Add trailing dot

Conflicts:
	control_plane/src/pageserver.rs
    => pick ours

.github/workflows/build_and_test.yml

@@ -236,6 +236,27 @@ 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

Cargo.lock

@@ -4372,7 +4372,6 @@ dependencies = [
  "hyper 1.2.0",
  "hyper-tungstenite",
  "hyper-util",
- "indexmap 2.0.1",
  "ipnet",
  "itertools",
  "lasso",

Cargo.toml

@@ -99,7 +99,6 @@ 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"

Makefile

@@ -81,14 +81,11 @@ $(POSTGRES_INSTALL_DIR)/build/%/config.status:
 		echo "'git submodule update --init --recursive --depth 2 --progress .' in project root.\n"; \
 		exit 1; }
 	mkdir -p $(POSTGRES_INSTALL_DIR)/build/$*
-
-	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-$$VERSION/configure \
+	(cd $(POSTGRES_INSTALL_DIR)/build/$* && \
+	env PATH="$(EXTRA_PATH_OVERRIDES):$$PATH" $(ROOT_PROJECT_DIR)/vendor/postgres-$*/configure \
 		CFLAGS='$(PG_CFLAGS)' \
-		$(PG_CONFIGURE_OPTS) --with-extra-version=" ($$EXTRA_VERSION)" \
-		--prefix=$(abspath $(POSTGRES_INSTALL_DIR))/$$VERSION > configure.log)
+		$(PG_CONFIGURE_OPTS) \
+		--prefix=$(abspath $(POSTGRES_INSTALL_DIR))/$* > configure.log)
 
 # nicer alias to run 'configure'
 # Note: I've been unable to use templates for this part of our configuration.

compute_tools/src/bin/compute_ctl.rs

@@ -51,7 +51,6 @@ 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,
@@ -70,34 +69,6 @@ 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])?;
@@ -112,15 +83,9 @@ fn init() -> Result<(String, clap::ArgMatches)> {
         .to_string();
     info!("build_tag: {build_tag}");
 
-    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 matches = cli().get_matches();
+    let pgbin_default = String::from("postgres");
+    let pgbin = matches.get_one::<String>("pgbin").unwrap_or(&pgbin_default);
 
     let ext_remote_storage = matches
         .get_one::<String>("remote-ext-config")
@@ -148,30 +113,6 @@ fn process_cli(matches: &clap::ArgMatches) -> Result<ProcessCliResult> {
     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.
@@ -208,7 +149,7 @@ fn startup_context_from_env() -> Option<opentelemetry::ContextGuard> {
     if let Ok(val) = std::env::var("TRACESTATE") {
         startup_tracing_carrier.insert("tracestate".to_string(), val);
     }
-    if !startup_tracing_carrier.is_empty() {
+    let startup_context_guard = if !startup_tracing_carrier.is_empty() {
         use opentelemetry::propagation::TextMapPropagator;
         use opentelemetry::sdk::propagation::TraceContextPropagator;
         let guard = TraceContextPropagator::new()
@@ -218,17 +159,8 @@ fn startup_context_from_env() -> Option<opentelemetry::ContextGuard> {
         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");
 
@@ -269,34 +201,6 @@ fn try_spec_from_cli(
         }
     };
 
-    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;
 
@@ -335,6 +239,8 @@ fn wait_spec(
     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");
@@ -363,29 +269,6 @@ fn wait_spec(
         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;
@@ -435,10 +318,10 @@ fn start_postgres(
         }
     }
 
-    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),
@@ -493,7 +376,7 @@ fn start_postgres(
             // 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(),
@@ -506,41 +389,12 @@ fn start_postgres(
         }
     }
 
-    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");
@@ -555,25 +409,6 @@ fn wait_postgres(pg: Option<PostgresHandle>) -> Result<WaitPostgresResult> {
         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.
@@ -615,19 +450,13 @@ fn cleanup_after_postgres_exit(
         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:

control_plane/src/bin/neon_local.rs

@@ -382,7 +382,7 @@ fn handle_init(init_match: &ArgMatches) -> anyhow::Result<LocalEnv> {
     // Initialize pageserver, create initial tenant and timeline.
     for ps_conf in &env.pageservers {
         PageServerNode::from_env(&env, ps_conf)
-            .initialize(pageserver_config.clone())
+            .initialize(&pageserver_config)
             .unwrap_or_else(|e| {
                 eprintln!("pageserver init failed: {e:?}");
                 exit(1);
@@ -835,8 +835,6 @@ 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,
@@ -854,9 +852,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                 _ => {}
             }
 
-            if !allow_multiple {
-                cplane.check_conflicting_endpoints(mode, tenant_id, timeline_id)?;
-            }
+            cplane.check_conflicting_endpoints(mode, tenant_id, timeline_id)?;
 
             cplane.new_endpoint(
                 &endpoint_id,
@@ -885,8 +881,6 @@ 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") {
@@ -912,13 +906,11 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                 .cloned()
                 .unwrap_or_default();
 
-            if !allow_multiple {
-                cplane.check_conflicting_endpoints(
-                    endpoint.mode,
-                    endpoint.tenant_id,
-                    endpoint.timeline_id,
-                )?;
-            }
+            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();
@@ -1434,12 +1426,6 @@ 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)
@@ -1602,7 +1588,6 @@ 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.")
@@ -1611,7 +1596,6 @@ 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")

control_plane/src/local_env.rs

@@ -562,10 +562,6 @@ impl LocalEnv {
             fs::create_dir_all(SafekeeperNode::datadir_path_by_id(self, safekeeper.id))?;
         }
 
-        for ps in &self.pageservers {
-            fs::create_dir(self.pageserver_data_dir(ps.id))?;
-        }
-
         self.persist_config(base_path)
     }
 

control_plane/src/pageserver.rs

@@ -10,15 +10,14 @@ use std::io;
 use std::io::Write;
 use std::num::NonZeroU64;
 use std::path::PathBuf;
-use std::str::FromStr;
+use std::process::Command;
 use std::time::Duration;
 
 use anyhow::{bail, Context};
 use camino::Utf8PathBuf;
 use futures::SinkExt;
 use pageserver_api::models::{
-    self, AuxFilePolicy, LocationConfig, ShardParameters, TenantHistorySize, TenantInfo,
-    TimelineInfo,
+    self, LocationConfig, ShardParameters, TenantHistorySize, TenantInfo, TimelineInfo,
 };
 use pageserver_api::shard::TenantShardId;
 use pageserver_client::mgmt_api;
@@ -74,12 +73,10 @@ impl PageServerNode {
         }
     }
 
-    fn pageserver_init_make_toml(
-        &self,
-        cli_overrides: toml_edit::Document,
-    ) -> anyhow::Result<toml_edit::Document> {
-        // TODO: this is a legacy code, it should be refactored to use toml_edit directly.
-
+    /// 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> {
         // FIXME: the paths should be shell-escaped to handle paths with spaces, quotas etc.
         let pg_distrib_dir_param = format!(
             "pg_distrib_dir='{}'",
@@ -174,21 +171,12 @@ impl PageServerNode {
         // Apply the user-provided overrides
         overrides.push(cli_overrides.to_string());
 
-        // 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)
+        overrides
     }
 
     /// Initializes a pageserver node by creating its config with the overrides provided.
-    pub fn initialize(&self, config_overrides: toml_edit::Document) -> anyhow::Result<()> {
+    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)
             .with_context(|| format!("Failed to run init for pageserver node {}", self.conf.id))
     }
@@ -209,7 +197,7 @@ impl PageServerNode {
         self.start_node().await
     }
 
-    fn pageserver_init(&self, cli_overrides: toml_edit::Document) -> anyhow::Result<()> {
+    fn pageserver_init(&self, config_overrides: &toml_edit::Document) -> anyhow::Result<()> {
         let datadir = self.repo_path();
         let node_id = self.conf.id;
         println!(
@@ -220,20 +208,36 @@ impl PageServerNode {
         );
         io::stdout().flush()?;
 
-        let config = self
-            .pageserver_init_make_toml(cli_overrides)
-            .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
+        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),
+        );
 
         // Write metadata file, used by pageserver on startup to register itself with
         // the storage controller
@@ -425,11 +429,11 @@ impl PageServerNode {
                 .map(serde_json::from_str)
                 .transpose()
                 .context("parse `timeline_get_throttle` from json")?,
-            switch_aux_file_policy: settings
-                .remove("switch_aux_file_policy")
-                .map(|x| x.parse::<AuxFilePolicy>())
+            switch_to_aux_file_v2: settings
+                .remove("switch_to_aux_file_v2")
+                .map(|x| x.parse::<bool>())
                 .transpose()
-                .context("Failed to parse 'switch_aux_file_policy'")?,
+                .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")
@@ -548,11 +552,11 @@ impl PageServerNode {
                     .map(serde_json::from_str)
                     .transpose()
                     .context("parse `timeline_get_throttle` from json")?,
-                switch_aux_file_policy: settings
-                    .remove("switch_aux_file_policy")
-                    .map(|x| x.parse::<AuxFilePolicy>())
+                switch_to_aux_file_v2: settings
+                    .remove("switch_to_aux_file_v2")
+                    .map(|x| x.parse::<bool>())
                     .transpose()
-                    .context("Failed to parse 'switch_aux_file_policy'")?,
+                    .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
             }
         };
 

libs/metrics/src/lib.rs

@@ -480,15 +480,6 @@ 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>

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.
-    pub fn raw_size(range: &Range<Key>) -> u32 {
+    fn raw_size(range: &Range<Key>) -> u32 {
         if is_contiguous_range(range) {
             contiguous_range_len(range)
         } else {

libs/pageserver_api/src/models.rs

@@ -1,4 +1,3 @@
-pub mod detach_ancestor;
 pub mod partitioning;
 pub mod utilization;
 
@@ -9,7 +8,6 @@ use std::{
     collections::HashMap,
     io::{BufRead, Read},
     num::{NonZeroU64, NonZeroUsize},
-    str::FromStr,
     time::{Duration, SystemTime},
 };
 
@@ -305,31 +303,7 @@ 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_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)
-        }
-    }
+    pub switch_to_aux_file_v2: Option<bool>,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]

libs/pageserver_api/src/models/detach_ancestor.rs

@@ -1,6 +0,0 @@
-use utils::id::TimelineId;
-
-#[derive(Default, serde::Serialize)]
-pub struct AncestorDetached {
-    pub reparented_timelines: Vec<TimelineId>,
-}

libs/postgres_ffi/src/xlog_utils.rs

@@ -331,10 +331,7 @@ 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 =
@@ -370,16 +367,8 @@ 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;
-    // 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,
-        )
+    let (shdr_rem_len, infoflags) = if first_page_only {
+        (seg_off, pg_constants::XLP_FIRST_IS_CONTRECORD)
     } else {
         (0, 0)
     };
@@ -408,22 +397,20 @@ 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,
+            xlp_info: if page_off >= pg_constants::SIZE_OF_PAGE_HEADER as u64 {
+                pg_constants::XLP_FIRST_IS_CONTRECORD
+            } else {
+                0
+            },
             xlp_tli: PG_TLI,
             xlp_pageaddr: lsn.page_lsn().0,
-            xlp_rem_len,
+            xlp_rem_len: if page_off >= pg_constants::SIZE_OF_PAGE_HEADER as u64 {
+                page_off as u32
+            } else {
+                0u32
+            },
             ..Default::default() // Put 0 in padding fields.
         };
         let hdr_bytes = header.encode()?;

libs/walproposer/src/api_bindings.rs

@@ -50,14 +50,6 @@ 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;
@@ -399,7 +391,6 @@ 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),
@@ -430,32 +421,6 @@ 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)

libs/walproposer/src/walproposer.rs

@@ -1,5 +1,8 @@
 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::{
@@ -7,8 +10,6 @@ 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.
@@ -27,10 +28,6 @@ pub trait ApiImpl {
         todo!()
     }
 
-    fn update_donor(&self, _donor: &mut Safekeeper, _donor_lsn: u64) {
-        todo!()
-    }
-
     fn get_current_timestamp(&self) -> i64 {
         todo!()
     }
@@ -277,7 +274,6 @@ 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};
@@ -301,8 +297,6 @@ 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 {
@@ -333,22 +327,11 @@ 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,
@@ -524,7 +507,6 @@ mod tests {
             ],
             replies_ptr: AtomicUsize::new(0),
             sync_channel: sender,
-            shmem: UnsafeCell::new(crate::api_bindings::empty_shmem()),
         });
         let config = crate::walproposer::Config {
             ttid,

pageserver/src/aux_file.rs

@@ -1,4 +1,3 @@
-use bytes::{Buf, BufMut, Bytes};
 use pageserver_api::key::{Key, AUX_KEY_PREFIX, METADATA_KEY_SIZE};
 use tracing::warn;
 
@@ -62,84 +61,6 @@ 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::*;
@@ -188,21 +109,4 @@ 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()
-        );
-    }
 }

pageserver/src/disk_usage_eviction_task.rs

@@ -540,12 +540,7 @@ 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,
-                                layer.metadata,
-                            )
+                            .evict_layer(tenant_manager.get_conf(), layer.timeline_id, layer.name)
                             .await;
                         Ok(file_size)
                     });

pageserver/src/http/routes.rs

@@ -63,7 +63,6 @@ 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::LayerFileName;
 use crate::tenant::timeline::CompactFlags;
 use crate::tenant::timeline::Timeline;
 use crate::tenant::SpawnMode;
@@ -1229,15 +1228,13 @@ 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 = LayerFileName::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_name)
+        .download_layer(layer_file_name)
         .await
         .map_err(ApiError::InternalServerError)?;
 
@@ -1261,14 +1258,11 @@ 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 = LayerFileName::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_name)
+        .evict_layer(layer_file_name)
         .await
         .map_err(ApiError::InternalServerError)?;
 
@@ -1833,75 +1827,6 @@ 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,
@@ -2590,10 +2515,6 @@ 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)
         })

pageserver/src/metrics.rs

@@ -1512,80 +1512,29 @@ static REMOTE_TIMELINE_CLIENT_BYTES_FINISHED_COUNTER: Lazy<IntCounterVec> = Lazy
 });
 
 pub(crate) struct TenantManagerMetrics {
-    tenant_slots_attached: UIntGauge,
-    tenant_slots_secondary: UIntGauge,
-    tenant_slots_inprogress: UIntGauge,
+    pub(crate) tenant_slots: 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(|| {
-    let tenant_slots = register_uint_gauge_vec!(
+    TenantManagerMetrics {
+    tenant_slots: register_uint_gauge!(
         "pageserver_tenant_manager_slots",
         "How many slots currently exist, including all attached, secondary and in-progress operations",
-        &["mode"]
     )
-    .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"),
-    }
+    .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"),
+}
 });
 
 pub(crate) struct DeletionQueueMetrics {
@@ -2326,7 +2275,6 @@ 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 {
@@ -2929,8 +2877,6 @@ 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| {

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;
@@ -24,7 +24,6 @@ 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;
@@ -671,7 +670,7 @@ impl Timeline {
         self.get(CHECKPOINT_KEY, lsn, ctx).await
     }
 
-    async fn list_aux_files_v1(
+    pub(crate) async fn list_aux_files(
         &self,
         lsn: Lsn,
         ctx: &RequestContext,
@@ -689,63 +688,6 @@ 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.
     ///
@@ -1447,9 +1389,6 @@ 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(),
         })?;
@@ -1465,122 +1404,90 @@ impl<'a> DatadirModification<'a> {
         content: &[u8],
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        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 {
-                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 file_path = path.to_string();
+        let content = if content.is_empty() {
+            None
+        } else {
+            Some(Bytes::copy_from_slice(content))
+        };
 
-        if let AuxFilePolicy::V1 | AuxFilePolicy::CrossValidation = policy {
-            let file_path = path.to_string();
-            let content = if content.is_empty() {
-                None
+        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;
             } else {
-                Some(Bytes::copy_from_slice(content))
-            };
+                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 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 {
+                    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")?,
                         )),
                     );
-                    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);
-                    }
+                    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(())
     }
 

pageserver/src/repository.rs

@@ -33,6 +33,7 @@ impl Value {
     }
 }
 
+#[cfg(test)]
 #[derive(Debug, PartialEq)]
 pub(crate) enum InvalidInput {
     TooShortValue,
@@ -41,8 +42,10 @@ 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 {

pageserver/src/task_mgr.rs

@@ -319,9 +319,6 @@ 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,
 
@@ -370,8 +367,6 @@ pub enum TaskKind {
 
     #[cfg(test)]
     UnitTest,
-
-    DetachAncestor,
 }
 
 #[derive(Default)]

pageserver/src/tenant.rs

@@ -322,9 +322,6 @@ 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 {
@@ -1679,34 +1676,6 @@ 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()
     }
@@ -2560,7 +2529,6 @@ impl Tenant {
                 &crate::metrics::tenant_throttling::TIMELINE_GET,
             )),
             tenant_conf: Arc::new(ArcSwap::from_pointee(attached_conf)),
-            ongoing_timeline_detach: std::sync::Mutex::default(),
         }
     }
 
@@ -3758,7 +3726,7 @@ pub(crate) mod harness {
                 image_layer_creation_check_threshold: Some(
                     tenant_conf.image_layer_creation_check_threshold,
                 ),
-                switch_aux_file_policy: Some(tenant_conf.switch_aux_file_policy),
+                switch_to_aux_file_v2: Some(tenant_conf.switch_to_aux_file_v2),
             }
         }
     }

pageserver/src/tenant/config.rs

@@ -9,7 +9,6 @@
 //! 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};
@@ -371,9 +370,9 @@ pub struct TenantConf {
     // Expresed in multiples of checkpoint distance.
     pub image_layer_creation_check_threshold: u8,
 
-    /// Switch to a new aux file policy. Switching this flag requires the user has not written any aux file into
+    /// Switch to aux file v2. 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_aux_file_policy: AuxFilePolicy,
+    pub switch_to_aux_file_v2: bool,
 }
 
 /// Same as TenantConf, but this struct preserves the information about
@@ -472,7 +471,7 @@ pub struct TenantConfOpt {
 
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(default)]
-    pub switch_aux_file_policy: Option<AuxFilePolicy>,
+    pub switch_to_aux_file_v2: Option<bool>,
 }
 
 impl TenantConfOpt {
@@ -530,9 +529,9 @@ impl TenantConfOpt {
             image_layer_creation_check_threshold: self
                 .image_layer_creation_check_threshold
                 .unwrap_or(global_conf.image_layer_creation_check_threshold),
-            switch_aux_file_policy: self
-                .switch_aux_file_policy
-                .unwrap_or(global_conf.switch_aux_file_policy),
+            switch_to_aux_file_v2: self
+                .switch_to_aux_file_v2
+                .unwrap_or(global_conf.switch_to_aux_file_v2),
         }
     }
 }
@@ -574,7 +573,7 @@ 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_aux_file_policy: AuxFilePolicy::V1,
+            switch_to_aux_file_v2: false,
         }
     }
 }
@@ -649,7 +648,7 @@ 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_aux_file_policy: value.switch_aux_file_policy,
+            switch_to_aux_file_v2: value.switch_to_aux_file_v2,
         }
     }
 }

pageserver/src/tenant/delete.rs

@@ -585,20 +585,9 @@ 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)
-
-                    // 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
-                        }
-                    }
+                    crate::metrics::TENANT_MANAGER
+                        .tenant_slots
+                        .set(locked.len() as u64);
 
                     match removed {
                         TenantsMapRemoveResult::Occupied(TenantSlot::Attached(tenant)) => {

pageserver/src/tenant/metadata.rs

@@ -207,24 +207,6 @@ 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
     }

pageserver/src/tenant/mgr.rs

@@ -56,7 +56,6 @@ 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
@@ -247,7 +246,6 @@ impl TenantsMap {
         }
     }
 
-    #[cfg(all(debug_assertions, not(test)))]
     pub(crate) fn len(&self) -> usize {
         match self {
             TenantsMap::Initializing => 0,
@@ -748,7 +746,6 @@ pub async fn init_tenant_mgr(
             }
         };
 
-        METRICS.slot_inserted(&slot);
         tenants.insert(tenant_shard_id, slot);
     }
 
@@ -756,7 +753,7 @@ 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 {
@@ -827,14 +824,6 @@ 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();
@@ -2008,101 +1997,6 @@ 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)]
@@ -2534,13 +2428,10 @@ impl SlotGuard {
                 TenantsMap::Open(m) => m,
             };
 
-            METRICS.slot_inserted(&new_value);
-
             let replaced = m.insert(self.tenant_shard_id, new_value);
             self.upserted = true;
-            if let Some(replaced) = replaced.as_ref() {
-                METRICS.slot_removed(replaced);
-            }
+
+            METRICS.tenant_slots.set(m.len() as u64);
 
             replaced
         };
@@ -2650,13 +2541,9 @@ impl Drop for SlotGuard {
                 }
 
                 if self.old_value_is_shutdown() {
-                    METRICS.slot_removed(entry.get());
                     entry.remove();
                 } else {
-                    let inserting = self.old_value.take().unwrap();
-                    METRICS.slot_inserted(&inserting);
-                    let replaced = entry.insert(inserting);
-                    METRICS.slot_removed(&replaced);
+                    entry.insert(self.old_value.take().unwrap());
                 }
             }
             Entry::Vacant(_) => {
@@ -2667,6 +2554,8 @@ impl Drop for SlotGuard {
                 );
             }
         }
+
+        METRICS.tenant_slots.set(m.len() as u64);
     }
 }
 
@@ -2746,9 +2635,7 @@ fn tenant_map_acquire_slot_impl(
             }
             _ => {
                 let (completion, barrier) = utils::completion::channel();
-                let inserting = TenantSlot::InProgress(barrier);
-                METRICS.slot_inserted(&inserting);
-                v.insert(inserting);
+                v.insert(TenantSlot::InProgress(barrier));
                 tracing::debug!("Vacant, inserted InProgress");
                 Ok(SlotGuard::new(*tenant_shard_id, None, completion))
             }
@@ -2784,10 +2671,7 @@ 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 in_progress = TenantSlot::InProgress(barrier);
-                    METRICS.slot_inserted(&in_progress);
-                    let old_value = o.insert(in_progress);
-                    METRICS.slot_removed(&old_value);
+                    let old_value = o.insert(TenantSlot::InProgress(barrier));
                     tracing::debug!("Occupied, replaced with InProgress");
                     Ok(SlotGuard::new(
                         *tenant_shard_id,

pageserver/src/tenant/remote_timeline_client.rs

@@ -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);
+        self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
 
         Ok(())
     }
@@ -591,7 +591,7 @@ impl RemoteTimelineClient {
 
         upload_queue.latest_metadata.apply(update);
 
-        self.schedule_index_upload(upload_queue);
+        self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
 
         Ok(())
     }
@@ -611,14 +611,18 @@ 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);
+            self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
         }
 
         Ok(())
     }
 
     /// Launch an index-file upload operation in the background (internal function)
-    fn schedule_index_upload(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
+    fn schedule_index_upload(
+        self: &Arc<Self>,
+        upload_queue: &mut UploadQueueInitialized,
+        metadata: TimelineMetadata,
+    ) {
         let disk_consistent_lsn = upload_queue.latest_metadata.disk_consistent_lsn();
 
         info!(
@@ -627,7 +631,11 @@ impl RemoteTimelineClient {
             upload_queue.latest_files_changes_since_metadata_upload_scheduled,
         );
 
-        let index_part = IndexPart::from(&*upload_queue);
+        let index_part = IndexPart::new(
+            upload_queue.latest_files.clone(),
+            disk_consistent_lsn,
+            metadata,
+        );
         let op = UploadOp::UploadMetadata(index_part, disk_consistent_lsn);
         self.metric_begin(&op);
         upload_queue.queued_operations.push_back(op);
@@ -637,61 +645,9 @@ 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.
     ///
-    /// 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().filename(), 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.
+    /// Launch an upload operation in the background.
+    ///
     pub(crate) fn schedule_layer_file_upload(
         self: &Arc<Self>,
         layer: ResidentLayer,
@@ -717,11 +673,9 @@ impl RemoteTimelineClient {
         upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
 
         info!(
-            gen=?metadata.generation,
-            shard=?metadata.shard,
-            "scheduled layer file upload {layer}",
+            "scheduled layer file upload {layer} gen={:?} shard={:?}",
+            metadata.generation, metadata.shard
         );
-
         let op = UploadOp::UploadLayer(layer, metadata);
         self.metric_begin(&op);
         upload_queue.queued_operations.push_back(op);
@@ -784,6 +738,10 @@ impl RemoteTimelineClient {
     where
         I: IntoIterator<Item = LayerFileName>,
     {
+        // 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.
@@ -822,7 +780,7 @@ 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);
+            self.schedule_index_upload(upload_queue, metadata);
         }
 
         with_metadata
@@ -924,18 +882,12 @@ impl RemoteTimelineClient {
 
     /// Wait for all previously scheduled uploads/deletions to complete
     pub(crate) async fn wait_completion(self: &Arc<Self>) -> anyhow::Result<()> {
-        let receiver = {
+        let mut 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");
         }
@@ -1051,7 +1003,8 @@ impl RemoteTimelineClient {
             let deleted_at = Utc::now().naive_utc();
             stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
 
-            let mut index_part = IndexPart::from(&stopped.upload_queue_for_deletion);
+            let mut index_part = IndexPart::try_from(&stopped.upload_queue_for_deletion)
+                .context("IndexPart serialize")?;
             index_part.deleted_at = Some(deleted_at);
             index_part
         };
@@ -1132,93 +1085,6 @@ 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().filename(),
-            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().filename(),
-            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().filename(),
-            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,
@@ -1390,7 +1256,7 @@ 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
                 }
@@ -1517,25 +1383,13 @@ impl RemoteTimelineClient {
 
             let upload_result: anyhow::Result<()> = match &task.op {
                 UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
-                    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().filename(),
-                        layer_metadata.generation,
-                    );
-
+                    let path = layer.local_path();
                     upload::upload_timeline_layer(
+                        self.conf,
                         &self.storage_impl,
-                        local_path,
-                        &remote_path,
-                        layer_metadata.file_size(),
+                        path,
+                        layer_metadata,
+                        self.generation,
                         &self.cancel,
                     )
                     .measure_remote_op(
@@ -1964,6 +1818,29 @@ 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::*;
@@ -1971,7 +1848,6 @@ mod tests {
         context::RequestContext,
         tenant::{
             harness::{TenantHarness, TIMELINE_ID},
-            storage_layer::layer::local_layer_path,
             Tenant, Timeline,
         },
         DEFAULT_PG_VERSION,
@@ -2154,20 +2030,11 @@ mod tests {
         ]
         .into_iter()
         .map(|(name, contents): (LayerFileName, 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();
+            std::fs::write(timeline_path.join(name.file_name()), &contents).unwrap();
 
             Layer::for_resident(
                 harness.conf,
                 &timeline,
-                local_path,
                 name,
                 LayerFileMetadata::new(contents.len() as u64, generation, shard),
             )
@@ -2304,22 +2171,19 @@ 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 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(&local_path, &content_1).unwrap();
+        std::fs::write(
+            timeline_path.join(layer_file_name_1.file_name()),
+            &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),
         );
@@ -2388,7 +2252,12 @@ 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_index_part = IndexPart::example();
+        let example_metadata = TimelineMetadata::example();
+        let example_index_part = IndexPart::new(
+            HashMap::new(),
+            example_metadata.disk_consistent_lsn(),
+            example_metadata,
+        );
 
         let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
 

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -21,7 +21,6 @@ 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::layer::local_layer_path;
 use crate::tenant::storage_layer::LayerFileName;
 use crate::tenant::Generation;
 use crate::virtual_file::{on_fatal_io_error, MaybeFatalIo, VirtualFile};
@@ -56,13 +55,7 @@ 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 = local_layer_path(
-        conf,
-        &tenant_shard_id,
-        &timeline_id,
-        layer_file_name,
-        &layer_metadata.generation,
-    );
+    let local_path = timeline_path.join(layer_file_name.file_name());
 
     let remote_path = remote_layer_path(
         &tenant_shard_id.tenant_id,

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -6,6 +6,7 @@ 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;
@@ -103,14 +104,15 @@ impl IndexPart {
 
     pub const FILE_NAME: &'static str = "index_part.json";
 
-    fn new(
-        layers_and_metadata: &HashMap<LayerFileName, LayerFileMetadata>,
+    pub fn new(
+        layers_and_metadata: HashMap<LayerFileName, LayerFileMetadata>,
         disk_consistent_lsn: Lsn,
         metadata: TimelineMetadata,
     ) -> Self {
+        // Transform LayerFileMetadata into IndexLayerMetadata
         let layer_metadata = layers_and_metadata
-            .iter()
-            .map(|(k, v)| (k.to_owned(), IndexLayerMetadata::from(v)))
+            .into_iter()
+            .map(|(k, v)| (k, IndexLayerMetadata::from(v)))
             .collect();
 
         Self {
@@ -139,24 +141,20 @@ 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 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();
+impl TryFrom<&UploadQueueInitialized> for IndexPart {
+    type Error = SerializeError;
 
-        Self::new(&uq.latest_files, disk_consistent_lsn, metadata)
+    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,
+        ))
     }
 }
 
@@ -174,8 +172,8 @@ 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,

pageserver/src/tenant/remote_timeline_client/upload.rs

@@ -12,13 +12,18 @@ use tokio_util::sync::CancellationToken;
 use utils::backoff;
 
 use super::Generation;
-use crate::tenant::remote_timeline_client::{
-    index::IndexPart, remote_index_path, remote_initdb_archive_path,
-    remote_initdb_preserved_archive_path,
+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 remote_storage::{GenericRemoteStorage, RemotePath, TimeTravelError};
+use remote_storage::{GenericRemoteStorage, 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.
@@ -60,10 +65,11 @@ 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,
-    local_path: &'a Utf8Path,
-    remote_path: &'a RemotePath,
-    metadata_size: u64,
+    source_path: &'a Utf8Path,
+    known_metadata: &'a LayerFileMetadata,
+    generation: Generation,
     cancel: &CancellationToken,
 ) -> anyhow::Result<()> {
     fail_point!("before-upload-layer", |_| {
@@ -72,7 +78,8 @@ pub(super) async fn upload_timeline_layer<'a>(
 
     pausable_failpoint!("before-upload-layer-pausable");
 
-    let source_file_res = fs::File::open(&local_path).await;
+    let storage_path = remote_path(conf, source_path, generation)?;
+    let source_file_res = fs::File::open(&source_path).await;
     let source_file = match source_file_res {
         Ok(source_file) => source_file,
         Err(e) if e.kind() == ErrorKind::NotFound => {
@@ -83,49 +90,34 @@ 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 = %local_path, "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
+            info!(path = %source_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 {local_path:?}"))?,
+        Err(e) => {
+            Err(e).with_context(|| format!("open a source file for layer {source_path:?}"))?
+        }
     };
 
     let fs_size = source_file
         .metadata()
         .await
-        .with_context(|| format!("get the source file metadata for layer {local_path:?}"))?
+        .with_context(|| format!("get the source file metadata for layer {source_path:?}"))?
         .len();
 
+    let metadata_size = known_metadata.file_size();
     if metadata_size != fs_size {
-        bail!("File {local_path:?} has its current FS size {fs_size} diferent from initially determined {metadata_size}");
+        bail!("File {source_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 {local_path:?} size {fs_size} usize"))?;
+        .with_context(|| format!("convert {source_path:?} size {fs_size} usize"))?;
 
     let reader = tokio_util::io::ReaderStream::with_capacity(source_file, super::BUFFER_SIZE);
 
     storage
-        .upload(reader, fs_size, remote_path, None, cancel)
+        .upload(reader, fs_size, &storage_path, None, cancel)
         .await
-        .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}"))
+        .with_context(|| format!("upload layer from local path '{source_path}'"))
 }
 
 /// Uploads the given `initdb` data to the remote storage.

pageserver/src/tenant/secondary.rs

@@ -21,9 +21,8 @@ use self::{
 use super::{
     config::{SecondaryLocationConfig, TenantConfOpt},
     mgr::TenantManager,
-    remote_timeline_client::LayerFileMetadata,
     span::debug_assert_current_span_has_tenant_id,
-    storage_layer::{layer::local_layer_path, LayerFileName},
+    storage_layer::LayerFileName,
 };
 
 use pageserver_api::{
@@ -183,7 +182,6 @@ impl SecondaryTenant {
         conf: &PageServerConf,
         timeline_id: TimelineId,
         name: LayerFileName,
-        metadata: LayerFileMetadata,
     ) {
         debug_assert_current_span_has_tenant_id();
 
@@ -197,13 +195,9 @@ impl SecondaryTenant {
 
         let now = SystemTime::now();
 
-        let local_path = local_layer_path(
-            conf,
-            &self.tenant_shard_id,
-            &timeline_id,
-            &name,
-            &metadata.generation,
-        );
+        let path = conf
+            .timeline_path(&self.tenant_shard_id, &timeline_id)
+            .join(name.file_name());
 
         let this = self.clone();
 
@@ -214,7 +208,7 @@ 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(local_path);
+            let deleted = std::fs::remove_file(path);
 
             let not_found = deleted
                 .as_ref()

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::{layer::local_layer_path, LayerFileName},
+        storage_layer::LayerFileName,
         tasks::{warn_when_period_overrun, BackgroundLoopKind},
     },
     virtual_file::{on_fatal_io_error, MaybeFatalIo, VirtualFile},
@@ -621,12 +621,12 @@ impl<'a> TenantDownloader<'a> {
                 let layers_in_heatmap = heatmap_timeline
                     .layers
                     .iter()
-                    .map(|l| (&l.name, l.metadata.generation))
+                    .map(|l| &l.name)
                     .collect::<HashSet<_>>();
                 let layers_on_disk = timeline_state
                     .on_disk_layers
                     .iter()
-                    .map(|l| (l.0, l.1.metadata.generation))
+                    .map(|l| l.0)
                     .collect::<HashSet<_>>();
 
                 let mut layer_count = layers_on_disk.len();
@@ -637,24 +637,16 @@ impl<'a> TenantDownloader<'a> {
                     .sum();
 
                 // Remove on-disk layers that are no longer present in heatmap
-                for (layer_file_name, generation) in layers_on_disk.difference(&layers_in_heatmap) {
+                for layer in layers_on_disk.difference(&layers_in_heatmap) {
                     layer_count -= 1;
                     layer_byte_count -= timeline_state
                         .on_disk_layers
-                        .get(layer_file_name)
+                        .get(layer)
                         .unwrap()
                         .metadata
                         .file_size();
 
-                    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));
+                    delete_layers.push((*timeline_id, (*layer).clone()));
                 }
 
                 progress.bytes_downloaded += layer_byte_count;
@@ -669,7 +661,11 @@ impl<'a> TenantDownloader<'a> {
         }
 
         // Execute accumulated deletions
-        for (timeline_id, layer_name, local_path) in delete_layers {
+        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());
             tracing::info!(timeline_id=%timeline_id, "Removing secondary local layer {layer_name} because it's absent in heatmap",);
 
             tokio::fs::remove_file(&local_path)
@@ -758,6 +754,9 @@ 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();
@@ -807,14 +806,10 @@ 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 = local_layer_path(
-                        self.conf,
-                        tenant_shard_id,
-                        &timeline.timeline_id,
-                        &layer.name,
-                        &layer.metadata.generation,
-                    );
-
+                    let local_path = self
+                        .conf
+                        .timeline_path(tenant_shard_id, &timeline.timeline_id)
+                        .join(layer.name.file_name());
                     match tokio::fs::metadata(&local_path).await {
                         Ok(meta) => {
                             tracing::debug!(
@@ -908,13 +903,7 @@ impl<'a> TenantDownloader<'a> {
             };
 
             if downloaded_bytes != layer.metadata.file_size {
-                let local_path = local_layer_path(
-                    self.conf,
-                    tenant_shard_id,
-                    &timeline.timeline_id,
-                    &layer.name,
-                    &layer.metadata.generation,
-                );
+                let local_path = timeline_path.join(layer.name.to_string());
 
                 tracing::warn!(
                     "Downloaded layer {} with unexpected size {} != {}.  Removing download.",

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -1139,15 +1139,15 @@ impl DeltaLayerInner {
         Ok(all_keys)
     }
 
-    /// 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.
+    /// Using the given writer, write out a truncated version, where LSNs higher than the
+    /// truncate_at are missing.
+    #[cfg(test)]
     pub(super) async fn copy_prefix(
         &self,
         writer: &mut DeltaLayerWriter,
-        until: Lsn,
+        truncate_at: Lsn,
         ctx: &RequestContext,
-    ) -> anyhow::Result<usize> {
+    ) -> anyhow::Result<()> {
         use crate::tenant::vectored_blob_io::{
             BlobMeta, VectoredReadBuilder, VectoredReadExtended,
         };
@@ -1211,8 +1211,6 @@ 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
@@ -1231,7 +1229,7 @@ impl DeltaLayerInner {
 
             prev = Option::from(item);
 
-            let actionable = actionable.filter(|x| x.0.lsn < until);
+            let actionable = actionable.filter(|x| x.0.lsn < truncate_at);
 
             let builder = if let Some((meta, offsets)) = actionable {
                 // extend or create a new builder
@@ -1299,7 +1297,7 @@ 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, %key, %lsn, "failed to parse will_init out of serialized value");
+                            tracing::error!(data=?utils::Hex(data), err=?_e, "failed to parse will_init out of serialized value");
                         })
                         .unwrap_or(false);
 
@@ -1316,10 +1314,7 @@ impl DeltaLayerInner {
                         )
                         .await;
                     per_blob_copy = tmp;
-
                     res?;
-
-                    records += 1;
                 }
 
                 buffer = Some(res.buf);
@@ -1331,7 +1326,7 @@ impl DeltaLayerInner {
             "with the sentinel above loop should had handled all"
         );
 
-        Ok(records)
+        Ok(())
     }
 
     pub(super) async fn dump(&self, ctx: &RequestContext) -> anyhow::Result<()> {
@@ -1404,6 +1399,7 @@ impl DeltaLayerInner {
         Ok(())
     }
 
+    #[cfg(test)]
     fn stream_index_forwards<'a, R>(
         &'a self,
         reader: &'a DiskBtreeReader<R, DELTA_KEY_SIZE>,

pageserver/src/tenant/storage_layer/filename.rs

@@ -2,13 +2,11 @@
 //! 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;
@@ -76,19 +74,10 @@ 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()?;
 
@@ -193,10 +182,6 @@ 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()?;
 
@@ -274,22 +259,9 @@ impl From<DeltaFileName> for LayerFileName {
 impl FromStr for LayerFileName {
     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 gen_suffix_regex = Regex::new("^(?<base>.+)-(?<gen>[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 = DeltaFileName::parse_str(&file_name);
-        let image = ImageFileName::parse_str(&file_name);
+        let delta = DeltaFileName::parse_str(value);
+        let image = ImageFileName::parse_str(value);
         let ok = match (delta, image) {
             (None, None) => {
                 return Err(format!(
@@ -343,42 +315,3 @@ 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 = LayerFileName::Image(ImageFileName {
-            key_range: Key::from_i128(0)
-                ..Key::from_hex("000000067F00000001000004DF0000000006").unwrap(),
-            lsn: Lsn::from_hex("00000000014FED58").unwrap(),
-        });
-        let parsed = LayerFileName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-00000001").map_err(|s| anyhow::anyhow!(s))?;
-        assert_eq!(parsed, expected,);
-
-        // Omitting generation suffix is valid
-        let parsed = LayerFileName::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 = LayerFileName::Delta(DeltaFileName {
-            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 = LayerFileName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481-00000001").map_err(|s| anyhow::anyhow!(s))?;
-        assert_eq!(parsed, expected);
-
-        // Omitting generation suffix is valid
-        let parsed = LayerFileName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481").map_err(|s| anyhow::anyhow!(s))?;
-        assert_eq!(parsed, expected);
-
-        Ok(())
-    }
-}

pageserver/src/tenant/storage_layer/layer.rs

@@ -4,13 +4,12 @@ use pageserver_api::keyspace::KeySpace;
 use pageserver_api::models::{
     HistoricLayerInfo, LayerAccessKind, LayerResidenceEventReason, LayerResidenceStatus,
 };
-use pageserver_api::shard::{ShardIndex, TenantShardId};
+use pageserver_api::shard::ShardIndex;
 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;
 
@@ -124,25 +123,6 @@ impl PartialEq for Layer {
     }
 }
 
-pub(crate) fn local_layer_path(
-    conf: &PageServerConf,
-    tenant_shard_id: &TenantShardId,
-    timeline_id: &TimelineId,
-    layer_file_name: &LayerFileName,
-    _generation: &Generation,
-) -> Utf8PathBuf {
-    let timeline_path = conf.timeline_path(tenant_shard_id, timeline_id);
-
-    timeline_path.join(layer_file_name.file_name())
-
-    // TOOD: include generation in the name in now+1 releases.
-    // timeline_path.join(format!(
-    //     "{}{}",
-    //     layer_file_name.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(
@@ -151,14 +131,6 @@ impl Layer {
         file_name: LayerFileName,
         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,
@@ -171,7 +143,6 @@ impl Layer {
         let owner = Layer(Arc::new(LayerInner::new(
             conf,
             timeline,
-            local_path,
             access_stats,
             desc,
             None,
@@ -188,7 +159,6 @@ impl Layer {
     pub(crate) fn for_resident(
         conf: &'static PageServerConf,
         timeline: &Arc<Timeline>,
-        local_path: Utf8PathBuf,
         file_name: LayerFileName,
         metadata: LayerFileMetadata,
     ) -> ResidentLayer {
@@ -214,7 +184,6 @@ impl Layer {
             LayerInner::new(
                 conf,
                 timeline,
-                local_path,
                 access_stats,
                 desc,
                 Some(inner),
@@ -256,19 +225,9 @@ impl Layer {
                 LayerResidenceStatus::Resident,
                 LayerResidenceEventReason::LayerCreate,
             );
-
-            let local_path = local_layer_path(
-                conf,
-                &timeline.tenant_shard_id,
-                &timeline.timeline_id,
-                &desc.filename(),
-                &timeline.generation,
-            );
-
             LayerInner::new(
                 conf,
                 timeline,
-                local_path,
                 access_stats,
                 desc,
                 Some(inner),
@@ -451,13 +410,6 @@ 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<()> {
@@ -757,17 +709,19 @@ 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);
@@ -783,7 +737,7 @@ impl LayerInner {
         LayerInner {
             conf,
             debug_str: { format!("timelines/{}/{}", timeline.timeline_id, desc.filename()).into() },
-            path: local_path,
+            path,
             desc,
             timeline: Arc::downgrade(timeline),
             have_remote_client: timeline.remote_client.is_some(),
@@ -1843,23 +1797,25 @@ impl ResidentLayer {
         }
     }
 
-    /// Returns the amount of keys and values written to the writer.
-    pub(crate) async fn copy_delta_prefix(
+    /// 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(
         &self,
         writer: &mut super::delta_layer::DeltaLayerWriter,
-        until: Lsn,
+        truncate_at: Lsn,
         ctx: &RequestContext,
-    ) -> anyhow::Result<usize> {
+    ) -> anyhow::Result<()> {
         use LayerKind::*;
 
         let owner = &self.owner.0;
 
         match self.downloaded.get(owner, ctx).await? {
             Delta(ref d) => d
-                .copy_prefix(writer, until, ctx)
+                .copy_prefix(writer, truncate_at, ctx)
                 .await
-                .with_context(|| format!("copy_delta_prefix until {until} of {self}")),
-            Image(_) => anyhow::bail!(format!("cannot copy_lsn_prefix of image layer {self}")),
+                .with_context(|| format!("truncate {self}")),
+            Image(_) => anyhow::bail!(format!("cannot truncate image layer {self}")),
         }
     }
 

pageserver/src/tenant/tasks.rs

@@ -2,7 +2,6 @@
 //! such as compaction and GC
 
 use std::ops::ControlFlow;
-use std::str::FromStr;
 use std::sync::Arc;
 use std::time::{Duration, Instant};
 
@@ -10,11 +9,9 @@ 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};
@@ -47,7 +44,6 @@ pub(crate) enum BackgroundLoopKind {
     Compaction,
     Gc,
     Eviction,
-    IngestHouseKeeping,
     ConsumptionMetricsCollectMetrics,
     ConsumptionMetricsSyntheticSizeWorker,
     InitialLogicalSizeCalculation,
@@ -136,30 +132,6 @@ 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(())
-            }
-        },
-    );
 }
 
 ///
@@ -407,61 +379,6 @@ 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 {
@@ -503,6 +420,8 @@ pub(crate) async fn random_init_delay(
     period: Duration,
     cancel: &CancellationToken,
 ) -> Result<(), Cancelled> {
+    use rand::Rng;
+
     if period == Duration::ZERO {
         return Ok(());
     }

pageserver/src/tenant/timeline.rs

@@ -1,6 +1,5 @@
 mod compaction;
 pub mod delete;
-pub(crate) mod detach_ancestor;
 mod eviction_task;
 mod init;
 pub mod layer_manager;
@@ -23,9 +22,8 @@ use pageserver_api::{
     },
     keyspace::{KeySpaceAccum, SparseKeyPartitioning},
     models::{
-        AuxFilePolicy, CompactionAlgorithm, DownloadRemoteLayersTaskInfo,
-        DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy, InMemoryLayerInfo, LayerMapInfo,
-        TimelineState,
+        CompactionAlgorithm, DownloadRemoteLayersTaskInfo, DownloadRemoteLayersTaskSpawnRequest,
+        EvictionPolicy, InMemoryLayerInfo, LayerMapInfo, TimelineState,
     },
     reltag::BlockNumber,
     shard::{ShardIdentity, ShardNumber, TenantShardId},
@@ -60,7 +58,6 @@ use std::{
     ops::ControlFlow,
 };
 
-use crate::tenant::storage_layer::layer::local_layer_path;
 use crate::tenant::{
     layer_map::{LayerMap, SearchResult},
     metadata::TimelineMetadata,
@@ -865,13 +862,9 @@ impl Timeline {
                 // Initialise the reconstruct state for the key with the cache
                 // entry returned above.
                 let mut reconstruct_state = ValuesReconstructState::new();
-
-                // 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 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)
@@ -1083,7 +1076,7 @@ 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"
@@ -1501,21 +1494,15 @@ 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
     }
 
-    // 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) {
+    /// 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) {
         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.
@@ -1542,11 +1529,13 @@ 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() {
-                    // 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
+                    // 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
                     // sees its LSN advance: we only do this if we've been layer-less
                     // for some time.
-                    tracing::debug!(
+                    tracing::info!(
                         "Advancing disk_consistent_lsn past WAL ingest gap {} -> {}",
                         disk_consistent_lsn,
                         last_record_lsn
@@ -1636,6 +1625,11 @@ 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! {
@@ -1905,7 +1899,7 @@ 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: &LayerFileName,
+        layer_file_name: &str,
     ) -> anyhow::Result<Option<bool>> {
         let Some(layer) = self.find_layer(layer_file_name).await else {
             return Ok(None);
@@ -1923,10 +1917,7 @@ 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: &LayerFileName,
-    ) -> anyhow::Result<Option<bool>> {
+    pub(crate) async fn evict_layer(&self, layer_file_name: &str) -> anyhow::Result<Option<bool>> {
         let _gate = self
             .gate
             .enter()
@@ -2000,12 +1991,13 @@ const REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE: u64 = 10;
 
 // Private functions
 impl Timeline {
-    pub(crate) fn get_switch_aux_file_policy(&self) -> AuxFilePolicy {
+    #[allow(dead_code)]
+    pub(crate) fn get_switch_to_aux_file_v2(&self) -> bool {
         let tenant_conf = self.tenant_conf.load();
         tenant_conf
             .tenant_conf
-            .switch_aux_file_policy
-            .unwrap_or(self.conf.default_tenant_conf.switch_aux_file_policy)
+            .switch_to_aux_file_v2
+            .unwrap_or(self.conf.default_tenant_conf.switch_to_aux_file_v2)
     }
 
     pub(crate) fn get_lazy_slru_download(&self) -> bool {
@@ -2417,8 +2409,8 @@ impl Timeline {
 
                 for discovered in discovered {
                     let (name, kind) = match discovered {
-                        Discovered::Layer(layer_file_name, local_path, file_size) => {
-                            discovered_layers.push((layer_file_name, local_path, file_size));
+                        Discovered::Layer(file_name, file_size) => {
+                            discovered_layers.push((file_name, file_size));
                             continue;
                         }
                         Discovered::Metadata => {
@@ -2463,7 +2455,7 @@ impl Timeline {
                 let mut needs_cleanup = Vec::new();
                 let mut total_physical_size = 0;
 
-                for (name, local_path, decision) in decided {
+                for (name, decision) in decided {
                     let decision = match decision {
                         Ok(UseRemote { local, remote }) => {
                             // Remote is authoritative, but we may still choose to retain
@@ -2473,23 +2465,26 @@ impl Timeline {
                                 // the correct generation.
                                 UseLocal(remote)
                             } else {
-                                let local_path = local_path.as_ref().expect("Locally found layer must have path");
-                                init::cleanup_local_file_for_remote(local_path, &local, &remote)?;
+                                path.push(name.file_name());
+                                init::cleanup_local_file_for_remote(&path, &local, &remote)?;
+                                path.pop();
                                 UseRemote { local, remote }
                             }
                         }
                         Ok(decision) => decision,
                         Err(DismissedLayer::Future { local }) => {
                             if local.is_some() {
-                                let local_path = local_path.expect("Locally found layer must have path");
-                                init::cleanup_future_layer(&local_path, &name, disk_consistent_lsn)?;
+                                path.push(name.file_name());
+                                init::cleanup_future_layer(&path, &name, disk_consistent_lsn)?;
+                                path.pop();
                             }
                             needs_cleanup.push(name);
                             continue;
                         }
                         Err(DismissedLayer::LocalOnly(local)) => {
-                            let local_path = local_path.expect("Locally found layer must have path");
-                            init::cleanup_local_only_file(&local_path, &name, &local)?;
+                            path.push(name.file_name());
+                            init::cleanup_local_only_file(&path, &name, &local)?;
+                            path.pop();
                             // this file never existed remotely, we will have to do rework
                             continue;
                         }
@@ -2505,18 +2500,7 @@ impl Timeline {
                     let layer = match decision {
                         UseLocal(m) => {
                             total_physical_size += m.file_size();
-
-                            let local_path = local_path.unwrap_or_else(|| {
-                                local_layer_path(
-                                    conf,
-                                    &this.tenant_shard_id,
-                                    &this.timeline_id,
-                                    &name,
-                                    &m.generation,
-                                )
-                            });
-
-                            Layer::for_resident(conf, &this, local_path, name, m).drop_eviction_guard()
+                            Layer::for_resident(conf, &this, name, m).drop_eviction_guard()
                         }
                         Evicted(remote) | UseRemote { remote, .. } => {
                             Layer::for_evicted(conf, &this, name, remote)
@@ -2997,11 +2981,11 @@ impl Timeline {
         }
     }
 
-    async fn find_layer(&self, layer_name: &LayerFileName) -> Option<Layer> {
+    async fn find_layer(&self, layer_file_name: &str) -> 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();
-            if layer_name == &historic_layer_name {
+            let historic_layer_name = historic_layer.filename().file_name();
+            if layer_file_name == historic_layer_name {
                 return Some(guard.get_from_desc(&historic_layer));
             }
         }
@@ -3031,7 +3015,7 @@ impl Timeline {
 
             HeatMapLayer::new(
                 layer.layer_desc().filename(),
-                (&layer.metadata()).into(),
+                layer.metadata().into(),
                 last_activity_ts,
             )
         });
@@ -3533,7 +3517,7 @@ impl Timeline {
         Ok(ancestor)
     }
 
-    pub(crate) fn get_ancestor_timeline(&self) -> anyhow::Result<Arc<Timeline>> {
+    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 {:?}",
@@ -4349,49 +4333,6 @@ 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.
@@ -4500,24 +4441,6 @@ 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>,
@@ -4676,8 +4599,6 @@ 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();
 

pageserver/src/tenant/timeline/compaction.rs

@@ -15,8 +15,7 @@ use anyhow::{anyhow, Context};
 use enumset::EnumSet;
 use fail::fail_point;
 use itertools::Itertools;
-use pageserver_api::keyspace::ShardedRange;
-use pageserver_api::shard::{ShardCount, ShardIdentity, TenantShardId};
+use pageserver_api::shard::{ShardIdentity, TenantShardId};
 use tokio_util::sync::CancellationToken;
 use tracing::{debug, info, info_span, trace, warn, Instrument};
 use utils::id::TimelineId;
@@ -94,7 +93,7 @@ impl Timeline {
         // Define partitioning schema if needed
 
         // FIXME: the match should only cover repartitioning, not the next steps
-        let partition_count = match self
+        match self
             .repartition(
                 self.get_last_record_lsn(),
                 self.get_compaction_target_size(),
@@ -147,7 +146,6 @@ 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
@@ -159,150 +157,9 @@ 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(())
     }
 

pageserver/src/tenant/timeline/delete.rs

@@ -422,10 +422,6 @@ 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>);

pageserver/src/tenant/timeline/detach_ancestor.rs

@@ -1,550 +0,0 @@
-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().filename(),
-        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)
-}

pageserver/src/tenant/timeline/init.rs

@@ -12,7 +12,7 @@ use crate::{
     METADATA_FILE_NAME,
 };
 use anyhow::Context;
-use camino::{Utf8Path, Utf8PathBuf};
+use camino::Utf8Path;
 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, Utf8PathBuf, u64),
+    Layer(LayerFileName, u64),
     /// Old ephmeral files from previous launches, should be removed
     Ephemeral(String),
     /// Old temporary timeline files, unsure what these really are, should be removed
@@ -46,7 +46,7 @@ pub(super) fn scan_timeline_dir(path: &Utf8Path) -> anyhow::Result<Vec<Discovere
         let discovered = match LayerFileName::from_str(&file_name) {
             Ok(file_name) => {
                 let file_size = direntry.metadata()?.len();
-                Discovered::Layer(file_name, direntry.path().to_owned(), file_size)
+                Discovered::Layer(file_name, file_size)
             }
             Err(_) => {
                 if file_name == METADATA_FILE_NAME {
@@ -104,38 +104,26 @@ pub(super) enum DismissedLayer {
 
 /// Merges local discoveries and remote [`IndexPart`] to a collection of decisions.
 pub(super) fn reconcile(
-    discovered: Vec<(LayerFileName, Utf8PathBuf, u64)>,
+    discovered: Vec<(LayerFileName, u64)>,
     index_part: Option<&IndexPart>,
     disk_consistent_lsn: Lsn,
     generation: Generation,
     shard: ShardIndex,
-) -> Vec<(
-    LayerFileName,
-    Option<Utf8PathBuf>,
-    Result<Decision, DismissedLayer>,
-)> {
+) -> Vec<(LayerFileName, Result<Decision, DismissedLayer>)> {
     use Decision::*;
 
-    // name => (local_path, local_metadata, remote_metadata)
-    type Collected = HashMap<
-        LayerFileName,
-        (
-            Option<Utf8PathBuf>,
-            Option<LayerFileMetadata>,
-            Option<LayerFileMetadata>,
-        ),
-    >;
+    // name => (local, remote)
+    type Collected = HashMap<LayerFileName, (Option<LayerFileMetadata>, Option<LayerFileMetadata>)>;
 
     let mut discovered = discovered
         .into_iter()
-        .map(|(layer_name, local_path, file_size)| {
+        .map(|(name, file_size)| {
             (
-                layer_name,
+                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(local_path),
                     Some(LayerFileMetadata::new(file_size, generation, shard)),
                     None,
                 ),
@@ -152,15 +140,15 @@ pub(super) fn reconcile(
         .map(|(name, metadata)| (name, LayerFileMetadata::from(metadata)))
         .for_each(|(name, metadata)| {
             if let Some(existing) = discovered.get_mut(name) {
-                existing.2 = Some(metadata);
+                existing.1 = Some(metadata);
             } else {
-                discovered.insert(name.to_owned(), (None, None, Some(metadata)));
+                discovered.insert(name.to_owned(), (None, Some(metadata)));
             }
         });
 
     discovered
         .into_iter()
-        .map(|(name, (local_path, local, remote))| {
+        .map(|(name, (local, remote))| {
             let decision = if name.is_in_future(disk_consistent_lsn) {
                 Err(DismissedLayer::Future { local })
             } else {
@@ -177,7 +165,7 @@ pub(super) fn reconcile(
                 }
             };
 
-            (name, local_path, decision)
+            (name, decision)
         })
         .collect::<Vec<_>>()
 }

pageserver/src/tenant/timeline/layer_manager.rs

@@ -205,24 +205,6 @@ 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();

pgxn/neon/Makefile

@@ -14,8 +14,7 @@ OBJS = \
 	relsize_cache.o \
 	walproposer.o \
 	walproposer_pg.o \
-	control_plane_connector.o \
-	walsender_hooks.o
+	control_plane_connector.o
 
 PG_CPPFLAGS = -I$(libpq_srcdir)
 SHLIB_LINK_INTERNAL = $(libpq)

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 = 2;
+int         neon_protocol_version = 1;
 
 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,
-							2, /* use protocol version 2 */
+							1, /* default to old protocol for now */
 							1, /* min */
 							2, /* max */
 							PGC_SU_BACKEND,

pgxn/neon/neon.c

@@ -34,7 +34,6 @@
 #include "walproposer.h"
 #include "pagestore_client.h"
 #include "control_plane_connector.h"
-#include "walsender_hooks.h"
 
 PG_MODULE_MAGIC;
 void		_PG_init(void);
@@ -266,6 +265,7 @@ LogicalSlotsMonitorMain(Datum main_arg)
 	}
 }
 
+
 void
 _PG_init(void)
 {
@@ -279,7 +279,6 @@ _PG_init(void)
 
 	pg_init_libpagestore();
 	pg_init_walproposer();
-        WalSender_Custom_XLogReaderRoutines = NeonOnDemandXLogReaderRoutines;
 
 	InitLogicalReplicationMonitor();
 

pgxn/neon/neon_walreader.c

@@ -36,7 +36,10 @@
 
 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);
 
@@ -79,9 +82,8 @@ struct NeonWALReader
 	XLogRecPtr	req_lsn;
 	Size		req_len;
 	Size		req_progress;
-	char		donor_conninfo[MAXCONNINFO];
+	WalProposer *wp;			/* we learn donor through walproposer */
 	char		donor_name[64]; /* saved donor safekeeper name for logging */
-	XLogRecPtr	donor_lsn;
 	/* state of connection to safekeeper */
 	NeonWALReaderRemoteState rem_state;
 	WalProposerConn *wp_conn;
@@ -105,7 +107,7 @@ struct NeonWALReader
 
 /* palloc and initialize NeonWALReader */
 NeonWALReader *
-NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, char *log_prefix)
+NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, WalProposer *wp, char *log_prefix)
 {
 	NeonWALReader *reader;
 
@@ -121,6 +123,8 @@ NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, char *log_
 	reader->seg.ws_tli = 0;
 	reader->segcxt.ws_segsize = wal_segment_size;
 
+	reader->wp = wp;
+
 	reader->rem_state = RS_NONE;
 
 	if (log_prefix)
@@ -200,16 +204,21 @@ NeonWALReadRemote(NeonWALReader *state, char *buf, XLogRecPtr startptr, Size cou
 {
 	if (state->rem_state == RS_NONE)
 	{
-		if (!NeonWALReaderUpdateDonor(state))
+		XLogRecPtr	donor_lsn;
+
+		/* no connection yet; start one */
+		Safekeeper *donor = GetDonor(state->wp, &donor_lsn);
+
+		if (donor == NULL)
 		{
 			snprintf(state->err_msg, sizeof(state->err_msg),
 					 "failed to establish remote connection to fetch WAL: no donor available");
 			return NEON_WALREAD_ERROR;
-
 		}
-		/* 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);
+		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);
 		if (PQstatus(state->wp_conn->pg_conn) == CONNECTION_BAD)
 		{
 			snprintf(state->err_msg, sizeof(state->err_msg),
@@ -242,22 +251,10 @@ 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), term);
+							 LSN_FORMAT_ARGS(startptr), state->wp->propTerm);
 					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))
@@ -407,10 +404,6 @@ 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;
 		}
 	}
@@ -533,7 +526,7 @@ err:
 }
 
 /* reset remote connection and request in progress */
-void
+static void
 NeonWALReaderResetRemote(NeonWALReader *state)
 {
 	state->req_lsn = InvalidXLogRecPtr;
@@ -698,25 +691,13 @@ 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
  */
-bool
+static bool
 neon_wal_segment_open(NeonWALReader *state, XLogSegNo nextSegNo,
 					  TimeLineID *tli_p)
 {
@@ -743,7 +724,7 @@ is_wal_segment_exists(XLogSegNo segno, int segsize, TimeLineID tli)
 }
 
 /* copy of vanilla wal_segment_close with NeonWALReader */
-void
+static void
 neon_wal_segment_close(NeonWALReader *state)
 {
 	if (state->seg.ws_file >= 0)
@@ -759,19 +740,3 @@ 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';
-}

pgxn/neon/neon_walreader.h

@@ -19,19 +19,12 @@ typedef enum
 	NEON_WALREAD_ERROR,
 } NeonWALReadResult;
 
-extern NeonWALReader *NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, char *log_prefix);
+extern NeonWALReader *NeonWALReaderAllocate(int wal_segment_size, XLogRecPtr available_lsn, WalProposer *wp, 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__ */

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)
@@ -922,8 +922,7 @@ static void
 DetermineEpochStartLsn(WalProposer *wp)
 {
 	TermHistory *dth;
-	int			n_ready = 0;
-	WalproposerShmemState *walprop_shared;
+	int          n_ready = 0;
 
 	wp->propEpochStartLsn = InvalidXLogRecPtr;
 	wp->donorEpoch = 0;
@@ -965,18 +964,16 @@ 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)
 	{
@@ -987,12 +984,11 @@ 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) ||
@@ -1026,9 +1022,10 @@ 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.
@@ -1043,7 +1040,7 @@ DetermineEpochStartLsn(WalProposer *wp)
 			 * compute (who could generate WAL) is ok.
 			 */
 			if (!((dth->n_entries >= 1) && (dth->entries[dth->n_entries - 1].term ==
-											pg_atomic_read_u64(&walprop_shared->mineLastElectedTerm))))
+											walprop_shared->mineLastElectedTerm)))
 			{
 				/*
 				 * Panic to restart PG as we need to retake basebackup.
@@ -1057,8 +1054,8 @@ 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);
 }
 
 /*
@@ -1108,13 +1105,9 @@ 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
@@ -1184,12 +1177,6 @@ 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);
 }
@@ -1581,17 +1568,17 @@ 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.
  */
-static void
-UpdateDonorShmem(WalProposer *wp)
+Safekeeper *
+GetDonor(WalProposer *wp, XLogRecPtr *donor_lsn)
 {
 	Safekeeper *donor = NULL;
 	int			i;
-	XLogRecPtr	donor_lsn = InvalidXLogRecPtr;
+	*donor_lsn = InvalidXLogRecPtr;
 
 	if (wp->n_votes < wp->quorum)
 	{
-		wp_log(WARNING, "UpdateDonorShmem called before elections are won");
-		return;
+		wp_log(WARNING, "GetDonor called before elections are won");
+		return NULL;
 	}
 
 	/*
@@ -1602,7 +1589,7 @@ UpdateDonorShmem(WalProposer *wp)
 	if (wp->safekeeper[wp->donor].state >= SS_IDLE)
 	{
 		donor = &wp->safekeeper[wp->donor];
-		donor_lsn = wp->propEpochStartLsn;
+		*donor_lsn = wp->propEpochStartLsn;
 	}
 
 	/*
@@ -1614,19 +1601,13 @@ UpdateDonorShmem(WalProposer *wp)
 	{
 		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;
 		}
 	}
-
-	if (donor == NULL)
-	{
-		wp_log(WARNING, "UpdateDonorShmem didn't find a suitable donor, skipping");
-		return;
-	}
-	wp->api.update_donor(wp, donor, donor_lsn);
+	return donor;
 }
 
 /*
@@ -1636,7 +1617,7 @@ static void
 HandleSafekeeperResponse(WalProposer *wp, Safekeeper *sk)
 {
 	XLogRecPtr	candidateTruncateLsn;
-	XLogRecPtr	newCommitLsn;
+	XLogRecPtr  newCommitLsn;
 
 	newCommitLsn = GetAcknowledgedByQuorumWALPosition(wp);
 	if (newCommitLsn > wp->commitLsn)
@@ -1646,7 +1627,7 @@ 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

pgxn/neon/walproposer.h

@@ -284,19 +284,14 @@ typedef struct PageserverFeedback
 
 typedef struct WalproposerShmemState
 {
-	pg_atomic_uint64 propEpochStartLsn;
-	char		donor_name[64];
-	char		donor_conninfo[MAXCONNINFO];
-	XLogRecPtr	donor_lsn;
-
 	slock_t		mutex;
-	pg_atomic_uint64 mineLastElectedTerm;
+	term_t		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;
@@ -470,9 +465,6 @@ 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);
 
@@ -505,7 +497,7 @@ 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);
@@ -553,14 +545,13 @@ 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);
 
@@ -580,9 +571,9 @@ 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);
 
@@ -725,14 +716,12 @@ 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 TimeLineID walprop_pg_get_timeline_id(void);
+extern Safekeeper *GetDonor(WalProposer *wp, XLogRecPtr *donor_lsn);
 
 
 #define WPEVENT		1337		/* special log level for walproposer internal

pgxn/neon/walproposer_pg.c

@@ -85,6 +85,7 @@ 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;
@@ -93,8 +94,6 @@ 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);
@@ -137,7 +136,6 @@ WalProposerSync(int argc, char *argv[])
 	WalProposer *wp;
 
 	init_walprop_config(true);
-	WalproposerShmemInit_SyncSafekeeper();
 	walprop_pg_init_standalone_sync_safekeepers();
 	walprop_pg_load_libpqwalreceiver();
 
@@ -283,8 +281,6 @@ 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);
 	}
@@ -293,17 +289,6 @@ 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
@@ -414,13 +399,6 @@ nwp_shmem_startup_hook(void)
 	WalproposerShmemInit();
 }
 
-WalproposerShmemState *
-GetWalpropShmemState()
-{
-	Assert(walprop_shared != NULL);
-	return walprop_shared;
-}
-
 static WalproposerShmemState *
 walprop_pg_get_shmem_state(WalProposer *wp)
 {
@@ -453,15 +431,14 @@ 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;
 		}
@@ -574,7 +551,6 @@ static void
 walprop_sigusr2(SIGNAL_ARGS)
 {
 	int			save_errno = errno;
-
 	got_SIGUSR2 = true;
 	SetLatch(MyLatch);
 	errno = save_errno;
@@ -622,7 +598,7 @@ walprop_pg_get_current_timestamp(WalProposer *wp)
 	return GetCurrentTimestamp();
 }
 
-TimeLineID
+static TimeLineID
 walprop_pg_get_timeline_id(void)
 {
 #if PG_VERSION_NUM >= 150000
@@ -641,20 +617,6 @@ 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)
@@ -755,6 +717,7 @@ walprop_connect_start(Safekeeper *sk)
 {
 	Assert(sk->conn == NULL);
 	sk->conn = libpqwp_connect_start(sk->conninfo);
+
 }
 
 static WalProposerConnectPollStatusType
@@ -1128,7 +1091,7 @@ 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)
@@ -1332,13 +1295,116 @@ XLogBroadcastWalProposer(WalProposer *wp)
 	}
 }
 
-/*
-  Used to download WAL before basebackup for logical walsenders from sk, no longer
-  needed because walsender always uses neon_walreader.
- */
+/* Download WAL before basebackup for logical walsenders from sk, if needed */
 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;
 }
 
@@ -1479,7 +1545,7 @@ 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, log_prefix);
+	sk->xlogreader = NeonWALReaderAllocate(wal_segment_size, sk->wp->propEpochStartLsn, sk->wp, log_prefix);
 	if (sk->xlogreader == NULL)
 		wpg_log(FATAL, "failed to allocate xlog reader");
 }
@@ -1894,8 +1960,8 @@ 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;
@@ -2029,25 +2095,22 @@ 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,

pgxn/neon/walsender_hooks.c

@@ -1,172 +0,0 @@
-/*-------------------------------------------------------------------------
- *
- * 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;
-}

pgxn/neon/walsender_hooks.h

@@ -1,7 +0,0 @@
-#ifndef __WALSENDER_HOOKS_H__
-#define __WALSENDER_HOOKS_H__
-
-struct XLogReaderRoutine;
-void		NeonOnDemandXLogReaderRoutines(struct XLogReaderRoutine *xlr);
-
-#endif

poetry.lock

@@ -1001,17 +1001,18 @@ dotenv = ["python-dotenv"]
 
 [[package]]
 name = "flask-cors"
-version = "4.0.1"
+version = "3.0.10"
 description = "A Flask extension adding a decorator for CORS support"
 optional = false
 python-versions = "*"
 files = [
-    {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"},
+    {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"},
 ]
 
 [package.dependencies]
 Flask = ">=0.9"
+Six = "*"
 
 [[package]]
 name = "frozenlist"
@@ -1242,13 +1243,13 @@ files = [
 
 [[package]]
 name = "jinja2"
-version = "3.1.4"
+version = "3.1.3"
 description = "A very fast and expressive template engine."
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "jinja2-3.1.4-py3-none-any.whl", hash = "sha256:bc5dd2abb727a5319567b7a813e6a2e7318c39f4f487cfe6c89c6f9c7d25197d"},
-    {file = "jinja2-3.1.4.tar.gz", hash = "sha256:4a3aee7acbbe7303aede8e9648d13b8bf88a429282aa6122a993f0ac800cb369"},
+    {file = "Jinja2-3.1.3-py3-none-any.whl", hash = "sha256:7d6d50dd97d52cbc355597bd845fabfbac3f551e1f99619e39a35ce8c370b5fa"},
+    {file = "Jinja2-3.1.3.tar.gz", hash = "sha256:ac8bd6544d4bb2c9792bf3a159e80bba8fda7f07e81bc3aed565432d5925ba90"},
 ]
 
 [package.dependencies]
@@ -2611,13 +2612,13 @@ files = [
 
 [[package]]
 name = "werkzeug"
-version = "3.0.3"
+version = "3.0.1"
 description = "The comprehensive WSGI web application library."
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "werkzeug-3.0.3-py3-none-any.whl", hash = "sha256:fc9645dc43e03e4d630d23143a04a7f947a9a3b5727cd535fdfe155a17cc48c8"},
-    {file = "werkzeug-3.0.3.tar.gz", hash = "sha256:097e5bfda9f0aba8da6b8545146def481d06aa7d3266e7448e2cccf67dd8bd18"},
+    {file = "werkzeug-3.0.1-py3-none-any.whl", hash = "sha256:90a285dc0e42ad56b34e696398b8122ee4c681833fb35b8334a095d82c56da10"},
+    {file = "werkzeug-3.0.1.tar.gz", hash = "sha256:507e811ecea72b18a404947aded4b3390e1db8f826b494d76550ef45bb3b1dcc"},
 ]
 
 [package.dependencies]
@@ -2899,4 +2900,4 @@ cffi = ["cffi (>=1.11)"]
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.9"
-content-hash = "496d6d9f722983bda4d1265370bc8ba75560da74ab5d6b68c94a03290815e1eb"
+content-hash = "b3452b50901123fd5f2c385ce8a0c1c492296393b8a7926a322b6df0ea3ac572"

proxy/Cargo.toml

@@ -40,7 +40,6 @@ 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"] }

proxy/src/bin/proxy.rs

@@ -27,7 +27,6 @@ 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;
 
@@ -244,12 +243,6 @@ 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]
@@ -349,7 +342,7 @@ async fn main() -> anyhow::Result<()> {
 
     let cancel_map = CancelMap::default();
 
-    let redis_publisher = match &redis_notifications_client {
+    let redis_publisher = match &regional_redis_client {
         Some(redis_publisher) => Some(Arc::new(Mutex::new(RedisPublisherClient::new(
             redis_publisher.clone(),
             args.region.clone(),
@@ -606,8 +599,6 @@ 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,

proxy/src/compute.rs

@@ -1,7 +1,7 @@
 use crate::{
     auth::parse_endpoint_param,
     cancellation::CancelClosure,
-    console::{errors::WakeComputeError, messages::MetricsAuxInfo, provider::ApiLockError},
+    console::{errors::WakeComputeError, messages::MetricsAuxInfo},
     context::RequestMonitoring,
     error::{ReportableError, UserFacingError},
     metrics::{Metrics, NumDbConnectionsGuard},
@@ -34,9 +34,6 @@ pub enum ConnectionError {
 
     #[error("{COULD_NOT_CONNECT}: {0}")]
     WakeComputeError(#[from] WakeComputeError),
-
-    #[error("error acquiring resource permit: {0}")]
-    TooManyConnectionAttempts(#[from] ApiLockError),
 }
 
 impl UserFacingError for ConnectionError {
@@ -60,9 +57,6 @@ 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(),
         }
     }
@@ -78,7 +72,6 @@ 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(),
         }
     }
 }

proxy/src/config.rs

@@ -2,7 +2,7 @@ use crate::{
     auth::{self, backend::AuthRateLimiter},
     console::locks::ApiLocks,
     rate_limiter::RateBucketInfo,
-    serverless::{cancel_set::CancelSet, GlobalConnPoolOptions},
+    serverless::GlobalConnPoolOptions,
     Host,
 };
 use anyhow::{bail, ensure, Context, Ok};
@@ -56,8 +56,6 @@ 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 {
@@ -538,9 +536,9 @@ pub struct RetryConfig {
 impl RetryConfig {
     /// Default options for RetryConfig.
 
-    /// Total delay for 5 retries with 200ms base delay and 2 backoff factor is about 6s.
+    /// Total delay for 8 retries with 100ms base delay and 1.6 backoff factor is about 7s.
     pub const CONNECT_TO_COMPUTE_DEFAULT_VALUES: &'static str =
-        "num_retries=5,base_retry_wait_duration=200ms,retry_wait_exponent_base=2";
+        "num_retries=8,base_retry_wait_duration=100ms,retry_wait_exponent_base=1.6";
     /// 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 =
@@ -594,7 +592,7 @@ 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=10,epoch=10m,timeout=10ms";
+        "shards=64,permits=50,epoch=10m,timeout=500ms";
 
     // pub const DEFAULT_OPTIONS_WAKE_COMPUTE_LOCK: &'static str = "shards=32,permits=4,epoch=10m,timeout=1s";
 

proxy/src/console/provider.rs

@@ -12,7 +12,6 @@ use crate::{
     compute,
     config::{CacheOptions, EndpointCacheConfig, ProjectInfoCacheOptions},
     context::RequestMonitoring,
-    error::ReportableError,
     intern::ProjectIdInt,
     metrics::ApiLockMetrics,
     scram, EndpointCacheKey,
@@ -31,8 +30,6 @@ 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";
 
@@ -79,7 +76,7 @@ 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 temporarily unavailable. Check your quotas and/or contact our support.")
+                        format!("{REQUEST_FAILED}: endpoint is temporary unavailable. check your quotas and/or contact our support")
                     }
                     _ => REQUEST_FAILED.to_owned(),
                 },
@@ -214,8 +211,8 @@ pub mod errors {
         #[error("Too many connections attempts")]
         TooManyConnections,
 
-        #[error("error acquiring resource permit: {0}")]
-        TooManyConnectionAttempts(#[from] ApiLockError),
+        #[error("Timeout waiting to acquire wake compute lock")]
+        TimeoutError,
     }
 
     // This allows more useful interactions than `#[from]`.
@@ -225,6 +222,17 @@ 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::*;
@@ -237,9 +245,7 @@ pub mod errors {
 
                 TooManyConnections => self.to_string(),
 
-                TooManyConnectionAttempts(_) => {
-                    "Failed to acquire permit to connect to the database. Too many database connection attempts are currently ongoing.".to_owned()
-                }
+                TimeoutError => "timeout while acquiring the compute resource lock".to_owned(),
             }
         }
     }
@@ -250,7 +256,7 @@ pub mod errors {
                 WakeComputeError::BadComputeAddress(_) => crate::error::ErrorKind::ControlPlane,
                 WakeComputeError::ApiError(e) => e.get_error_kind(),
                 WakeComputeError::TooManyConnections => crate::error::ErrorKind::RateLimit,
-                WakeComputeError::TooManyConnectionAttempts(e) => e.get_error_kind(),
+                WakeComputeError::TimeoutError => crate::error::ErrorKind::ServiceRateLimit,
             }
         }
     }
@@ -450,23 +456,6 @@ 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,
@@ -486,7 +475,7 @@ impl<K: Hash + Eq + Clone> ApiLocks<K> {
         })
     }
 
-    pub async fn get_permit(&self, key: &K) -> Result<WakeComputePermit, ApiLockError> {
+    pub async fn get_permit(&self, key: &K) -> Result<WakeComputePermit, errors::WakeComputeError> {
         if self.permits == 0 {
             return Ok(WakeComputePermit { permit: None });
         }

proxy/src/proxy/retry.rs

@@ -86,8 +86,6 @@ 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,
         }
     }

proxy/src/proxy/wake_compute.rs

@@ -119,7 +119,7 @@ fn report_error(e: &WakeComputeError, retry: bool) {
             WakeupFailureKind::ApiConsoleOtherError
         }
         WakeComputeError::TooManyConnections => WakeupFailureKind::ApiConsoleLocked,
-        WakeComputeError::TooManyConnectionAttempts(_) => WakeupFailureKind::TimeoutError,
+        WakeComputeError::TimeoutError => WakeupFailureKind::TimeoutError,
     };
     Metrics::get()
         .proxy

proxy/src/serverless.rs

@@ -3,7 +3,6 @@
 //! Handles both SQL over HTTP and SQL over Websockets.
 
 mod backend;
-pub mod cancel_set;
 mod conn_pool;
 mod http_util;
 mod json;
@@ -110,37 +109,20 @@ pub async fn task_main(
         let conn_id = uuid::Uuid::new_v4();
         let http_conn_span = tracing::info_span!("http_conn", ?conn_id);
 
-        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.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),
+        );
     }
 
     connections.wait().await;
@@ -261,7 +243,6 @@ 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
         }

proxy/src/serverless/backend.rs

@@ -10,7 +10,6 @@ use crate::{
     console::{
         errors::{GetAuthInfoError, WakeComputeError},
         locks::ApiLocks,
-        provider::ApiLockError,
         CachedNodeInfo,
     },
     context::RequestMonitoring,
@@ -132,8 +131,6 @@ 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 {
@@ -144,7 +141,6 @@ 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(),
         }
     }
 }
@@ -157,9 +153,6 @@ 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()
-            }
         }
     }
 }
@@ -172,15 +165,6 @@ 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,
         }
     }
 }

proxy/src/serverless/cancel_set.rs

@@ -1,102 +0,0 @@
-//! 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);
-        }
-    }
-}

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::{serverless::cancel_set::CancelSet, BranchId, EndpointId, ProjectId};
+    use crate::{BranchId, EndpointId, ProjectId};
 
     use super::*;
 
@@ -767,8 +767,6 @@ 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 {

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::ClientDisconnect,
-            SqlOverHttpCancel::Connect => ErrorKind::ClientDisconnect,
+            SqlOverHttpCancel::Postgres => ErrorKind::RateLimit,
+            SqlOverHttpCancel::Connect => ErrorKind::ServiceRateLimit,
         }
     }
 }

pyproject.toml

@@ -14,7 +14,7 @@ requests = "^2.31.0"
 pytest-xdist = "^3.3.1"
 asyncpg = "^0.29.0"
 aiopg = "^1.4.0"
-Jinja2 = "^3.1.4"
+Jinja2 = "^3.1.3"
 types-requests = "^2.31.0.0"
 types-psycopg2 = "^2.9.21.10"
 boto3 = "^1.34.11"
@@ -24,7 +24,7 @@ backoff = "^2.2.1"
 pytest-lazy-fixture = "^0.6.3"
 prometheus-client = "^0.14.1"
 pytest-timeout = "^2.1.0"
-Werkzeug = "^3.0.3"
+Werkzeug = "^3.0.1"
 pytest-order = "^1.1.0"
 allure-pytest = "^2.13.2"
 pytest-asyncio = "^0.21.0"

safekeeper/src/safekeeper.rs

@@ -725,18 +725,6 @@ 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
 
@@ -1019,10 +1007,6 @@ 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(())

safekeeper/src/send_wal.rs

@@ -506,8 +506,6 @@ 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
@@ -586,22 +584,14 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
     async fn wait_wal(&mut self) -> Result<(), CopyStreamHandlerEnd> {
         loop {
             self.end_pos = self.end_watch.get();
-            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 {
+            if self.end_pos > self.start_pos {
+                // We 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) = self.wait_for_lsn().await? {
+            if let Some(lsn) = wait_for_lsn(&mut self.end_watch, self.term, self.start_pos).await? {
                 self.end_pos = lsn;
                 trace!("got end_pos {:?}, streaming", self.end_pos);
                 return Ok(());
@@ -638,54 +628,6 @@ 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.
@@ -743,6 +685,47 @@ 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};

safekeeper/src/wal_storage.rs

@@ -38,12 +38,6 @@ 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<()>;
 
@@ -84,8 +78,6 @@ 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.
@@ -177,8 +169,6 @@ 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,
@@ -334,20 +324,6 @@ 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.

safekeeper/tests/walproposer_sim/safekeeper_disk.rs

@@ -182,10 +182,6 @@ 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 {

safekeeper/tests/walproposer_sim/walproposer_api.rs

@@ -17,7 +17,8 @@ use utils::lsn::Lsn;
 use walproposer::{
     api_bindings::Level,
     bindings::{
-        NeonWALReadResult, SafekeeperStateDesiredEvents, WL_SOCKET_READABLE, WL_SOCKET_WRITEABLE,
+        pg_atomic_uint64, NeonWALReadResult, PageserverFeedback, SafekeeperStateDesiredEvents,
+        WL_SOCKET_READABLE, WL_SOCKET_WRITEABLE,
     },
     walproposer::{ApiImpl, Config},
 };
@@ -223,13 +224,31 @@ 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::api_bindings::empty_shmem()),
+            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,
+            }),
             config: args.config,
             event_set: RefCell::new(None),
         }
@@ -255,12 +274,6 @@ 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,

test_runner/fixtures/neon_fixtures.py

@@ -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, LayerFileName, parse_layer_file_name
+from fixtures.pageserver.types import IndexPartDump
 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_dict,
+    remote_storage_to_toml_inline_table,
 )
 from fixtures.safekeeper.http import SafekeeperHttpClient
 from fixtures.safekeeper.utils import are_walreceivers_absent
@@ -82,7 +82,6 @@ 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
@@ -466,7 +465,6 @@ 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
@@ -490,7 +488,6 @@ class NeonEnvBuilder:
         self.env: Optional[NeonEnv] = None
         self.keep_remote_storage_contents: bool = True
         self.neon_binpath = neon_binpath
-        self.neon_local_binpath = neon_binpath
         self.pg_distrib_dir = pg_distrib_dir
         self.pg_version = pg_version
         self.preserve_database_files = preserve_database_files
@@ -522,8 +519,6 @@ 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"
@@ -569,7 +564,6 @@ 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
@@ -638,11 +632,17 @@ class NeonEnvBuilder:
     def from_repo_dir(
         self,
         repo_dir: Path,
+        neon_binpath: Optional[Path] = None,
+        pg_distrib_dir: Optional[Path] = None,
     ) -> NeonEnv:
         """
         A simple method to import data into the current NeonEnvBuilder from a snapshot of a repo dir.
         """
 
+        # Setting custom `neon_binpath` and `pg_distrib_dir` is useful for compatibility tests
+        self.neon_binpath = neon_binpath or self.neon_binpath
+        self.pg_distrib_dir = pg_distrib_dir or self.pg_distrib_dir
+
         # Get the initial tenant and timeline from the snapshot config
         snapshot_config_toml = repo_dir / "config"
         with snapshot_config_toml.open("r") as f:
@@ -982,7 +982,7 @@ class NeonEnv:
 
     Some notable functions and fields in NeonEnv:
 
-    endpoints - A factory object for creating postgres compute nodes.
+    postgres - A factory object for creating postgres compute nodes.
 
     pageservers - An array containing objects representing the pageservers
 
@@ -1017,10 +1017,9 @@ class NeonEnv:
         self.pg_version = config.pg_version
         # Binary path for pageserver, safekeeper, etc
         self.neon_binpath = config.neon_binpath
-        # Binary path for neon_local test-specific binaries
-        self.neon_local_binpath = config.neon_local_binpath
-        if self.neon_local_binpath is None:
-            self.neon_local_binpath = self.neon_binpath
+        # Binary path for neon_local test-specific binaries: may be overridden
+        # after construction for compat testing
+        self.neon_local_binpath = config.neon_binpath
         self.pg_distrib_dir = config.pg_distrib_dir
         self.endpoint_counter = 0
         self.storage_controller_config = config.storage_controller_config
@@ -1052,7 +1051,6 @@ 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
         cfg: Dict[str, Any] = {
@@ -1289,7 +1287,6 @@ 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
@@ -1320,7 +1317,6 @@ 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()
 
@@ -1360,7 +1356,6 @@ 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.
@@ -1394,7 +1389,6 @@ 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
 
@@ -1554,7 +1548,6 @@ 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.
@@ -1578,16 +1571,6 @@ 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")
 
@@ -1730,7 +1713,8 @@ class NeonCli(AbstractNeonCli):
 
         ps_config = {}
         if remote_storage is not None:
-            ps_config["remote_storage"] = remote_storage_to_toml_dict(remote_storage)
+            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(";"):
@@ -1832,7 +1816,6 @@ class NeonCli(AbstractNeonCli):
         hot_standby: bool = False,
         lsn: Optional[Lsn] = None,
         pageserver_id: Optional[int] = None,
-        allow_multiple=False,
     ) -> "subprocess.CompletedProcess[str]":
         args = [
             "endpoint",
@@ -1856,8 +1839,6 @@ 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()
@@ -1869,7 +1850,6 @@ 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",
@@ -1884,8 +1864,6 @@ 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()
@@ -2449,16 +2427,20 @@ 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]], None]):
+    def edit_config_toml(self, edit_fn: Callable[[Dict[str, Any]], bool]):
         """
         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)
-        edit_fn(config)
-        with open(path, "w") as f:
-            toml.dump(config, f)
+        save = edit_fn(config)
+        if save:
+            with open(path, "w") as f:
+                toml.dump(config, f)
 
     def patch_config_toml_nonrecursive(self, patch: Dict[str, Any]) -> Dict[str, Any]:
         """
@@ -2469,12 +2451,13 @@ class NeonPageserver(PgProtocol, LogUtils):
         """
         replacements = {}
 
-        def doit(config: Dict[str, Any]):
+        def doit(config: Dict[str, Any]) -> bool:
             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
@@ -2652,37 +2635,6 @@ 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: LayerFileName
-    ) -> 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"""
@@ -3371,7 +3323,6 @@ 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.
@@ -3394,7 +3345,6 @@ 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)
@@ -3411,10 +3361,7 @@ class Endpoint(PgProtocol):
         return self
 
     def start(
-        self,
-        remote_ext_config: Optional[str] = None,
-        pageserver_id: Optional[int] = None,
-        allow_multiple: bool = False,
+        self, remote_ext_config: Optional[str] = None, pageserver_id: Optional[int] = None
     ) -> "Endpoint":
         """
         Start the Postgres instance.
@@ -3430,7 +3377,6 @@ class Endpoint(PgProtocol):
             safekeepers=self.active_safekeepers,
             remote_ext_config=remote_ext_config,
             pageserver_id=pageserver_id,
-            allow_multiple=allow_multiple,
         )
         self.running = True
 
@@ -3560,7 +3506,6 @@ 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.
@@ -3576,12 +3521,7 @@ class Endpoint(PgProtocol):
             hot_standby=hot_standby,
             lsn=lsn,
             pageserver_id=pageserver_id,
-            allow_multiple=allow_multiple,
-        ).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)
 
         log.info(f"Postgres startup took {time.time() - started_at} seconds")
 

test_runner/fixtures/pageserver/allowed_errors.py

@@ -89,8 +89,6 @@ DEFAULT_PAGESERVER_ALLOWED_ERRORS = (
     # 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.*",
-    # 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",
 )
 
 

test_runner/fixtures/pageserver/http.py

@@ -819,23 +819,6 @@ 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
     ):

test_runner/fixtures/pageserver/types.py

@@ -1,4 +1,3 @@
-import re
 from dataclasses import dataclass
 from typing import Any, Dict, Tuple, Union
 
@@ -48,36 +47,46 @@ class InvalidFileName(Exception):
     pass
 
 
-IMAGE_LAYER_FILE_NAME = re.compile("^([A-F0-9]{36})-([A-F0-9]{36})__([A-F0-9]{16})(-[a-f0-9]{8})?$")
-
-
 def parse_image_layer(f_name: str) -> Tuple[int, int, int]:
     """Parse an image layer file name. Return key start, key end, and snapshot lsn"""
-
-    match = IMAGE_LAYER_FILE_NAME.match(f_name)
-    if match is None:
-        raise InvalidFileName(f"'{f_name}' is not an image layer filename")
-
-    return int(match.group(1), 16), int(match.group(2), 16), int(match.group(3), 16)
-
-
-DELTA_LAYER_FILE_NAME = re.compile(
-    "^([A-F0-9]{36})-([A-F0-9]{36})__([A-F0-9]{16})-([A-F0-9]{16})(-[a-f0-9]{8})?$"
-)
+    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
 
 
 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"""
-    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),
-    )
+    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
 
 
 def parse_layer_file_name(file_name: str) -> LayerFileName:

test_runner/fixtures/parametrize.py

@@ -5,7 +5,6 @@ import pytest
 from _pytest.python import Metafunc
 
 from fixtures.pg_version import PgVersion
-from fixtures.utils import AuxFileStore
 
 """
 Dynamically parametrize tests by different parameters
@@ -32,11 +31,6 @@ 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"]

test_runner/fixtures/remote_storage.py

@@ -141,13 +141,11 @@ class LocalFsStorage:
         with self.heatmap_path(tenant_id).open("r") as f:
             return json.load(f)
 
-    def to_toml_dict(self) -> Dict[str, Any]:
-        return {
+    def to_toml_inline_table(self) -> str:
+        rv = {
             "local_path": str(self.root),
         }
-
-    def to_toml_inline_table(self) -> str:
-        return toml.TomlEncoder().dump_inline_table(self.to_toml_dict())
+        return toml.TomlEncoder().dump_inline_table(rv)
 
     def cleanup(self):
         # no cleanup is done here, because there's NeonEnvBuilder.cleanup_local_storage which will remove everything, including localfs files
@@ -196,7 +194,7 @@ class S3Storage:
             }
         )
 
-    def to_toml_dict(self) -> Dict[str, Any]:
+    def to_toml_inline_table(self) -> str:
         rv = {
             "bucket_name": self.bucket_name,
             "bucket_region": self.bucket_region,
@@ -208,10 +206,7 @@ class S3Storage:
         if self.endpoint is not None:
             rv["endpoint"] = self.endpoint
 
-        return rv
-
-    def to_toml_inline_table(self) -> str:
-        return toml.TomlEncoder().dump_inline_table(self.to_toml_dict())
+        return toml.TomlEncoder().dump_inline_table(rv)
 
     def do_cleanup(self):
         if not self.cleanup:
@@ -419,13 +414,6 @@ 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)):

test_runner/fixtures/utils.py

@@ -1,5 +1,4 @@
 import contextlib
-import enum
 import json
 import os
 import re
@@ -485,16 +484,3 @@ 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}'"

test_runner/performance/test_branch_creation.py

@@ -141,10 +141,9 @@ 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):
-        tenant_config = config.get("tenant_config", {})
-        tenant_config["compaction_period"] = "3s"
-        tenant_config["gc_period"] = "0s"
-        config["tenant_config"] = tenant_config
+        config["compaction_period"] = "3s"
+        config["gc_period"] = "0s"
+        return True
 
     env.pageserver.edit_config_toml(patch_default_tenant_config)
     env.pageserver.start(

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_aux_file_policy": "CrossValidation",
+        "switch_to_aux_file_v2": True,
     }
 
     ps_http = env.pageserver.http_client()

test_runner/regress/test_compatibility.py

@@ -233,18 +233,17 @@ def test_forward_compatibility(
         neon_env_builder.pageserver_validate_vectored_get = None
 
         neon_env_builder.num_safekeepers = 3
-
-        # Use previous version's production binaries (pageserver, safekeeper, pg_distrib_dir, etc.).
-        # But always use the current version's neon_local binary.
-        # This is because we want to test the compatibility of the data format, not the compatibility of the neon_local CLI.
-        neon_env_builder.neon_binpath = compatibility_neon_bin
-        neon_env_builder.pg_distrib_dir = compatibility_postgres_distrib_dir
-        neon_env_builder.neon_local_binpath = neon_env_builder.neon_local_binpath
-
+        neon_local_binpath = neon_env_builder.neon_binpath
         env = neon_env_builder.from_repo_dir(
             compatibility_snapshot_dir / "repo",
+            neon_binpath=compatibility_neon_bin,
+            pg_distrib_dir=compatibility_postgres_distrib_dir,
         )
 
+        # Use current neon_local even though we're using old binaries for
+        # everything else: our test code is written for latest CLI args.
+        env.neon_local_binpath = neon_local_binpath
+
         neon_env_builder.start()
 
         check_neon_works(

test_runner/regress/test_crafted_wal_end.py

@@ -19,12 +19,6 @@ from fixtures.neon_fixtures import NeonEnvBuilder, WalCraft
 def test_crafted_wal_end(neon_env_builder: NeonEnvBuilder, wal_type: str):
     env = neon_env_builder.init_start()
     env.neon_cli.create_branch("test_crafted_wal_end")
-    env.pageserver.allowed_errors.extend(
-        [
-            # seems like pageserver stop triggers these
-            ".*initial size calculation failed.*Bad state (not active).*",
-        ]
-    )
 
     endpoint = env.endpoints.create("test_crafted_wal_end")
     wal_craft = WalCraft(env)

test_runner/regress/test_disk_usage_eviction.py

@@ -47,9 +47,8 @@ def test_min_resident_size_override_handling(
     if config_level_override is not None:
 
         def set_min_resident_size(config):
-            tenant_config = config.get("tenant_config", {})
-            tenant_config["min_resident_size_override"] = config_level_override
-            config["tenant_config"] = tenant_config
+            config["tenant_config"] = {"min_resident_size": config_level_override}
+            return True
 
         env.pageserver.edit_config_toml(set_min_resident_size)
     env.pageserver.stop()

test_runner/regress/test_duplicate_layers.py

@@ -2,7 +2,6 @@ import time
 
 import pytest
 from fixtures.neon_fixtures import NeonEnvBuilder, PgBin, wait_for_last_flush_lsn
-from fixtures.pageserver.types import parse_layer_file_name
 from fixtures.pageserver.utils import (
     wait_for_last_record_lsn,
     wait_for_upload_queue_empty,
@@ -87,7 +86,14 @@ def test_actually_duplicated_l1(neon_env_builder: NeonEnvBuilder, pg_bin: PgBin)
 
     # path = env.remote_storage.timeline_path(tenant_id, timeline_id)
     l1_found = None
-    for path in env.pageserver.list_layers(tenant_id, timeline_id):
+    for path in env.pageserver.timeline_dir(tenant_id, timeline_id).iterdir():
+        if path.name == "metadata" or path.name.startswith("ephemeral-"):
+            continue
+
+        if len(path.suffixes) > 0:
+            # temp files
+            continue
+
         [key_range, lsn_range] = path.name.split("__", maxsplit=1)
 
         if "-" not in lsn_range:
@@ -102,21 +108,19 @@ def test_actually_duplicated_l1(neon_env_builder: NeonEnvBuilder, pg_bin: PgBin)
 
         if l1_found is not None:
             raise RuntimeError(f"found multiple L1: {l1_found.name} and {path.name}")
-        l1_found = parse_layer_file_name(path.name)
+        l1_found = path
 
     assert l1_found is not None, "failed to find L1 locally"
 
     uploaded = env.pageserver_remote_storage.remote_layer_path(
-        tenant_id, timeline_id, l1_found.to_str()
+        tenant_id, timeline_id, l1_found.name
     )
     assert not uploaded.exists(), "to-be-overwritten should not yet be uploaded"
 
     env.pageserver.start()
     wait_until_tenant_active(pageserver_http, tenant_id)
 
-    assert not env.pageserver.layer_exists(
-        tenant_id, timeline_id, l1_found
-    ), "partial compaction result should had been removed during startup"
+    assert not l1_found.exists(), "partial compaction result should had been removed during startup"
 
     # wait for us to catch up again
     wait_for_last_record_lsn(pageserver_http, tenant_id, timeline_id, lsn)
@@ -126,18 +130,18 @@ def test_actually_duplicated_l1(neon_env_builder: NeonEnvBuilder, pg_bin: PgBin)
     # give time for log flush
     time.sleep(1)
 
-    message = f".*duplicated L1 layer layer={l1_found}"
+    message = f".*duplicated L1 layer layer={l1_found.name}"
     found_msg = env.pageserver.log_contains(message)
     # resident or evicted, it should not be overwritten, however it should had been non-existing at startup
     assert (
         found_msg is None
     ), "layer should had been removed during startup, did it live on as evicted?"
 
-    assert env.pageserver.layer_exists(tenant_id, timeline_id, l1_found), "the L1 reappears"
+    assert l1_found.exists(), "the L1 reappears"
 
     wait_for_upload_queue_empty(pageserver_http, tenant_id, timeline_id)
 
     uploaded = env.pageserver_remote_storage.remote_layer_path(
-        tenant_id, timeline_id, l1_found.to_str()
+        tenant_id, timeline_id, l1_found.name
     )
     assert uploaded.exists(), "the L1 is uploaded"

test_runner/regress/test_layer_eviction.py

@@ -7,7 +7,6 @@ from fixtures.neon_fixtures import (
     flush_ep_to_pageserver,
     wait_for_last_flush_lsn,
 )
-from fixtures.pageserver.types import parse_layer_file_name
 from fixtures.pageserver.utils import wait_for_upload
 from fixtures.remote_storage import RemoteStorageKind
 
@@ -58,9 +57,9 @@ def test_basic_eviction(
     for sk in env.safekeepers:
         sk.stop()
 
-    initial_local_layers = dict(
-        (parse_layer_file_name(path.name), path)
-        for path in env.pageserver.list_layers(tenant_id, timeline_id)
+    timeline_path = env.pageserver.timeline_dir(tenant_id, timeline_id)
+    initial_local_layers = sorted(
+        list(filter(lambda path: path.name != "metadata", timeline_path.glob("*")))
     )
     assert (
         len(initial_local_layers) > 1
@@ -74,7 +73,6 @@ def test_basic_eviction(
     assert len(initial_local_layers) == len(
         initial_layer_map_info.historic_layers
     ), "Should have the same layers in memory and on disk"
-
     for returned_layer in initial_layer_map_info.historic_layers:
         assert (
             returned_layer.kind == "Delta"
@@ -83,29 +81,27 @@ def test_basic_eviction(
             not returned_layer.remote
         ), f"All created layers should be present locally, but got {returned_layer}"
 
-        returned_layer_name = parse_layer_file_name(returned_layer.layer_file_name)
-        assert (
-            returned_layer_name in initial_local_layers
-        ), f"Did not find returned layer {returned_layer_name} in local layers {list(initial_local_layers.keys())}"
-
-        local_layer_path = (
-            env.pageserver.timeline_dir(tenant_id, timeline_id)
-            / initial_local_layers[returned_layer_name]
+        local_layers = list(
+            filter(lambda layer: layer.name == returned_layer.layer_file_name, initial_local_layers)
         )
         assert (
-            returned_layer.layer_file_size == local_layer_path.stat().st_size
-        ), f"Returned layer {returned_layer} has a different file size than local layer {local_layer_path}"
+            len(local_layers) == 1
+        ), f"Did not find returned layer {returned_layer} in local layers {initial_local_layers}"
+        local_layer = local_layers[0]
+        assert (
+            returned_layer.layer_file_size == local_layer.stat().st_size
+        ), f"Returned layer {returned_layer} has a different file size than local layer {local_layer}"
 
     # Detach all layers, ensre they are not in the local FS, but are still dumped as part of the layer map
-    for local_layer_name, local_layer_path in initial_local_layers.items():
+    for local_layer in initial_local_layers:
         client.evict_layer(
-            tenant_id=tenant_id, timeline_id=timeline_id, layer_name=local_layer_path.name
+            tenant_id=tenant_id, timeline_id=timeline_id, layer_name=local_layer.name
         )
-        assert not env.pageserver.layer_exists(
-            tenant_id, timeline_id, local_layer_name
-        ), f"Did not expect to find {local_layer_name} layer after evicting"
+        assert not any(
+            new_local_layer.name == local_layer.name for new_local_layer in timeline_path.glob("*")
+        ), f"Did not expect to find {local_layer} layer after evicting"
 
-    empty_layers = env.pageserver.list_layers(tenant_id, timeline_id)
+    empty_layers = list(filter(lambda path: path.name != "metadata", timeline_path.glob("*")))
     assert not empty_layers, f"After evicting all layers, timeline {tenant_id}/{timeline_id} should have no layers locally, but got: {empty_layers}"
 
     evicted_layer_map_info = client.layer_map_info(tenant_id=tenant_id, timeline_id=timeline_id)
@@ -122,15 +118,15 @@ def test_basic_eviction(
         assert (
             returned_layer.remote
         ), f"All layers should be evicted and not present locally, but got {returned_layer}"
-        returned_layer_name = parse_layer_file_name(returned_layer.layer_file_name)
-        assert (
-            returned_layer_name in initial_local_layers
+        assert any(
+            local_layer.name == returned_layer.layer_file_name
+            for local_layer in initial_local_layers
         ), f"Did not find returned layer {returned_layer} in local layers {initial_local_layers}"
 
     # redownload all evicted layers and ensure the initial state is restored
-    for local_layer_name, _local_layer_path in initial_local_layers.items():
+    for local_layer in initial_local_layers:
         client.download_layer(
-            tenant_id=tenant_id, timeline_id=timeline_id, layer_name=local_layer_name.to_str()
+            tenant_id=tenant_id, timeline_id=timeline_id, layer_name=local_layer.name
         )
     client.timeline_download_remote_layers(
         tenant_id,
@@ -141,9 +137,8 @@ def test_basic_eviction(
         at_least_one_download=False,
     )
 
-    redownloaded_layers = dict(
-        (parse_layer_file_name(path.name), path)
-        for path in env.pageserver.list_layers(tenant_id, timeline_id)
+    redownloaded_layers = sorted(
+        list(filter(lambda path: path.name != "metadata", timeline_path.glob("*")))
     )
     assert (
         redownloaded_layers == initial_local_layers

test_runner/regress/test_logical_replication.py

@@ -6,9 +6,7 @@ from string import ascii_lowercase
 import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import (
-    AuxFileStore,
     NeonEnv,
-    NeonEnvBuilder,
     logical_replication_sync,
     wait_for_last_flush_lsn,
 )
@@ -20,19 +18,6 @@ def random_string(n: int):
     return "".join([choice(ascii_lowercase) for _ in range(n)])
 
 
-@pytest.mark.parametrize(
-    "pageserver_aux_file_policy", [AuxFileStore.V1, AuxFileStore.V2, AuxFileStore.CrossValidation]
-)
-def test_aux_file_v2_flag(neon_simple_env: NeonEnv, pageserver_aux_file_policy: AuxFileStore):
-    env = neon_simple_env
-    with env.pageserver.http_client() as client:
-        tenant_config = client.tenant_config(env.initial_tenant).effective_config
-        assert pageserver_aux_file_policy == tenant_config["switch_aux_file_policy"]
-
-
-@pytest.mark.parametrize(
-    "pageserver_aux_file_policy", [AuxFileStore.V1, AuxFileStore.CrossValidation]
-)
 def test_logical_replication(neon_simple_env: NeonEnv, vanilla_pg):
     env = neon_simple_env
 
@@ -174,9 +159,6 @@ COMMIT;
 
 
 # Test that neon.logical_replication_max_snap_files works
-@pytest.mark.parametrize(
-    "pageserver_aux_file_policy", [AuxFileStore.V1, AuxFileStore.CrossValidation]
-)
 def test_obsolete_slot_drop(neon_simple_env: NeonEnv, vanilla_pg):
     def slot_removed(ep):
         assert (
@@ -221,86 +203,8 @@ def test_obsolete_slot_drop(neon_simple_env: NeonEnv, vanilla_pg):
     wait_until(number_of_iterations=10, interval=2, func=partial(slot_removed, endpoint))
 
 
-# Tests that walsender correctly blocks until WAL is downloaded from safekeepers
-def test_lr_with_slow_safekeeper(neon_env_builder: NeonEnvBuilder, vanilla_pg):
-    neon_env_builder.num_safekeepers = 3
-    env = neon_env_builder.init_start()
-
-    env.neon_cli.create_branch("init")
-    endpoint = env.endpoints.create_start("init")
-
-    with endpoint.connect().cursor() as cur:
-        cur.execute("create table wal_generator (id serial primary key, data text)")
-        cur.execute(
-            """
-INSERT INTO wal_generator (data)
-SELECT repeat('A', 1024) -- Generates a kilobyte of data per row
-FROM generate_series(1, 16384) AS seq; -- Inserts enough rows to exceed 16MB of data
-"""
-        )
-        cur.execute("create table t(a int)")
-        cur.execute("create publication pub for table t")
-        cur.execute("insert into t values (1)")
-
-    vanilla_pg.start()
-    vanilla_pg.safe_psql("create table t(a int)")
-    connstr = endpoint.connstr().replace("'", "''")
-    vanilla_pg.safe_psql(f"create subscription sub1 connection '{connstr}' publication pub")
-    logical_replication_sync(vanilla_pg, endpoint)
-    vanilla_pg.stop()
-
-    # Pause the safekeepers so that they can't send WAL (except to pageserver)
-    for sk in env.safekeepers:
-        sk_http = sk.http_client()
-        sk_http.configure_failpoints([("sk-pause-send", "return")])
-
-    # Insert a 2
-    with endpoint.connect().cursor() as cur:
-        cur.execute("insert into t values (2)")
-
-    endpoint.stop_and_destroy()
-
-    # This new endpoint should contain [1, 2], but it can't access WAL from safekeeper
-    endpoint = env.endpoints.create_start("init")
-    with endpoint.connect().cursor() as cur:
-        cur.execute("select * from t")
-        res = [r[0] for r in cur.fetchall()]
-        assert res == [1, 2]
-
-    # Reconnect subscriber
-    vanilla_pg.start()
-    connstr = endpoint.connstr().replace("'", "''")
-    vanilla_pg.safe_psql(f"alter subscription sub1 connection '{connstr}'")
-
-    time.sleep(5)
-    # Make sure the 2 isn't replicated
-    assert [r[0] for r in vanilla_pg.safe_psql("select * from t")] == [1]
-
-    # Re-enable WAL download
-    for sk in env.safekeepers:
-        sk_http = sk.http_client()
-        sk_http.configure_failpoints([("sk-pause-send", "off")])
-
-    logical_replication_sync(vanilla_pg, endpoint)
-    assert [r[0] for r in vanilla_pg.safe_psql("select * from t")] == [1, 2]
-
-    # Check that local reads also work
-    with endpoint.connect().cursor() as cur:
-        cur.execute("insert into t values (3)")
-    logical_replication_sync(vanilla_pg, endpoint)
-    assert [r[0] for r in vanilla_pg.safe_psql("select * from t")] == [1, 2, 3]
-
-    log_path = vanilla_pg.pgdatadir / "pg.log"
-    with open(log_path, "r") as log_file:
-        logs = log_file.read()
-        assert "could not receive data from WAL stream" not in logs
-
-
 # Test compute start at LSN page of which starts with contrecord
 # https://github.com/neondatabase/neon/issues/5749
-@pytest.mark.parametrize(
-    "pageserver_aux_file_policy", [AuxFileStore.V1, AuxFileStore.CrossValidation]
-)
 def test_wal_page_boundary_start(neon_simple_env: NeonEnv, vanilla_pg):
     env = neon_simple_env
 
@@ -391,9 +295,6 @@ def test_wal_page_boundary_start(neon_simple_env: NeonEnv, vanilla_pg):
 # logical replication bug as such, but without logical replication,
 # records passed ot the WAL redo process are never large enough to hit
 # the bug.
-@pytest.mark.parametrize(
-    "pageserver_aux_file_policy", [AuxFileStore.V1, AuxFileStore.CrossValidation]
-)
 def test_large_records(neon_simple_env: NeonEnv, vanilla_pg):
     env = neon_simple_env
 
@@ -465,9 +366,6 @@ def test_slots_and_branching(neon_simple_env: NeonEnv):
     ws_cur.execute("select pg_create_logical_replication_slot('my_slot', 'pgoutput')")
 
 
-@pytest.mark.parametrize(
-    "pageserver_aux_file_policy", [AuxFileStore.V1, AuxFileStore.CrossValidation]
-)
 def test_replication_shutdown(neon_simple_env: NeonEnv):
     # Ensure Postgres can exit without stuck when a replication job is active + neon extension installed
     env = neon_simple_env

test_runner/regress/test_lsn_mapping.py

@@ -119,11 +119,11 @@ def test_ts_of_lsn_api(neon_env_builder: NeonEnvBuilder):
 
     cur = endpoint_main.connect().cursor()
     # Create table, and insert rows, each in a separate transaction
-    # Enable synchronous commit as we are timing sensitive
+    # Disable synchronous_commit to make this initialization go faster.
     #
     # Each row contains current insert LSN and the current timestamp, when
     # the row was inserted.
-    cur.execute("SET synchronous_commit=on")
+    cur.execute("SET synchronous_commit=off")
     cur.execute("CREATE TABLE foo (x integer)")
     tbl = []
     for i in range(1000):
@@ -132,7 +132,7 @@ def test_ts_of_lsn_api(neon_env_builder: NeonEnvBuilder):
         after_timestamp = query_scalar(cur, "SELECT clock_timestamp()").replace(tzinfo=timezone.utc)
         after_lsn = query_scalar(cur, "SELECT pg_current_wal_lsn()")
         tbl.append([i, after_timestamp, after_lsn])
-        time.sleep(0.02)
+        time.sleep(0.005)
 
     # Execute one more transaction with synchronous_commit enabled, to flush
     # all the previous transactions

test_runner/regress/test_pageserver_generations.py

@@ -10,7 +10,6 @@ of the pageserver are:
 """
 
 import enum
-import os
 import re
 import time
 from typing import Optional
@@ -701,50 +700,3 @@ def test_multi_attach(
 
     # All data we wrote while multi-attached remains readable
     workload.validate(pageservers[2].id)
-
-
-@pytest.mark.skip(reason="To be enabled after release with new local path style")
-def test_upgrade_generationless_local_file_paths(
-    neon_env_builder: NeonEnvBuilder,
-):
-    """
-    Test pageserver behavior when startup up with local layer paths without
-    generation numbers: it should accept these layer files, and avoid doing
-    a delete/download cycle on them.
-    """
-    env = neon_env_builder.init_start(initial_tenant_conf=TENANT_CONF)
-    tenant_id = env.initial_tenant
-    timeline_id = env.initial_timeline
-
-    workload = Workload(env, tenant_id, timeline_id)
-    workload.init()
-    workload.write_rows(1000)
-
-    env.pageserver.stop()
-
-    # Rename the local paths to legacy format, to simulate what
-    # we would see when upgrading
-    timeline_dir = env.pageserver.timeline_dir(tenant_id, timeline_id)
-    files_renamed = 0
-    for filename in os.listdir(timeline_dir):
-        path = os.path.join(timeline_dir, filename)
-        log.info(f"Found file {path}")
-        if path.endswith("-00000001"):
-            new_path = path[:-9]
-            os.rename(path, new_path)
-            log.info(f"Renamed {path} -> {new_path}")
-            files_renamed += 1
-
-    assert files_renamed > 0
-
-    env.pageserver.start()
-
-    workload.validate()
-
-    # Assert that there were no on-demand downloads
-    assert (
-        env.pageserver.http_client().get_metric_value(
-            "pageserver_remote_ondemand_downloaded_layers_total"
-        )
-        == 0
-    )

test_runner/regress/test_pageserver_restart.py

@@ -20,10 +20,7 @@ def test_pageserver_restart(neon_env_builder: NeonEnvBuilder):
     endpoint = env.endpoints.create_start("main")
     pageserver_http = env.pageserver.http_client()
 
-    assert (
-        pageserver_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"})
-        == 1
-    )
+    assert pageserver_http.get_metric_value("pageserver_tenant_manager_slots") == 1
 
     pg_conn = endpoint.connect()
     cur = pg_conn.cursor()
@@ -58,10 +55,7 @@ def test_pageserver_restart(neon_env_builder: NeonEnvBuilder):
     env.pageserver.start()
 
     # We reloaded our tenant
-    assert (
-        pageserver_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"})
-        == 1
-    )
+    assert pageserver_http.get_metric_value("pageserver_tenant_manager_slots") == 1
 
     cur.execute("SELECT count(*) FROM foo")
     assert cur.fetchone() == (100000,)

test_runner/regress/test_pageserver_secondary.py

@@ -2,12 +2,12 @@ import json
 import os
 import random
 import time
+from pathlib import Path
 from typing import Any, Dict, Optional
 
 import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnvBuilder, NeonPageserver, S3Scrubber
-from fixtures.pageserver.types import parse_layer_file_name
 from fixtures.pageserver.utils import (
     assert_prefix_empty,
     poll_for_remote_storage_iterations,
@@ -51,13 +51,9 @@ def evict_random_layers(
         if "ephemeral" in layer.name or "temp_download" in layer.name:
             continue
 
-        layer_name = parse_layer_file_name(layer.name)
-
         if rng.choice([True, False]):
-            log.info(f"Evicting layer {tenant_id}/{timeline_id} {layer_name.to_str()}")
-            client.evict_layer(
-                tenant_id=tenant_id, timeline_id=timeline_id, layer_name=layer_name.to_str()
-            )
+            log.info(f"Evicting layer {tenant_id}/{timeline_id} {layer.name}")
+            client.evict_layer(tenant_id=tenant_id, timeline_id=timeline_id, layer_name=layer.name)
 
 
 @pytest.mark.parametrize("seed", [1, 2, 3])
@@ -406,6 +402,32 @@ def test_heatmap_uploads(neon_env_builder: NeonEnvBuilder):
     validate_heatmap(heatmap_second)
 
 
+def list_layers(pageserver, 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 = pageserver.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 test_secondary_downloads(neon_env_builder: NeonEnvBuilder):
     """
     Test the overall data flow in secondary mode:
@@ -460,8 +482,8 @@ def test_secondary_downloads(neon_env_builder: NeonEnvBuilder):
 
     ps_secondary.http_client().tenant_secondary_download(tenant_id)
 
-    assert ps_attached.list_layers(tenant_id, timeline_id) == ps_secondary.list_layers(
-        tenant_id, timeline_id
+    assert list_layers(ps_attached, tenant_id, timeline_id) == list_layers(
+        ps_secondary, tenant_id, timeline_id
     )
 
     # Make changes on attached pageserver, check secondary downloads them
@@ -478,8 +500,8 @@ def test_secondary_downloads(neon_env_builder: NeonEnvBuilder):
     ps_secondary.http_client().tenant_secondary_download(tenant_id)
 
     try:
-        assert ps_attached.list_layers(tenant_id, timeline_id) == ps_secondary.list_layers(
-            tenant_id, timeline_id
+        assert list_layers(ps_attached, tenant_id, timeline_id) == list_layers(
+            ps_secondary, tenant_id, timeline_id
         )
     except:
         # Do a full listing of the secondary location on errors, to help debug of
@@ -501,8 +523,8 @@ def test_secondary_downloads(neon_env_builder: NeonEnvBuilder):
     # ==================================================================
     try:
         log.info("Evicting a layer...")
-        layer_to_evict = ps_attached.list_layers(tenant_id, timeline_id)[0]
-        some_other_layer = ps_attached.list_layers(tenant_id, timeline_id)[1]
+        layer_to_evict = list_layers(ps_attached, tenant_id, timeline_id)[0]
+        some_other_layer = list_layers(ps_attached, tenant_id, timeline_id)[1]
         log.info(f"Victim layer: {layer_to_evict.name}")
         ps_attached.http_client().evict_layer(
             tenant_id, timeline_id, layer_name=layer_to_evict.name
@@ -515,13 +537,13 @@ def test_secondary_downloads(neon_env_builder: NeonEnvBuilder):
             layer["name"] for layer in heatmap_after_eviction["timelines"][0]["layers"]
         )
         assert layer_to_evict.name not in heatmap_layers
-        assert parse_layer_file_name(some_other_layer.name).to_str() in heatmap_layers
+        assert some_other_layer.name in heatmap_layers
 
         ps_secondary.http_client().tenant_secondary_download(tenant_id)
 
-        assert layer_to_evict not in ps_attached.list_layers(tenant_id, timeline_id)
-        assert ps_attached.list_layers(tenant_id, timeline_id) == ps_secondary.list_layers(
-            tenant_id, timeline_id
+        assert layer_to_evict not in list_layers(ps_attached, tenant_id, timeline_id)
+        assert list_layers(ps_attached, tenant_id, timeline_id) == list_layers(
+            ps_secondary, tenant_id, timeline_id
         )
     except:
         # On assertion failures, log some details to help with debugging
@@ -608,7 +630,7 @@ def test_secondary_background_downloads(neon_env_builder: NeonEnvBuilder):
         for timeline_id in timelines:
             log.info(f"Checking for secondary timeline {timeline_id} on node {ps_secondary.id}")
             # One or more layers should be present for all timelines
-            assert ps_secondary.list_layers(tenant_id, timeline_id)
+            assert list_layers(ps_secondary, tenant_id, timeline_id)
 
         # Delete the second timeline: this should be reflected later on the secondary
         env.storage_controller.pageserver_api().timeline_delete(tenant_id, timelines[1])
@@ -623,10 +645,10 @@ def test_secondary_background_downloads(neon_env_builder: NeonEnvBuilder):
         ps_secondary = next(p for p in env.pageservers if p != ps_attached)
 
         # This one was not deleted
-        assert ps_secondary.list_layers(tenant_id, timelines[0])
+        assert list_layers(ps_secondary, tenant_id, timelines[0])
 
         # This one was deleted
-        assert not ps_secondary.list_layers(tenant_id, timelines[1])
+        assert not list_layers(ps_secondary, tenant_id, timelines[1])
 
     t_end = time.time()
 
@@ -686,7 +708,7 @@ def test_slow_secondary_downloads(neon_env_builder: NeonEnvBuilder, via_controll
     ps_attached.http_client().timeline_checkpoint(tenant_id, timeline_id)
 
     # Expect lots of layers
-    assert len(ps_attached.list_layers(tenant_id, timeline_id)) > 10
+    assert len(list_layers(ps_attached, tenant_id, timeline_id)) > 10
 
     # Simulate large data by making layer downloads artifically slow
     for ps in env.pageservers:

test_runner/regress/test_pg_waldump.py

@@ -1,28 +1,13 @@
 import os
-import shutil
 
 from fixtures.neon_fixtures import NeonEnv, PgBin
 from fixtures.utils import subprocess_capture
 
 
-def check_wal_segment(pg_waldump_path: str, segment_path: str, test_output_dir):
-    # use special --ignore option to ignore the validation checks in pg_waldump
-    # this is necessary, because neon WAL files contain gap at the beginning
-    output_path, _, _ = subprocess_capture(
-        test_output_dir, [pg_waldump_path, "--ignore", segment_path]
-    )
-
-    with open(f"{output_path}.stdout", "r") as f:
-        stdout = f.read()
-        assert "ABORT" in stdout
-        assert "COMMIT" in stdout
-
-
 # Simple test to check that pg_waldump works with neon WAL files
 def test_pg_waldump(neon_simple_env: NeonEnv, test_output_dir, pg_bin: PgBin):
     env = neon_simple_env
-    tenant_id = env.initial_tenant
-    timeline_id = env.neon_cli.create_branch("test_pg_waldump", "empty")
+    env.neon_cli.create_branch("test_pg_waldump", "empty")
     endpoint = env.endpoints.create_start("test_pg_waldump")
 
     cur = endpoint.connect().cursor()
@@ -50,12 +35,12 @@ def test_pg_waldump(neon_simple_env: NeonEnv, test_output_dir, pg_bin: PgBin):
     assert endpoint.pgdata_dir
     wal_path = os.path.join(endpoint.pgdata_dir, "pg_wal/000000010000000000000001")
     pg_waldump_path = os.path.join(pg_bin.pg_bin_path, "pg_waldump")
-    # check segment on compute
-    check_wal_segment(pg_waldump_path, wal_path, test_output_dir)
 
-    # Check file on safekeepers as well. pg_waldump is strict about file naming, so remove .partial suffix.
-    sk = env.safekeepers[0]
-    sk_tli_dir = sk.timeline_dir(tenant_id, timeline_id)
-    non_partial_path = os.path.join(sk_tli_dir, "000000010000000000000001")
-    shutil.copyfile(os.path.join(sk_tli_dir, "000000010000000000000001.partial"), non_partial_path)
-    check_wal_segment(pg_waldump_path, non_partial_path, test_output_dir)
+    # use special --ignore option to ignore the validation checks in pg_waldump
+    # this is necessary, because neon WAL files contain gap at the beginning
+    output_path, _, _ = subprocess_capture(test_output_dir, [pg_waldump_path, "--ignore", wal_path])
+
+    with open(f"{output_path}.stdout", "r") as f:
+        stdout = f.read()
+        assert "ABORT" in stdout
+        assert "COMMIT" in stdout

test_runner/regress/test_postgres_version.py

@@ -1,35 +0,0 @@
-import json
-import re
-from pathlib import Path
-
-from fixtures.neon_fixtures import PgBin
-from fixtures.pg_version import PgVersion
-
-
-def test_postgres_version(base_dir: Path, pg_bin: PgBin, pg_version: PgVersion):
-    """Test that Postgres version matches the one we expect"""
-
-    with (base_dir / "vendor" / "revisions.json").open() as f:
-        expected_revisions = json.load(f)
-
-    output_prefix = pg_bin.run_capture(["postgres", "--version"], with_command_header=False)
-    stdout = Path(f"{output_prefix}.stdout")
-    assert stdout.exists(), "postgres --version didn't print anything to stdout"
-
-    with stdout.open() as f:
-        output = f.read().strip()
-
-    # `postgres --version` prints something like "postgres (PostgreSQL) 15.6 (85d809c124a898847a97d66a211f7d5ef4f8e0cb)".
-    pattern = r"postgres \(PostgreSQL\) (?P<version>\d+\.\d+) \((?P<commit>[0-9a-f]{40})\)"
-    match = re.search(pattern, output, re.IGNORECASE)
-    assert match is not None, f"Can't parse {output} with {pattern}"
-
-    version = match.group("version")
-    commit = match.group("commit")
-
-    assert (
-        pg_version.v_prefixed in expected_revisions
-    ), f"Version `{pg_version.v_prefixed}` doesn't exist in `vendor/revisions.json`, please update it if these changes are intentional"
-
-    msg = f"Unexpected Postgres {pg_version} version: `{output}`, please update `vendor/revisions.json` if these changes are intentional"
-    assert [version, commit] == expected_revisions[pg_version.v_prefixed], msg

test_runner/regress/test_remote_storage.py

@@ -12,7 +12,6 @@ from fixtures.neon_fixtures import (
     wait_for_last_flush_lsn,
 )
 from fixtures.pageserver.http import PageserverApiException, PageserverHttpClient
-from fixtures.pageserver.types import parse_layer_file_name
 from fixtures.pageserver.utils import (
     timeline_delete_wait_completed,
     wait_for_last_record_lsn,
@@ -830,9 +829,8 @@ def test_compaction_waits_for_upload(
     assert len(upload_stuck_layers) > 0
 
     for name in upload_stuck_layers:
-        assert env.pageserver.layer_exists(
-            tenant_id, timeline_id, parse_layer_file_name(name)
-        ), "while uploads are stuck the layers should be present on disk"
+        path = env.pageserver.timeline_dir(tenant_id, timeline_id) / name
+        assert path.exists(), "while uploads are stuck the layers should be present on disk"
 
     # now this will do the L0 => L1 compaction and want to remove
     # upload_stuck_layers and the original initdb L0
@@ -840,9 +838,8 @@ def test_compaction_waits_for_upload(
 
     # as uploads are paused, the upload_stuck_layers should still be with us
     for name in upload_stuck_layers:
-        assert env.pageserver.layer_exists(
-            tenant_id, timeline_id, parse_layer_file_name(name)
-        ), "uploads are stuck still over compaction"
+        path = env.pageserver.timeline_dir(tenant_id, timeline_id) / name
+        assert path.exists(), "uploads are stuck still over compaction"
 
     compacted_layers = client.layer_map_info(tenant_id, timeline_id).historic_by_name()
     overlap = compacted_layers.intersection(upload_stuck_layers)
@@ -876,8 +873,9 @@ def test_compaction_waits_for_upload(
     wait_until(10, 1, until_layer_deletes_completed)
 
     for name in upload_stuck_layers:
-        assert not env.pageserver.layer_exists(
-            tenant_id, timeline_id, parse_layer_file_name(name)
+        path = env.pageserver.timeline_dir(tenant_id, timeline_id) / name
+        assert (
+            not path.exists()
         ), "l0 should now be removed because of L0 => L1 compaction and completed uploads"
 
     # We should not have hit the error handling path in uploads where a uploaded file is gone

test_runner/regress/test_s3_restore.py

@@ -47,7 +47,7 @@ def test_tenant_s3_restore(
     tenant_id = env.initial_tenant
 
     # Default tenant and the one we created
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 1
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 1
 
     # create two timelines one being the parent of another, both with non-trivial data
     parent = None
@@ -72,13 +72,13 @@ def test_tenant_s3_restore(
     time.sleep(4)
 
     assert (
-        ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 1
+        ps_http.get_metric_value("pageserver_tenant_manager_slots") == 1
     ), "tenant removed before we deletion was issued"
     iterations = poll_for_remote_storage_iterations(remote_storage_kind)
     tenant_delete_wait_completed(ps_http, tenant_id, iterations)
     ps_http.deletion_queue_flush(execute=True)
     assert (
-        ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 0
+        ps_http.get_metric_value("pageserver_tenant_manager_slots") == 0
     ), "tenant removed before we deletion was issued"
     env.storage_controller.attach_hook_drop(tenant_id)
 
@@ -116,4 +116,4 @@ def test_tenant_s3_restore(
             # There might be some activity that advances the lsn so we can't use a strict equality check
             assert last_flush_lsn >= expected_last_flush_lsn, "last_flush_lsn too old"
 
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 1
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 1

test_runner/regress/test_sharding.py

@@ -177,67 +177,6 @@ def test_sharding_split_unsharded(
     env.storage_controller.consistency_check()
 
 
-def test_sharding_split_compaction(neon_env_builder: NeonEnvBuilder):
-    """
-    Test that after a split, we clean up parent layer data in the child shards via compaction.
-    """
-    TENANT_CONF = {
-        # small checkpointing and compaction targets to ensure we generate many upload operations
-        "checkpoint_distance": f"{128 * 1024}",
-        "compaction_threshold": "1",
-        "compaction_target_size": f"{128 * 1024}",
-        # no PITR horizon, we specify the horizon when we request on-demand GC
-        "pitr_interval": "3600s",
-        # disable background compaction and GC. We invoke it manually when we want it to happen.
-        "gc_period": "0s",
-        "compaction_period": "0s",
-        # create image layers eagerly, so that GC can remove some layers
-        "image_creation_threshold": "1",
-        "image_layer_creation_check_threshold": "0",
-    }
-
-    env = neon_env_builder.init_start(initial_tenant_conf=TENANT_CONF)
-    tenant_id = env.initial_tenant
-    timeline_id = env.initial_timeline
-
-    # Check that we created with an unsharded TenantShardId: this is the default,
-    # but check it in case we change the default in future
-    assert env.storage_controller.inspect(TenantShardId(tenant_id, 0, 0)) is not None
-
-    workload = Workload(env, tenant_id, timeline_id, branch_name="main")
-    workload.init()
-    workload.write_rows(256)
-    workload.validate()
-    workload.stop()
-
-    # Split one shard into two
-    shards = env.storage_controller.tenant_shard_split(tenant_id, shard_count=2)
-
-    # Check we got the shard IDs we expected
-    assert env.storage_controller.inspect(TenantShardId(tenant_id, 0, 2)) is not None
-    assert env.storage_controller.inspect(TenantShardId(tenant_id, 1, 2)) is not None
-
-    workload.validate()
-    workload.stop()
-
-    env.storage_controller.consistency_check()
-
-    # Cleanup part 1: while layers are still in PITR window, we should only drop layers that are fully redundant
-    for shard in shards:
-        ps = env.get_tenant_pageserver(shard)
-
-        # Invoke compaction: this should drop any layers that don't overlap with the shard's key stripes
-        detail_before = ps.http_client().timeline_detail(shard, timeline_id)
-        ps.http_client().timeline_compact(shard, timeline_id)
-        detail_after = ps.http_client().timeline_detail(shard, timeline_id)
-
-        # Physical size should shrink because some layers have been dropped
-        assert detail_after["current_physical_size"] < detail_before["current_physical_size"]
-
-    # Compaction shouldn't make anything unreadable
-    workload.validate()
-
-
 def test_sharding_split_smoke(
     neon_env_builder: NeonEnvBuilder,
 ):

test_runner/regress/test_tenant_delete.py

@@ -64,7 +64,7 @@ def test_tenant_delete_smoke(
     )
 
     # Default tenant and the one we created
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 2
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 2
 
     # create two timelines one being the parent of another
     parent = None
@@ -90,9 +90,9 @@ def test_tenant_delete_smoke(
 
     iterations = poll_for_remote_storage_iterations(remote_storage_kind)
 
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 2
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 2
     tenant_delete_wait_completed(ps_http, tenant_id, iterations)
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 1
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 1
 
     tenant_path = env.pageserver.tenant_dir(tenant_id)
     assert not tenant_path.exists()
@@ -108,7 +108,7 @@ def test_tenant_delete_smoke(
     )
 
     # Deletion updates the tenant count: the one default tenant remains
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 1
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 1
 
 
 class Check(enum.Enum):
@@ -532,9 +532,7 @@ def test_tenant_delete_concurrent(
 
         # The TenantSlot is still present while the original request is hung before
         # final removal
-        assert (
-            ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 1
-        )
+        assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 1
 
         # Permit the original request to run to success
         ps_http.configure_failpoints((BEFORE_REMOVE_FAILPOINT, "off"))
@@ -558,8 +556,7 @@ def test_tenant_delete_concurrent(
     )
 
     # Zero tenants remain (we deleted the default tenant)
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 0
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "inprogress"}) == 0
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 0
 
 
 def test_tenant_delete_races_timeline_creation(
@@ -676,7 +673,7 @@ def test_tenant_delete_races_timeline_creation(
     )
 
     # Zero tenants remain (we deleted the default tenant)
-    assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 0
+    assert ps_http.get_metric_value("pageserver_tenant_manager_slots") == 0
 
 
 def test_tenant_delete_scrubber(pg_bin: PgBin, neon_env_builder: NeonEnvBuilder):

test_runner/regress/test_tenant_size.py

@@ -668,9 +668,9 @@ def test_synthetic_size_while_deleting(neon_env_builder: NeonEnvBuilder):
 
         client.configure_failpoints((failpoint, "off"))
 
-        # accept both, because the deletion might still complete before
-        matcher = "(Failed to refresh gc_info before gathering inputs|NotFound: tenant)"
-        with pytest.raises(PageserverApiException, match=matcher):
+        with pytest.raises(
+            PageserverApiException, match="Failed to refresh gc_info before gathering inputs"
+        ):
             completion.result()
 
     # this happens on both cases

test_runner/regress/test_tenants_with_remote_storage.py

@@ -18,7 +18,6 @@ from fixtures.neon_fixtures import (
     NeonEnvBuilder,
     last_flush_lsn_upload,
 )
-from fixtures.pageserver.types import parse_layer_file_name
 from fixtures.pageserver.utils import (
     assert_tenant_state,
     wait_for_last_record_lsn,
@@ -247,10 +246,7 @@ def test_tenant_redownloads_truncated_file_on_startup(
 
     # ensure the same size is found from the index_part.json
     index_part = env.pageserver_remote_storage.index_content(tenant_id, timeline_id)
-    assert (
-        index_part["layer_metadata"][parse_layer_file_name(path.name).to_str()]["file_size"]
-        == expected_size
-    )
+    assert index_part["layer_metadata"][path.name]["file_size"] == expected_size
 
     ## Start the pageserver. It will notice that the file size doesn't match, and
     ## rename away the local file. It will be re-downloaded when it's needed.
@@ -280,7 +276,7 @@ def test_tenant_redownloads_truncated_file_on_startup(
 
     # the remote side of local_layer_truncated
     remote_layer_path = env.pageserver_remote_storage.remote_layer_path(
-        tenant_id, timeline_id, parse_layer_file_name(path.name).to_str()
+        tenant_id, timeline_id, path.name
     )
 
     # if the upload ever was ongoing, this check would be racy, but at least one

test_runner/regress/test_timeline_detach_ancestor.py

@@ -1,403 +0,0 @@
-import enum
-from concurrent.futures import ThreadPoolExecutor
-from queue import Empty, Queue
-from threading import Barrier
-from typing import List
-
-import pytest
-from fixtures.log_helper import log
-from fixtures.neon_fixtures import (
-    NeonEnvBuilder,
-    wait_for_last_flush_lsn,
-)
-from fixtures.pageserver.http import HistoricLayerInfo
-from fixtures.pageserver.utils import wait_timeline_detail_404
-from fixtures.types import Lsn, TimelineId
-
-
-def by_end_lsn(info: HistoricLayerInfo) -> Lsn:
-    assert info.lsn_end is not None
-    return Lsn(info.lsn_end)
-
-
-def layer_name(info: HistoricLayerInfo) -> str:
-    return info.layer_file_name
-
-
-@enum.unique
-class Branchpoint(str, enum.Enum):
-    """
-    Have branches at these Lsns possibly relative to L0 layer boundary.
-    """
-
-    EARLIER = "earlier"
-    AT_L0 = "at"
-    AFTER_L0 = "after"
-    LAST_RECORD_LSN = "head"
-
-    def __str__(self) -> str:
-        return self.value
-
-    @staticmethod
-    def all() -> List["Branchpoint"]:
-        return [
-            Branchpoint.EARLIER,
-            Branchpoint.AT_L0,
-            Branchpoint.AFTER_L0,
-            Branchpoint.LAST_RECORD_LSN,
-        ]
-
-
-SHUTDOWN_ALLOWED_ERRORS = [
-    ".*initial size calculation failed: downloading failed, possibly for shutdown",
-    ".*failed to freeze and flush: cannot flush frozen layers when flush_loop is not running, state is Exited",
-]
-
-
-@pytest.mark.parametrize("branchpoint", Branchpoint.all())
-@pytest.mark.parametrize("restart_after", [True, False])
-def test_ancestor_detach_branched_from(
-    neon_env_builder: NeonEnvBuilder, branchpoint: Branchpoint, restart_after: bool
-):
-    """
-    Creates a branch relative to L0 lsn boundary according to Branchpoint. Later the timeline is detached.
-    """
-    # TODO: parametrize; currently unimplemented over at pageserver
-    write_to_branch_first = True
-
-    env = neon_env_builder.init_start()
-
-    env.pageserver.allowed_errors.extend(SHUTDOWN_ALLOWED_ERRORS)
-
-    client = env.pageserver.http_client()
-
-    with env.endpoints.create_start("main", tenant_id=env.initial_tenant) as ep:
-        ep.safe_psql("CREATE TABLE foo (i BIGINT);")
-
-        after_first_tx = wait_for_last_flush_lsn(env, ep, env.initial_tenant, env.initial_timeline)
-
-        ep.safe_psql("INSERT INTO foo SELECT i::bigint FROM generate_series(0, 8191) g(i);")
-
-        # create a single layer for us to remote copy
-        wait_for_last_flush_lsn(env, ep, env.initial_tenant, env.initial_timeline)
-        client.timeline_checkpoint(env.initial_tenant, env.initial_timeline)
-
-        ep.safe_psql("INSERT INTO foo SELECT i::bigint FROM generate_series(8192, 16383) g(i);")
-        wait_for_last_flush_lsn(env, ep, env.initial_tenant, env.initial_timeline)
-
-    deltas = client.layer_map_info(env.initial_tenant, env.initial_timeline).delta_layers()
-    # there is also the in-mem layer, but ignore it for now
-    assert len(deltas) == 2, "expecting there to be two deltas: initdb and checkpointed"
-    later_delta = max(deltas, key=by_end_lsn)
-    assert later_delta.lsn_end is not None
-
-    # -1 as the lsn_end is exclusive.
-    last_lsn = Lsn(later_delta.lsn_end).lsn_int - 1
-
-    if branchpoint == Branchpoint.EARLIER:
-        branch_at = after_first_tx
-        rows = 0
-        truncated_layers = 1
-    elif branchpoint == Branchpoint.AT_L0:
-        branch_at = Lsn(last_lsn)
-        rows = 8192
-        truncated_layers = 0
-    elif branchpoint == Branchpoint.AFTER_L0:
-        branch_at = Lsn(last_lsn + 8)
-        rows = 8192
-        # as there is no 8 byte walrecord, nothing should get copied from the straddling layer
-        truncated_layers = 0
-    else:
-        # this case also covers the implicit flush of ancestor as the inmemory hasn't been flushed yet
-        assert branchpoint == Branchpoint.LAST_RECORD_LSN
-        branch_at = None
-        rows = 16384
-        truncated_layers = 0
-
-    name = "new main"
-
-    timeline_id = env.neon_cli.create_branch(
-        name, "main", env.initial_tenant, ancestor_start_lsn=branch_at
-    )
-
-    recorded = Lsn(client.timeline_detail(env.initial_tenant, timeline_id)["ancestor_lsn"])
-    if branch_at is None:
-        # fix it up if we need it later (currently unused)
-        branch_at = recorded
-    else:
-        assert branch_at == recorded, "the test should not use unaligned lsns"
-
-    if write_to_branch_first:
-        with env.endpoints.create_start(name, tenant_id=env.initial_tenant) as ep:
-            assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == rows
-            # make sure the ep is writable
-            # with BEFORE_L0, AFTER_L0 there will be a gap in Lsns caused by accurate end_lsn on straddling layers
-            ep.safe_psql("CREATE TABLE audit AS SELECT 1 as starts;")
-            wait_for_last_flush_lsn(env, ep, env.initial_tenant, timeline_id)
-
-        # branch must have a flush for "PREV_LSN: none"
-        client.timeline_checkpoint(env.initial_tenant, timeline_id)
-        branch_layers = set(
-            map(layer_name, client.layer_map_info(env.initial_tenant, timeline_id).historic_layers)
-        )
-    else:
-        branch_layers = set()
-
-    all_reparented = client.detach_ancestor(env.initial_tenant, timeline_id)
-    assert all_reparented == set()
-
-    if restart_after:
-        env.pageserver.stop()
-        env.pageserver.start()
-
-    with env.endpoints.create_start("main", tenant_id=env.initial_tenant) as ep:
-        assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == 16384
-
-    with env.endpoints.create_start(name, tenant_id=env.initial_tenant) as ep:
-        assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == rows
-
-    old_main_info = client.layer_map_info(env.initial_tenant, env.initial_timeline)
-    old_main = set(map(layer_name, old_main_info.historic_layers))
-
-    new_main_info = client.layer_map_info(env.initial_tenant, timeline_id)
-    new_main = set(map(layer_name, new_main_info.historic_layers))
-
-    new_main_copied_or_truncated = new_main - branch_layers
-    new_main_truncated = new_main_copied_or_truncated - old_main
-
-    assert len(new_main_truncated) == truncated_layers
-    # could additionally check that the symmetric difference has layers starting at the same lsn
-    # but if nothing was copied, then there is no nice rule.
-    # there could be a hole in LSNs between copied from the "old main" and the first branch layer.
-
-    client.timeline_delete(env.initial_tenant, env.initial_timeline)
-    wait_timeline_detail_404(client, env.initial_tenant, env.initial_timeline, 10, 1.0)
-
-
-@pytest.mark.parametrize("restart_after", [True, False])
-def test_ancestor_detach_reparents_earlier(neon_env_builder: NeonEnvBuilder, restart_after: bool):
-    """
-    The case from RFC:
-
-                              +-> another branch with same ancestor_lsn as new main
-                              |
-    old main -------|---------X--------->
-                    |         |         |
-                    |         |         +-> after
-                    |         |
-                    |         +-> new main
-                    |
-                    +-> reparented
-
-    Ends up as:
-
-    old main --------------------------->
-                                        |
-                                        +-> after
-
-                              +-> another branch with same ancestor_lsn as new main
-                              |
-    new main -------|---------|->
-                    |
-                    +-> reparented
-
-    We confirm the end result by being able to delete "old main" after deleting "after".
-    """
-
-    # TODO: support not yet implemented for these
-    write_to_branch_first = True
-
-    env = neon_env_builder.init_start()
-
-    env.pageserver.allowed_errors.extend(SHUTDOWN_ALLOWED_ERRORS)
-
-    client = env.pageserver.http_client()
-
-    with env.endpoints.create_start("main", tenant_id=env.initial_tenant) as ep:
-        ep.safe_psql("CREATE TABLE foo (i BIGINT);")
-        ep.safe_psql("CREATE TABLE audit AS SELECT 1 as starts;")
-
-        branchpoint_pipe = wait_for_last_flush_lsn(
-            env, ep, env.initial_tenant, env.initial_timeline
-        )
-
-        ep.safe_psql("INSERT INTO foo SELECT i::bigint FROM generate_series(0, 8191) g(i);")
-
-        branchpoint_x = wait_for_last_flush_lsn(env, ep, env.initial_tenant, env.initial_timeline)
-        client.timeline_checkpoint(env.initial_tenant, env.initial_timeline)
-
-        ep.safe_psql("INSERT INTO foo SELECT i::bigint FROM generate_series(8192, 16383) g(i);")
-        wait_for_last_flush_lsn(env, ep, env.initial_tenant, env.initial_timeline)
-
-    # as this only gets reparented, we don't need to write to it like new main
-    reparented = env.neon_cli.create_branch(
-        "reparented", "main", env.initial_tenant, ancestor_start_lsn=branchpoint_pipe
-    )
-
-    same_branchpoint = env.neon_cli.create_branch(
-        "same_branchpoint", "main", env.initial_tenant, ancestor_start_lsn=branchpoint_x
-    )
-
-    timeline_id = env.neon_cli.create_branch(
-        "new main", "main", env.initial_tenant, ancestor_start_lsn=branchpoint_x
-    )
-
-    after = env.neon_cli.create_branch("after", "main", env.initial_tenant, ancestor_start_lsn=None)
-
-    if write_to_branch_first:
-        with env.endpoints.create_start("new main", tenant_id=env.initial_tenant) as ep:
-            assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == 8192
-            with ep.cursor() as cur:
-                cur.execute("UPDATE audit SET starts = starts + 1")
-                assert cur.rowcount == 1
-            wait_for_last_flush_lsn(env, ep, env.initial_tenant, timeline_id)
-
-        client.timeline_checkpoint(env.initial_tenant, timeline_id)
-
-    all_reparented = client.detach_ancestor(env.initial_tenant, timeline_id)
-    assert all_reparented == {reparented, same_branchpoint}
-
-    if restart_after:
-        env.pageserver.stop()
-        env.pageserver.start()
-
-    env.pageserver.quiesce_tenants()
-
-    # checking the ancestor after is much faster than waiting for the endpoint not start
-    expected_result = [
-        ("main", env.initial_timeline, None, 16384, 1),
-        ("after", after, env.initial_timeline, 16384, 1),
-        ("new main", timeline_id, None, 8192, 2),
-        ("same_branchpoint", same_branchpoint, timeline_id, 8192, 1),
-        ("reparented", reparented, timeline_id, 0, 1),
-    ]
-
-    for _, timeline_id, expected_ancestor, _, _ in expected_result:
-        details = client.timeline_detail(env.initial_tenant, timeline_id)
-        ancestor_timeline_id = details["ancestor_timeline_id"]
-        if expected_ancestor is None:
-            assert ancestor_timeline_id is None
-        else:
-            assert TimelineId(ancestor_timeline_id) == expected_ancestor
-
-    for name, _, _, rows, starts in expected_result:
-        with env.endpoints.create_start(name, tenant_id=env.initial_tenant) as ep:
-            assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == rows
-            assert ep.safe_psql(f"SELECT count(*) FROM audit WHERE starts = {starts}")[0][0] == 1
-
-    # delete the timelines to confirm detach actually worked
-    client.timeline_delete(env.initial_tenant, after)
-    wait_timeline_detail_404(client, env.initial_tenant, after, 10, 1.0)
-
-    client.timeline_delete(env.initial_tenant, env.initial_timeline)
-    wait_timeline_detail_404(client, env.initial_tenant, env.initial_timeline, 10, 1.0)
-
-
-@pytest.mark.parametrize("restart_after", [True, False])
-def test_detached_receives_flushes_while_being_detached(
-    neon_env_builder: NeonEnvBuilder, restart_after: bool
-):
-    """
-    Makes sure that the timeline is able to receive writes through-out the detach process.
-    """
-    write_to_branch_first = True
-
-    env = neon_env_builder.init_start()
-
-    client = env.pageserver.http_client()
-
-    # row counts have been manually verified to cause reconnections and getpage
-    # requests when restart_after=False with pg16
-    def insert_rows(n: int, ep) -> int:
-        ep.safe_psql(
-            f"INSERT INTO foo SELECT i::bigint, 'more info!! this is a long string' || i FROM generate_series(0, {n - 1}) g(i);"
-        )
-        return n
-
-    with env.endpoints.create_start("main", tenant_id=env.initial_tenant) as ep:
-        ep.safe_psql("CREATE EXTENSION neon_test_utils;")
-        ep.safe_psql("CREATE TABLE foo (i BIGINT, aux TEXT NOT NULL);")
-
-        rows = insert_rows(256, ep)
-
-        branchpoint = wait_for_last_flush_lsn(env, ep, env.initial_tenant, env.initial_timeline)
-
-    timeline_id = env.neon_cli.create_branch(
-        "new main", "main", tenant_id=env.initial_tenant, ancestor_start_lsn=branchpoint
-    )
-
-    log.info("starting the new main endpoint")
-    ep = env.endpoints.create_start("new main", tenant_id=env.initial_tenant)
-    assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == rows
-
-    if write_to_branch_first:
-        rows += insert_rows(256, ep)
-        wait_for_last_flush_lsn(env, ep, env.initial_tenant, timeline_id)
-        client.timeline_checkpoint(env.initial_tenant, timeline_id)
-        log.info("completed {write_to_branch_first=}")
-
-    def small_txs(ep, queue: Queue[str], barrier):
-        extra_rows = 0
-
-        with ep.connect() as conn:
-            while True:
-                try:
-                    queue.get_nowait()
-                    break
-                except Empty:
-                    pass
-
-                if barrier is not None:
-                    barrier.wait()
-                    barrier = None
-
-                cursor = conn.cursor()
-                cursor.execute(
-                    "INSERT INTO foo(i, aux) VALUES (1, 'more info!! this is a long string' || 1);"
-                )
-                extra_rows += 1
-        return extra_rows
-
-    with ThreadPoolExecutor(max_workers=1) as exec:
-        queue: Queue[str] = Queue()
-        barrier = Barrier(2)
-
-        completion = exec.submit(small_txs, ep, queue, barrier)
-        barrier.wait()
-
-        reparented = client.detach_ancestor(env.initial_tenant, timeline_id)
-        assert len(reparented) == 0
-
-        if restart_after:
-            # ep and row production is kept alive on purpose
-            env.pageserver.stop()
-            env.pageserver.start()
-
-        env.pageserver.quiesce_tenants()
-
-        queue.put("done")
-        extra_rows = completion.result()
-        assert extra_rows > 0, "some rows should had been written"
-        rows += extra_rows
-
-    assert client.timeline_detail(env.initial_tenant, timeline_id)["ancestor_timeline_id"] is None
-
-    assert ep.safe_psql("SELECT clear_buffer_cache();")
-    assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == rows
-    assert ep.safe_psql("SELECT SUM(LENGTH(aux)) FROM foo")[0][0] != 0
-    ep.stop()
-
-    # finally restart the endpoint and make sure we still have the same answer
-    with env.endpoints.create_start("new main", tenant_id=env.initial_tenant) as ep:
-        assert ep.safe_psql("SELECT count(*) FROM foo;")[0][0] == rows
-
-    env.pageserver.allowed_errors.extend(SHUTDOWN_ALLOWED_ERRORS)
-
-
-# TODO:
-# - after starting the operation, tenant is deleted
-# - after starting the operation, pageserver is shutdown, restarted
-# - after starting the operation, bottom-most timeline is deleted, pageserver is restarted, gc is inhibited
-# - deletion of reparented while reparenting should fail once, then succeed (?)
-# - branch near existing L1 boundary, image layers?
-# - investigate: why are layers started at uneven lsn? not just after branching, but in general.