file = "regex-2024.4.28-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:28e1f28d07220c0f3da0e8fcd5a115bbb53f8b55cecf9bec0c946eb9a059a94c"},
+    {file = "regex-2024.4.28-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:aaa179975a64790c1f2701ac562b5eeb733946eeb036b5bcca05c8d928a62f10"},
+    {file = "regex-2024.4.28-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:6f435946b7bf7a1b438b4e6b149b947c837cb23c704e780c19ba3e6855dbbdd3"},
+    {file = "regex-2024.4.28-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:19d6c11bf35a6ad077eb23852827f91c804eeb71ecb85db4ee1386825b9dc4db"},
+    {file = "regex-2024.4.28-cp310-cp310-musllinux_1_1_ppc64le.whl", hash = "sha256:fdae0120cddc839eb8e3c15faa8ad541cc6d906d3eb24d82fb041cfe2807bc1e"},
+    {file = "regex-2024.4.28-cp310-cp310-musllinux_1_1_s390x.whl", hash = "sha256:e672cf9caaf669053121f1766d659a8813bd547edef6e009205378faf45c67b8"},
+    {file = "regex-2024.4.28-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:f57515750d07e14743db55d59759893fdb21d2668f39e549a7d6cad5d70f9fea"},
+    {file = "regex-2024.4.28-cp310-cp310-win32.whl", hash = "sha256:a1409c4eccb6981c7baabc8888d3550df518add6e06fe74fa1d9312c1838652d"},
+    {file = "regex-2024.4.28-cp310-cp310-win_amd64.whl", hash = "sha256:1f687a28640f763f23f8a9801fe9e1b37338bb1ca5d564ddd41619458f1f22d1"},
+    {file = "regex-2024.4.28-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:84077821c85f222362b72fdc44f7a3a13587a013a45cf14534df1cbbdc9a6796"},
+    {file = "regex-2024.4.28-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:b45d4503de8f4f3dc02f1d28a9b039e5504a02cc18906cfe744c11def942e9eb"},
+    {file = "regex-2024.4.28-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:457c2cd5a646dd4ed536c92b535d73548fb8e216ebee602aa9f48e068fc393f3"},
+    {file = "regex-2024.4.28-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2b51739ddfd013c6f657b55a508de8b9ea78b56d22b236052c3a85a675102dc6"},
+    {file = "regex-2024.4.28-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:459226445c7d7454981c4c0ce0ad1a72e1e751c3e417f305722bbcee6697e06a"},
+    {file = "regex-2024.4.28-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:670fa596984b08a4a769491cbdf22350431970d0112e03d7e4eeaecaafcd0fec"},
+    {file = "regex-2024.4.28-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fe00f4fe11c8a521b173e6324d862ee7ee3412bf7107570c9b564fe1119b56fb"},
+    {file = "regex-2024.4.28-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:36f392dc7763fe7924575475736bddf9ab9f7a66b920932d0ea50c2ded2f5636"},
+    {file = "regex-2024.4.28-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:23a412b7b1a7063f81a742463f38821097b6a37ce1e5b89dd8e871d14dbfd86b"},
+    {file = "regex-2024.4.28-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:f1d6e4b7b2ae3a6a9df53efbf199e4bfcff0959dbdb5fd9ced34d4407348e39a"},
+    {file = "regex-2024.4.28-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:499334ad139557de97cbc4347ee921c0e2b5e9c0f009859e74f3f77918339257"},
+    {file = "regex-2024.4.28-cp311-cp311-musllinux_1_1_s390x.whl", hash = "sha256:0940038bec2fe9e26b203d636c44d31dd8766abc1fe66262da6484bd82461ccf"},
+    {file = "regex-2024.4.28-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:66372c2a01782c5fe8e04bff4a2a0121a9897e19223d9eab30c54c50b2ebeb7f"},
+    {file = "regex-2024.4.28-cp311-cp311-win32.whl", hash = "sha256:c77d10ec3c1cf328b2f501ca32583625987ea0f23a0c2a49b37a39ee5c4c4630"},
+    {file = "regex-2024.4.28-cp311-cp311-win_amd64.whl", hash = "sha256:fc0916c4295c64d6890a46e02d4482bb5ccf33bf1a824c0eaa9e83b148291f90"},
+    {file = "regex-2024.4.28-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:08a1749f04fee2811c7617fdd46d2e46d09106fa8f475c884b65c01326eb15c5"},
+    {file = "regex-2024.4.28-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:b8eb28995771c087a73338f695a08c9abfdf723d185e57b97f6175c5051ff1ae"},
+    {file = "regex-2024.4.28-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:dd7ef715ccb8040954d44cfeff17e6b8e9f79c8019daae2fd30a8806ef5435c0"},
+    {file = "regex-2024.4.28-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:fb0315a2b26fde4005a7c401707c5352df274460f2f85b209cf6024271373013"},
+    {file = "regex-2024.4.28-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f2fc053228a6bd3a17a9b0a3f15c3ab3cf95727b00557e92e1cfe094b88cc662"},
+    {file = "regex-2024.4.28-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:7fe9739a686dc44733d52d6e4f7b9c77b285e49edf8570754b322bca6b85b4cc"},
+    {file = "regex-2024.4.28-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a74fcf77d979364f9b69fcf8200849ca29a374973dc193a7317698aa37d8b01c"},
+    {file = "regex-2024.4.28-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:965fd0cf4694d76f6564896b422724ec7b959ef927a7cb187fc6b3f4e4f59833"},
+    {file = "regex-2024.4.28-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:2fef0b38c34ae675fcbb1b5db760d40c3fc3612cfa186e9e50df5782cac02bcd"},
+    {file = "regex-2024.4.28-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:bc365ce25f6c7c5ed70e4bc674f9137f52b7dd6a125037f9132a7be52b8a252f"},
+    {file = "regex-2024.4.28-cp312-cp312-musllinux_1_1_ppc64le.whl", hash = "sha256:ac69b394764bb857429b031d29d9604842bc4cbfd964d764b1af1868eeebc4f0"},
+    {file = "regex-2024.4.28-cp312-cp312-musllinux_1_1_s390x.whl", hash = "sha256:144a1fc54765f5c5c36d6d4b073299832aa1ec6a746a6452c3ee7b46b3d3b11d"},
+    {file = "regex-2024.4.28-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:2630ca4e152c221072fd4a56d4622b5ada876f668ecd24d5ab62544ae6793ed6"},
+    {file = "regex-2024.4.28-cp312-cp312-win32.whl", hash = "sha256:7f3502f03b4da52bbe8ba962621daa846f38489cae5c4a7b5d738f15f6443d17"},
+    {file = "regex-2024.4.28-cp312-cp312-win_amd64.whl", hash = "sha256:0dd3f69098511e71880fb00f5815db9ed0ef62c05775395968299cb400aeab82"},
+    {file = "regex-2024.4.28-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:374f690e1dd0dbdcddea4a5c9bdd97632cf656c69113f7cd6a361f2a67221cb6"},
+    {file = "regex-2024.4.28-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:25f87ae6b96374db20f180eab083aafe419b194e96e4f282c40191e71980c666"},
+    {file = "regex-2024.4.28-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:5dbc1bcc7413eebe5f18196e22804a3be1bfdfc7e2afd415e12c068624d48247"},
+    {file = "regex-2024.4.28-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f85151ec5a232335f1be022b09fbbe459042ea1951d8a48fef251223fc67eee1"},
+    {file = "regex-2024.4.28-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:57ba112e5530530fd175ed550373eb263db4ca98b5f00694d73b18b9a02e7185"},
+    {file = "regex-2024.4.28-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:224803b74aab56aa7be313f92a8d9911dcade37e5f167db62a738d0c85fdac4b"},
+    {file = "regex-2024.4.28-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0a54a047b607fd2d2d52a05e6ad294602f1e0dec2291152b745870afc47c1397"},
+    {file = "regex-2024.4.28-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0a2a512d623f1f2d01d881513af9fc6a7c46e5cfffb7dc50c38ce959f9246c94"},
+    {file = "regex-2024.4.28-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:c06bf3f38f0707592898428636cbb75d0a846651b053a1cf748763e3063a6925"},
+    {file = "regex-2024.4.28-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:1031a5e7b048ee371ab3653aad3030ecfad6ee9ecdc85f0242c57751a05b0ac4"},
+    {file = "regex-2024.4.28-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:d7a353ebfa7154c871a35caca7bfd8f9e18666829a1dc187115b80e35a29393e"},
+    {file = "regex-2024.4.28-cp38-cp38-musllinux_1_1_ppc64le.whl", hash = "sha256:7e76b9cfbf5ced1aca15a0e5b6f229344d9b3123439ffce552b11faab0114a02"},
+    {file = "regex-2024.4.28-cp38-cp38-musllinux_1_1_s390x.whl", hash = "sha256:5ce479ecc068bc2a74cb98dd8dba99e070d1b2f4a8371a7dfe631f85db70fe6e"},
+    {file = "regex-2024.4.28-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:7d77b6f63f806578c604dca209280e4c54f0fa9a8128bb8d2cc5fb6f99da4150"},
+    {file = "regex-2024.4.28-cp38-cp38-win32.whl", hash = "sha256:d84308f097d7a513359757c69707ad339da799e53b7393819ec2ea36bc4beb58"},
+    {file = "regex-2024.4.28-cp38-cp38-win_amd64.whl", hash = "sha256:2cc1b87bba1dd1a898e664a31012725e48af826bf3971e786c53e32e02adae6c"},
+    {file = "regex-2024.4.28-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:7413167c507a768eafb5424413c5b2f515c606be5bb4ef8c5dee43925aa5718b"},
+    {file = "regex-2024.4.28-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:108e2dcf0b53a7c4ab8986842a8edcb8ab2e59919a74ff51c296772e8e74d0ae"},
+    {file = "regex-2024.4.28-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:f1c5742c31ba7d72f2dedf7968998730664b45e38827637e0f04a2ac7de2f5f1"},
+    {file = "regex-2024.4.28-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:ecc6148228c9ae25ce403eade13a0961de1cb016bdb35c6eafd8e7b87ad028b1"},
+    {file = "regex-2024.4.28-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:b7d893c8cf0e2429b823ef1a1d360a25950ed11f0e2a9df2b5198821832e1947"},
+    {file = "regex-2024.4.28-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:4290035b169578ffbbfa50d904d26bec16a94526071ebec3dadbebf67a26b25e"},
+    {file = "regex-2024.4.28-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:44a22ae1cfd82e4ffa2066eb3390777dc79468f866f0625261a93e44cdf6482b"},
+    {file = "regex-2024.4.28-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:fd24fd140b69f0b0bcc9165c397e9b2e89ecbeda83303abf2a072609f60239e2"},
+    {file = "regex-2024.4.28-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:39fb166d2196413bead229cd64a2ffd6ec78ebab83fff7d2701103cf9f4dfd26"},
+    {file = "regex-2024.4.28-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:9301cc6db4d83d2c0719f7fcda37229691745168bf6ae849bea2e85fc769175d"},
+    {file = "regex-2024.4.28-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:7c3d389e8d76a49923683123730c33e9553063d9041658f23897f0b396b2386f"},
+    {file = "regex-2024.4.28-cp39-cp39-musllinux_1_1_ppc64le.whl", hash = "sha256:99ef6289b62042500d581170d06e17f5353b111a15aa6b25b05b91c6886df8fc"},
+    {file = "regex-2024.4.28-cp39-cp39-musllinux_1_1_s390x.whl", hash = "sha256:b91d529b47798c016d4b4c1d06cc826ac40d196da54f0de3c519f5a297c5076a"},
+    {file = "regex-2024.4.28-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:43548ad74ea50456e1c68d3c67fff3de64c6edb85bcd511d1136f9b5376fc9d1"},
+    {file = "regex-2024.4.28-cp39-cp39-win32.whl", hash = "sha256:05d9b6578a22db7dedb4df81451f360395828b04f4513980b6bd7a1412c679cc"},
+    {file = "regex-2024.4.28-cp39-cp39-win_amd64.whl", hash = "sha256:3986217ec830c2109875be740531feb8ddafe0dfa49767cdcd072ed7e8927962"},
+    {file = "regex-2024.4.28.tar.gz", hash = "sha256:83ab366777ea45d58f72593adf35d36ca911ea8bd838483c1823b883a121b0e4"},
+]
+
 [[package]]
 name = "requests"
 version = "2.31.0"
@@ -2257,18 +2567,18 @@ urllib3 = ">=1.25.10"
 tests = ["coverage (>=6.0.0)", "flake8", "mypy", "pytest (>=7.0.0)", "pytest-asyncio", "pytest-cov", "pytest-localserver", "types-mock", "types-requests"]

 [[package]]
-name = "rsa"
-version = "4.9"
-description = "Pure-Python RSA implementation"
+name = "rfc3339-validator"
+version = "0.1.4"
+description = "A pure python RFC3339 validator"
 optional = false
-python-versions = ">=3.6,<4"
+python-versions = ">=2.7, !=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*"
 files = [
-    {file = "rsa-4.9-py3-none-any.whl", hash = "sha256:90260d9058e514786967344d0ef75fa8727eed8a7d2e43ce9f4bcf1b536174f7"},
-    {file = "rsa-4.9.tar.gz", hash = "sha256:e38464a49c6c85d7f1351b0126661487a7e0a14a50f1675ec50eb34d4f20ef21"},
+    {file = "rfc3339_validator-0.1.4-py2.py3-none-any.whl", hash = "sha256:24f6ec1eda14ef823da9e36ec7113124b39c04d50a4d3d3a3c2859577e7791fa"},
+    {file = "rfc3339_validator-0.1.4.tar.gz", hash = "sha256:138a2abdf93304ad60530167e51d2dfb9549521a836871b88d7f4695d0022f6b"},
 ]

 [package.dependencies]
-pyasn1 = ">=0.1.3"
+six = "*"

 [[package]]
 name = "ruff"
@@ -2367,22 +2677,18 @@ files = [
 ]

 [[package]]
-name = "sshpubkeys"
-version = "3.3.1"
-description = "SSH public key parser"
+name = "sympy"
+version = "1.12"
+description = "Computer algebra system (CAS) in Python"
 optional = false
-python-versions = ">=3"
+python-versions = ">=3.8"
 files = [
-    {file = "sshpubkeys-3.3.1-py2.py3-none-any.whl", hash = "sha256:946f76b8fe86704b0e7c56a00d80294e39bc2305999844f079a217885060b1ac"},
-    {file = "sshpubkeys-3.3.1.tar.gz", hash = "sha256:3020ed4f8c846849299370fbe98ff4157b0ccc1accec105e07cfa9ae4bb55064"},
+    {file = "sympy-1.12-py3-none-any.whl", hash = "sha256:c3588cd4295d0c0f603d0f2ae780587e64e2efeedb3521e46b9bb1d08d184fa5"},
+    {file = "sympy-1.12.tar.gz", hash = "sha256:ebf595c8dac3e0fdc4152c51878b498396ec7f30e7a914d6071e674d49420fb8"},
 ]

 [package.dependencies]
-cryptography = ">=2.1.4"
-ecdsa = ">=0.13"
-
-[package.extras]
-dev = ["twine", "wheel", "yapf"]
+mpmath = ">=0.19"

 [[package]]
 name = "toml"
@@ -2612,13 +2918,13 @@ files = [

 [[package]]
 name = "werkzeug"
-version = "3.0.1"
+version = "3.0.3"
 description = "The comprehensive WSGI web application library."
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "werkzeug-3.0.1-py3-none-any.whl", hash = "sha256:90a285dc0e42ad56b34e696398b8122ee4c681833fb35b8334a095d82c56da10"},
-    {file = "werkzeug-3.0.1.tar.gz", hash = "sha256:507e811ecea72b18a404947aded4b3390e1db8f826b494d76550ef45bb3b1dcc"},
+    {file = "werkzeug-3.0.3-py3-none-any.whl", hash = "sha256:fc9645dc43e03e4d630d23143a04a7f947a9a3b5727cd535fdfe155a17cc48c8"},
+    {file = "werkzeug-3.0.3.tar.gz", hash = "sha256:097e5bfda9f0aba8da6b8545146def481d06aa7d3266e7448e2cccf67dd8bd18"},
 ]

 [package.dependencies]
@@ -2653,16 +2959,6 @@ files = [
    {file = "wrapt-1.14.1-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:8ad85f7f4e20964db4daadcab70b47ab05c7c1cf2a7c1e51087bfaa83831854c"},
    {file = "wrapt-1.14.1-cp310-cp310-win32.whl", hash = "sha256:a9a52172be0b5aae932bef82a79ec0a0ce87288c7d132946d645eba03f0ad8a8"},
    {file = "wrapt-1.14.1-cp310-cp310-win_amd64.whl", hash = "sha256:6d323e1554b3d22cfc03cd3243b5bb815a51f5249fdcbb86fda4bf62bab9e164"},
-    {file = "wrapt-1.14.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:ecee4132c6cd2ce5308e21672015ddfed1ff975ad0ac8d27168ea82e71413f55"},
-    {file = "wrapt-1.14.1-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:2020f391008ef874c6d9e208b24f28e31bcb85ccff4f335f15a3251d222b92d9"},
-    {file = "wrapt-1.14.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2feecf86e1f7a86517cab34ae6c2f081fd2d0dac860cb0c0ded96d799d20b335"},
-    {file = "wrapt-1.14.1-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:240b1686f38ae665d1b15475966fe0472f78e71b1b4903c143a842659c8e4cb9"},
-    {file = "wrapt-1.14.1-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a9008dad07d71f68487c91e96579c8567c98ca4c3881b9b113bc7b33e9fd78b8"},
-    {file = "wrapt-1.14.1-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:6447e9f3ba72f8e2b985a1da758767698efa72723d5b59accefd716e9e8272bf"},
-    {file = "wrapt-1.14.1-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:acae32e13a4153809db37405f5eba5bac5fbe2e2ba61ab227926a22901051c0a"},
-    {file = "wrapt-1.14.1-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:49ef582b7a1152ae2766557f0550a9fcbf7bbd76f43fbdc94dd3bf07cc7168be"},
-    {file = "wrapt-1.14.1-cp311-cp311-win32.whl", hash = "sha256:358fe87cc899c6bb0ddc185bf3dbfa4ba646f05b1b0b9b5a27c2cb92c2cea204"},
-    {file = "wrapt-1.14.1-cp311-cp311-win_amd64.whl", hash = "sha256:26046cd03936ae745a502abf44dac702a5e6880b2b01c29aea8ddf3353b68224"},
    {file = "wrapt-1.14.1-cp35-cp35m-manylinux1_i686.whl", hash = "sha256:43ca3bbbe97af00f49efb06e352eae40434ca9d915906f77def219b88e85d907"},
    {file = "wrapt-1.14.1-cp35-cp35m-manylinux1_x86_64.whl", hash = "sha256:6b1a564e6cb69922c7fe3a678b9f9a3c54e72b469875aa8018f18b4d1dd1adf3"},
    {file = "wrapt-1.14.1-cp35-cp35m-manylinux2010_i686.whl", hash = "sha256:00b6d4ea20a906c0ca56d84f93065b398ab74b927a7a3dbd470f6fc503f95dc3"},
@@ -2900,4 +3196,4 @@ cffi = ["cffi (>=1.11)"]
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.9"
-content-hash = "b3452b50901123fd5f2c385ce8a0c1c492296393b8a7926a322b6df0ea3ac572"
+content-hash = "dcde14c58a32bda5f123319a069352c458b3719f3c62977991eebb9803a46a9e"
--- a/proxy/Cargo.toml
+++ b/proxy/Cargo.toml
@@ -40,6 +40,7 @@ hyper.workspace = true
 hyper1 = { package = "hyper", version = "1.2", features = ["server"] }
 hyper-util = { version = "0.1", features = ["server", "http1", "http2", "tokio"] }
 http-body-util = { version = "0.1" }
+indexmap.workspace = true
 ipnet.workspace = true
 itertools.workspace = true
 lasso = { workspace = true, features = ["multi-threaded"] }
--- a/proxy/src/bin/proxy.rs
+++ b/proxy/src/bin/proxy.rs
@@ -27,6 +27,7 @@ use proxy::redis::cancellation_publisher::RedisPublisherClient;
 use proxy::redis::connection_with_credentials_provider::ConnectionWithCredentialsProvider;
 use proxy::redis::elasticache;
 use proxy::redis::notifications;
+use proxy::serverless::cancel_set::CancelSet;
 use proxy::serverless::GlobalConnPoolOptions;
 use proxy::usage_metrics;

@@ -243,6 +244,12 @@ struct SqlOverHttpArgs {
    /// increase memory used by the pool
    #[clap(long, default_value_t = 128)]
    sql_over_http_pool_shards: usize,
+
+    #[clap(long, default_value_t = 10000)]
+    sql_over_http_client_conn_threshold: u64,
+
+    #[clap(long, default_value_t = 64)]
+    sql_over_http_cancel_set_shards: usize,
 }

 #[tokio::main]
@@ -599,6 +606,8 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
            opt_in: args.sql_over_http.sql_over_http_pool_opt_in,
            max_total_conns: args.sql_over_http.sql_over_http_pool_max_total_conns,
        },
+        cancel_set: CancelSet::new(args.sql_over_http.sql_over_http_cancel_set_shards),
+        client_conn_threshold: args.sql_over_http.sql_over_http_client_conn_threshold,
    };
    let authentication_config = AuthenticationConfig {
        scram_protocol_timeout: args.scram_protocol_timeout,
--- a/proxy/src/compute.rs
+++ b/proxy/src/compute.rs
@@ -1,7 +1,7 @@
 use crate::{
    auth::parse_endpoint_param,
    cancellation::CancelClosure,
-    console::{errors::WakeComputeError, messages::MetricsAuxInfo},
+    console::{errors::WakeComputeError, messages::MetricsAuxInfo, provider::ApiLockError},
    context::RequestMonitoring,
    error::{ReportableError, UserFacingError},
    metrics::{Metrics, NumDbConnectionsGuard},
@@ -34,6 +34,9 @@ pub enum ConnectionError {

    #[error("{COULD_NOT_CONNECT}: {0}")]
    WakeComputeError(#[from] WakeComputeError),
+
+    #[error("error acquiring resource permit: {0}")]
+    TooManyConnectionAttempts(#[from] ApiLockError),
 }

 impl UserFacingError for ConnectionError {
@@ -57,6 +60,9 @@ impl UserFacingError for ConnectionError {
                None => err.to_string(),
            },
            WakeComputeError(err) => err.to_string_client(),
+            TooManyConnectionAttempts(_) => {
+                "Failed to acquire permit to connect to the database. Too many database connection attempts are currently ongoing.".to_owned()
+            }
            _ => COULD_NOT_CONNECT.to_owned(),
        }
    }
@@ -72,6 +78,7 @@ impl ReportableError for ConnectionError {
            ConnectionError::CouldNotConnect(_) => crate::error::ErrorKind::Compute,
            ConnectionError::TlsError(_) => crate::error::ErrorKind::Compute,
            ConnectionError::WakeComputeError(e) => e.get_error_kind(),
+            ConnectionError::TooManyConnectionAttempts(e) => e.get_error_kind(),
        }
    }
 }
--- a/proxy/src/config.rs
+++ b/proxy/src/config.rs
@@ -2,7 +2,7 @@ use crate::{
    auth::{self, backend::AuthRateLimiter},
    console::locks::ApiLocks,
    rate_limiter::RateBucketInfo,
-    serverless::GlobalConnPoolOptions,
+    serverless::{cancel_set::CancelSet, GlobalConnPoolOptions},
    Host,
 };
 use anyhow::{bail, ensure, Context, Ok};
@@ -56,6 +56,8 @@ pub struct TlsConfig {
 pub struct HttpConfig {
    pub request_timeout: tokio::time::Duration,
    pub pool_options: GlobalConnPoolOptions,
+    pub cancel_set: CancelSet,
+    pub client_conn_threshold: u64,
 }

 pub struct AuthenticationConfig {
@@ -536,9 +538,9 @@ pub struct RetryConfig {
 impl RetryConfig {
    /// Default options for RetryConfig.

-    /// Total delay for 8 retries with 100ms base delay and 1.6 backoff factor is about 7s.
+    /// Total delay for 5 retries with 200ms base delay and 2 backoff factor is about 6s.
    pub const CONNECT_TO_COMPUTE_DEFAULT_VALUES: &'static str =
-        "num_retries=8,base_retry_wait_duration=100ms,retry_wait_exponent_base=1.6";
+        "num_retries=5,base_retry_wait_duration=200ms,retry_wait_exponent_base=2";
    /// Total delay for 8 retries with 100ms base delay and 1.6 backoff factor is about 7s.
    /// Cplane has timeout of 60s on each request. 8m7s in total.
    pub const WAKE_COMPUTE_DEFAULT_VALUES: &'static str =
@@ -592,7 +594,7 @@ impl ConcurrencyLockOptions {
    pub const DEFAULT_OPTIONS_WAKE_COMPUTE_LOCK: &'static str = "permits=0";
    /// Default options for [`crate::console::provider::ApiLocks`].
    pub const DEFAULT_OPTIONS_CONNECT_COMPUTE_LOCK: &'static str =
-        "shards=64,permits=50,epoch=10m,timeout=500ms";
+        "shards=64,permits=10,epoch=10m,timeout=10ms";

    // pub const DEFAULT_OPTIONS_WAKE_COMPUTE_LOCK: &'static str = "shards=32,permits=4,epoch=10m,timeout=1s";

--- a/proxy/src/console/provider.rs
+++ b/proxy/src/console/provider.rs
@@ -12,6 +12,7 @@ use crate::{
    compute,
    config::{CacheOptions, EndpointCacheConfig, ProjectInfoCacheOptions},
    context::RequestMonitoring,
+    error::ReportableError,
    intern::ProjectIdInt,
    metrics::ApiLockMetrics,
    scram, EndpointCacheKey,
@@ -30,6 +31,8 @@ pub mod errors {
    };
    use thiserror::Error;

+    use super::ApiLockError;
+
    /// A go-to error message which doesn't leak any detail.
    const REQUEST_FAILED: &str = "Console request failed";

@@ -76,7 +79,7 @@ pub mod errors {
                    }
                    http::StatusCode::LOCKED | http::StatusCode::UNPROCESSABLE_ENTITY => {
                        // Status 423: project might be in maintenance mode (or bad state), or quotas exceeded.
-                        format!("{REQUEST_FAILED}: endpoint is temporary unavailable. check your quotas and/or contact our support")
+                        format!("{REQUEST_FAILED}: endpoint is temporarily unavailable. Check your quotas and/or contact our support.")
                    }
                    _ => REQUEST_FAILED.to_owned(),
                },
@@ -211,8 +214,8 @@ pub mod errors {
        #[error("Too many connections attempts")]
        TooManyConnections,

-        #[error("Timeout waiting to acquire wake compute lock")]
-        TimeoutError,
+        #[error("error acquiring resource permit: {0}")]
+        TooManyConnectionAttempts(#[from] ApiLockError),
    }

    // This allows more useful interactions than `#[from]`.
@@ -222,17 +225,6 @@ pub mod errors {
        }
    }

-    impl From<tokio::sync::AcquireError> for WakeComputeError {
-        fn from(_: tokio::sync::AcquireError) -> Self {
-            WakeComputeError::TimeoutError
-        }
-    }
-    impl From<tokio::time::error::Elapsed> for WakeComputeError {
-        fn from(_: tokio::time::error::Elapsed) -> Self {
-            WakeComputeError::TimeoutError
-        }
-    }
-
    impl UserFacingError for WakeComputeError {
        fn to_string_client(&self) -> String {
            use WakeComputeError::*;
@@ -245,7 +237,9 @@ pub mod errors {

                TooManyConnections => self.to_string(),

-                TimeoutError => "timeout while acquiring the compute resource lock".to_owned(),
+                TooManyConnectionAttempts(_) => {
+                    "Failed to acquire permit to connect to the database. Too many database connection attempts are currently ongoing.".to_owned()
+                }
            }
        }
    }
@@ -256,7 +250,7 @@ pub mod errors {
                WakeComputeError::BadComputeAddress(_) => crate::error::ErrorKind::ControlPlane,
                WakeComputeError::ApiError(e) => e.get_error_kind(),
                WakeComputeError::TooManyConnections => crate::error::ErrorKind::RateLimit,
-                WakeComputeError::TimeoutError => crate::error::ErrorKind::ServiceRateLimit,
+                WakeComputeError::TooManyConnectionAttempts(e) => e.get_error_kind(),
            }
        }
    }
@@ -456,6 +450,23 @@ pub struct ApiLocks<K> {
    metrics: &'static ApiLockMetrics,
 }

+#[derive(Debug, thiserror::Error)]
+pub enum ApiLockError {
+    #[error("lock was closed")]
+    AcquireError(#[from] tokio::sync::AcquireError),
+    #[error("permit could not be acquired")]
+    TimeoutError(#[from] tokio::time::error::Elapsed),
+}
+
+impl ReportableError for ApiLockError {
+    fn get_error_kind(&self) -> crate::error::ErrorKind {
+        match self {
+            ApiLockError::AcquireError(_) => crate::error::ErrorKind::Service,
+            ApiLockError::TimeoutError(_) => crate::error::ErrorKind::RateLimit,
+        }
+    }
+}
+
 impl<K: Hash + Eq + Clone> ApiLocks<K> {
    pub fn new(
        name: &'static str,
@@ -475,7 +486,7 @@ impl<K: Hash + Eq + Clone> ApiLocks<K> {
        })
    }

-    pub async fn get_permit(&self, key: &K) -> Result<WakeComputePermit, errors::WakeComputeError> {
+    pub async fn get_permit(&self, key: &K) -> Result<WakeComputePermit, ApiLockError> {
        if self.permits == 0 {
            return Ok(WakeComputePermit { permit: None });
        }
--- a/proxy/src/proxy/retry.rs
+++ b/proxy/src/proxy/retry.rs
@@ -86,6 +86,8 @@ impl ShouldRetry for compute::ConnectionError {
        match self {
            compute::ConnectionError::Postgres(err) => err.should_retry_database_address(),
            compute::ConnectionError::CouldNotConnect(err) => err.should_retry_database_address(),
+            // the cache entry was not checked for validity
+            compute::ConnectionError::TooManyConnectionAttempts(_) => false,
            _ => true,
        }
    }
--- a/proxy/src/proxy/wake_compute.rs
+++ b/proxy/src/proxy/wake_compute.rs
@@ -119,7 +119,7 @@ fn report_error(e: &WakeComputeError, retry: bool) {
            WakeupFailureKind::ApiConsoleOtherError
        }
        WakeComputeError::TooManyConnections => WakeupFailureKind::ApiConsoleLocked,
-        WakeComputeError::TimeoutError => WakeupFailureKind::TimeoutError,
+        WakeComputeError::TooManyConnectionAttempts(_) => WakeupFailureKind::TimeoutError,
    };
    Metrics::get()
        .proxy
--- a/proxy/src/serverless.rs
+++ b/proxy/src/serverless.rs
@@ -3,6 +3,7 @@
 //! Handles both SQL over HTTP and SQL over Websockets.

 mod backend;
+pub mod cancel_set;
 mod conn_pool;
 mod http_util;
 mod json;
@@ -109,20 +110,37 @@ pub async fn task_main(
        let conn_id = uuid::Uuid::new_v4();
        let http_conn_span = tracing::info_span!("http_conn", ?conn_id);

-        connections.spawn(
-            connection_handler(
-                config,
-                backend.clone(),
-                connections.clone(),
-                cancellation_handler.clone(),
-                cancellation_token.clone(),
-                server.clone(),
-                tls_acceptor.clone(),
-                conn,
-                peer_addr,
-            )
-            .instrument(http_conn_span),
-        );
+        let n_connections = Metrics::get()
+            .proxy
+            .client_connections
+            .sample(crate::metrics::Protocol::Http);
+        tracing::trace!(?n_connections, threshold = ?config.http_config.client_conn_threshold, "check");
+        if n_connections > config.http_config.client_conn_threshold {
+            tracing::trace!("attempting to cancel a random connection");
+            if let Some(token) = config.http_config.cancel_set.take() {
+                tracing::debug!("cancelling a random connection");
+                token.cancel()
+            }
+        }
+
+        let conn_token = cancellation_token.child_token();
+        let conn = connection_handler(
+            config,
+            backend.clone(),
+            connections.clone(),
+            cancellation_handler.clone(),
+            conn_token.clone(),
+            server.clone(),
+            tls_acceptor.clone(),
+            conn,
+            peer_addr,
+        )
+        .instrument(http_conn_span);
+
+        connections.spawn(async move {
+            let _cancel_guard = config.http_config.cancel_set.insert(conn_id, conn_token);
+            conn.await
+        });
    }

    connections.wait().await;
@@ -243,6 +261,7 @@ async fn connection_handler(
    // On cancellation, trigger the HTTP connection handler to shut down.
    let res = match select(pin!(cancellation_token.cancelled()), pin!(conn)).await {
        Either::Left((_cancelled, mut conn)) => {
+            tracing::debug!(%peer_addr, "cancelling connection");
            conn.as_mut().graceful_shutdown();
            conn.await
        }
--- a/proxy/src/serverless/backend.rs
+++ b/proxy/src/serverless/backend.rs
@@ -10,6 +10,7 @@ use crate::{
    console::{
        errors::{GetAuthInfoError, WakeComputeError},
        locks::ApiLocks,
+        provider::ApiLockError,
        CachedNodeInfo,
    },
    context::RequestMonitoring,
@@ -131,6 +132,8 @@ pub enum HttpConnError {
    AuthError(#[from] AuthError),
    #[error("wake_compute returned error")]
    WakeCompute(#[from] WakeComputeError),
+    #[error("error acquiring resource permit: {0}")]
+    TooManyConnectionAttempts(#[from] ApiLockError),
 }

 impl ReportableError for HttpConnError {
@@ -141,6 +144,7 @@ impl ReportableError for HttpConnError {
            HttpConnError::GetAuthInfo(a) => a.get_error_kind(),
            HttpConnError::AuthError(a) => a.get_error_kind(),
            HttpConnError::WakeCompute(w) => w.get_error_kind(),
+            HttpConnError::TooManyConnectionAttempts(w) => w.get_error_kind(),
        }
    }
 }
@@ -153,6 +157,9 @@ impl UserFacingError for HttpConnError {
            HttpConnError::GetAuthInfo(c) => c.to_string_client(),
            HttpConnError::AuthError(c) => c.to_string_client(),
            HttpConnError::WakeCompute(c) => c.to_string_client(),
+            HttpConnError::TooManyConnectionAttempts(_) => {
+                "Failed to acquire permit to connect to the database. Too many database connection attempts are currently ongoing.".to_owned()
+            }
        }
    }
 }
@@ -165,6 +172,15 @@ impl ShouldRetry for HttpConnError {
            HttpConnError::GetAuthInfo(_) => false,
            HttpConnError::AuthError(_) => false,
            HttpConnError::WakeCompute(_) => false,
+            HttpConnError::TooManyConnectionAttempts(_) => false,
+        }
+    }
+    fn should_retry_database_address(&self) -> bool {
+        match self {
+            HttpConnError::ConnectionError(e) => e.should_retry_database_address(),
+            // we never checked cache validity
+            HttpConnError::TooManyConnectionAttempts(_) => false,
+            _ => true,
        }
    }
 }
--- a/proxy/src/serverless/cancel_set.rs
+++ b/proxy/src/serverless/cancel_set.rs
@@ -0,0 +1,102 @@
+//! A set for cancelling random http connections
+
+use std::{
+    hash::{BuildHasher, BuildHasherDefault},
+    num::NonZeroUsize,
+    time::Duration,
+};
+
+use indexmap::IndexMap;
+use parking_lot::Mutex;
+use rand::{thread_rng, Rng};
+use rustc_hash::FxHasher;
+use tokio::time::Instant;
+use tokio_util::sync::CancellationToken;
+use uuid::Uuid;
+
+type Hasher = BuildHasherDefault<FxHasher>;
+
+pub struct CancelSet {
+    shards: Box<[Mutex<CancelShard>]>,
+    // keyed by random uuid, fxhasher is fine
+    hasher: Hasher,
+}
+
+pub struct CancelShard {
+    tokens: IndexMap<uuid::Uuid, (Instant, CancellationToken), Hasher>,
+}
+
+impl CancelSet {
+    pub fn new(shards: usize) -> Self {
+        CancelSet {
+            shards: (0..shards)
+                .map(|_| {
+                    Mutex::new(CancelShard {
+                        tokens: IndexMap::with_hasher(Hasher::default()),
+                    })
+                })
+                .collect(),
+            hasher: Hasher::default(),
+        }
+    }
+
+    pub fn take(&self) -> Option<CancellationToken> {
+        for _ in 0..4 {
+            if let Some(token) = self.take_raw(thread_rng().gen()) {
+                return Some(token);
+            }
+            tracing::trace!("failed to get cancel token");
+        }
+        None
+    }
+
+    pub fn take_raw(&self, rng: usize) -> Option<CancellationToken> {
+        NonZeroUsize::new(self.shards.len())
+            .and_then(|len| self.shards[rng % len].lock().take(rng / len))
+    }
+
+    pub fn insert(&self, id: uuid::Uuid, token: CancellationToken) -> CancelGuard<'_> {
+        let shard = NonZeroUsize::new(self.shards.len()).map(|len| {
+            let hash = self.hasher.hash_one(id) as usize;
+            let shard = &self.shards[hash % len];
+            shard.lock().insert(id, token);
+            shard
+        });
+        CancelGuard { shard, id }
+    }
+}
+
+impl CancelShard {
+    fn take(&mut self, rng: usize) -> Option<CancellationToken> {
+        NonZeroUsize::new(self.tokens.len()).and_then(|len| {
+            // 10 second grace period so we don't cancel new connections
+            if self.tokens.get_index(rng % len)?.1 .0.elapsed() < Duration::from_secs(10) {
+                return None;
+            }
+
+            let (_key, (_insert, token)) = self.tokens.swap_remove_index(rng % len)?;
+            Some(token)
+        })
+    }
+
+    fn remove(&mut self, id: uuid::Uuid) {
+        self.tokens.swap_remove(&id);
+    }
+
+    fn insert(&mut self, id: uuid::Uuid, token: CancellationToken) {
+        self.tokens.insert(id, (Instant::now(), token));
+    }
+}
+
+pub struct CancelGuard<'a> {
+    shard: Option<&'a Mutex<CancelShard>>,
+    id: Uuid,
+}
+
+impl Drop for CancelGuard<'_> {
+    fn drop(&mut self) {
+        if let Some(shard) = self.shard {
+            shard.lock().remove(self.id);
+        }
+    }
+}
--- a/proxy/src/serverless/conn_pool.rs
+++ b/proxy/src/serverless/conn_pool.rs
@@ -716,7 +716,7 @@ impl<C: ClientInnerExt> Drop for Client<C> {
 mod tests {
    use std::{mem, sync::atomic::AtomicBool};

-    use crate::{BranchId, EndpointId, ProjectId};
+    use crate::{serverless::cancel_set::CancelSet, BranchId, EndpointId, ProjectId};

    use super::*;

@@ -767,6 +767,8 @@ mod tests {
                max_total_conns: 3,
            },
            request_timeout: Duration::from_secs(1),
+            cancel_set: CancelSet::new(0),
+            client_conn_threshold: u64::MAX,
        }));
        let pool = GlobalConnPool::new(config);
        let conn_info = ConnInfo {
--- a/proxy/src/serverless/sql_over_http.rs
+++ b/proxy/src/serverless/sql_over_http.rs
@@ -424,8 +424,8 @@ pub enum SqlOverHttpCancel {
 impl ReportableError for SqlOverHttpCancel {
    fn get_error_kind(&self) -> ErrorKind {
        match self {
-            SqlOverHttpCancel::Postgres => ErrorKind::RateLimit,
-            SqlOverHttpCancel::Connect => ErrorKind::ServiceRateLimit,
+            SqlOverHttpCancel::Postgres => ErrorKind::ClientDisconnect,
+            SqlOverHttpCancel::Connect => ErrorKind::ClientDisconnect,
        }
    }
 }
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -14,17 +14,17 @@ requests = "^2.31.0"
 pytest-xdist = "^3.3.1"
 asyncpg = "^0.29.0"
 aiopg = "^1.4.0"
-Jinja2 = "^3.1.3"
+Jinja2 = "^3.1.4"
 types-requests = "^2.31.0.0"
 types-psycopg2 = "^2.9.21.10"
 boto3 = "^1.34.11"
 boto3-stubs = {extras = ["s3"], version = "^1.26.16"}
-moto = {extras = ["server"], version = "^4.1.2"}
+moto = {extras = ["server"], version = "^5.0.6"}
 backoff = "^2.2.1"
 pytest-lazy-fixture = "^0.6.3"
 prometheus-client = "^0.14.1"
 pytest-timeout = "^2.1.0"
-Werkzeug = "^3.0.1"
+Werkzeug = "^3.0.3"
 pytest-order = "^1.1.0"
 allure-pytest = "^2.13.2"
 pytest-asyncio = "^0.21.0"
--- a/s3_scrubber/src/checks.rs
+++ b/s3_scrubber/src/checks.rs
@@ -13,7 +13,7 @@ use crate::metadata_stream::stream_listing;
 use crate::{download_object_with_retries, RootTarget, TenantShardTimelineId};
 use futures_util::StreamExt;
 use pageserver::tenant::remote_timeline_client::parse_remote_index_path;
-use pageserver::tenant::storage_layer::LayerFileName;
+use pageserver::tenant::storage_layer::LayerName;
 use pageserver::tenant::IndexPart;
 use remote_storage::RemotePath;

@@ -110,7 +110,7 @@ pub(crate) fn branch_cleanup_and_check_errors(
                    for (layer, metadata) in index_part.layer_metadata {
                        if metadata.file_size == 0 {
                            result.errors.push(format!(
-                                "index_part.json contains a layer {} that has 0 size in its layer metadata", layer.file_name(),
+                                "index_part.json contains a layer {} that has 0 size in its layer metadata", layer,
                            ))
                        }

@@ -121,7 +121,7 @@ pub(crate) fn branch_cleanup_and_check_errors(
                            // layer we think is missing.
                            result.errors.push(format!(
                                "index_part.json contains a layer {}{} (shard {}) that is not present in remote storage",
-                                layer.file_name(),
+                                layer,
                                metadata.generation.get_suffix(),
                                metadata.shard
                            ))
@@ -170,8 +170,7 @@ pub(crate) struct LayerRef {
 /// the tenant to query whether an object exists.
 #[derive(Default)]
 pub(crate) struct TenantObjectListing {
-    shard_timelines:
-        HashMap<(ShardIndex, TimelineId), HashMap<(LayerFileName, Generation), LayerRef>>,
+    shard_timelines: HashMap<(ShardIndex, TimelineId), HashMap<(LayerName, Generation), LayerRef>>,
 }

 impl TenantObjectListing {
@@ -180,7 +179,7 @@ impl TenantObjectListing {
    pub(crate) fn push(
        &mut self,
        ttid: TenantShardTimelineId,
-        layers: HashSet<(LayerFileName, Generation)>,
+        layers: HashSet<(LayerName, Generation)>,
    ) {
        let shard_index = ShardIndex::new(
            ttid.tenant_shard_id.shard_number,
@@ -208,7 +207,7 @@ impl TenantObjectListing {
    pub(crate) fn check_ref(
        &mut self,
        timeline_id: TimelineId,
-        layer_file: &LayerFileName,
+        layer_file: &LayerName,
        metadata: &IndexLayerMetadata,
    ) -> bool {
        let Some(shard_tl) = self.shard_timelines.get_mut(&(metadata.shard, timeline_id)) else {
@@ -224,7 +223,7 @@ impl TenantObjectListing {
        true
    }

-    pub(crate) fn get_orphans(&self) -> Vec<(ShardIndex, TimelineId, LayerFileName, Generation)> {
+    pub(crate) fn get_orphans(&self) -> Vec<(ShardIndex, TimelineId, LayerName, Generation)> {
        let mut result = Vec::new();
        for ((shard_index, timeline_id), layers) in &self.shard_timelines {
            for ((layer_file, generation), layer_ref) in layers {
@@ -249,23 +248,23 @@ pub(crate) enum BlobDataParseResult {
    Parsed {
        index_part: IndexPart,
        index_part_generation: Generation,
-        s3_layers: HashSet<(LayerFileName, Generation)>,
+        s3_layers: HashSet<(LayerName, Generation)>,
    },
    /// The remains of a deleted Timeline (i.e. an initdb archive only)
    Relic,
    Incorrect(Vec<String>),
 }

-fn parse_layer_object_name(name: &str) -> Result<(LayerFileName, Generation), String> {
+fn parse_layer_object_name(name: &str) -> Result<(LayerName, Generation), String> {
    match name.rsplit_once('-') {
        // FIXME: this is gross, just use a regex?
        Some((layer_filename, gen)) if gen.len() == 8 => {
-            let layer = layer_filename.parse::<LayerFileName>()?;
+            let layer = layer_filename.parse::<LayerName>()?;
            let gen =
                Generation::parse_suffix(gen).ok_or("Malformed generation suffix".to_string())?;
            Ok((layer, gen))
        }
-        _ => Ok((name.parse::<LayerFileName>()?, Generation::none())),
+        _ => Ok((name.parse::<LayerName>()?, Generation::none())),
    }
 }

--- a/s3_scrubber/src/lib.rs
+++ b/s3_scrubber/src/lib.rs
@@ -312,7 +312,10 @@ pub fn init_s3_client(account_id: Option<String>, bucket_region: Region) -> Clie
    let sleep_impl: Arc<dyn AsyncSleep> = Arc::new(TokioSleep::new());

    let mut builder = Config::builder()
-        .behavior_version(BehaviorVersion::v2023_11_09())
+        .behavior_version(
+            #[allow(deprecated)] /* TODO: https://github.com/neondatabase/neon/issues/7665 */
+            BehaviorVersion::v2023_11_09(),
+        )
        .region(bucket_region)
        .retry_config(RetryConfig::adaptive().with_max_attempts(3))
        .sleep_impl(SharedAsyncSleep::from(sleep_impl))
--- a/s3_scrubber/src/tenant_snapshot.rs
+++ b/s3_scrubber/src/tenant_snapshot.rs
@@ -12,7 +12,7 @@ use aws_sdk_s3::Client;
 use camino::Utf8PathBuf;
 use futures::{StreamExt, TryStreamExt};
 use pageserver::tenant::remote_timeline_client::index::IndexLayerMetadata;
-use pageserver::tenant::storage_layer::LayerFileName;
+use pageserver::tenant::storage_layer::LayerName;
 use pageserver::tenant::IndexPart;
 use pageserver_api::shard::TenantShardId;
 use utils::generation::Generation;
@@ -48,16 +48,16 @@ impl SnapshotDownloader {
    async fn download_layer(
        &self,
        ttid: TenantShardTimelineId,
-        layer_name: LayerFileName,
+        layer_name: LayerName,
        layer_metadata: IndexLayerMetadata,
-    ) -> anyhow::Result<(LayerFileName, IndexLayerMetadata)> {
+    ) -> anyhow::Result<(LayerName, IndexLayerMetadata)> {
        // Note this is local as in a local copy of S3 data, not local as in the pageserver's local format.  They use
        // different layer names (remote-style has the generation suffix)
        let local_path = self.output_path.join(format!(
            "{}/timelines/{}/{}{}",
            ttid.tenant_shard_id,
            ttid.timeline_id,
-            layer_name.file_name(),
+            layer_name,
            layer_metadata.generation.get_suffix()
        ));

@@ -76,7 +76,7 @@ impl SnapshotDownloader {
            let remote_layer_path = format!(
                "{}{}{}",
                timeline_root.prefix_in_bucket,
-                layer_name.file_name(),
+                layer_name,
                layer_metadata.generation.get_suffix()
            );

@@ -110,7 +110,7 @@ impl SnapshotDownloader {
    async fn download_layers(
        &self,
        ttid: TenantShardTimelineId,
-        layers: Vec<(LayerFileName, IndexLayerMetadata)>,
+        layers: Vec<(LayerName, IndexLayerMetadata)>,
    ) -> anyhow::Result<()> {
        let layer_count = layers.len();
        tracing::info!("Downloading {} layers for timeline {ttid}...", layer_count);
@@ -138,7 +138,7 @@ impl SnapshotDownloader {
                    tracing::info!(
                        "[{download_count}/{layer_count}] OK: {} bytes {ttid} {}",
                        layer_metadata.file_size,
-                        layer_name.file_name()
+                        layer_name
                    );
                }
                Err(e) => {
@@ -163,7 +163,7 @@ impl SnapshotDownloader {
        index_part_generation: Generation,
        ancestor_layers: &mut HashMap<
            TenantShardTimelineId,
-            HashMap<LayerFileName, IndexLayerMetadata>,
+            HashMap<LayerName, IndexLayerMetadata>,
        >,
    ) -> anyhow::Result<()> {
        let index_bytes = serde_json::to_string(&index_part).unwrap();
@@ -234,7 +234,7 @@ impl SnapshotDownloader {
        // happen if this tenant has been split at some point)
        let mut ancestor_layers: HashMap<
            TenantShardTimelineId,
-            HashMap<LayerFileName, IndexLayerMetadata>,
+            HashMap<LayerName, IndexLayerMetadata>,
        > = Default::default();

        for shard in shards.into_iter().filter(|s| s.shard_count == shard_count) {
--- a/safekeeper/src/safekeeper.rs
+++ b/safekeeper/src/safekeeper.rs
@@ -725,6 +725,18 @@ where
            self.state.inmem.commit_lsn
        );

+        // Before first WAL write initialize its segment. It makes first segment
+        // pg_waldump'able because stream from compute doesn't include its
+        // segment and page headers.
+        //
+        // If we fail before first WAL write flush this action would be
+        // repeated, that's ok because it is idempotent.
+        if self.wal_store.flush_lsn() == Lsn::INVALID {
+            self.wal_store
+                .initialize_first_segment(msg.start_streaming_at)
+                .await?;
+        }
+
        // TODO: cross check divergence point, check if msg.start_streaming_at corresponds to
        // intersection of our history and history from msg

@@ -1007,6 +1019,10 @@ mod tests {
            self.lsn
        }

+        async fn initialize_first_segment(&mut self, _init_lsn: Lsn) -> Result<()> {
+            Ok(())
+        }
+
        async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
            self.lsn = startpos + buf.len() as u64;
            Ok(())
--- a/safekeeper/src/send_wal.rs
+++ b/safekeeper/src/send_wal.rs
@@ -506,6 +506,8 @@ struct WalSender<'a, IO> {
    send_buf: [u8; MAX_SEND_SIZE],
 }

+const POLL_STATE_TIMEOUT: Duration = Duration::from_secs(1);
+
 impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
    /// Send WAL until
    /// - an error occurs
@@ -584,14 +586,22 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
    async fn wait_wal(&mut self) -> Result<(), CopyStreamHandlerEnd> {
        loop {
            self.end_pos = self.end_watch.get();
-            if self.end_pos > self.start_pos {
-                // We have something to send.
+            let have_something_to_send = (|| {
+                fail::fail_point!(
+                    "sk-pause-send",
+                    self.appname.as_deref() != Some("pageserver"),
+                    |_| { false }
+                );
+                self.end_pos > self.start_pos
+            })();
+
+            if have_something_to_send {
                trace!("got end_pos {:?}, streaming", self.end_pos);
                return Ok(());
            }

            // Wait for WAL to appear, now self.end_pos == self.start_pos.
-            if let Some(lsn) = wait_for_lsn(&mut self.end_watch, self.term, self.start_pos).await? {
+            if let Some(lsn) = self.wait_for_lsn().await? {
                self.end_pos = lsn;
                trace!("got end_pos {:?}, streaming", self.end_pos);
                return Ok(());
@@ -628,6 +638,54 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
                .await?;
        }
    }
+
+    /// Wait until we have available WAL > start_pos or timeout expires. Returns
+    /// - Ok(Some(end_pos)) if needed lsn is successfully observed;
+    /// - Ok(None) if timeout expired;
+    /// - Err in case of error -- only if 1) term changed while fetching in recovery
+    ///   mode 2) watch channel closed, which must never happen.
+    async fn wait_for_lsn(&mut self) -> anyhow::Result<Option<Lsn>> {
+        let fp = (|| {
+            fail::fail_point!(
+                "sk-pause-send",
+                self.appname.as_deref() != Some("pageserver"),
+                |_| { true }
+            );
+            false
+        })();
+        if fp {
+            tokio::time::sleep(POLL_STATE_TIMEOUT).await;
+            return Ok(None);
+        }
+
+        let res = timeout(POLL_STATE_TIMEOUT, async move {
+            loop {
+                let end_pos = self.end_watch.get();
+                if end_pos > self.start_pos {
+                    return Ok(end_pos);
+                }
+                if let EndWatch::Flush(rx) = &self.end_watch {
+                    let curr_term = rx.borrow().term;
+                    if let Some(client_term) = self.term {
+                        if curr_term != client_term {
+                            bail!("term changed: requested {}, now {}", client_term, curr_term);
+                        }
+                    }
+                }
+                self.end_watch.changed().await?;
+            }
+        })
+        .await;
+
+        match res {
+            // success
+            Ok(Ok(commit_lsn)) => Ok(Some(commit_lsn)),
+            // error inside closure
+            Ok(Err(err)) => Err(err),
+            // timeout
+            Err(_) => Ok(None),
+        }
+    }
 }

 /// A half driving receiving replies.
@@ -685,47 +743,6 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> ReplyReader<IO> {
    }
 }

-const POLL_STATE_TIMEOUT: Duration = Duration::from_secs(1);
-
-/// Wait until we have available WAL > start_pos or timeout expires. Returns
-/// - Ok(Some(end_pos)) if needed lsn is successfully observed;
-/// - Ok(None) if timeout expired;
-/// - Err in case of error -- only if 1) term changed while fetching in recovery
-///   mode 2) watch channel closed, which must never happen.
-async fn wait_for_lsn(
-    rx: &mut EndWatch,
-    client_term: Option<Term>,
-    start_pos: Lsn,
-) -> anyhow::Result<Option<Lsn>> {
-    let res = timeout(POLL_STATE_TIMEOUT, async move {
-        loop {
-            let end_pos = rx.get();
-            if end_pos > start_pos {
-                return Ok(end_pos);
-            }
-            if let EndWatch::Flush(rx) = rx {
-                let curr_term = rx.borrow().term;
-                if let Some(client_term) = client_term {
-                    if curr_term != client_term {
-                        bail!("term changed: requested {}, now {}", client_term, curr_term);
-                    }
-                }
-            }
-            rx.changed().await?;
-        }
-    })
-    .await;
-
-    match res {
-        // success
-        Ok(Ok(commit_lsn)) => Ok(Some(commit_lsn)),
-        // error inside closure
-        Ok(Err(err)) => Err(err),
-        // timeout
-        Err(_) => Ok(None),
-    }
-}
-
 #[cfg(test)]
 mod tests {
    use utils::id::{TenantId, TimelineId};
--- a/safekeeper/src/wal_storage.rs
+++ b/safekeeper/src/wal_storage.rs
@@ -38,6 +38,12 @@ pub trait Storage {
    /// LSN of last durably stored WAL record.
    fn flush_lsn(&self) -> Lsn;

+    /// Initialize segment by creating proper long header at the beginning of
+    /// the segment and short header at the page of given LSN. This is only used
+    /// for timeline initialization because compute will stream data only since
+    /// init_lsn. Other segment headers are included in compute stream.
+    async fn initialize_first_segment(&mut self, init_lsn: Lsn) -> Result<()>;
+
    /// Write piece of WAL from buf to disk, but not necessarily sync it.
    async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()>;

@@ -78,6 +84,8 @@ pub struct PhysicalStorage {

    /// Size of WAL segment in bytes.
    wal_seg_size: usize,
+    pg_version: u32,
+    system_id: u64,

    /// Written to disk, but possibly still in the cache and not fully persisted.
    /// Also can be ahead of record_lsn, if happen to be in the middle of a WAL record.
@@ -169,6 +177,8 @@ impl PhysicalStorage {
            timeline_dir,
            conf: conf.clone(),
            wal_seg_size,
+            pg_version: state.server.pg_version,
+            system_id: state.server.system_id,
            write_lsn,
            write_record_lsn: write_lsn,
            flush_record_lsn: flush_lsn,
@@ -324,6 +334,20 @@ impl Storage for PhysicalStorage {
        self.flush_record_lsn
    }

+    async fn initialize_first_segment(&mut self, init_lsn: Lsn) -> Result<()> {
+        let segno = init_lsn.segment_number(self.wal_seg_size);
+        let (mut file, _) = self.open_or_create(segno).await?;
+        let major_pg_version = self.pg_version / 10000;
+        let wal_seg =
+            postgres_ffi::generate_wal_segment(segno, self.system_id, major_pg_version, init_lsn)?;
+        file.seek(SeekFrom::Start(0)).await?;
+        file.write_all(&wal_seg).await?;
+        file.flush().await?;
+        info!("initialized segno {} at lsn {}", segno, init_lsn);
+        // note: file is *not* fsynced
+        Ok(())
+    }
+
    /// Write WAL to disk.
    async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
        // Disallow any non-sequential writes, which can result in gaps or overwrites.
--- a/safekeeper/tests/walproposer_sim/safekeeper_disk.rs
+++ b/safekeeper/tests/walproposer_sim/safekeeper_disk.rs
@@ -182,6 +182,10 @@ impl wal_storage::Storage for DiskWALStorage {
        self.flush_record_lsn
    }

+    async fn initialize_first_segment(&mut self, _init_lsn: Lsn) -> Result<()> {
+        Ok(())
+    }
+
    /// Write piece of WAL from buf to disk, but not necessarily sync it.
    async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
        if self.write_lsn != startpos {
--- a/safekeeper/tests/walproposer_sim/walproposer_api.rs
+++ b/safekeeper/tests/walproposer_sim/walproposer_api.rs
@@ -17,8 +17,7 @@ use utils::lsn::Lsn;
 use walproposer::{
    api_bindings::Level,
    bindings::{
-        pg_atomic_uint64, NeonWALReadResult, PageserverFeedback, SafekeeperStateDesiredEvents,
-        WL_SOCKET_READABLE, WL_SOCKET_WRITEABLE,
+        NeonWALReadResult, SafekeeperStateDesiredEvents, WL_SOCKET_READABLE, WL_SOCKET_WRITEABLE,
    },
    walproposer::{ApiImpl, Config},
 };
@@ -224,31 +223,13 @@ impl SimulationApi {
            })
            .collect::<Vec<_>>();

-        let empty_feedback = PageserverFeedback {
-            present: false,
-            currentClusterSize: 0,
-            last_received_lsn: 0,
-            disk_consistent_lsn: 0,
-            remote_consistent_lsn: 0,
-            replytime: 0,
-            shard_number: 0,
-        };
-
        Self {
            os: args.os,
            safekeepers: RefCell::new(sk_conns),
            disk: args.disk,
            redo_start_lsn: args.redo_start_lsn,
            last_logged_commit_lsn: 0,
-            shmem: UnsafeCell::new(walproposer::bindings::WalproposerShmemState {
-                mutex: 0,
-                mineLastElectedTerm: 0,
-                backpressureThrottlingTime: pg_atomic_uint64 { value: 0 },
-                currentClusterSize: pg_atomic_uint64 { value: 0 },
-                shard_ps_feedback: [empty_feedback; 128],
-                num_shards: 0,
-                min_ps_feedback: empty_feedback,
-            }),
+            shmem: UnsafeCell::new(walproposer::api_bindings::empty_shmem()),
            config: args.config,
            event_set: RefCell::new(None),
        }
@@ -274,6 +255,12 @@ impl ApiImpl for SimulationApi {
        self.os.now() as i64 * 1000
    }

+    fn update_donor(&self, donor: &mut walproposer::bindings::Safekeeper, donor_lsn: u64) {
+        let mut shmem = unsafe { *self.get_shmem_state() };
+        shmem.propEpochStartLsn.value = donor_lsn;
+        shmem.donor_conninfo = donor.conninfo;
+    }
+
    fn conn_status(
        &self,
        _: &mut walproposer::bindings::Safekeeper,
--- a/storage_controller/src/service.rs
+++ b/storage_controller/src/service.rs
@@ -4745,7 +4745,7 @@ impl Service {
                    // them in an optimization
                    const DOWNLOAD_FRESHNESS_THRESHOLD: u64 = 10 * 1024 * 1024 * 1024;

-                    if progress.bytes_total == 0
+                    if progress.heatmap_mtime.is_none()
                        || progress.bytes_total < DOWNLOAD_FRESHNESS_THRESHOLD
                            && progress.bytes_downloaded != progress.bytes_total
                        || progress.bytes_total - progress.bytes_downloaded
--- a/test_runner/README.md
+++ b/test_runner/README.md
@@ -92,6 +92,166 @@ Exit after the first test failure:
 `./scripts/pytest -x ...`
 (there are many more pytest options; run `pytest -h` to see them.)

+#### Running Python tests against real S3 or S3-compatible services
+
+Neon's `libs/remote_storage` supports multiple implementations of remote storage.
+At the time of writing, that is
+```rust
+pub enum RemoteStorageKind {
+    /// Storage based on local file system.
+    /// Specify a root folder to place all stored files into.
+    LocalFs(Utf8PathBuf),
+    /// AWS S3 based storage, storing all files in the S3 bucket
+    /// specified by the config
+    AwsS3(S3Config),
+    /// Azure Blob based storage, storing all files in the container
+    /// specified by the config
+    AzureContainer(AzureConfig),
+}
+```
+
+The test suite has a Python enum with equal name but different meaning:
+
+```python
+@enum.unique
+class RemoteStorageKind(str, enum.Enum):
+    LOCAL_FS = "local_fs"
+    MOCK_S3 = "mock_s3"
+    REAL_S3 = "real_s3"
+```
+
+* `LOCAL_FS` => `LocalFs`
+* `MOCK_S3`: starts [`moto`](https://github.com/getmoto/moto)'s S3 implementation, then configures Pageserver with `AwsS3`
+* `REAL_S3` => configure `AwsS3` as detailed below
+
+When a test in the test suite needs an `AwsS3`, it is supposed to call `remote_storage.s3_storage()`.
+That function checks env var `ENABLE_REAL_S3_REMOTE_STORAGE`:
+* If it is not set, use `MOCK_S3`
+* If it is set, use `REAL_S3`.
+
+For `REAL_S3`, the test suite creates the dict/toml representation of the `RemoteStorageKind::AwsS3` based on env vars:
+
+```rust
+pub struct S3Config {
+    // test suite env var: REMOTE_STORAGE_S3_BUCKET
+    pub bucket_name: String,
+    // test suite env var: REMOTE_STORAGE_S3_REGION
+    pub bucket_region: String,
+    // test suite determines this
+    pub prefix_in_bucket: Option<String>,
+    // no env var exists; test suite sets it for MOCK_S3, because that's how moto works
+    pub endpoint: Option<String>,
+    ...
+}
+```
+
+*Credentials* are not part of the config, but discovered by the AWS SDK.
+See the `libs/remote_storage` Rust code.
+We're documenting two mechanism here:
+
+The test suite supports two mechanisms (`remote_storage.py`):
+
+**Credential mechanism 1**: env vars `AWS_ACCESS_KEY_ID` and `AWS_SECRET_ACCESS_KEY`.
+Populate the env vars with AWS access keys that you created in IAM.
+Our CI uses this mechanism.
+However, it is _not_ recommended for interactive use by developers ([learn more](https://docs.aws.amazon.com/sdkref/latest/guide/access-users.html#credentials-long-term)).
+Instead, use profiles (next section).
+
+**Credential mechanism 2**: env var `AWS_PROFILE`.
+This uses the AWS SDK's (and CLI's) profile mechanism.
+Learn more about it [in the official docs](https://docs.aws.amazon.com/sdkref/latest/guide/file-format.html).
+After configuring a profile (e.g. via the aws CLI), set the env var to its name.
+
+In conclusion, the full command line is:
+
+```bash
+# with long-term AWS access keys
+ENABLE_REAL_S3_REMOTE_STORAGE=true \
+REMOTE_STORAGE_S3_BUCKET=mybucket \
+REMOTE_STORAGE_S3_REGION=eu-central-1 \
+AWS_ACCESS_KEY_ID=... \
+AWS_SECRET_ACCESS_KEY=... \
+./scripts/pytest
+```
+<!-- Don't forget to update the Minio example when changing these -->
+```bash
+# with AWS PROFILE
+ENABLE_REAL_S3_REMOTE_STORAGE=true \
+REMOTE_STORAGE_S3_BUCKET=mybucket \
+REMOTE_STORAGE_S3_REGION=eu-central-1 \
+AWS_PROFILE=... \
+./scripts/pytest
+```
+
+If you're using SSO, make sure to `aws sso login --profile $AWS_PROFILE` first.
+
+##### Minio
+
+If you want to run test without the cloud setup, we recommend [minio](https://min.io/docs/minio/linux/index.html).
+
+```bash
+# Start in Terminal 1
+mkdir /tmp/minio_data
+minio server /tmp/minio_data --console-address 127.0.0.1:9001 --address 127.0.0.1:9000
+```
+
+In another terminal, create an `aws` CLI profile for it:
+
+```ini
+# append to ~/.aws/config
+[profile local-minio]
+services = local-minio-services
+[services local-minio-services]
+s3 =
+  endpoint_url=http://127.0.0.1:9000/
+```
+
+
+Now configure the credentials (this is going to write `~/.aws/credentials` for you).
+It's an interactive prompt.
+
+```bash
+# Terminal 2
+$ aws --profile local-minio configure
+AWS Access Key ID [None]: minioadmin
+AWS Secret Access Key [None]: minioadmin
+Default region name [None]:
+Default output format [None]:
+```
+
+Now create a bucket `testbucket` using the CLI.
+
+```bash
+# (don't forget to have AWS_PROFILE env var set; or use --profile)
+aws --profile local-minio s3 mb s3://mybucket
+```
+
+(If it doesn't work, make sure you update your AWS CLI to a recent version.
+ The [service-specific endpoint feature](https://docs.aws.amazon.com/sdkref/latest/guide/feature-ss-endpoints.html)
+ that we're using is quite new.)
+
+```bash
+# with AWS PROFILE
+ENABLE_REAL_S3_REMOTE_STORAGE=true \
+REMOTE_STORAGE_S3_BUCKET=mybucket \
+REMOTE_STORAGE_S3_REGION=doesntmatterforminio \
+AWS_PROFILE=local-minio \
+./scripts/pytest
+```
+
+NB: you can avoid the `--profile` by setting the `AWS_PROFILE` variable.
+Just like the AWS SDKs, the `aws` CLI is sensible to it.
+
+#### Running Rust tests against real S3 or S3-compatible services
+
+We have some Rust tests that only run against real S3, e.g., [here](https://github.com/neondatabase/neon/blob/c18d3340b5e3c978a81c3db8b6f1e83cd9087e8a/libs/remote_storage/tests/test_real_s3.rs#L392-L397).
+
+They use the same env vars as the Python test suite (see previous section)
+but interpret them on their own.
+However, at this time, the interpretation is identical.
+
+So, above instructions apply to the Rust test as well.
+
 ### Writing a test

 Every test needs a Neon Environment, or NeonEnv to operate in. A Neon Environment
--- a/test_runner/fixtures/neon_fixtures.py
+++ b/test_runner/fixtures/neon_fixtures.py
@@ -14,7 +14,7 @@ import textwrap
 import threading
 import time
 import uuid
-from contextlib import ExitStack, closing, contextmanager
+from contextlib import closing, contextmanager
 from dataclasses import dataclass
 from datetime import datetime
 from enum import Enum
@@ -54,7 +54,7 @@ from fixtures.pageserver.allowed_errors import (
    DEFAULT_STORAGE_CONTROLLER_ALLOWED_ERRORS,
 )
 from fixtures.pageserver.http import PageserverHttpClient
-from fixtures.pageserver.types import IndexPartDump
+from fixtures.pageserver.types import IndexPartDump, LayerName, parse_layer_file_name
 from fixtures.pageserver.utils import (
    wait_for_last_record_lsn,
    wait_for_upload,
@@ -68,7 +68,7 @@ from fixtures.remote_storage import (
    RemoteStorageUser,
    S3Storage,
    default_remote_storage,
-    remote_storage_to_toml_inline_table,
+    remote_storage_to_toml_dict,
 )
 from fixtures.safekeeper.http import SafekeeperHttpClient
 from fixtures.safekeeper.utils import are_walreceivers_absent
@@ -82,6 +82,7 @@ from fixtures.utils import (
    subprocess_capture,
    wait_until,
 )
+from fixtures.utils import AuxFileStore as AuxFileStore  # reexport

 """
 This file contains pytest fixtures. A fixture is a test resource that can be
@@ -465,6 +466,7 @@ class NeonEnvBuilder:
        initial_tenant: Optional[TenantId] = None,
        initial_timeline: Optional[TimelineId] = None,
        pageserver_virtual_file_io_engine: Optional[str] = None,
+        pageserver_aux_file_policy: Optional[AuxFileStore] = None,
    ):
        self.repo_dir = repo_dir
        self.rust_log_override = rust_log_override
@@ -519,6 +521,8 @@ class NeonEnvBuilder:
            self.pageserver_validate_vectored_get = bool(validate)
            log.debug(f'Overriding pageserver validate_vectored_get config to "{validate}"')

+        self.pageserver_aux_file_policy = pageserver_aux_file_policy
+
        assert test_name.startswith(
            "test_"
        ), "Unexpectedly instantiated from outside a test function"
@@ -564,6 +568,7 @@ class NeonEnvBuilder:
            timeline_id=env.initial_timeline,
            shard_count=initial_tenant_shard_count,
            shard_stripe_size=initial_tenant_shard_stripe_size,
+            aux_file_v2=self.pageserver_aux_file_policy,
        )
        assert env.initial_tenant == initial_tenant
        assert env.initial_timeline == initial_timeline
@@ -982,7 +987,7 @@ class NeonEnv:

    Some notable functions and fields in NeonEnv:

-    postgres - A factory object for creating postgres compute nodes.
+    endpoints - A factory object for creating postgres compute nodes.

    pageservers - An array containing objects representing the pageservers

@@ -1051,15 +1056,16 @@ class NeonEnv:
        )

        self.pageserver_virtual_file_io_engine = config.pageserver_virtual_file_io_engine
+        self.pageserver_aux_file_policy = config.pageserver_aux_file_policy

-        # Create a config file corresponding to the options
+        # Create the neon_local's `NeonLocalInitConf`
        cfg: Dict[str, Any] = {
            "default_tenant_id": str(self.initial_tenant),
            "broker": {
                "listen_addr": self.broker.listen_addr(),
            },
-            "pageservers": [],
            "safekeepers": [],
+            "pageservers": [],
        }

        if self.control_plane_api is not None:
@@ -1098,6 +1104,17 @@ class NeonEnv:
            if config.pageserver_validate_vectored_get is not None:
                ps_cfg["validate_vectored_get"] = config.pageserver_validate_vectored_get

+            if self.pageserver_remote_storage is not None:
+                ps_cfg["remote_storage"] = remote_storage_to_toml_dict(
+                    self.pageserver_remote_storage
+                )
+
+            if config.pageserver_config_override is not None:
+                for o in config.pageserver_config_override.split(";"):
+                    override = toml.loads(o)
+                    for key, value in override.items():
+                        ps_cfg[key] = value
+
            # Create a corresponding NeonPageserver object
            self.pageservers.append(
                NeonPageserver(
@@ -1134,7 +1151,6 @@ class NeonEnv:
        self.neon_cli.init(
            cfg,
            force=config.config_init_force,
-            pageserver_config_override=config.pageserver_config_override,
        )

    def start(self):
@@ -1287,6 +1303,7 @@ def _shared_simple_env(
    pg_distrib_dir: Path,
    pg_version: PgVersion,
    pageserver_virtual_file_io_engine: str,
+    pageserver_aux_file_policy: Optional[AuxFileStore],
 ) -> Iterator[NeonEnv]:
    """
    # Internal fixture backing the `neon_simple_env` fixture. If TEST_SHARED_FIXTURES
@@ -1317,6 +1334,7 @@ def _shared_simple_env(
        test_name=request.node.name,
        test_output_dir=test_output_dir,
        pageserver_virtual_file_io_engine=pageserver_virtual_file_io_engine,
+        pageserver_aux_file_policy=pageserver_aux_file_policy,
    ) as builder:
        env = builder.init_start()

@@ -1356,6 +1374,7 @@ def neon_env_builder(
    test_overlay_dir: Path,
    top_output_dir: Path,
    pageserver_virtual_file_io_engine: str,
+    pageserver_aux_file_policy: Optional[AuxFileStore] = None,
 ) -> Iterator[NeonEnvBuilder]:
    """
    Fixture to create a Neon environment for test.
@@ -1389,6 +1408,7 @@ def neon_env_builder(
        test_name=request.node.name,
        test_output_dir=test_output_dir,
        test_overlay_dir=test_overlay_dir,
+        pageserver_aux_file_policy=pageserver_aux_file_policy,
    ) as builder:
        yield builder

@@ -1548,6 +1568,7 @@ class NeonCli(AbstractNeonCli):
        shard_stripe_size: Optional[int] = None,
        placement_policy: Optional[str] = None,
        set_default: bool = False,
+        aux_file_v2: Optional[AuxFileStore] = None,
    ) -> Tuple[TenantId, TimelineId]:
        """
        Creates a new tenant, returns its id and its initial timeline's id.
@@ -1571,6 +1592,16 @@ class NeonCli(AbstractNeonCli):
                    product(["-c"], (f"{key}:{value}" for key, value in conf.items()))
                )
            )
+
+        if aux_file_v2 is AuxFileStore.V2:
+            args.extend(["-c", "switch_aux_file_policy:v2"])
+
+        if aux_file_v2 is AuxFileStore.V1:
+            args.extend(["-c", "switch_aux_file_policy:v1"])
+
+        if aux_file_v2 is AuxFileStore.CrossValidation:
+            args.extend(["-c", "switch_aux_file_policy:cross_validation"])
+
        if set_default:
            args.append("--set-default")

@@ -1705,47 +1736,22 @@ class NeonCli(AbstractNeonCli):

    def init(
        self,
-        config: Dict[str, Any],
+        init_config: Dict[str, Any],
        force: Optional[str] = None,
-        pageserver_config_override: Optional[str] = None,
    ) -> "subprocess.CompletedProcess[str]":
-        remote_storage = self.env.pageserver_remote_storage
-
-        ps_config = {}
-        if remote_storage is not None:
-            remote_storage_toml_table = remote_storage_to_toml_inline_table(remote_storage)
-            ps_config["remote_storage"] = remote_storage_toml_table
-
-        if pageserver_config_override is not None:
-            for o in pageserver_config_override.split(";"):
-                override = toml.loads(o)
-                for key, value in override.items():
-                    ps_config[key] = value
-
-        with ExitStack() as stack:
-            ps_config_file = stack.enter_context(tempfile.NamedTemporaryFile(mode="w+"))
-            ps_config_file.write(toml.dumps(ps_config))
-            ps_config_file.flush()
-
-            neon_local_config = stack.enter_context(tempfile.NamedTemporaryFile(mode="w+"))
-            neon_local_config.write(toml.dumps(config))
-            neon_local_config.flush()
+        with tempfile.NamedTemporaryFile(mode="w+") as init_config_tmpfile:
+            init_config_tmpfile.write(toml.dumps(init_config))
+            init_config_tmpfile.flush()

            cmd = [
                "init",
-                f"--config={neon_local_config.name}",
-                "--pg-version",
-                self.env.pg_version,
-                f"--pageserver-config={ps_config_file.name}",
+                f"--config={init_config_tmpfile.name}",
            ]

            if force is not None:
                cmd.extend(["--force", force])

-            s3_env_vars = None
-            if isinstance(remote_storage, S3Storage):
-                s3_env_vars = remote_storage.access_env_vars()
-            res = self.raw_cli(cmd, extra_env_vars=s3_env_vars)
+            res = self.raw_cli(cmd)
            res.check_returncode()
        return res

@@ -1816,6 +1822,7 @@ class NeonCli(AbstractNeonCli):
        hot_standby: bool = False,
        lsn: Optional[Lsn] = None,
        pageserver_id: Optional[int] = None,
+        allow_multiple=False,
    ) -> "subprocess.CompletedProcess[str]":
        args = [
            "endpoint",
@@ -1839,6 +1846,8 @@ class NeonCli(AbstractNeonCli):
            args.extend(["--hot-standby", "true"])
        if pageserver_id is not None:
            args.extend(["--pageserver-id", str(pageserver_id)])
+        if allow_multiple:
+            args.extend(["--allow-multiple"])

        res = self.raw_cli(args)
        res.check_returncode()
@@ -1850,6 +1859,7 @@ class NeonCli(AbstractNeonCli):
        safekeepers: Optional[List[int]] = None,
        remote_ext_config: Optional[str] = None,
        pageserver_id: Optional[int] = None,
+        allow_multiple=False,
    ) -> "subprocess.CompletedProcess[str]":
        args = [
            "endpoint",
@@ -1864,6 +1874,8 @@ class NeonCli(AbstractNeonCli):
            args.append(endpoint_id)
        if pageserver_id is not None:
            args.extend(["--pageserver-id", str(pageserver_id)])
+        if allow_multiple:
+            args.extend(["--allow-multiple"])

        res = self.raw_cli(args)
        res.check_returncode()
@@ -2427,20 +2439,16 @@ class NeonPageserver(PgProtocol, LogUtils):
    def config_toml_path(self) -> Path:
        return self.workdir / "pageserver.toml"

-    def edit_config_toml(self, edit_fn: Callable[[Dict[str, Any]], bool]):
+    def edit_config_toml(self, edit_fn: Callable[[Dict[str, Any]], None]):
        """
        Edit the pageserver's config toml file in place.
-
-        The `edit_fn` is to manipulate the dict, and if it returns True, the file will be written.
-        If it returns False, no changes are made to the file system.
        """
        path = self.config_toml_path
        with open(path, "r") as f:
            config = toml.load(f)
-        save = edit_fn(config)
-        if save:
-            with open(path, "w") as f:
-                toml.dump(config, f)
+        edit_fn(config)
+        with open(path, "w") as f:
+            toml.dump(config, f)

    def patch_config_toml_nonrecursive(self, patch: Dict[str, Any]) -> Dict[str, Any]:
        """
@@ -2451,13 +2459,12 @@ class NeonPageserver(PgProtocol, LogUtils):
        """
        replacements = {}

-        def doit(config: Dict[str, Any]) -> bool:
+        def doit(config: Dict[str, Any]):
            while len(patch) > 0:
                key, new = patch.popitem()
                old = config.get(key, None)
                config[key] = new
                replacements[key] = old
-            return True

        self.edit_config_toml(doit)
        return replacements
@@ -2635,6 +2642,37 @@ class NeonPageserver(PgProtocol, LogUtils):
            tenant_id, generation=self.env.storage_controller.attach_hook_issue(tenant_id, self.id)
        )

+    def list_layers(self, tenant_id: TenantId, timeline_id: TimelineId) -> list[Path]:
+        """
+        Inspect local storage on a pageserver to discover which layer files are present.
+
+        :return: list of relative paths to layers, from the timeline root.
+        """
+        timeline_path = self.timeline_dir(tenant_id, timeline_id)
+
+        def relative(p: Path) -> Path:
+            return p.relative_to(timeline_path)
+
+        return sorted(
+            list(
+                map(
+                    relative,
+                    filter(
+                        lambda path: path.name != "metadata"
+                        and "ephemeral" not in path.name
+                        and "temp" not in path.name,
+                        timeline_path.glob("*"),
+                    ),
+                )
+            )
+        )
+
+    def layer_exists(
+        self, tenant_id: TenantId, timeline_id: TimelineId, layer_name: LayerName
+    ) -> bool:
+        layers = self.list_layers(tenant_id, timeline_id)
+        return layer_name in [parse_layer_file_name(p.name) for p in layers]
+

 class PgBin:
    """A helper class for executing postgres binaries"""
@@ -3323,6 +3361,7 @@ class Endpoint(PgProtocol):
        lsn: Optional[Lsn] = None,
        config_lines: Optional[List[str]] = None,
        pageserver_id: Optional[int] = None,
+        allow_multiple: bool = False,
    ) -> "Endpoint":
        """
        Create a new Postgres endpoint.
@@ -3345,6 +3384,7 @@ class Endpoint(PgProtocol):
            pg_port=self.pg_port,
            http_port=self.http_port,
            pageserver_id=pageserver_id,
+            allow_multiple=allow_multiple,
        )
        path = Path("endpoints") / self.endpoint_id / "pgdata"
        self.pgdata_dir = os.path.join(self.env.repo_dir, path)
@@ -3361,7 +3401,10 @@ class Endpoint(PgProtocol):
        return self

    def start(
-        self, remote_ext_config: Optional[str] = None, pageserver_id: Optional[int] = None
+        self,
+        remote_ext_config: Optional[str] = None,
+        pageserver_id: Optional[int] = None,
+        allow_multiple: bool = False,
    ) -> "Endpoint":
        """
        Start the Postgres instance.
@@ -3377,6 +3420,7 @@ class Endpoint(PgProtocol):
            safekeepers=self.active_safekeepers,
            remote_ext_config=remote_ext_config,
            pageserver_id=pageserver_id,
+            allow_multiple=allow_multiple,
        )
        self.running = True

@@ -3506,6 +3550,7 @@ class Endpoint(PgProtocol):
        config_lines: Optional[List[str]] = None,
        remote_ext_config: Optional[str] = None,
        pageserver_id: Optional[int] = None,
+        allow_multiple=False,
    ) -> "Endpoint":
        """
        Create an endpoint, apply config, and start Postgres.
@@ -3521,7 +3566,12 @@ class Endpoint(PgProtocol):
            hot_standby=hot_standby,
            lsn=lsn,
            pageserver_id=pageserver_id,
-        ).start(remote_ext_config=remote_ext_config, pageserver_id=pageserver_id)
+            allow_multiple=allow_multiple,
+        ).start(
+            remote_ext_config=remote_ext_config,
+            pageserver_id=pageserver_id,
+            allow_multiple=allow_multiple,
+        )

        log.info(f"Postgres startup took {time.time() - started_at} seconds")

--- a/test_runner/fixtures/pageserver/allowed_errors.py
+++ b/test_runner/fixtures/pageserver/allowed_errors.py
@@ -88,7 +88,9 @@ DEFAULT_PAGESERVER_ALLOWED_ERRORS = (
    ".*Flushed oversized open layer with size.*",
    # During teardown, we stop the storage controller before the pageservers, so pageservers
    # can experience connection errors doing background deletion queue work.
-    ".*WARN deletion backend: calling control plane generation validation API failed.*Connection refused.*",
+    ".*WARN deletion backend: calling control plane generation validation API failed.*error sending request.*",
+    # Can happen when the test shuts down the storage controller while it is calling the utilization API
+    ".*WARN.*path=/v1/utilization .*request was dropped before completing",
 )


--- a/test_runner/fixtures/pageserver/http.py
+++ b/test_runner/fixtures/pageserver/http.py
@@ -819,6 +819,23 @@ class PageserverHttpClient(requests.Session, MetricsGetter):
                continue
            self.download_layer(tenant_id, timeline_id, layer.layer_file_name)

+    def detach_ancestor(
+        self,
+        tenant_id: Union[TenantId, TenantShardId],
+        timeline_id: TimelineId,
+        batch_size: int | None = None,
+    ) -> Set[TimelineId]:
+        params = {}
+        if batch_size is not None:
+            params["batch_size"] = batch_size
+        res = self.put(
+            f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline/{timeline_id}/detach_ancestor",
+            params=params,
+        )
+        self.verbose_error(res)
+        json = res.json()
+        return set(map(TimelineId, json["reparented_timelines"]))
+
    def evict_layer(
        self, tenant_id: Union[TenantId, TenantShardId], timeline_id: TimelineId, layer_name: str
    ):
--- a/test_runner/fixtures/pageserver/types.py
+++ b/test_runner/fixtures/pageserver/types.py
@@ -1,3 +1,4 @@
+import re
 from dataclasses import dataclass
 from typing import Any, Dict, Tuple, Union

@@ -11,7 +12,7 @@ class IndexLayerMetadata:


@dataclass(frozen=True)
-class ImageLayerFileName:
+class ImageLayerName:
    lsn: Lsn
    key_start: Key
    key_end: Key
@@ -25,7 +26,7 @@ class ImageLayerFileName:


@dataclass(frozen=True)
-class DeltaLayerFileName:
+class DeltaLayerName:
    lsn_start: Lsn
    lsn_end: Lsn
    key_start: Key
@@ -40,65 +41,57 @@ class DeltaLayerFileName:
        return ret


-LayerFileName = Union[ImageLayerFileName, DeltaLayerFileName]
+LayerName = Union[ImageLayerName, DeltaLayerName]


 class InvalidFileName(Exception):
    pass


+IMAGE_LAYER_FILE_NAME = re.compile(
+    "^([A-F0-9]{36})-([A-F0-9]{36})__([A-F0-9]{16})(-v1-[a-f0-9]{8})?$"
+)
+
+
 def parse_image_layer(f_name: str) -> Tuple[int, int, int]:
    """Parse an image layer file name. Return key start, key end, and snapshot lsn"""
-    parts = f_name.split("__")
-    if len(parts) != 2:
-        raise InvalidFileName(f"expecting two parts separated by '__', got: {parts}")
-    key_parts = parts[0].split("-")
-    if len(key_parts) != 2:
-        raise InvalidFileName(
-            f"expecting two key parts separated by '--' in parts[0], got: {key_parts}"
-        )
-    try:
-        return int(key_parts[0], 16), int(key_parts[1], 16), int(parts[1], 16)
-    except ValueError as e:
-        raise InvalidFileName(f"conversion error: {f_name}") from e
+
+    match = IMAGE_LAYER_FILE_NAME.match(f_name)
+    if match is None:
+        raise InvalidFileName(f"'{f_name}' is not an image layer filename")
+
+    return int(match.group(1), 16), int(match.group(2), 16), int(match.group(3), 16)
+
+
+DELTA_LAYER_FILE_NAME = re.compile(
+    "^([A-F0-9]{36})-([A-F0-9]{36})__([A-F0-9]{16})-([A-F0-9]{16})(-v1-[a-f0-9]{8})?$"
+)


 def parse_delta_layer(f_name: str) -> Tuple[int, int, int, int]:
    """Parse a delta layer file name. Return key start, key end, lsn start, and lsn end"""
-    parts = f_name.split("__")
-    if len(parts) != 2:
-        raise InvalidFileName(f"expecting two parts separated by '__', got: {parts}")
-    key_parts = parts[0].split("-")
-    if len(key_parts) != 2:
-        raise InvalidFileName(
-            f"expecting two key parts separated by '--' in parts[0], got: {key_parts}"
-        )
-    lsn_parts = parts[1].split("-")
-    if len(lsn_parts) != 2:
-        raise InvalidFileName(
-            f"expecting two lsn parts separated by '--' in parts[1], got: {lsn_parts}"
-        )
-    try:
-        return (
-            int(key_parts[0], 16),
-            int(key_parts[1], 16),
-            int(lsn_parts[0], 16),
-            int(lsn_parts[1], 16),
-        )
-    except ValueError as e:
-        raise InvalidFileName(f"conversion error: {f_name}") from e
+    match = DELTA_LAYER_FILE_NAME.match(f_name)
+    if match is None:
+        raise InvalidFileName(f"'{f_name}' is not an delta layer filename")
+
+    return (
+        int(match.group(1), 16),
+        int(match.group(2), 16),
+        int(match.group(3), 16),
+        int(match.group(4), 16),
+    )


-def parse_layer_file_name(file_name: str) -> LayerFileName:
+def parse_layer_file_name(file_name: str) -> LayerName:
    try:
        key_start, key_end, lsn = parse_image_layer(file_name)
-        return ImageLayerFileName(lsn=Lsn(lsn), key_start=Key(key_start), key_end=Key(key_end))
+        return ImageLayerName(lsn=Lsn(lsn), key_start=Key(key_start), key_end=Key(key_end))
    except InvalidFileName:
        pass

    try:
        key_start, key_end, lsn_start, lsn_end = parse_delta_layer(file_name)
-        return DeltaLayerFileName(
+        return DeltaLayerName(
            lsn_start=Lsn(lsn_start),
            lsn_end=Lsn(lsn_end),
            key_start=Key(key_start),
@@ -110,18 +103,15 @@ def parse_layer_file_name(file_name: str) -> LayerFileName:
    raise InvalidFileName("neither image nor delta layer")


-def is_future_layer(layer_file_name: LayerFileName, disk_consistent_lsn: Lsn):
+def is_future_layer(layer_file_name: LayerName, disk_consistent_lsn: Lsn):
    """
    Determines if this layer file is considered to be in future meaning we will discard these
    layers during timeline initialization from the given disk_consistent_lsn.
    """
-    if (
-        isinstance(layer_file_name, ImageLayerFileName)
-        and layer_file_name.lsn > disk_consistent_lsn
-    ):
+    if isinstance(layer_file_name, ImageLayerName) and layer_file_name.lsn > disk_consistent_lsn:
        return True
    elif (
-        isinstance(layer_file_name, DeltaLayerFileName)
+        isinstance(layer_file_name, DeltaLayerName)
        and layer_file_name.lsn_end > disk_consistent_lsn + 1
    ):
        return True
@@ -131,7 +121,7 @@ def is_future_layer(layer_file_name: LayerFileName, disk_consistent_lsn: Lsn):

@dataclass
 class IndexPartDump:
-    layer_metadata: Dict[LayerFileName, IndexLayerMetadata]
+    layer_metadata: Dict[LayerName, IndexLayerMetadata]
    disk_consistent_lsn: Lsn

    @classmethod
--- a/test_runner/fixtures/parametrize.py
+++ b/test_runner/fixtures/parametrize.py
@@ -5,6 +5,7 @@ import pytest
 from _pytest.python import Metafunc

 from fixtures.pg_version import PgVersion
+from fixtures.utils import AuxFileStore

 """
 Dynamically parametrize tests by different parameters
@@ -31,6 +32,11 @@ def pageserver_virtual_file_io_engine() -> Optional[str]:
    return os.getenv("PAGESERVER_VIRTUAL_FILE_IO_ENGINE")


+@pytest.fixture(scope="function", autouse=True)
+def pageserver_aux_file_policy() -> Optional[AuxFileStore]:
+    return None
+
+
 def pytest_generate_tests(metafunc: Metafunc):
    if (bt := os.getenv("BUILD_TYPE")) is None:
        build_types = ["debug", "release"]
--- a/test_runner/fixtures/remote_storage.py
+++ b/test_runner/fixtures/remote_storage.py
@@ -50,7 +50,7 @@ class MockS3Server:
        # XXX: do not use `shell=True` or add `exec ` to the command here otherwise.
        # We use `self.subprocess.kill()` to shut down the server, which would not "just" work in Linux
        # if a process is started from the shell process.
-        self.subprocess = subprocess.Popen(["poetry", "run", "moto_server", "s3", f"-p{port}"])
+        self.subprocess = subprocess.Popen(["poetry", "run", "moto_server", f"-p{port}"])
        error = None
        try:
            return_code = self.subprocess.poll()
@@ -141,11 +141,13 @@ class LocalFsStorage:
        with self.heatmap_path(tenant_id).open("r") as f:
            return json.load(f)

-    def to_toml_inline_table(self) -> str:
-        rv = {
+    def to_toml_dict(self) -> Dict[str, Any]:
+        return {
            "local_path": str(self.root),
        }
-        return toml.TomlEncoder().dump_inline_table(rv)
+
+    def to_toml_inline_table(self) -> str:
+        return toml.TomlEncoder().dump_inline_table(self.to_toml_dict())

    def cleanup(self):
        # no cleanup is done here, because there's NeonEnvBuilder.cleanup_local_storage which will remove everything, including localfs files
@@ -194,7 +196,7 @@ class S3Storage:
            }
        )

-    def to_toml_inline_table(self) -> str:
+    def to_toml_dict(self) -> Dict[str, Any]:
        rv = {
            "bucket_name": self.bucket_name,
            "bucket_region": self.bucket_region,
@@ -206,7 +208,10 @@ class S3Storage:
        if self.endpoint is not None:
            rv["endpoint"] = self.endpoint

-        return toml.TomlEncoder().dump_inline_table(rv)
+        return rv
+
+    def to_toml_inline_table(self) -> str:
+        return toml.TomlEncoder().dump_inline_table(self.to_toml_dict())

    def do_cleanup(self):
        if not self.cleanup:
@@ -414,6 +419,13 @@ def default_remote_storage() -> RemoteStorageKind:
    return RemoteStorageKind.LOCAL_FS


+def remote_storage_to_toml_dict(remote_storage: RemoteStorage) -> Dict[str, Any]:
+    if not isinstance(remote_storage, (LocalFsStorage, S3Storage)):
+        raise Exception("invalid remote storage type")
+
+    return remote_storage.to_toml_dict()
+
+
 # serialize as toml inline table
 def remote_storage_to_toml_inline_table(remote_storage: RemoteStorage) -> str:
    if not isinstance(remote_storage, (LocalFsStorage, S3Storage)):
--- a/test_runner/fixtures/utils.py
+++ b/test_runner/fixtures/utils.py
@@ -1,4 +1,5 @@
 import contextlib
+import enum
 import json
 import os
 import re
@@ -484,3 +485,16 @@ def assert_no_errors(log_file, service, allowed_errors):
        log.info(f"not allowed {service} error: {error.strip()}")

    assert not errors, f"Log errors on {service}: {errors[0]}"
+
+
+@enum.unique
+class AuxFileStore(str, enum.Enum):
+    V1 = "V1"
+    V2 = "V2"
+    CrossValidation = "CrossValidation"
+
+    def __repr__(self) -> str:
+        return f"'aux-{self.value}'"
+
+    def __str__(self) -> str:
+        return f"'aux-{self.value}'"
--- a/test_runner/performance/test_branch_creation.py
+++ b/test_runner/performance/test_branch_creation.py
@@ -141,9 +141,10 @@ def test_branch_creation_many(neon_compare: NeonCompare, n_branches: int, shape:
    # start without gc so we can time compaction with less noise; use shorter
    # period for compaction so it starts earlier
    def patch_default_tenant_config(config):
-        config["compaction_period"] = "3s"
-        config["gc_period"] = "0s"
-        return True
+        tenant_config = config.get("tenant_config", {})
+        tenant_config["compaction_period"] = "3s"
+        tenant_config["gc_period"] = "0s"
+        config["tenant_config"] = tenant_config

    env.pageserver.edit_config_toml(patch_default_tenant_config)
    env.pageserver.start(
--- a/test_runner/performance/test_storage_controller_scale.py
+++ b/test_runner/performance/test_storage_controller_scale.py
@@ -102,6 +102,9 @@ def test_storage_controller_many_tenants(
                tenant_id,
                shard_count,
                stripe_size,
+                # Upload heatmaps fast, so that secondary downloads happen promptly, enabling
+                # the controller's optimization migrations to proceed promptly.
+                tenant_config={"heatmap_period": "10s"},
                placement_policy={"Attached": 1},
            )
            futs.append(f)
--- a/test_runner/regress/test_attach_tenant_config.py
+++ b/test_runner/regress/test_attach_tenant_config.py
@@ -190,7 +190,7 @@ def test_fully_custom_config(positive_env: NeonEnv):
        "trace_read_requests": True,
        "walreceiver_connect_timeout": "13m",
        "image_layer_creation_check_threshold": 1,
-        "switch_to_aux_file_v2": True,
+        "switch_aux_file_policy": "CrossValidation",
    }

    ps_http = env.pageserver.http_client()
--- a/Show More
+++ b/Show More