permit for unauthenticated connection attempts

Cargo.lock

@@ -233,7 +233,7 @@ checksum = "16e62a023e7c117e27523144c5d2459f4397fcc3cab0085af8e2224f643a0193"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -244,7 +244,7 @@ checksum = "b9ccdd8f2a161be9bd5c023df56f1b2a0bd1d83872ae53b71a84a12c9bf6e842"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -881,7 +881,7 @@ dependencies = [
  "regex",
  "rustc-hash",
  "shlex",
- "syn 2.0.32",
+ "syn 2.0.28",
  "which",
 ]
 
@@ -1095,7 +1095,7 @@ dependencies = [
  "heck",
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -1245,19 +1245,16 @@ name = "control_plane"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "async-trait",
  "camino",
  "clap",
  "comfy-table",
  "compute_api",
- "futures",
  "git-version",
  "hex",
  "hyper",
  "nix 0.26.2",
  "once_cell",
  "pageserver_api",
- "pageserver_client",
  "postgres",
  "postgres_backend",
  "postgres_connection",
@@ -1271,8 +1268,6 @@ dependencies = [
  "tar",
  "thiserror",
  "tokio",
- "tokio-postgres",
- "tokio-util",
  "toml",
  "tracing",
  "url",
@@ -1486,7 +1481,7 @@ dependencies = [
  "proc-macro2",
  "quote",
  "strsim",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -1497,7 +1492,7 @@ checksum = "29a358ff9f12ec09c3e61fef9b5a9902623a695a46a917b07f269bff1445611a"
 dependencies = [
  "darling_core",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -1572,7 +1567,7 @@ checksum = "487585f4d0c6655fe74905e2504d8ad6908e4db67f744eb140876906c2f3175d"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -1666,7 +1661,7 @@ dependencies = [
  "darling",
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -1920,7 +1915,7 @@ checksum = "89ca545a94061b6365f2c7355b4b32bd20df3ff95f02da9329b34ccc3bd6ee72"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -2906,7 +2901,7 @@ checksum = "a948666b637a0f465e8564c73e89d4dde00d72d4d473cc972f390fc3dcee7d9c"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -3108,7 +3103,6 @@ dependencies = [
  "humantime-serde",
  "hyper",
  "itertools",
- "md5",
  "metrics",
  "nix 0.26.2",
  "num-traits",
@@ -3167,7 +3161,6 @@ dependencies = [
  "enum-map",
  "hex",
  "postgres_ffi",
- "rand 0.8.5",
  "serde",
  "serde_json",
  "serde_with",
@@ -3178,19 +3171,6 @@ dependencies = [
  "workspace_hack",
 ]
 
-[[package]]
-name = "pageserver_client"
-version = "0.1.0"
-dependencies = [
- "async-trait",
- "pageserver_api",
- "reqwest",
- "serde",
- "thiserror",
- "utils",
- "workspace_hack",
-]
-
 [[package]]
 name = "parking"
 version = "2.1.1"
@@ -3350,7 +3330,7 @@ checksum = "39407670928234ebc5e6e580247dd567ad73a3578460c5990f9503df207e8f07"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -3557,7 +3537,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3b69d39aab54d069e7f2fe8cb970493e7834601ca2d8c65fd7bbd183578080d1"
 dependencies = [
  "proc-macro2",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -4165,7 +4145,7 @@ dependencies = [
  "regex",
  "relative-path",
  "rustc_version",
- "syn 2.0.32",
+ "syn 2.0.28",
  "unicode-ident",
 ]
 
@@ -4311,7 +4291,6 @@ dependencies = [
  "histogram",
  "itertools",
  "pageserver",
- "pageserver_api",
  "rand 0.8.5",
  "remote_storage",
  "reqwest",
@@ -4600,7 +4579,7 @@ checksum = "aafe972d60b0b9bee71a91b92fee2d4fb3c9d7e8f6b179aa99f27203d99a4816"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -4681,7 +4660,7 @@ dependencies = [
  "darling",
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -4948,9 +4927,9 @@ dependencies = [
 
 [[package]]
 name = "syn"
-version = "2.0.32"
+version = "2.0.28"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "239814284fd6f1a4ffe4ca893952cdd93c224b6a1571c9a9eadd670295c0c9e2"
+checksum = "04361975b3f5e348b2189d8dc55bc942f278b2d482a6a0365de5bdd62d351567"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -5080,7 +5059,7 @@ checksum = "f9456a42c5b0d803c8cd86e73dd7cc9edd429499f37a3550d286d5e86720569f"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -5198,7 +5177,7 @@ checksum = "5b8a1e28f2deaa14e508979454cb3a223b10b938b45af148bc0986de36f1923b"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -5499,7 +5478,7 @@ checksum = "0f57e3ca2a01450b1a921183a9c9cbfda207fd822cef4ccb00a65402cbba7a74"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]
@@ -5785,7 +5764,6 @@ dependencies = [
  "serde",
  "serde_assert",
  "serde_json",
- "serde_path_to_error",
  "serde_with",
  "signal-hook",
  "strum",
@@ -5944,7 +5922,7 @@ dependencies = [
  "once_cell",
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
  "wasm-bindgen-shared",
 ]
 
@@ -5978,7 +5956,7 @@ checksum = "e128beba882dd1eb6200e1dc92ae6c5dbaa4311aa7bb211ca035779e5efc39f8"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
  "wasm-bindgen-backend",
  "wasm-bindgen-shared",
 ]
@@ -6315,7 +6293,7 @@ dependencies = [
  "smallvec",
  "subtle",
  "syn 1.0.109",
- "syn 2.0.32",
+ "syn 2.0.28",
  "time",
  "time-macros",
  "tokio",
@@ -6377,22 +6355,22 @@ dependencies = [
 
 [[package]]
 name = "zerocopy"
-version = "0.7.31"
+version = "0.7.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1c4061bedbb353041c12f413700357bec76df2c7e2ca8e4df8bac24c6bf68e3d"
+checksum = "7a7af71d8643341260a65f89fa60c0eeaa907f34544d8f6d9b0df72f069b5e74"
 dependencies = [
  "zerocopy-derive",
 ]
 
 [[package]]
 name = "zerocopy-derive"
-version = "0.7.31"
+version = "0.7.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b3c129550b3e6de3fd0ba67ba5c81818f9805e58b8d7fee80a3a59d2c9fc601a"
+checksum = "9731702e2f0617ad526794ae28fbc6f6ca8849b5ba729666c2a5bc4b6ddee2cd"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 2.0.32",
+ "syn 2.0.28",
 ]
 
 [[package]]

Cargo.toml

@@ -5,7 +5,6 @@ members = [
     "control_plane",
     "pageserver",
     "pageserver/ctl",
-    "pageserver/client",
     "proxy",
     "safekeeper",
     "storage_broker",
@@ -183,7 +182,6 @@ compute_api = { version = "0.1", path = "./libs/compute_api/" }
 consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
 metrics = { version = "0.1", path = "./libs/metrics/" }
 pageserver_api = { version = "0.1", path = "./libs/pageserver_api/" }
-pageserver_client = { path = "./pageserver/client" }
 postgres_backend = { version = "0.1", path = "./libs/postgres_backend/" }
 postgres_connection = { version = "0.1", path = "./libs/postgres_connection/" }
 postgres_ffi = { version = "0.1", path = "./libs/postgres_ffi/" }

compute_tools/src/spec.rs

@@ -298,7 +298,7 @@ pub fn handle_roles(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
                 // safe to add more permissions here. BYPASSRLS and REPLICATION are inherited
                 // from neon_superuser.
                 let mut query: String = format!(
-                    "CREATE ROLE {} INHERIT CREATEROLE CREATEDB BYPASSRLS REPLICATION IN ROLE neon_superuser",
+                    "CREATE ROLE {} INHERIT CREATEROLE CREATEDB IN ROLE neon_superuser",
                     name.pg_quote()
                 );
                 info!("role create query: '{}'", &query);

control_plane/Cargo.toml

@@ -6,11 +6,9 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
-async-trait.workspace = true
 camino.workspace = true
 clap.workspace = true
 comfy-table.workspace = true
-futures.workspace = true
 git-version.workspace = true
 nix.workspace = true
 once_cell.workspace = true
@@ -26,11 +24,10 @@ tar.workspace = true
 thiserror.workspace = true
 toml.workspace = true
 tokio.workspace = true
-tokio-postgres.workspace = true
-tokio-util.workspace = true
 url.workspace = true
+# Note: Do not directly depend on pageserver or safekeeper; use pageserver_api or safekeeper_api
+# instead, so that recompile times are better.
 pageserver_api.workspace = true
-pageserver_client.workspace = true
 postgres_backend.workspace = true
 safekeeper_api.workspace = true
 postgres_connection.workspace = true

control_plane/src/attachment_service.rs

@@ -9,7 +9,7 @@ pub struct AttachmentService {
     env: LocalEnv,
     listen: String,
     path: PathBuf,
-    client: reqwest::Client,
+    client: reqwest::blocking::Client,
 }
 
 const COMMAND: &str = "attachment_service";
@@ -53,7 +53,7 @@ impl AttachmentService {
             env: env.clone(),
             path,
             listen,
-            client: reqwest::ClientBuilder::new()
+            client: reqwest::blocking::ClientBuilder::new()
                 .build()
                 .expect("Failed to construct http client"),
         }
@@ -64,7 +64,7 @@ impl AttachmentService {
             .expect("non-Unicode path")
     }
 
-    pub async fn start(&self) -> anyhow::Result<Child> {
+    pub fn start(&self) -> anyhow::Result<Child> {
         let path_str = self.path.to_string_lossy();
 
         background_process::start_process(
@@ -73,11 +73,10 @@ impl AttachmentService {
             &self.env.attachment_service_bin(),
             ["-l", &self.listen, "-p", &path_str],
             [],
-            background_process::InitialPidFile::Create(self.pid_file()),
+            background_process::InitialPidFile::Create(&self.pid_file()),
             // TODO: a real status check
-            || async move { anyhow::Ok(true) },
+            || Ok(true),
         )
-        .await
     }
 
     pub fn stop(&self, immediate: bool) -> anyhow::Result<()> {
@@ -85,7 +84,7 @@ impl AttachmentService {
     }
 
     /// Call into the attach_hook API, for use before handing out attachments to pageservers
-    pub async fn attach_hook(
+    pub fn attach_hook(
         &self,
         tenant_id: TenantId,
         pageserver_id: NodeId,
@@ -105,16 +104,16 @@ impl AttachmentService {
             node_id: Some(pageserver_id),
         };
 
-        let response = self.client.post(url).json(&request).send().await?;
+        let response = self.client.post(url).json(&request).send()?;
         if response.status() != StatusCode::OK {
             return Err(anyhow!("Unexpected status {}", response.status()));
         }
 
-        let response = response.json::<AttachHookResponse>().await?;
+        let response = response.json::<AttachHookResponse>()?;
         Ok(response.gen)
     }
 
-    pub async fn inspect(&self, tenant_id: TenantId) -> anyhow::Result<Option<(u32, NodeId)>> {
+    pub fn inspect(&self, tenant_id: TenantId) -> anyhow::Result<Option<(u32, NodeId)>> {
         use hyper::StatusCode;
 
         let url = self
@@ -127,12 +126,12 @@ impl AttachmentService {
 
         let request = InspectRequest { tenant_id };
 
-        let response = self.client.post(url).json(&request).send().await?;
+        let response = self.client.post(url).json(&request).send()?;
         if response.status() != StatusCode::OK {
             return Err(anyhow!("Unexpected status {}", response.status()));
         }
 
-        let response = response.json::<InspectResponse>().await?;
+        let response = response.json::<InspectResponse>()?;
         Ok(response.attachment)
     }
 }

control_plane/src/background_process.rs

@@ -44,15 +44,15 @@ const NOTICE_AFTER_RETRIES: u64 = 50;
 
 /// Argument to `start_process`, to indicate whether it should create pidfile or if the process creates
 /// it itself.
-pub enum InitialPidFile {
+pub enum InitialPidFile<'t> {
     /// Create a pidfile, to allow future CLI invocations to manipulate the process.
-    Create(Utf8PathBuf),
+    Create(&'t Utf8Path),
     /// The process will create the pidfile itself, need to wait for that event.
-    Expect(Utf8PathBuf),
+    Expect(&'t Utf8Path),
 }
 
 /// Start a background child process using the parameters given.
-pub async fn start_process<F, Fut, AI, A, EI>(
+pub fn start_process<F, AI, A, EI>(
     process_name: &str,
     datadir: &Path,
     command: &Path,
@@ -62,8 +62,7 @@ pub async fn start_process<F, Fut, AI, A, EI>(
     process_status_check: F,
 ) -> anyhow::Result<Child>
 where
-    F: Fn() -> Fut,
-    Fut: std::future::Future<Output = anyhow::Result<bool>>,
+    F: Fn() -> anyhow::Result<bool>,
     AI: IntoIterator<Item = A>,
     A: AsRef<OsStr>,
     // Not generic AsRef<OsStr>, otherwise empty `envs` prevents type inference
@@ -90,7 +89,7 @@ where
     let filled_cmd = fill_remote_storage_secrets_vars(fill_rust_env_vars(background_command));
     filled_cmd.envs(envs);
 
-    let pid_file_to_check = match &initial_pid_file {
+    let pid_file_to_check = match initial_pid_file {
         InitialPidFile::Create(path) => {
             pre_exec_create_pidfile(filled_cmd, path);
             path
@@ -108,7 +107,7 @@ where
     );
 
     for retries in 0..RETRIES {
-        match process_started(pid, pid_file_to_check, &process_status_check).await {
+        match process_started(pid, Some(pid_file_to_check), &process_status_check) {
             Ok(true) => {
                 println!("\n{process_name} started, pid: {pid}");
                 return Ok(spawned_process);
@@ -317,20 +316,22 @@ where
     cmd
 }
 
-async fn process_started<F, Fut>(
+fn process_started<F>(
     pid: Pid,
-    pid_file_to_check: &Utf8Path,
+    pid_file_to_check: Option<&Utf8Path>,
     status_check: &F,
 ) -> anyhow::Result<bool>
 where
-    F: Fn() -> Fut,
-    Fut: std::future::Future<Output = anyhow::Result<bool>>,
+    F: Fn() -> anyhow::Result<bool>,
 {
-    match status_check().await {
-        Ok(true) => match pid_file::read(pid_file_to_check)? {
-            PidFileRead::NotExist => Ok(false),
-            PidFileRead::LockedByOtherProcess(pid_in_file) => Ok(pid_in_file == pid),
-            PidFileRead::NotHeldByAnyProcess(_) => Ok(false),
+    match status_check() {
+        Ok(true) => match pid_file_to_check {
+            Some(pid_file_path) => match pid_file::read(pid_file_path)? {
+                PidFileRead::NotExist => Ok(false),
+                PidFileRead::LockedByOtherProcess(pid_in_file) => Ok(pid_in_file == pid),
+                PidFileRead::NotHeldByAnyProcess(_) => Ok(false),
+            },
+            None => Ok(true),
         },
         Ok(false) => Ok(false),
         Err(e) => anyhow::bail!("process failed to start: {e}"),

control_plane/src/bin/neon_local.rs

@@ -120,20 +120,15 @@ fn main() -> Result<()> {
         let mut env = LocalEnv::load_config().context("Error loading config")?;
         let original_env = env.clone();
 
-        let rt = tokio::runtime::Builder::new_current_thread()
-            .enable_all()
-            .build()
-            .unwrap();
-
         let subcommand_result = match sub_name {
-            "tenant" => rt.block_on(handle_tenant(sub_args, &mut env)),
-            "timeline" => rt.block_on(handle_timeline(sub_args, &mut env)),
-            "start" => rt.block_on(handle_start_all(sub_args, &env)),
+            "tenant" => handle_tenant(sub_args, &mut env),
+            "timeline" => handle_timeline(sub_args, &mut env),
+            "start" => handle_start_all(sub_args, &env),
             "stop" => handle_stop_all(sub_args, &env),
-            "pageserver" => rt.block_on(handle_pageserver(sub_args, &env)),
-            "attachment_service" => rt.block_on(handle_attachment_service(sub_args, &env)),
-            "safekeeper" => rt.block_on(handle_safekeeper(sub_args, &env)),
-            "endpoint" => rt.block_on(handle_endpoint(sub_args, &env)),
+            "pageserver" => handle_pageserver(sub_args, &env),
+            "attachment_service" => handle_attachment_service(sub_args, &env),
+            "safekeeper" => handle_safekeeper(sub_args, &env),
+            "endpoint" => handle_endpoint(sub_args, &env),
             "mappings" => handle_mappings(sub_args, &mut env),
             "pg" => bail!("'pg' subcommand has been renamed to 'endpoint'"),
             _ => bail!("unexpected subcommand {sub_name}"),
@@ -274,13 +269,12 @@ fn print_timeline(
 
 /// Returns a map of timeline IDs to timeline_id@lsn strings.
 /// Connects to the pageserver to query this information.
-async fn get_timeline_infos(
+fn get_timeline_infos(
     env: &local_env::LocalEnv,
     tenant_id: &TenantId,
 ) -> Result<HashMap<TimelineId, TimelineInfo>> {
     Ok(get_default_pageserver(env)
-        .timeline_list(tenant_id)
-        .await?
+        .timeline_list(tenant_id)?
         .into_iter()
         .map(|timeline_info| (timeline_info.timeline_id, timeline_info))
         .collect())
@@ -379,14 +373,11 @@ fn pageserver_config_overrides(init_match: &ArgMatches) -> Vec<&str> {
         .collect()
 }
 
-async fn handle_tenant(
-    tenant_match: &ArgMatches,
-    env: &mut local_env::LocalEnv,
-) -> anyhow::Result<()> {
+fn handle_tenant(tenant_match: &ArgMatches, env: &mut local_env::LocalEnv) -> anyhow::Result<()> {
     let pageserver = get_default_pageserver(env);
     match tenant_match.subcommand() {
         Some(("list", _)) => {
-            for t in pageserver.tenant_list().await? {
+            for t in pageserver.tenant_list()? {
                 println!("{} {:?}", t.id, t.state);
             }
         }
@@ -403,16 +394,12 @@ async fn handle_tenant(
                 // We must register the tenant with the attachment service, so
                 // that when the pageserver restarts, it will be re-attached.
                 let attachment_service = AttachmentService::from_env(env);
-                attachment_service
-                    .attach_hook(tenant_id, pageserver.conf.id)
-                    .await?
+                attachment_service.attach_hook(tenant_id, pageserver.conf.id)?
             } else {
                 None
             };
 
-            pageserver
-                .tenant_create(tenant_id, generation, tenant_conf)
-                .await?;
+            pageserver.tenant_create(tenant_id, generation, tenant_conf)?;
             println!("tenant {tenant_id} successfully created on the pageserver");
 
             // Create an initial timeline for the new tenant
@@ -422,16 +409,14 @@ async fn handle_tenant(
                 .copied()
                 .context("Failed to parse postgres version from the argument string")?;
 
-            let timeline_info = pageserver
-                .timeline_create(
-                    tenant_id,
-                    new_timeline_id,
-                    None,
-                    None,
-                    Some(pg_version),
-                    None,
-                )
-                .await?;
+            let timeline_info = pageserver.timeline_create(
+                tenant_id,
+                new_timeline_id,
+                None,
+                None,
+                Some(pg_version),
+                None,
+            )?;
             let new_timeline_id = timeline_info.timeline_id;
             let last_record_lsn = timeline_info.last_record_lsn;
 
@@ -465,7 +450,6 @@ async fn handle_tenant(
 
             pageserver
                 .tenant_config(tenant_id, tenant_conf)
-                .await
                 .with_context(|| format!("Tenant config failed for tenant with id {tenant_id}"))?;
             println!("tenant {tenant_id} successfully configured on the pageserver");
         }
@@ -474,7 +458,7 @@ async fn handle_tenant(
             let new_pageserver = get_pageserver(env, matches)?;
             let new_pageserver_id = new_pageserver.conf.id;
 
-            migrate_tenant(env, tenant_id, new_pageserver).await?;
+            migrate_tenant(env, tenant_id, new_pageserver)?;
             println!("tenant {tenant_id} migrated to {}", new_pageserver_id);
         }
 
@@ -484,13 +468,13 @@ async fn handle_tenant(
     Ok(())
 }
 
-async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
+fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
     let pageserver = get_default_pageserver(env);
 
     match timeline_match.subcommand() {
         Some(("list", list_match)) => {
             let tenant_id = get_tenant_id(list_match, env)?;
-            let timelines = pageserver.timeline_list(&tenant_id).await?;
+            let timelines = pageserver.timeline_list(&tenant_id)?;
             print_timelines_tree(timelines, env.timeline_name_mappings())?;
         }
         Some(("create", create_match)) => {
@@ -506,16 +490,14 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
 
             let new_timeline_id_opt = parse_timeline_id(create_match)?;
 
-            let timeline_info = pageserver
-                .timeline_create(
-                    tenant_id,
-                    new_timeline_id_opt,
-                    None,
-                    None,
-                    Some(pg_version),
-                    None,
-                )
-                .await?;
+            let timeline_info = pageserver.timeline_create(
+                tenant_id,
+                new_timeline_id_opt,
+                None,
+                None,
+                Some(pg_version),
+                None,
+            )?;
             let new_timeline_id = timeline_info.timeline_id;
 
             let last_record_lsn = timeline_info.last_record_lsn;
@@ -560,9 +542,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
 
             let mut cplane = ComputeControlPlane::load(env.clone())?;
             println!("Importing timeline into pageserver ...");
-            pageserver
-                .timeline_import(tenant_id, timeline_id, base, pg_wal, pg_version)
-                .await?;
+            pageserver.timeline_import(tenant_id, timeline_id, base, pg_wal, pg_version)?;
             env.register_branch_mapping(name.to_string(), tenant_id, timeline_id)?;
 
             println!("Creating endpoint for imported timeline ...");
@@ -598,16 +578,14 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
                 .map(|lsn_str| Lsn::from_str(lsn_str))
                 .transpose()
                 .context("Failed to parse ancestor start Lsn from the request")?;
-            let timeline_info = pageserver
-                .timeline_create(
-                    tenant_id,
-                    None,
-                    start_lsn,
-                    Some(ancestor_timeline_id),
-                    None,
-                    None,
-                )
-                .await?;
+            let timeline_info = pageserver.timeline_create(
+                tenant_id,
+                None,
+                start_lsn,
+                Some(ancestor_timeline_id),
+                None,
+                None,
+            )?;
             let new_timeline_id = timeline_info.timeline_id;
 
             let last_record_lsn = timeline_info.last_record_lsn;
@@ -626,7 +604,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
     Ok(())
 }
 
-async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     let (sub_name, sub_args) = match ep_match.subcommand() {
         Some(ep_subcommand_data) => ep_subcommand_data,
         None => bail!("no endpoint subcommand provided"),
@@ -636,12 +614,10 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
     match sub_name {
         "list" => {
             let tenant_id = get_tenant_id(sub_args, env)?;
-            let timeline_infos = get_timeline_infos(env, &tenant_id)
-                .await
-                .unwrap_or_else(|e| {
-                    eprintln!("Failed to load timeline info: {}", e);
-                    HashMap::new()
-                });
+            let timeline_infos = get_timeline_infos(env, &tenant_id).unwrap_or_else(|e| {
+                eprintln!("Failed to load timeline info: {}", e);
+                HashMap::new()
+            });
 
             let timeline_name_mappings = env.timeline_name_mappings();
 
@@ -815,9 +791,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
             };
 
             println!("Starting existing endpoint {endpoint_id}...");
-            endpoint
-                .start(&auth_token, safekeepers, remote_ext_config)
-                .await?;
+            endpoint.start(&auth_token, safekeepers, remote_ext_config)?;
         }
         "reconfigure" => {
             let endpoint_id = sub_args
@@ -835,7 +809,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                 } else {
                     None
                 };
-            endpoint.reconfigure(pageserver_id).await?;
+            endpoint.reconfigure(pageserver_id)?;
         }
         "stop" => {
             let endpoint_id = sub_args
@@ -901,12 +875,11 @@ fn get_pageserver(env: &local_env::LocalEnv, args: &ArgMatches) -> Result<PageSe
     ))
 }
 
-async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     match sub_match.subcommand() {
         Some(("start", subcommand_args)) => {
             if let Err(e) = get_pageserver(env, subcommand_args)?
                 .start(&pageserver_config_overrides(subcommand_args))
-                .await
             {
                 eprintln!("pageserver start failed: {e}");
                 exit(1);
@@ -933,10 +906,7 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
                 exit(1);
             }
 
-            if let Err(e) = pageserver
-                .start(&pageserver_config_overrides(subcommand_args))
-                .await
-            {
+            if let Err(e) = pageserver.start(&pageserver_config_overrides(subcommand_args)) {
                 eprintln!("pageserver start failed: {e}");
                 exit(1);
             }
@@ -950,17 +920,14 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
                 exit(1);
             }
 
-            if let Err(e) = pageserver
-                .start(&pageserver_config_overrides(subcommand_args))
-                .await
-            {
+            if let Err(e) = pageserver.start(&pageserver_config_overrides(subcommand_args)) {
                 eprintln!("pageserver start failed: {e}");
                 exit(1);
             }
         }
 
         Some(("status", subcommand_args)) => {
-            match get_pageserver(env, subcommand_args)?.check_status().await {
+            match get_pageserver(env, subcommand_args)?.check_status() {
                 Ok(_) => println!("Page server is up and running"),
                 Err(err) => {
                     eprintln!("Page server is not available: {}", err);
@@ -975,14 +942,11 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
     Ok(())
 }
 
-async fn handle_attachment_service(
-    sub_match: &ArgMatches,
-    env: &local_env::LocalEnv,
-) -> Result<()> {
+fn handle_attachment_service(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     let svc = AttachmentService::from_env(env);
     match sub_match.subcommand() {
         Some(("start", _start_match)) => {
-            if let Err(e) = svc.start().await {
+            if let Err(e) = svc.start() {
                 eprintln!("start failed: {e}");
                 exit(1);
             }
@@ -1023,7 +987,7 @@ fn safekeeper_extra_opts(init_match: &ArgMatches) -> Vec<String> {
         .collect()
 }
 
-async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     let (sub_name, sub_args) = match sub_match.subcommand() {
         Some(safekeeper_command_data) => safekeeper_command_data,
         None => bail!("no safekeeper subcommand provided"),
@@ -1041,7 +1005,7 @@ async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
         "start" => {
             let extra_opts = safekeeper_extra_opts(sub_args);
 
-            if let Err(e) = safekeeper.start(extra_opts).await {
+            if let Err(e) = safekeeper.start(extra_opts) {
                 eprintln!("safekeeper start failed: {}", e);
                 exit(1);
             }
@@ -1067,7 +1031,7 @@ async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
             }
 
             let extra_opts = safekeeper_extra_opts(sub_args);
-            if let Err(e) = safekeeper.start(extra_opts).await {
+            if let Err(e) = safekeeper.start(extra_opts) {
                 eprintln!("safekeeper start failed: {}", e);
                 exit(1);
             }
@@ -1080,15 +1044,15 @@ async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
     Ok(())
 }
 
-async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> anyhow::Result<()> {
+fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> anyhow::Result<()> {
     // Endpoints are not started automatically
 
-    broker::start_broker_process(env).await?;
+    broker::start_broker_process(env)?;
 
     // Only start the attachment service if the pageserver is configured to need it
     if env.control_plane_api.is_some() {
         let attachment_service = AttachmentService::from_env(env);
-        if let Err(e) = attachment_service.start().await {
+        if let Err(e) = attachment_service.start() {
             eprintln!("attachment_service start failed: {:#}", e);
             try_stop_all(env, true);
             exit(1);
@@ -1097,10 +1061,7 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
 
     for ps_conf in &env.pageservers {
         let pageserver = PageServerNode::from_env(env, ps_conf);
-        if let Err(e) = pageserver
-            .start(&pageserver_config_overrides(sub_match))
-            .await
-        {
+        if let Err(e) = pageserver.start(&pageserver_config_overrides(sub_match)) {
             eprintln!("pageserver {} start failed: {:#}", ps_conf.id, e);
             try_stop_all(env, true);
             exit(1);
@@ -1109,7 +1070,7 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
 
     for node in env.safekeepers.iter() {
         let safekeeper = SafekeeperNode::from_env(env, node);
-        if let Err(e) = safekeeper.start(vec![]).await {
+        if let Err(e) = safekeeper.start(vec![]) {
             eprintln!("safekeeper {} start failed: {:#}", safekeeper.id, e);
             try_stop_all(env, false);
             exit(1);

control_plane/src/broker.rs

@@ -11,7 +11,7 @@ use camino::Utf8PathBuf;
 
 use crate::{background_process, local_env};
 
-pub async fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
+pub fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
     let broker = &env.broker;
     let listen_addr = &broker.listen_addr;
 
@@ -19,15 +19,15 @@ pub async fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<(
 
     let args = [format!("--listen-addr={listen_addr}")];
 
-    let client = reqwest::Client::new();
+    let client = reqwest::blocking::Client::new();
     background_process::start_process(
         "storage_broker",
         &env.base_data_dir,
         &env.storage_broker_bin(),
         args,
         [],
-        background_process::InitialPidFile::Create(storage_broker_pid_file_path(env)),
-        || async {
+        background_process::InitialPidFile::Create(&storage_broker_pid_file_path(env)),
+        || {
             let url = broker.client_url();
             let status_url = url.join("status").with_context(|| {
                 format!("Failed to append /status path to broker endpoint {url}")
@@ -36,13 +36,12 @@ pub async fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<(
                 .get(status_url)
                 .build()
                 .with_context(|| format!("Failed to construct request to broker endpoint {url}"))?;
-            match client.execute(request).await {
+            match client.execute(request) {
                 Ok(resp) => Ok(resp.status().is_success()),
                 Err(_) => Ok(false),
             }
         },
     )
-    .await
     .context("Failed to spawn storage_broker subprocess")?;
     Ok(())
 }

control_plane/src/endpoint.rs

@@ -464,7 +464,7 @@ impl Endpoint {
         }
     }
 
-    pub async fn start(
+    pub fn start(
         &self,
         auth_token: &Option<String>,
         safekeepers: Vec<NodeId>,
@@ -587,7 +587,7 @@ impl Endpoint {
         const MAX_ATTEMPTS: u32 = 10 * 30; // Wait up to 30 s
         loop {
             attempt += 1;
-            match self.get_status().await {
+            match self.get_status() {
                 Ok(state) => {
                     match state.status {
                         ComputeStatus::Init => {
@@ -629,8 +629,8 @@ impl Endpoint {
     }
 
     // Call the /status HTTP API
-    pub async fn get_status(&self) -> Result<ComputeState> {
-        let client = reqwest::Client::new();
+    pub fn get_status(&self) -> Result<ComputeState> {
+        let client = reqwest::blocking::Client::new();
 
         let response = client
             .request(
@@ -641,17 +641,16 @@ impl Endpoint {
                     self.http_address.port()
                 ),
             )
-            .send()
-            .await?;
+            .send()?;
 
         // Interpret the response
         let status = response.status();
         if !(status.is_client_error() || status.is_server_error()) {
-            Ok(response.json().await?)
+            Ok(response.json()?)
         } else {
             // reqwest does not export its error construction utility functions, so let's craft the message ourselves
             let url = response.url().to_owned();
-            let msg = match response.text().await {
+            let msg = match response.text() {
                 Ok(err_body) => format!("Error: {}", err_body),
                 Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
             };
@@ -659,7 +658,7 @@ impl Endpoint {
         }
     }
 
-    pub async fn reconfigure(&self, pageserver_id: Option<NodeId>) -> Result<()> {
+    pub fn reconfigure(&self, pageserver_id: Option<NodeId>) -> Result<()> {
         let mut spec: ComputeSpec = {
             let spec_path = self.endpoint_path().join("spec.json");
             let file = std::fs::File::open(spec_path)?;
@@ -688,7 +687,7 @@ impl Endpoint {
             spec.pageserver_connstring = Some(format!("postgresql://no_user@{host}:{port}"));
         }
 
-        let client = reqwest::Client::new();
+        let client = reqwest::blocking::Client::new();
         let response = client
             .post(format!(
                 "http://{}:{}/configure",
@@ -699,15 +698,14 @@ impl Endpoint {
                 "{{\"spec\":{}}}",
                 serde_json::to_string_pretty(&spec)?
             ))
-            .send()
-            .await?;
+            .send()?;
 
         let status = response.status();
         if !(status.is_client_error() || status.is_server_error()) {
             Ok(())
         } else {
             let url = response.url().to_owned();
-            let msg = match response.text().await {
+            let msg = match response.text() {
                 Ok(err_body) => format!("Error: {}", err_body),
                 Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
             };

control_plane/src/pageserver.rs

@@ -6,24 +6,28 @@
 //!
 use std::borrow::Cow;
 use std::collections::HashMap;
-
-use std::io;
-use std::io::Write;
+use std::fs::File;
+use std::io::{BufReader, Write};
 use std::num::NonZeroU64;
 use std::path::PathBuf;
 use std::process::{Child, Command};
 use std::time::Duration;
+use std::{io, result};
 
 use anyhow::{bail, Context};
 use camino::Utf8PathBuf;
-use futures::SinkExt;
-use pageserver_api::models::{self, LocationConfig, TenantInfo, TimelineInfo};
+use pageserver_api::models::{
+    self, LocationConfig, TenantInfo, TenantLocationConfigRequest, TimelineInfo,
+};
 use pageserver_api::shard::TenantShardId;
-use pageserver_client::mgmt_api;
 use postgres_backend::AuthType;
 use postgres_connection::{parse_host_port, PgConnectionConfig};
+use reqwest::blocking::{Client, RequestBuilder, Response};
+use reqwest::{IntoUrl, Method};
+use thiserror::Error;
 use utils::auth::{Claims, Scope};
 use utils::{
+    http::error::HttpErrorBody,
     id::{TenantId, TimelineId},
     lsn::Lsn,
 };
@@ -34,6 +38,45 @@ use crate::{background_process, local_env::LocalEnv};
 /// Directory within .neon which will be used by default for LocalFs remote storage.
 pub const PAGESERVER_REMOTE_STORAGE_DIR: &str = "local_fs_remote_storage/pageserver";
 
+#[derive(Error, Debug)]
+pub enum PageserverHttpError {
+    #[error("Reqwest error: {0}")]
+    Transport(#[from] reqwest::Error),
+
+    #[error("Error: {0}")]
+    Response(String),
+}
+
+impl From<anyhow::Error> for PageserverHttpError {
+    fn from(e: anyhow::Error) -> Self {
+        Self::Response(e.to_string())
+    }
+}
+
+type Result<T> = result::Result<T, PageserverHttpError>;
+
+pub trait ResponseErrorMessageExt: Sized {
+    fn error_from_body(self) -> Result<Self>;
+}
+
+impl ResponseErrorMessageExt for Response {
+    fn error_from_body(self) -> Result<Self> {
+        let status = self.status();
+        if !(status.is_client_error() || status.is_server_error()) {
+            return Ok(self);
+        }
+
+        // reqwest does not export its error construction utility functions, so let's craft the message ourselves
+        let url = self.url().to_owned();
+        Err(PageserverHttpError::Response(
+            match self.json::<HttpErrorBody>() {
+                Ok(err_body) => format!("Error: {}", err_body.msg),
+                Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
+            },
+        ))
+    }
+}
+
 //
 // Control routines for pageserver.
 //
@@ -44,7 +87,8 @@ pub struct PageServerNode {
     pub pg_connection_config: PgConnectionConfig,
     pub conf: PageServerConf,
     pub env: LocalEnv,
-    pub http_client: mgmt_api::Client,
+    pub http_client: Client,
+    pub http_base_url: String,
 }
 
 impl PageServerNode {
@@ -56,19 +100,8 @@ impl PageServerNode {
             pg_connection_config: PgConnectionConfig::new_host_port(host, port),
             conf: conf.clone(),
             env: env.clone(),
-            http_client: mgmt_api::Client::new(
-                format!("http://{}", conf.listen_http_addr),
-                {
-                    match conf.http_auth_type {
-                        AuthType::Trust => None,
-                        AuthType::NeonJWT => Some(
-                            env.generate_auth_token(&Claims::new(None, Scope::PageServerApi))
-                                .unwrap(),
-                        ),
-                    }
-                }
-                .as_deref(),
-            ),
+            http_client: Client::new(),
+            http_base_url: format!("http://{}/v1", conf.listen_http_addr),
         }
     }
 
@@ -149,8 +182,8 @@ impl PageServerNode {
             .expect("non-Unicode path")
     }
 
-    pub async fn start(&self, config_overrides: &[&str]) -> anyhow::Result<Child> {
-        self.start_node(config_overrides, false).await
+    pub fn start(&self, config_overrides: &[&str]) -> anyhow::Result<Child> {
+        self.start_node(config_overrides, false)
     }
 
     fn pageserver_init(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
@@ -191,12 +224,7 @@ impl PageServerNode {
         Ok(())
     }
 
-    async fn start_node(
-        &self,
-        config_overrides: &[&str],
-        update_config: bool,
-    ) -> anyhow::Result<Child> {
-        // TODO: using a thread here because start_process() is not async but we need to call check_status()
+    fn start_node(&self, config_overrides: &[&str], update_config: bool) -> anyhow::Result<Child> {
         let datadir = self.repo_path();
         print!(
             "Starting pageserver node {} at '{}' in {:?}",
@@ -204,7 +232,7 @@ impl PageServerNode {
             self.pg_connection_config.raw_address(),
             datadir
         );
-        io::stdout().flush().context("flush stdout")?;
+        io::stdout().flush()?;
 
         let datadir_path_str = datadir.to_str().with_context(|| {
             format!(
@@ -216,23 +244,20 @@ impl PageServerNode {
         if update_config {
             args.push(Cow::Borrowed("--update-config"));
         }
+
         background_process::start_process(
             "pageserver",
             &datadir,
             &self.env.pageserver_bin(),
             args.iter().map(Cow::as_ref),
             self.pageserver_env_variables()?,
-            background_process::InitialPidFile::Expect(self.pid_file()),
-            || async {
-                let st = self.check_status().await;
-                match st {
-                    Ok(()) => Ok(true),
-                    Err(mgmt_api::Error::ReceiveBody(_)) => Ok(false),
-                    Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
-                }
+            background_process::InitialPidFile::Expect(&self.pid_file()),
+            || match self.check_status() {
+                Ok(()) => Ok(true),
+                Err(PageserverHttpError::Transport(_)) => Ok(false),
+                Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
             },
         )
-        .await
     }
 
     fn pageserver_basic_args<'a>(
@@ -278,12 +303,7 @@ impl PageServerNode {
         background_process::stop_process(immediate, "pageserver", &self.pid_file())
     }
 
-    pub async fn page_server_psql_client(
-        &self,
-    ) -> anyhow::Result<(
-        tokio_postgres::Client,
-        tokio_postgres::Connection<tokio_postgres::Socket, tokio_postgres::tls::NoTlsStream>,
-    )> {
+    pub fn page_server_psql_client(&self) -> anyhow::Result<postgres::Client> {
         let mut config = self.pg_connection_config.clone();
         if self.conf.pg_auth_type == AuthType::NeonJWT {
             let token = self
@@ -291,18 +311,36 @@ impl PageServerNode {
                 .generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
             config = config.set_password(Some(token));
         }
-        Ok(config.connect_no_tls().await?)
+        Ok(config.connect_no_tls()?)
     }
 
-    pub async fn check_status(&self) -> mgmt_api::Result<()> {
-        self.http_client.status().await
+    fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> anyhow::Result<RequestBuilder> {
+        let mut builder = self.http_client.request(method, url);
+        if self.conf.http_auth_type == AuthType::NeonJWT {
+            let token = self
+                .env
+                .generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
+            builder = builder.bearer_auth(token)
+        }
+        Ok(builder)
     }
 
-    pub async fn tenant_list(&self) -> mgmt_api::Result<Vec<TenantInfo>> {
-        self.http_client.list_tenants().await
+    pub fn check_status(&self) -> Result<()> {
+        self.http_request(Method::GET, format!("{}/status", self.http_base_url))?
+            .send()?
+            .error_from_body()?;
+        Ok(())
     }
 
-    pub async fn tenant_create(
+    pub fn tenant_list(&self) -> Result<Vec<TenantInfo>> {
+        Ok(self
+            .http_request(Method::GET, format!("{}/tenant", self.http_base_url))?
+            .send()?
+            .error_from_body()?
+            .json()?)
+    }
+
+    pub fn tenant_create(
         &self,
         new_tenant_id: TenantId,
         generation: Option<u32>,
@@ -369,7 +407,6 @@ impl PageServerNode {
                 .map(|x| x.parse::<bool>())
                 .transpose()
                 .context("Failed to parse 'gc_feedback' as bool")?,
-            heatmap_period: settings.remove("heatmap_period").map(|x| x.to_string()),
         };
 
         let request = models::TenantCreateRequest {
@@ -380,10 +417,23 @@ impl PageServerNode {
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")
         }
-        Ok(self.http_client.tenant_create(&request).await?)
+        self.http_request(Method::POST, format!("{}/tenant", self.http_base_url))?
+            .json(&request)
+            .send()?
+            .error_from_body()?
+            .json::<Option<String>>()
+            .with_context(|| {
+                format!("Failed to parse tenant creation response for tenant id: {new_tenant_id:?}")
+            })?
+            .context("No tenant id was found in the tenant creation response")
+            .and_then(|tenant_id_string| {
+                tenant_id_string.parse().with_context(|| {
+                    format!("Failed to parse response string as tenant id: '{tenant_id_string}'")
+                })
+            })
     }
 
-    pub async fn tenant_config(
+    pub fn tenant_config(
         &self,
         tenant_id: TenantId,
         mut settings: HashMap<&str, &str>,
@@ -454,7 +504,6 @@ impl PageServerNode {
                     .map(|x| x.parse::<bool>())
                     .transpose()
                     .context("Failed to parse 'gc_feedback' as bool")?,
-                heatmap_period: settings.remove("heatmap_period").map(|x| x.to_string()),
             }
         };
 
@@ -462,30 +511,54 @@ impl PageServerNode {
             bail!("Unrecognized tenant settings: {settings:?}")
         }
 
-        self.http_client
-            .tenant_config(&models::TenantConfigRequest { tenant_id, config })
-            .await?;
+        self.http_request(Method::PUT, format!("{}/tenant/config", self.http_base_url))?
+            .json(&models::TenantConfigRequest { tenant_id, config })
+            .send()?
+            .error_from_body()?;
 
         Ok(())
     }
 
-    pub async fn location_config(
+    pub fn location_config(
         &self,
         tenant_id: TenantId,
         config: LocationConfig,
         flush_ms: Option<Duration>,
     ) -> anyhow::Result<()> {
-        Ok(self
-            .http_client
-            .location_config(tenant_id, config, flush_ms)
-            .await?)
+        let req_body = TenantLocationConfigRequest { tenant_id, config };
+
+        let path = format!(
+            "{}/tenant/{}/location_config",
+            self.http_base_url, tenant_id
+        );
+        let path = if let Some(flush_ms) = flush_ms {
+            format!("{}?flush_ms={}", path, flush_ms.as_millis())
+        } else {
+            path
+        };
+
+        self.http_request(Method::PUT, path)?
+            .json(&req_body)
+            .send()?
+            .error_from_body()?;
+
+        Ok(())
     }
 
-    pub async fn timeline_list(&self, tenant_id: &TenantId) -> anyhow::Result<Vec<TimelineInfo>> {
-        Ok(self.http_client.list_timelines(*tenant_id).await?)
+    pub fn timeline_list(&self, tenant_id: &TenantId) -> anyhow::Result<Vec<TimelineInfo>> {
+        let timeline_infos: Vec<TimelineInfo> = self
+            .http_request(
+                Method::GET,
+                format!("{}/tenant/{}/timeline", self.http_base_url, tenant_id),
+            )?
+            .send()?
+            .error_from_body()?
+            .json()?;
+
+        Ok(timeline_infos)
     }
 
-    pub async fn timeline_create(
+    pub fn timeline_create(
         &self,
         tenant_id: TenantId,
         new_timeline_id: Option<TimelineId>,
@@ -496,14 +569,29 @@ impl PageServerNode {
     ) -> anyhow::Result<TimelineInfo> {
         // If timeline ID was not specified, generate one
         let new_timeline_id = new_timeline_id.unwrap_or(TimelineId::generate());
-        let req = models::TimelineCreateRequest {
+
+        self.http_request(
+            Method::POST,
+            format!("{}/tenant/{}/timeline", self.http_base_url, tenant_id),
+        )?
+        .json(&models::TimelineCreateRequest {
             new_timeline_id,
             ancestor_start_lsn,
             ancestor_timeline_id,
             pg_version,
             existing_initdb_timeline_id,
-        };
-        Ok(self.http_client.timeline_create(tenant_id, &req).await?)
+        })
+        .send()?
+        .error_from_body()?
+        .json::<Option<TimelineInfo>>()
+        .with_context(|| {
+            format!("Failed to parse timeline creation response for tenant id: {tenant_id}")
+        })?
+        .with_context(|| {
+            format!(
+                "No timeline id was found in the timeline creation response for tenant {tenant_id}"
+            )
+        })
     }
 
     /// Import a basebackup prepared using either:
@@ -515,7 +603,7 @@ impl PageServerNode {
     /// * `timeline_id` - id to assign to imported timeline
     /// * `base` - (start lsn of basebackup, path to `base.tar` file)
     /// * `pg_wal` - if there's any wal to import: (end lsn, path to `pg_wal.tar`)
-    pub async fn timeline_import(
+    pub fn timeline_import(
         &self,
         tenant_id: TenantId,
         timeline_id: TimelineId,
@@ -523,60 +611,36 @@ impl PageServerNode {
         pg_wal: Option<(Lsn, PathBuf)>,
         pg_version: u32,
     ) -> anyhow::Result<()> {
-        let (client, conn) = self.page_server_psql_client().await?;
-        // The connection object performs the actual communication with the database,
-        // so spawn it off to run on its own.
-        tokio::spawn(async move {
-            if let Err(e) = conn.await {
-                eprintln!("connection error: {}", e);
-            }
-        });
-        tokio::pin!(client);
+        let mut client = self.page_server_psql_client()?;
 
         // Init base reader
         let (start_lsn, base_tarfile_path) = base;
-        let base_tarfile = tokio::fs::File::open(base_tarfile_path).await?;
-        let base_tarfile = tokio_util::io::ReaderStream::new(base_tarfile);
+        let base_tarfile = File::open(base_tarfile_path)?;
+        let mut base_reader = BufReader::new(base_tarfile);
 
         // Init wal reader if necessary
         let (end_lsn, wal_reader) = if let Some((end_lsn, wal_tarfile_path)) = pg_wal {
-            let wal_tarfile = tokio::fs::File::open(wal_tarfile_path).await?;
-            let wal_reader = tokio_util::io::ReaderStream::new(wal_tarfile);
+            let wal_tarfile = File::open(wal_tarfile_path)?;
+            let wal_reader = BufReader::new(wal_tarfile);
             (end_lsn, Some(wal_reader))
         } else {
             (start_lsn, None)
         };
 
-        let copy_in = |reader, cmd| {
-            let client = &client;
-            async move {
-                let writer = client.copy_in(&cmd).await?;
-                let writer = std::pin::pin!(writer);
-                let mut writer = writer.sink_map_err(|e| {
-                    std::io::Error::new(std::io::ErrorKind::Other, format!("{e}"))
-                });
-                let mut reader = std::pin::pin!(reader);
-                writer.send_all(&mut reader).await?;
-                writer.into_inner().finish().await?;
-                anyhow::Ok(())
-            }
-        };
-
         // Import base
-        copy_in(
-            base_tarfile,
-            format!(
-                "import basebackup {tenant_id} {timeline_id} {start_lsn} {end_lsn} {pg_version}"
-            ),
-        )
-        .await?;
+        let import_cmd = format!(
+            "import basebackup {tenant_id} {timeline_id} {start_lsn} {end_lsn} {pg_version}"
+        );
+        let mut writer = client.copy_in(&import_cmd)?;
+        io::copy(&mut base_reader, &mut writer)?;
+        writer.finish()?;
+
         // Import wal if necessary
-        if let Some(wal_reader) = wal_reader {
-            copy_in(
-                wal_reader,
-                format!("import wal {tenant_id} {timeline_id} {start_lsn} {end_lsn}"),
-            )
-            .await?;
+        if let Some(mut wal_reader) = wal_reader {
+            let import_cmd = format!("import wal {tenant_id} {timeline_id} {start_lsn} {end_lsn}");
+            let mut writer = client.copy_in(&import_cmd)?;
+            io::copy(&mut wal_reader, &mut writer)?;
+            writer.finish()?;
         }
 
         Ok(())

control_plane/src/safekeeper.rs

@@ -13,6 +13,7 @@ use std::{io, result};
 use anyhow::Context;
 use camino::Utf8PathBuf;
 use postgres_connection::PgConnectionConfig;
+use reqwest::blocking::{Client, RequestBuilder, Response};
 use reqwest::{IntoUrl, Method};
 use thiserror::Error;
 use utils::{http::error::HttpErrorBody, id::NodeId};
@@ -33,14 +34,12 @@ pub enum SafekeeperHttpError {
 
 type Result<T> = result::Result<T, SafekeeperHttpError>;
 
-#[async_trait::async_trait]
 pub trait ResponseErrorMessageExt: Sized {
-    async fn error_from_body(self) -> Result<Self>;
+    fn error_from_body(self) -> Result<Self>;
 }
 
-#[async_trait::async_trait]
-impl ResponseErrorMessageExt for reqwest::Response {
-    async fn error_from_body(self) -> Result<Self> {
+impl ResponseErrorMessageExt for Response {
+    fn error_from_body(self) -> Result<Self> {
         let status = self.status();
         if !(status.is_client_error() || status.is_server_error()) {
             return Ok(self);
@@ -49,7 +48,7 @@ impl ResponseErrorMessageExt for reqwest::Response {
         // reqwest does not export its error construction utility functions, so let's craft the message ourselves
         let url = self.url().to_owned();
         Err(SafekeeperHttpError::Response(
-            match self.json::<HttpErrorBody>().await {
+            match self.json::<HttpErrorBody>() {
                 Ok(err_body) => format!("Error: {}", err_body.msg),
                 Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
             },
@@ -70,7 +69,7 @@ pub struct SafekeeperNode {
 
     pub pg_connection_config: PgConnectionConfig,
     pub env: LocalEnv,
-    pub http_client: reqwest::Client,
+    pub http_client: Client,
     pub http_base_url: String,
 }
 
@@ -81,7 +80,7 @@ impl SafekeeperNode {
             conf: conf.clone(),
             pg_connection_config: Self::safekeeper_connection_config(conf.pg_port),
             env: env.clone(),
-            http_client: reqwest::Client::new(),
+            http_client: Client::new(),
             http_base_url: format!("http://127.0.0.1:{}/v1", conf.http_port),
         }
     }
@@ -104,7 +103,7 @@ impl SafekeeperNode {
             .expect("non-Unicode path")
     }
 
-    pub async fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<Child> {
+    pub fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<Child> {
         print!(
             "Starting safekeeper at '{}' in '{}'",
             self.pg_connection_config.raw_address(),
@@ -192,16 +191,13 @@ impl SafekeeperNode {
             &self.env.safekeeper_bin(),
             &args,
             [],
-            background_process::InitialPidFile::Expect(self.pid_file()),
-            || async {
-                match self.check_status().await {
-                    Ok(()) => Ok(true),
-                    Err(SafekeeperHttpError::Transport(_)) => Ok(false),
-                    Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
-                }
+            background_process::InitialPidFile::Expect(&self.pid_file()),
+            || match self.check_status() {
+                Ok(()) => Ok(true),
+                Err(SafekeeperHttpError::Transport(_)) => Ok(false),
+                Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
             },
         )
-        .await
     }
 
     ///
@@ -220,7 +216,7 @@ impl SafekeeperNode {
         )
     }
 
-    fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> reqwest::RequestBuilder {
+    fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> RequestBuilder {
         // TODO: authentication
         //if self.env.auth_type == AuthType::NeonJWT {
         //    builder = builder.bearer_auth(&self.env.safekeeper_auth_token)
@@ -228,12 +224,10 @@ impl SafekeeperNode {
         self.http_client.request(method, url)
     }
 
-    pub async fn check_status(&self) -> Result<()> {
+    pub fn check_status(&self) -> Result<()> {
         self.http_request(Method::GET, format!("{}/{}", self.http_base_url, "status"))
-            .send()
-            .await?
-            .error_from_body()
-            .await?;
+            .send()?
+            .error_from_body()?;
         Ok(())
     }
 }

control_plane/src/tenant_migration.rs

@@ -19,11 +19,11 @@ use utils::{
 };
 
 /// Given an attached pageserver, retrieve the LSN for all timelines
-async fn get_lsns(
+fn get_lsns(
     tenant_id: TenantId,
     pageserver: &PageServerNode,
 ) -> anyhow::Result<HashMap<TimelineId, Lsn>> {
-    let timelines = pageserver.timeline_list(&tenant_id).await?;
+    let timelines = pageserver.timeline_list(&tenant_id)?;
     Ok(timelines
         .into_iter()
         .map(|t| (t.timeline_id, t.last_record_lsn))
@@ -32,13 +32,13 @@ async fn get_lsns(
 
 /// Wait for the timeline LSNs on `pageserver` to catch up with or overtake
 /// `baseline`.
-async fn await_lsn(
+fn await_lsn(
     tenant_id: TenantId,
     pageserver: &PageServerNode,
     baseline: HashMap<TimelineId, Lsn>,
 ) -> anyhow::Result<()> {
     loop {
-        let latest = match get_lsns(tenant_id, pageserver).await {
+        let latest = match get_lsns(tenant_id, pageserver) {
             Ok(l) => l,
             Err(e) => {
                 println!(
@@ -84,7 +84,7 @@ async fn await_lsn(
 ///  - Coordinate attach/secondary/detach on pageservers
 ///  - call into attachment_service for generations
 ///  - reconfigure compute endpoints to point to new attached pageserver
-pub async fn migrate_tenant(
+pub fn migrate_tenant(
     env: &LocalEnv,
     tenant_id: TenantId,
     dest_ps: PageServerNode,
@@ -108,18 +108,16 @@ pub async fn migrate_tenant(
         }
     }
 
-    let previous = attachment_service.inspect(tenant_id).await?;
+    let previous = attachment_service.inspect(tenant_id)?;
     let mut baseline_lsns = None;
     if let Some((generation, origin_ps_id)) = &previous {
         let origin_ps = PageServerNode::from_env(env, env.get_pageserver_conf(*origin_ps_id)?);
 
         if origin_ps_id == &dest_ps.conf.id {
             println!("🔁 Already attached to {origin_ps_id}, freshening...");
-            let gen = attachment_service
-                .attach_hook(tenant_id, dest_ps.conf.id)
-                .await?;
+            let gen = attachment_service.attach_hook(tenant_id, dest_ps.conf.id)?;
             let dest_conf = build_location_config(LocationConfigMode::AttachedSingle, gen, None);
-            dest_ps.location_config(tenant_id, dest_conf, None).await?;
+            dest_ps.location_config(tenant_id, dest_conf, None)?;
             println!("✅ Migration complete");
             return Ok(());
         }
@@ -128,24 +126,20 @@ pub async fn migrate_tenant(
 
         let stale_conf =
             build_location_config(LocationConfigMode::AttachedStale, Some(*generation), None);
-        origin_ps
-            .location_config(tenant_id, stale_conf, Some(Duration::from_secs(10)))
-            .await?;
+        origin_ps.location_config(tenant_id, stale_conf, Some(Duration::from_secs(10)))?;
 
-        baseline_lsns = Some(get_lsns(tenant_id, &origin_ps).await?);
+        baseline_lsns = Some(get_lsns(tenant_id, &origin_ps)?);
     }
 
-    let gen = attachment_service
-        .attach_hook(tenant_id, dest_ps.conf.id)
-        .await?;
+    let gen = attachment_service.attach_hook(tenant_id, dest_ps.conf.id)?;
     let dest_conf = build_location_config(LocationConfigMode::AttachedMulti, gen, None);
 
     println!("🔁 Attaching to pageserver {}", dest_ps.conf.id);
-    dest_ps.location_config(tenant_id, dest_conf, None).await?;
+    dest_ps.location_config(tenant_id, dest_conf, None)?;
 
     if let Some(baseline) = baseline_lsns {
         println!("🕑 Waiting for LSN to catch up...");
-        await_lsn(tenant_id, &dest_ps, baseline).await?;
+        await_lsn(tenant_id, &dest_ps, baseline)?;
     }
 
     let cplane = ComputeControlPlane::load(env.clone())?;
@@ -155,7 +149,7 @@ pub async fn migrate_tenant(
                 "🔁 Reconfiguring endpoint {} to use pageserver {}",
                 endpoint_name, dest_ps.conf.id
             );
-            endpoint.reconfigure(Some(dest_ps.conf.id)).await?;
+            endpoint.reconfigure(Some(dest_ps.conf.id))?;
         }
     }
 
@@ -165,7 +159,7 @@ pub async fn migrate_tenant(
         }
 
         let other_ps = PageServerNode::from_env(env, other_ps_conf);
-        let other_ps_tenants = other_ps.tenant_list().await?;
+        let other_ps_tenants = other_ps.tenant_list()?;
 
         // Check if this tenant is attached
         let found = other_ps_tenants
@@ -187,9 +181,7 @@ pub async fn migrate_tenant(
             "💤 Switching to secondary mode on pageserver {}",
             other_ps.conf.id
         );
-        other_ps
-            .location_config(tenant_id, secondary_conf, None)
-            .await?;
+        other_ps.location_config(tenant_id, secondary_conf, None)?;
     }
 
     println!(
@@ -197,7 +189,7 @@ pub async fn migrate_tenant(
         dest_ps.conf.id
     );
     let dest_conf = build_location_config(LocationConfigMode::AttachedSingle, gen, None);
-    dest_ps.location_config(tenant_id, dest_conf, None).await?;
+    dest_ps.location_config(tenant_id, dest_conf, None)?;
 
     println!("✅ Migration complete");
 

libs/metrics/src/lib.rs

@@ -3,11 +3,8 @@
 //! Otherwise, we might not see all metrics registered via
 //! a default registry.
 #![deny(clippy::undocumented_unsafe_blocks)]
-
 use once_cell::sync::Lazy;
-use prometheus::core::{
-    Atomic, AtomicU64, Collector, GenericCounter, GenericCounterVec, GenericGauge, GenericGaugeVec,
-};
+use prometheus::core::{AtomicU64, Collector, GenericGauge, GenericGaugeVec};
 pub use prometheus::opts;
 pub use prometheus::register;
 pub use prometheus::Error;
@@ -135,137 +132,3 @@ fn get_rusage_stats() -> libc::rusage {
         rusage.assume_init()
     }
 }
-
-/// Create an [`IntCounterPairVec`] and registers to default registry.
-#[macro_export(local_inner_macros)]
-macro_rules! register_int_counter_pair_vec {
-    ($NAME1:expr, $HELP1:expr, $NAME2:expr, $HELP2:expr, $LABELS_NAMES:expr $(,)?) => {{
-        match (
-            $crate::register_int_counter_vec!($NAME1, $HELP1, $LABELS_NAMES),
-            $crate::register_int_counter_vec!($NAME2, $HELP2, $LABELS_NAMES),
-        ) {
-            (Ok(inc), Ok(dec)) => Ok($crate::IntCounterPairVec::new(inc, dec)),
-            (Err(e), _) | (_, Err(e)) => Err(e),
-        }
-    }};
-}
-/// Create an [`IntCounterPair`] and registers to default registry.
-#[macro_export(local_inner_macros)]
-macro_rules! register_int_counter_pair {
-    ($NAME1:expr, $HELP1:expr, $NAME2:expr, $HELP2:expr $(,)?) => {{
-        match (
-            $crate::register_int_counter!($NAME1, $HELP1),
-            $crate::register_int_counter!($NAME2, $HELP2),
-        ) {
-            (Ok(inc), Ok(dec)) => Ok($crate::IntCounterPair::new(inc, dec)),
-            (Err(e), _) | (_, Err(e)) => Err(e),
-        }
-    }};
-}
-
-/// A Pair of [`GenericCounterVec`]s. Like an [`GenericGaugeVec`] but will always observe changes
-pub struct GenericCounterPairVec<P: Atomic> {
-    inc: GenericCounterVec<P>,
-    dec: GenericCounterVec<P>,
-}
-
-/// A Pair of [`GenericCounter`]s. Like an [`GenericGauge`] but will always observe changes
-pub struct GenericCounterPair<P: Atomic> {
-    inc: GenericCounter<P>,
-    dec: GenericCounter<P>,
-}
-
-impl<P: Atomic> GenericCounterPairVec<P> {
-    pub fn new(inc: GenericCounterVec<P>, dec: GenericCounterVec<P>) -> Self {
-        Self { inc, dec }
-    }
-
-    /// `get_metric_with_label_values` returns the [`GenericCounterPair<P>`] for the given slice
-    /// of label values (same order as the VariableLabels in Desc). If that combination of
-    /// label values is accessed for the first time, a new [`GenericCounterPair<P>`] is created.
-    ///
-    /// An error is returned if the number of label values is not the same as the
-    /// number of VariableLabels in Desc.
-    pub fn get_metric_with_label_values(&self, vals: &[&str]) -> Result<GenericCounterPair<P>> {
-        Ok(GenericCounterPair {
-            inc: self.inc.get_metric_with_label_values(vals)?,
-            dec: self.dec.get_metric_with_label_values(vals)?,
-        })
-    }
-
-    /// `with_label_values` works as `get_metric_with_label_values`, but panics if an error
-    /// occurs.
-    pub fn with_label_values(&self, vals: &[&str]) -> GenericCounterPair<P> {
-        self.get_metric_with_label_values(vals).unwrap()
-    }
-}
-
-impl<P: Atomic> GenericCounterPair<P> {
-    pub fn new(inc: GenericCounter<P>, dec: GenericCounter<P>) -> Self {
-        Self { inc, dec }
-    }
-
-    /// Increment the gauge by 1, returning a guard that decrements by 1 on drop.
-    pub fn guard(&self) -> GenericCounterPairGuard<P> {
-        self.inc.inc();
-        GenericCounterPairGuard(self.dec.clone())
-    }
-
-    /// Increment the gauge by n, returning a guard that decrements by n on drop.
-    pub fn guard_by(&self, n: P::T) -> GenericCounterPairGuardBy<P> {
-        self.inc.inc_by(n);
-        GenericCounterPairGuardBy(self.dec.clone(), n)
-    }
-
-    /// Increase the gauge by 1.
-    #[inline]
-    pub fn inc(&self) {
-        self.inc.inc();
-    }
-
-    /// Decrease the gauge by 1.
-    #[inline]
-    pub fn dec(&self) {
-        self.dec.inc();
-    }
-
-    /// Add the given value to the gauge. (The value can be
-    /// negative, resulting in a decrement of the gauge.)
-    #[inline]
-    pub fn inc_by(&self, v: P::T) {
-        self.inc.inc_by(v);
-    }
-
-    /// Subtract the given value from the gauge. (The value can be
-    /// negative, resulting in an increment of the gauge.)
-    #[inline]
-    pub fn dec_by(&self, v: P::T) {
-        self.dec.inc_by(v);
-    }
-}
-
-/// Guard returned by [`GenericCounterPair::guard`]
-pub struct GenericCounterPairGuard<P: Atomic>(GenericCounter<P>);
-
-impl<P: Atomic> Drop for GenericCounterPairGuard<P> {
-    fn drop(&mut self) {
-        self.0.inc();
-    }
-}
-/// Guard returned by [`GenericCounterPair::guard_by`]
-pub struct GenericCounterPairGuardBy<P: Atomic>(GenericCounter<P>, P::T);
-
-impl<P: Atomic> Drop for GenericCounterPairGuardBy<P> {
-    fn drop(&mut self) {
-        self.0.inc_by(self.1);
-    }
-}
-
-/// A Pair of [`IntCounterVec`]s. Like an [`IntGaugeVec`] but will always observe changes
-pub type IntCounterPairVec = GenericCounterPairVec<AtomicU64>;
-
-/// A Pair of [`IntCounter`]s. Like an [`IntGauge`] but will always observe changes
-pub type IntCounterPair = GenericCounterPair<AtomicU64>;
-
-/// A guard for [`IntCounterPair`] that will decrement the gauge on drop
-pub type IntCounterPairGuard = GenericCounterPairGuard<AtomicU64>;

libs/pageserver_api/Cargo.toml

@@ -24,4 +24,3 @@ workspace_hack.workspace = true
 
 [dev-dependencies]
 bincode.workspace = true
-rand.workspace = true

libs/pageserver_api/src/key.rs

@@ -144,37 +144,3 @@ impl Key {
 pub fn is_rel_block_key(key: &Key) -> bool {
     key.field1 == 0x00 && key.field4 != 0
 }
-
-impl std::str::FromStr for Key {
-    type Err = anyhow::Error;
-
-    fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
-        Self::from_hex(s)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::str::FromStr;
-
-    use crate::key::Key;
-
-    use rand::Rng;
-    use rand::SeedableRng;
-
-    #[test]
-    fn display_fromstr_bijection() {
-        let mut rng = rand::rngs::StdRng::seed_from_u64(42);
-
-        let key = Key {
-            field1: rng.gen(),
-            field2: rng.gen(),
-            field3: rng.gen(),
-            field4: rng.gen(),
-            field5: rng.gen(),
-            field6: rng.gen(),
-        };
-
-        assert_eq!(key, Key::from_str(&format!("{key}")).unwrap());
-    }
-}

libs/pageserver_api/src/lib.rs

@@ -5,7 +5,6 @@ use const_format::formatcp;
 /// Public API types
 pub mod control_api;
 pub mod key;
-pub mod keyspace;
 pub mod models;
 pub mod reltag;
 pub mod shard;

libs/pageserver_api/src/models.rs

@@ -1,5 +1,3 @@
-pub mod partitioning;
-
 use std::{
     collections::HashMap,
     num::{NonZeroU64, NonZeroUsize},
@@ -239,7 +237,6 @@ pub struct TenantConfig {
     pub min_resident_size_override: Option<u64>,
     pub evictions_low_residence_duration_metric_threshold: Option<String>,
     pub gc_feedback: Option<bool>,
-    pub heatmap_period: Option<String>,
 }
 
 /// A flattened analog of a `pagesever::tenant::LocationMode`, which

libs/pageserver_api/src/models/partitioning.rs

@@ -1,151 +0,0 @@
-use utils::lsn::Lsn;
-
-#[derive(Debug, PartialEq, Eq)]
-pub struct Partitioning {
-    pub keys: crate::keyspace::KeySpace,
-
-    pub at_lsn: Lsn,
-}
-
-impl serde::Serialize for Partitioning {
-    fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        pub struct KeySpace<'a>(&'a crate::keyspace::KeySpace);
-
-        impl<'a> serde::Serialize for KeySpace<'a> {
-            fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
-            where
-                S: serde::Serializer,
-            {
-                use serde::ser::SerializeSeq;
-                let mut seq = serializer.serialize_seq(Some(self.0.ranges.len()))?;
-                for kr in &self.0.ranges {
-                    seq.serialize_element(&KeyRange(kr))?;
-                }
-                seq.end()
-            }
-        }
-
-        use serde::ser::SerializeMap;
-        let mut map = serializer.serialize_map(Some(2))?;
-        map.serialize_key("keys")?;
-        map.serialize_value(&KeySpace(&self.keys))?;
-        map.serialize_key("at_lsn")?;
-        map.serialize_value(&WithDisplay(&self.at_lsn))?;
-        map.end()
-    }
-}
-
-pub struct WithDisplay<'a, T>(&'a T);
-
-impl<'a, T: std::fmt::Display> serde::Serialize for WithDisplay<'a, T> {
-    fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        serializer.collect_str(&self.0)
-    }
-}
-
-pub struct KeyRange<'a>(&'a std::ops::Range<crate::key::Key>);
-
-impl<'a> serde::Serialize for KeyRange<'a> {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        use serde::ser::SerializeTuple;
-        let mut t = serializer.serialize_tuple(2)?;
-        t.serialize_element(&WithDisplay(&self.0.start))?;
-        t.serialize_element(&WithDisplay(&self.0.end))?;
-        t.end()
-    }
-}
-
-impl<'a> serde::Deserialize<'a> for Partitioning {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'a>,
-    {
-        pub struct KeySpace(crate::keyspace::KeySpace);
-
-        impl<'de> serde::Deserialize<'de> for KeySpace {
-            fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-            where
-                D: serde::Deserializer<'de>,
-            {
-                #[serde_with::serde_as]
-                #[derive(serde::Deserialize)]
-                #[serde(transparent)]
-                struct Key(#[serde_as(as = "serde_with::DisplayFromStr")] crate::key::Key);
-
-                #[serde_with::serde_as]
-                #[derive(serde::Deserialize)]
-                struct Range(Key, Key);
-
-                let ranges: Vec<Range> = serde::Deserialize::deserialize(deserializer)?;
-                Ok(Self(crate::keyspace::KeySpace {
-                    ranges: ranges
-                        .into_iter()
-                        .map(|Range(start, end)| (start.0..end.0))
-                        .collect(),
-                }))
-            }
-        }
-
-        #[serde_with::serde_as]
-        #[derive(serde::Deserialize)]
-        struct De {
-            keys: KeySpace,
-            #[serde_as(as = "serde_with::DisplayFromStr")]
-            at_lsn: Lsn,
-        }
-
-        let de: De = serde::Deserialize::deserialize(deserializer)?;
-        Ok(Self {
-            at_lsn: de.at_lsn,
-            keys: de.keys.0,
-        })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_serialization_roundtrip() {
-        let reference = r#"
-        {
-            "keys": [
-              [
-                "000000000000000000000000000000000000",
-                "000000000000000000000000000000000001"
-              ],
-              [
-                "000000067F00000001000000000000000000",
-                "000000067F00000001000000000000000002"
-              ],
-              [
-                "030000000000000000000000000000000000",
-                "030000000000000000000000000000000003"
-              ]
-            ],
-            "at_lsn": "0/2240160"
-        }
-        "#;
-
-        let de: Partitioning = serde_json::from_str(reference).unwrap();
-
-        let ser = serde_json::to_string(&de).unwrap();
-
-        let ser_de: serde_json::Value = serde_json::from_str(&ser).unwrap();
-
-        assert_eq!(
-            ser_de,
-            serde_json::from_str::<'_, serde_json::Value>(reference).unwrap()
-        );
-    }
-}

libs/postgres_connection/src/lib.rs

@@ -163,18 +163,8 @@ impl PgConnectionConfig {
     }
 
     /// Connect using postgres protocol with TLS disabled.
-    pub async fn connect_no_tls(
-        &self,
-    ) -> Result<
-        (
-            tokio_postgres::Client,
-            tokio_postgres::Connection<tokio_postgres::Socket, tokio_postgres::tls::NoTlsStream>,
-        ),
-        postgres::Error,
-    > {
-        self.to_tokio_postgres_config()
-            .connect(postgres::NoTls)
-            .await
+    pub fn connect_no_tls(&self) -> Result<postgres::Client, postgres::Error> {
+        postgres::Config::from(self.to_tokio_postgres_config()).connect(postgres::NoTls)
     }
 }
 

libs/utils/Cargo.toml

@@ -50,8 +50,6 @@ const_format.workspace = true
 # why is it only here? no other crate should use it, streams are rarely needed.
 tokio-stream = { version = "0.1.14" }
 
-serde_path_to_error.workspace = true
-
 [dev-dependencies]
 byteorder.workspace = true
 bytes.workspace = true

libs/utils/src/http/json.rs

@@ -25,12 +25,8 @@ pub async fn json_request_or_empty_body<T: for<'de> Deserialize<'de>>(
     if body.remaining() == 0 {
         return Ok(None);
     }
-
-    let mut deser = serde_json::de::Deserializer::from_reader(body.reader());
-
-    serde_path_to_error::deserialize(&mut deser)
-        // intentionally stringify because the debug version is not helpful in python logs
-        .map_err(|e| anyhow::anyhow!("Failed to parse json request: {e}"))
+    serde_json::from_reader(body.reader())
+        .context("Failed to parse json request")
         .map(Some)
         .map_err(ApiError::BadRequest)
 }

libs/utils/src/logging.rs

@@ -1,7 +1,6 @@
 use std::str::FromStr;
 
 use anyhow::Context;
-use metrics::{IntCounter, IntCounterVec};
 use once_cell::sync::Lazy;
 use strum_macros::{EnumString, EnumVariantNames};
 
@@ -25,48 +24,16 @@ impl LogFormat {
     }
 }
 
-struct TracingEventCountMetric {
-    error: IntCounter,
-    warn: IntCounter,
-    info: IntCounter,
-    debug: IntCounter,
-    trace: IntCounter,
-}
-
-static TRACING_EVENT_COUNT_METRIC: Lazy<TracingEventCountMetric> = Lazy::new(|| {
-    let vec = metrics::register_int_counter_vec!(
+static TRACING_EVENT_COUNT: Lazy<metrics::IntCounterVec> = Lazy::new(|| {
+    metrics::register_int_counter_vec!(
         "libmetrics_tracing_event_count",
         "Number of tracing events, by level",
         &["level"]
     )
-    .expect("failed to define metric");
-    TracingEventCountMetric::new(vec)
+    .expect("failed to define metric")
 });
 
-impl TracingEventCountMetric {
-    fn new(vec: IntCounterVec) -> Self {
-        Self {
-            error: vec.with_label_values(&["error"]),
-            warn: vec.with_label_values(&["warn"]),
-            info: vec.with_label_values(&["info"]),
-            debug: vec.with_label_values(&["debug"]),
-            trace: vec.with_label_values(&["trace"]),
-        }
-    }
-
-    fn inc_for_level(&self, level: tracing::Level) {
-        let counter = match level {
-            tracing::Level::ERROR => &self.error,
-            tracing::Level::WARN => &self.warn,
-            tracing::Level::INFO => &self.info,
-            tracing::Level::DEBUG => &self.debug,
-            tracing::Level::TRACE => &self.trace,
-        };
-        counter.inc();
-    }
-}
-
-struct TracingEventCountLayer(&'static TracingEventCountMetric);
+struct TracingEventCountLayer(&'static metrics::IntCounterVec);
 
 impl<S> tracing_subscriber::layer::Layer<S> for TracingEventCountLayer
 where
@@ -77,7 +44,15 @@ where
         event: &tracing::Event<'_>,
         _ctx: tracing_subscriber::layer::Context<'_, S>,
     ) {
-        self.0.inc_for_level(*event.metadata().level());
+        let level = event.metadata().level();
+        let level = match *level {
+            tracing::Level::ERROR => "error",
+            tracing::Level::WARN => "warn",
+            tracing::Level::INFO => "info",
+            tracing::Level::DEBUG => "debug",
+            tracing::Level::TRACE => "trace",
+        };
+        self.0.with_label_values(&[level]).inc();
     }
 }
 
@@ -131,9 +106,7 @@ pub fn init(
         };
         log_layer.with_filter(rust_log_env_filter())
     });
-    let r = r.with(
-        TracingEventCountLayer(&TRACING_EVENT_COUNT_METRIC).with_filter(rust_log_env_filter()),
-    );
+    let r = r.with(TracingEventCountLayer(&TRACING_EVENT_COUNT).with_filter(rust_log_env_filter()));
     match tracing_error_layer_enablement {
         TracingErrorLayerEnablement::EnableWithRustLogFilter => r
             .with(tracing_error::ErrorLayer::default().with_filter(rust_log_env_filter()))
@@ -284,14 +257,14 @@ impl std::fmt::Debug for SecretString {
 mod tests {
     use metrics::{core::Opts, IntCounterVec};
 
-    use crate::logging::{TracingEventCountLayer, TracingEventCountMetric};
+    use super::TracingEventCountLayer;
 
     #[test]
     fn tracing_event_count_metric() {
         let counter_vec =
             IntCounterVec::new(Opts::new("testmetric", "testhelp"), &["level"]).unwrap();
-        let metric = Box::leak(Box::new(TracingEventCountMetric::new(counter_vec.clone())));
-        let layer = TracingEventCountLayer(metric);
+        let counter_vec = Box::leak(Box::new(counter_vec)); // make it 'static
+        let layer = TracingEventCountLayer(counter_vec);
         use tracing_subscriber::prelude::*;
 
         tracing::subscriber::with_default(tracing_subscriber::registry().with(layer), || {

libs/utils/src/timeout.rs

@@ -2,11 +2,8 @@ use std::time::Duration;
 
 use tokio_util::sync::CancellationToken;
 
-#[derive(thiserror::Error, Debug)]
 pub enum TimeoutCancellableError {
-    #[error("Timed out")]
     Timeout,
-    #[error("Cancelled")]
     Cancelled,
 }
 

libs/walproposer/build.rs

@@ -1,6 +1,3 @@
-//! Links with walproposer, pgcommon, pgport and runs bindgen on walproposer.h
-//! to generate Rust bindings for it.
-
 use std::{env, path::PathBuf, process::Command};
 
 use anyhow::{anyhow, Context};

libs/walproposer/src/api_bindings.rs

@@ -1,6 +1,3 @@
-//! A C-Rust shim: defines implementation of C walproposer API, assuming wp
-//! callback_data stores Box to some Rust implementation.
-
 #![allow(dead_code)]
 
 use std::ffi::CStr;

pageserver/Cargo.toml

@@ -36,7 +36,6 @@ humantime.workspace = true
 humantime-serde.workspace = true
 hyper.workspace = true
 itertools.workspace = true
-md5.workspace = true
 nix.workspace = true
 # hack to get the number of worker threads tokio uses
 num_cpus = { version = "1.15" }

pageserver/client/Cargo.toml

@@ -1,14 +0,0 @@
-[package]
-name = "pageserver_client"
-version = "0.1.0"
-edition.workspace = true
-license.workspace = true
-
-[dependencies]
-pageserver_api.workspace = true
-thiserror.workspace = true
-async-trait.workspace = true
-reqwest.workspace = true
-utils.workspace = true
-serde.workspace = true
-workspace_hack = { version = "0.1", path = "../../workspace_hack" }

pageserver/client/src/lib.rs

@@ -1 +0,0 @@
-pub mod mgmt_api;

pageserver/client/src/mgmt_api.rs

@@ -1,188 +0,0 @@
-use pageserver_api::models::*;
-use reqwest::{IntoUrl, Method};
-use utils::{
-    http::error::HttpErrorBody,
-    id::{TenantId, TimelineId},
-};
-
-#[derive(Debug)]
-pub struct Client {
-    mgmt_api_endpoint: String,
-    authorization_header: Option<String>,
-    client: reqwest::Client,
-}
-
-#[derive(thiserror::Error, Debug)]
-pub enum Error {
-    #[error("receive body: {0}")]
-    ReceiveBody(reqwest::Error),
-
-    #[error("receive error body: {0}")]
-    ReceiveErrorBody(String),
-
-    #[error("pageserver API: {0}")]
-    ApiError(String),
-}
-
-pub type Result<T> = std::result::Result<T, Error>;
-
-#[async_trait::async_trait]
-pub trait ResponseErrorMessageExt: Sized {
-    async fn error_from_body(self) -> Result<Self>;
-}
-
-#[async_trait::async_trait]
-impl ResponseErrorMessageExt for reqwest::Response {
-    async fn error_from_body(mut self) -> Result<Self> {
-        let status = self.status();
-        if !(status.is_client_error() || status.is_server_error()) {
-            return Ok(self);
-        }
-
-        let url = self.url().to_owned();
-        Err(match self.json::<HttpErrorBody>().await {
-            Ok(HttpErrorBody { msg }) => Error::ApiError(msg),
-            Err(_) => {
-                Error::ReceiveErrorBody(format!("Http error ({}) at {}.", status.as_u16(), url))
-            }
-        })
-    }
-}
-
-impl Client {
-    pub fn new(mgmt_api_endpoint: String, jwt: Option<&str>) -> Self {
-        Self {
-            mgmt_api_endpoint,
-            authorization_header: jwt.map(|jwt| format!("Bearer {jwt}")),
-            client: reqwest::Client::new(),
-        }
-    }
-
-    pub async fn list_tenants(&self) -> Result<Vec<pageserver_api::models::TenantInfo>> {
-        let uri = format!("{}/v1/tenant", self.mgmt_api_endpoint);
-        let resp = self.get(&uri).await?;
-        resp.json().await.map_err(Error::ReceiveBody)
-    }
-
-    pub async fn list_timelines(
-        &self,
-        tenant_id: TenantId,
-    ) -> Result<Vec<pageserver_api::models::TimelineInfo>> {
-        let uri = format!("{}/v1/tenant/{tenant_id}/timeline", self.mgmt_api_endpoint);
-        self.get(&uri)
-            .await?
-            .json()
-            .await
-            .map_err(Error::ReceiveBody)
-    }
-
-    pub async fn timeline_info(
-        &self,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-    ) -> Result<pageserver_api::models::TimelineInfo> {
-        let uri = format!(
-            "{}/v1/tenant/{tenant_id}/timeline/{timeline_id}",
-            self.mgmt_api_endpoint
-        );
-        self.get(&uri)
-            .await?
-            .json()
-            .await
-            .map_err(Error::ReceiveBody)
-    }
-
-    pub async fn keyspace(
-        &self,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-    ) -> Result<pageserver_api::models::partitioning::Partitioning> {
-        let uri = format!(
-            "{}/v1/tenant/{tenant_id}/timeline/{timeline_id}/keyspace",
-            self.mgmt_api_endpoint
-        );
-        self.get(&uri)
-            .await?
-            .json()
-            .await
-            .map_err(Error::ReceiveBody)
-    }
-
-    async fn get<U: IntoUrl>(&self, uri: U) -> Result<reqwest::Response> {
-        self.request(Method::GET, uri, ()).await
-    }
-
-    async fn request<B: serde::Serialize, U: reqwest::IntoUrl>(
-        &self,
-        method: Method,
-        uri: U,
-        body: B,
-    ) -> Result<reqwest::Response> {
-        let req = self.client.request(method, uri);
-        let req = if let Some(value) = &self.authorization_header {
-            req.header(reqwest::header::AUTHORIZATION, value)
-        } else {
-            req
-        };
-        let res = req.json(&body).send().await.map_err(Error::ReceiveBody)?;
-        let response = res.error_from_body().await?;
-        Ok(response)
-    }
-
-    pub async fn status(&self) -> Result<()> {
-        let uri = format!("{}/v1/status", self.mgmt_api_endpoint);
-        self.get(&uri).await?;
-        Ok(())
-    }
-
-    pub async fn tenant_create(&self, req: &TenantCreateRequest) -> Result<TenantId> {
-        let uri = format!("{}/v1/tenant", self.mgmt_api_endpoint);
-        self.request(Method::POST, &uri, req)
-            .await?
-            .json()
-            .await
-            .map_err(Error::ReceiveBody)
-    }
-
-    pub async fn tenant_config(&self, req: &TenantConfigRequest) -> Result<()> {
-        let uri = format!("{}/v1/tenant/config", self.mgmt_api_endpoint);
-        self.request(Method::PUT, &uri, req).await?;
-        Ok(())
-    }
-
-    pub async fn location_config(
-        &self,
-        tenant_id: TenantId,
-        config: LocationConfig,
-        flush_ms: Option<std::time::Duration>,
-    ) -> Result<()> {
-        let req_body = TenantLocationConfigRequest { tenant_id, config };
-        let path = format!(
-            "{}/v1/tenant/{}/location_config",
-            self.mgmt_api_endpoint, tenant_id
-        );
-        let path = if let Some(flush_ms) = flush_ms {
-            format!("{}?flush_ms={}", path, flush_ms.as_millis())
-        } else {
-            path
-        };
-        self.request(Method::PUT, &path, &req_body).await?;
-        Ok(())
-    }
-
-    pub async fn timeline_create(
-        &self,
-        tenant_id: TenantId,
-        req: &TimelineCreateRequest,
-    ) -> Result<TimelineInfo> {
-        let uri = format!(
-            "{}/v1/tenant/{}/timeline",
-            self.mgmt_api_endpoint, tenant_id
-        );
-        self.request(Method::POST, &uri, req)
-            .await?
-            .json()
-            .await
-            .map_err(Error::ReceiveBody)
-    }
-}

pageserver/src/bin/pageserver.rs

@@ -14,7 +14,7 @@ use pageserver::control_plane_client::ControlPlaneClient;
 use pageserver::disk_usage_eviction_task::{self, launch_disk_usage_global_eviction_task};
 use pageserver::metrics::{STARTUP_DURATION, STARTUP_IS_LOADING};
 use pageserver::task_mgr::WALRECEIVER_RUNTIME;
-use pageserver::tenant::{secondary, TenantSharedResources};
+use pageserver::tenant::TenantSharedResources;
 use remote_storage::GenericRemoteStorage;
 use tokio::time::Instant;
 use tracing::*;
@@ -504,17 +504,6 @@ fn start_pageserver(
         }
     });
 
-    let secondary_controller = if let Some(remote_storage) = &remote_storage {
-        secondary::spawn_tasks(
-            tenant_manager.clone(),
-            remote_storage.clone(),
-            background_jobs_barrier.clone(),
-            shutdown_pageserver.clone(),
-        )
-    } else {
-        secondary::null_controller()
-    };
-
     // shared state between the disk-usage backed eviction background task and the http endpoint
     // that allows triggering disk-usage based eviction manually. note that the http endpoint
     // is still accessible even if background task is not configured as long as remote storage has
@@ -544,7 +533,6 @@ fn start_pageserver(
                 broker_client.clone(),
                 disk_usage_eviction_state,
                 deletion_queue.new_client(),
-                secondary_controller,
             )
             .context("Failed to initialize router state")?,
         );

pageserver/src/config.rs

@@ -41,8 +41,6 @@ use crate::{
     TIMELINE_DELETE_MARK_SUFFIX, TIMELINE_UNINIT_MARK_SUFFIX,
 };
 
-use self::defaults::DEFAULT_CONCURRENT_TENANT_WARMUP;
-
 pub mod defaults {
     use crate::tenant::config::defaults::*;
     use const_format::formatcp;
@@ -63,8 +61,6 @@ pub mod defaults {
 
     pub const DEFAULT_LOG_FORMAT: &str = "plain";
 
-    pub const DEFAULT_CONCURRENT_TENANT_WARMUP: usize = 8;
-
     pub const DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES: usize =
         super::ConfigurableSemaphore::DEFAULT_INITIAL.get();
 
@@ -74,8 +70,6 @@ pub mod defaults {
     pub const DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL: &str = "10 min";
     pub const DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY: &str = "10s";
 
-    pub const DEFAULT_HEATMAP_UPLOAD_CONCURRENCY: usize = 8;
-
     ///
     /// Default built-in configuration file.
     ///
@@ -98,7 +92,6 @@ pub mod defaults {
 #log_format = '{DEFAULT_LOG_FORMAT}'
 
 #concurrent_tenant_size_logical_size_queries = '{DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES}'
-#concurrent_tenant_warmup = '{DEFAULT_CONCURRENT_TENANT_WARMUP}'
 
 #metric_collection_interval = '{DEFAULT_METRIC_COLLECTION_INTERVAL}'
 #cached_metric_collection_interval = '{DEFAULT_CACHED_METRIC_COLLECTION_INTERVAL}'
@@ -124,8 +117,6 @@ pub mod defaults {
 #evictions_low_residence_duration_metric_threshold = '{DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD}'
 #gc_feedback = false
 
-#heatmap_upload_concurrency = {DEFAULT_HEATMAP_UPLOAD_CONCURRENCY}
-
 [remote_storage]
 
 "#
@@ -185,11 +176,6 @@ pub struct PageServerConf {
 
     pub log_format: LogFormat,
 
-    /// Number of tenants which will be concurrently loaded from remote storage proactively on startup,
-    /// does not limit tenants loaded in response to client I/O.  A lower value implicitly deprioritizes
-    /// loading such tenants, vs. other work in the system.
-    pub concurrent_tenant_warmup: ConfigurableSemaphore,
-
     /// Number of concurrent [`Tenant::gather_size_inputs`](crate::tenant::Tenant::gather_size_inputs) allowed.
     pub concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore,
     /// Limit of concurrent [`Tenant::gather_size_inputs`] issued by module `eviction_task`.
@@ -229,10 +215,6 @@ pub struct PageServerConf {
     /// If true, pageserver will make best-effort to operate without a control plane: only
     /// for use in major incidents.
     pub control_plane_emergency_mode: bool,
-
-    /// How many heatmap uploads may be done concurrency: lower values implicitly deprioritize
-    /// heatmap uploads vs. other remote storage operations.
-    pub heatmap_upload_concurrency: usize,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -293,7 +275,6 @@ struct PageServerConfigBuilder {
 
     log_format: BuilderValue<LogFormat>,
 
-    concurrent_tenant_warmup: BuilderValue<NonZeroUsize>,
     concurrent_tenant_size_logical_size_queries: BuilderValue<NonZeroUsize>,
 
     metric_collection_interval: BuilderValue<Duration>,
@@ -312,8 +293,6 @@ struct PageServerConfigBuilder {
     control_plane_api: BuilderValue<Option<Url>>,
     control_plane_api_token: BuilderValue<Option<SecretString>>,
     control_plane_emergency_mode: BuilderValue<bool>,
-
-    heatmap_upload_concurrency: BuilderValue<usize>,
 }
 
 impl Default for PageServerConfigBuilder {
@@ -351,8 +330,6 @@ impl Default for PageServerConfigBuilder {
             .expect("cannot parse default keepalive interval")),
             log_format: Set(LogFormat::from_str(DEFAULT_LOG_FORMAT).unwrap()),
 
-            concurrent_tenant_warmup: Set(NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
-                .expect("Invalid default constant")),
             concurrent_tenant_size_logical_size_queries: Set(
                 ConfigurableSemaphore::DEFAULT_INITIAL,
             ),
@@ -384,8 +361,6 @@ impl Default for PageServerConfigBuilder {
             control_plane_api: Set(None),
             control_plane_api_token: Set(None),
             control_plane_emergency_mode: Set(false),
-
-            heatmap_upload_concurrency: Set(DEFAULT_HEATMAP_UPLOAD_CONCURRENCY),
         }
     }
 }
@@ -466,10 +441,6 @@ impl PageServerConfigBuilder {
         self.log_format = BuilderValue::Set(log_format)
     }
 
-    pub fn concurrent_tenant_warmup(&mut self, u: NonZeroUsize) {
-        self.concurrent_tenant_warmup = BuilderValue::Set(u);
-    }
-
     pub fn concurrent_tenant_size_logical_size_queries(&mut self, u: NonZeroUsize) {
         self.concurrent_tenant_size_logical_size_queries = BuilderValue::Set(u);
     }
@@ -530,14 +501,7 @@ impl PageServerConfigBuilder {
         self.control_plane_emergency_mode = BuilderValue::Set(enabled)
     }
 
-    pub fn heatmap_upload_concurrency(&mut self, value: usize) {
-        self.heatmap_upload_concurrency = BuilderValue::Set(value)
-    }
-
     pub fn build(self) -> anyhow::Result<PageServerConf> {
-        let concurrent_tenant_warmup = self
-            .concurrent_tenant_warmup
-            .ok_or(anyhow!("missing concurrent_tenant_warmup"))?;
         let concurrent_tenant_size_logical_size_queries = self
             .concurrent_tenant_size_logical_size_queries
             .ok_or(anyhow!(
@@ -590,7 +554,6 @@ impl PageServerConfigBuilder {
                 .broker_keepalive_interval
                 .ok_or(anyhow!("No broker keepalive interval provided"))?,
             log_format: self.log_format.ok_or(anyhow!("missing log_format"))?,
-            concurrent_tenant_warmup: ConfigurableSemaphore::new(concurrent_tenant_warmup),
             concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::new(
                 concurrent_tenant_size_logical_size_queries,
             ),
@@ -632,10 +595,6 @@ impl PageServerConfigBuilder {
             control_plane_emergency_mode: self
                 .control_plane_emergency_mode
                 .ok_or(anyhow!("missing control_plane_emergency_mode"))?,
-
-            heatmap_upload_concurrency: self
-                .heatmap_upload_concurrency
-                .ok_or(anyhow!("missing heatmap_upload_concurrency"))?,
         })
     }
 }
@@ -828,11 +787,6 @@ impl PageServerConf {
                 "log_format" => builder.log_format(
                     LogFormat::from_config(&parse_toml_string(key, item)?)?
                 ),
-                "concurrent_tenant_warmup" => builder.concurrent_tenant_warmup({
-                    let input = parse_toml_string(key, item)?;
-                    let permits = input.parse::<usize>().context("expected a number of initial permits, not {s:?}")?;
-                    NonZeroUsize::new(permits).context("initial semaphore permits out of range: 0, use other configuration to disable a feature")?
-                }),
                 "concurrent_tenant_size_logical_size_queries" => builder.concurrent_tenant_size_logical_size_queries({
                     let input = parse_toml_string(key, item)?;
                     let permits = input.parse::<usize>().context("expected a number of initial permits, not {s:?}")?;
@@ -874,9 +828,7 @@ impl PageServerConf {
                 },
                 "control_plane_emergency_mode" => {
                     builder.control_plane_emergency_mode(parse_toml_bool(key, item)?)
-                },
-                "heatmap_upload_concurrency" => {
-                    builder.heatmap_upload_concurrency(parse_toml_u64(key, item)? as usize)
+
                 },
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
@@ -930,10 +882,6 @@ impl PageServerConf {
             broker_endpoint: storage_broker::DEFAULT_ENDPOINT.parse().unwrap(),
             broker_keepalive_interval: Duration::from_secs(5000),
             log_format: LogFormat::from_str(defaults::DEFAULT_LOG_FORMAT).unwrap(),
-            concurrent_tenant_warmup: ConfigurableSemaphore::new(
-                NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
-                    .expect("Invalid default constant"),
-            ),
             concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
             eviction_task_immitated_concurrent_logical_size_queries: ConfigurableSemaphore::default(
             ),
@@ -948,7 +896,6 @@ impl PageServerConf {
             control_plane_api: None,
             control_plane_api_token: None,
             control_plane_emergency_mode: false,
-            heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY,
         }
     }
 }
@@ -1152,9 +1099,6 @@ background_task_maximum_delay = '334 s'
                     storage_broker::DEFAULT_KEEPALIVE_INTERVAL
                 )?,
                 log_format: LogFormat::from_str(defaults::DEFAULT_LOG_FORMAT).unwrap(),
-                concurrent_tenant_warmup: ConfigurableSemaphore::new(
-                    NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP).unwrap()
-                ),
                 concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
                 eviction_task_immitated_concurrent_logical_size_queries:
                     ConfigurableSemaphore::default(),
@@ -1176,8 +1120,7 @@ background_task_maximum_delay = '334 s'
                 )?,
                 control_plane_api: None,
                 control_plane_api_token: None,
-                control_plane_emergency_mode: false,
-                heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY
+                control_plane_emergency_mode: false
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -1221,9 +1164,6 @@ background_task_maximum_delay = '334 s'
                 broker_endpoint: storage_broker::DEFAULT_ENDPOINT.parse().unwrap(),
                 broker_keepalive_interval: Duration::from_secs(5),
                 log_format: LogFormat::Json,
-                concurrent_tenant_warmup: ConfigurableSemaphore::new(
-                    NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP).unwrap()
-                ),
                 concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
                 eviction_task_immitated_concurrent_logical_size_queries:
                     ConfigurableSemaphore::default(),
@@ -1237,8 +1177,7 @@ background_task_maximum_delay = '334 s'
                 background_task_maximum_delay: Duration::from_secs(334),
                 control_plane_api: None,
                 control_plane_api_token: None,
-                control_plane_emergency_mode: false,
-                heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY
+                control_plane_emergency_mode: false
             },
             "Should be able to parse all basic config values correctly"
         );

pageserver/src/http/mod.rs

@@ -1,2 +1,4 @@
 pub mod routes;
 pub use routes::make_router;
+
+pub use pageserver_api::models;

pageserver/src/http/openapi_spec.yml

@@ -992,8 +992,8 @@ paths:
           type: string
     post:
       description: |
-        Create a timeline. Returns new timeline id on success.
-        Recreating the same timeline will succeed if the parameters match the existing timeline.
+        Create a timeline. Returns new timeline id on success.\
+        If no new timeline id is specified in parameters, it would be generated. It's an error to recreate the same timeline.
         If no pg_version is specified, assume DEFAULT_PG_VERSION hardcoded in the pageserver.
       requestBody:
         content:
@@ -1405,8 +1405,6 @@ components:
           type: integer
         trace_read_requests:
           type: boolean
-        heatmap_period:
-          type: integer
     TenantConfigResponse:
       type: object
       properties:

pageserver/src/http/routes.rs

@@ -28,18 +28,20 @@ use utils::http::endpoint::request_span;
 use utils::http::json::json_request_or_empty_body;
 use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
 
+use super::models::{
+    StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
+    TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
+};
 use crate::context::{DownloadBehavior, RequestContext};
 use crate::deletion_queue::DeletionQueueClient;
 use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::task_mgr::TaskKind;
 use crate::tenant::config::{LocationConf, TenantConfOpt};
-use crate::tenant::mgr::GetActiveTenantError;
 use crate::tenant::mgr::{
     GetTenantError, SetNewTenantConfigError, TenantManager, TenantMapError, TenantMapInsertError,
     TenantSlotError, TenantSlotUpsertError, TenantStateError,
 };
-use crate::tenant::secondary::SecondaryController;
 use crate::tenant::size::ModelInputs;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
 use crate::tenant::timeline::CompactFlags;
@@ -47,10 +49,6 @@ use crate::tenant::timeline::Timeline;
 use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError, TenantSharedResources};
 use crate::{config::PageServerConf, tenant::mgr};
 use crate::{disk_usage_eviction_task, tenant};
-use pageserver_api::models::{
-    StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
-    TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
-};
 use utils::{
     auth::SwappableJwtAuth,
     generation::Generation,
@@ -66,12 +64,7 @@ use utils::{
 };
 
 // Imports only used for testing APIs
-use pageserver_api::models::ConfigureFailpointsRequest;
-
-// For APIs that require an Active tenant, how long should we block waiting for that state?
-// This is not functionally necessary (clients will retry), but avoids generating a lot of
-// failed API calls while tenants are activating.
-const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
+use super::models::ConfigureFailpointsRequest;
 
 pub struct State {
     conf: &'static PageServerConf,
@@ -82,11 +75,9 @@ pub struct State {
     broker_client: storage_broker::BrokerClientChannel,
     disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
     deletion_queue_client: DeletionQueueClient,
-    secondary_controller: SecondaryController,
 }
 
 impl State {
-    #[allow(clippy::too_many_arguments)]
     pub fn new(
         conf: &'static PageServerConf,
         tenant_manager: Arc<TenantManager>,
@@ -95,7 +86,6 @@ impl State {
         broker_client: storage_broker::BrokerClientChannel,
         disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
         deletion_queue_client: DeletionQueueClient,
-        secondary_controller: SecondaryController,
     ) -> anyhow::Result<Self> {
         let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml", "/metrics"]
             .iter()
@@ -110,7 +100,6 @@ impl State {
             broker_client,
             disk_usage_eviction_state,
             deletion_queue_client,
-            secondary_controller,
         })
     }
 
@@ -239,19 +228,6 @@ impl From<GetTenantError> for ApiError {
     }
 }
 
-impl From<GetActiveTenantError> for ApiError {
-    fn from(e: GetActiveTenantError) -> ApiError {
-        match e {
-            GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
-            GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
-            GetActiveTenantError::NotFound(gte) => gte.into(),
-            GetActiveTenantError::WaitForActiveTimeout { .. } => {
-                ApiError::ResourceUnavailable(format!("{}", e).into())
-            }
-        }
-    }
-}
-
 impl From<SetNewTenantConfigError> for ApiError {
     fn from(e: SetNewTenantConfigError) -> ApiError {
         match e {
@@ -454,10 +430,7 @@ async fn timeline_create_handler(
     let state = get_state(&request);
 
     async {
-        let tenant = state.tenant_manager.get_attached_tenant_shard(tenant_shard_id, false)?;
-
-        tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
-
+        let tenant = state.tenant_manager.get_attached_tenant_shard(tenant_shard_id, true)?;
         match tenant.create_timeline(
             new_timeline_id,
             request_data.ancestor_timeline_id.map(TimelineId::from),
@@ -475,7 +448,7 @@ async fn timeline_create_handler(
                     .map_err(ApiError::InternalServerError)?;
                 json_response(StatusCode::CREATED, timeline_info)
             }
-            Err(tenant::CreateTimelineError::Conflict | tenant::CreateTimelineError::AlreadyCreating) => {
+            Err(tenant::CreateTimelineError::AlreadyExists) => {
                 json_response(StatusCode::CONFLICT, ())
             }
             Err(tenant::CreateTimelineError::AncestorLsn(err)) => {
@@ -716,23 +689,11 @@ async fn timeline_delete_handler(
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
 
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
     let state = get_state(&request);
 
-    let tenant = state
-        .tenant_manager
-        .get_attached_tenant_shard(tenant_shard_id, false)
-        .map_err(|e| {
-            match e {
-                // GetTenantError has a built-in conversion to ApiError, but in this context we don't
-                // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
-                GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
-                    "Requested tenant is missing".to_string().into_boxed_str(),
-                ),
-                e => e.into(),
-            }
-        })?;
-    tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
-    tenant.delete_timeline(timeline_id).instrument(info_span!("timeline_delete", tenant_id=%tenant_shard_id.tenant_id, shard=%tenant_shard_id.shard_slug(), %timeline_id))
+    state.tenant_manager.delete_timeline(tenant_shard_id, timeline_id, &ctx)
+        .instrument(info_span!("timeline_delete", tenant_id=%tenant_shard_id.tenant_id, shard=%tenant_shard_id.shard_slug(), %timeline_id))
         .await?;
 
     json_response(StatusCode::ACCEPTED, ())
@@ -1170,10 +1131,7 @@ async fn tenant_create_handler(
 
     // We created the tenant. Existing API semantics are that the tenant
     // is Active when this function returns.
-    if let res @ Err(_) = new_tenant
-        .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
-        .await
-    {
+    if let res @ Err(_) = new_tenant.wait_to_become_active().await {
         // This shouldn't happen because we just created the tenant directory
         // in tenant::mgr::create_tenant, and there aren't any remote timelines
         // to load, so, nothing can really fail during load.
@@ -1524,6 +1482,69 @@ async fn timeline_collect_keyspace(
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
 
+    struct Partitioning {
+        keys: crate::keyspace::KeySpace,
+
+        at_lsn: Lsn,
+    }
+
+    impl serde::Serialize for Partitioning {
+        fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
+        where
+            S: serde::Serializer,
+        {
+            use serde::ser::SerializeMap;
+            let mut map = serializer.serialize_map(Some(2))?;
+            map.serialize_key("keys")?;
+            map.serialize_value(&KeySpace(&self.keys))?;
+            map.serialize_key("at_lsn")?;
+            map.serialize_value(&WithDisplay(&self.at_lsn))?;
+            map.end()
+        }
+    }
+
+    struct WithDisplay<'a, T>(&'a T);
+
+    impl<'a, T: std::fmt::Display> serde::Serialize for WithDisplay<'a, T> {
+        fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
+        where
+            S: serde::Serializer,
+        {
+            serializer.collect_str(&self.0)
+        }
+    }
+
+    struct KeySpace<'a>(&'a crate::keyspace::KeySpace);
+
+    impl<'a> serde::Serialize for KeySpace<'a> {
+        fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
+        where
+            S: serde::Serializer,
+        {
+            use serde::ser::SerializeSeq;
+            let mut seq = serializer.serialize_seq(Some(self.0.ranges.len()))?;
+            for kr in &self.0.ranges {
+                seq.serialize_element(&KeyRange(kr))?;
+            }
+            seq.end()
+        }
+    }
+
+    struct KeyRange<'a>(&'a std::ops::Range<crate::repository::Key>);
+
+    impl<'a> serde::Serialize for KeyRange<'a> {
+        fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+        where
+            S: serde::Serializer,
+        {
+            use serde::ser::SerializeTuple;
+            let mut t = serializer.serialize_tuple(2)?;
+            t.serialize_element(&WithDisplay(&self.0.start))?;
+            t.serialize_element(&WithDisplay(&self.0.end))?;
+            t.end()
+        }
+    }
+
     let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
 
     async {
@@ -1535,9 +1556,7 @@ async fn timeline_collect_keyspace(
             .await
             .map_err(|e| ApiError::InternalServerError(e.into()))?;
 
-        let res = pageserver_api::models::partitioning::Partitioning { keys, at_lsn };
-
-        json_response(StatusCode::OK, res)
+        json_response(StatusCode::OK, Partitioning { keys, at_lsn })
     }
     .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
     .await
@@ -1597,7 +1616,9 @@ async fn disk_usage_eviction_run(
         }
     }
 
-    let config = json_request::<Config>(&mut r).await?;
+    let config = json_request::<Config>(&mut r)
+        .await
+        .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
 
     let usage = Usage {
         config,
@@ -1626,21 +1647,6 @@ async fn disk_usage_eviction_run(
     json_response(StatusCode::OK, res)
 }
 
-async fn secondary_upload_handler(
-    request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let state = get_state(&request);
-    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
-    state
-        .secondary_controller
-        .upload_tenant(tenant_shard_id)
-        .await
-        .map_err(ApiError::InternalServerError)?;
-
-    json_response(StatusCode::OK, ())
-}
-
 async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
     json_response(
         StatusCode::NOT_FOUND,
@@ -1900,9 +1906,6 @@ pub fn make_router(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
             |r| api_handler(r, evict_timeline_layer_handler),
         )
-        .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
-            api_handler(r, secondary_upload_handler)
-        })
         .put("/v1/disk_usage_eviction/run", |r| {
             api_handler(r, disk_usage_eviction_run)
         })

pageserver/src/keyspace.rs

@@ -1,12 +1,11 @@
+use crate::repository::{key_range_size, singleton_range, Key};
 use postgres_ffi::BLCKSZ;
 use std::ops::Range;
 
-use crate::key::Key;
-
 ///
 /// Represents a set of Keys, in a compact form.
 ///
-#[derive(Clone, Debug, Default, PartialEq, Eq)]
+#[derive(Clone, Debug, Default)]
 pub struct KeySpace {
     /// Contiguous ranges of keys that belong to the key space. In key order,
     /// and with no overlap.
@@ -187,33 +186,6 @@ impl KeySpaceRandomAccum {
     }
 }
 
-pub fn key_range_size(key_range: &Range<Key>) -> u32 {
-    let start = key_range.start;
-    let end = key_range.end;
-
-    if end.field1 != start.field1
-        || end.field2 != start.field2
-        || end.field3 != start.field3
-        || end.field4 != start.field4
-    {
-        return u32::MAX;
-    }
-
-    let start = (start.field5 as u64) << 32 | start.field6 as u64;
-    let end = (end.field5 as u64) << 32 | end.field6 as u64;
-
-    let diff = end - start;
-    if diff > u32::MAX as u64 {
-        u32::MAX
-    } else {
-        diff as u32
-    }
-}
-
-pub fn singleton_range(key: Key) -> Range<Key> {
-    key..key.next()
-}
-
 #[cfg(test)]
 mod tests {
     use super::*;

pageserver/src/lib.rs

@@ -10,7 +10,7 @@ pub mod deletion_queue;
 pub mod disk_usage_eviction_task;
 pub mod http;
 pub mod import_datadir;
-pub use pageserver_api::keyspace;
+pub mod keyspace;
 pub mod metrics;
 pub mod page_cache;
 pub mod page_service;

pageserver/src/metrics.rs

@@ -2,10 +2,9 @@ use enum_map::EnumMap;
 use metrics::metric_vec_duration::DurationResultObserver;
 use metrics::{
     register_counter_vec, register_gauge_vec, register_histogram, register_histogram_vec,
-    register_int_counter, register_int_counter_pair_vec, register_int_counter_vec,
-    register_int_gauge, register_int_gauge_vec, register_uint_gauge, register_uint_gauge_vec,
-    Counter, CounterVec, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPairVec,
-    IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
+    register_int_counter, register_int_counter_vec, register_int_gauge, register_int_gauge_vec,
+    register_uint_gauge, register_uint_gauge_vec, Counter, CounterVec, GaugeVec, Histogram,
+    HistogramVec, IntCounter, IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
 };
 use once_cell::sync::Lazy;
 use pageserver_api::shard::TenantShardId;
@@ -522,18 +521,14 @@ pub(crate) mod initial_logical_size {
     impl StartCalculation {
         pub(crate) fn first(&self, circumstances: StartCircumstances) -> OngoingCalculationGuard {
             let circumstances_label: &'static str = circumstances.into();
-            self.0
-                .with_label_values(&["first", circumstances_label])
-                .inc();
+            self.0.with_label_values(&["first", circumstances_label]);
             OngoingCalculationGuard {
                 inc_drop_calculation: Some(DROP_CALCULATION.first.clone()),
             }
         }
         pub(crate) fn retry(&self, circumstances: StartCircumstances) -> OngoingCalculationGuard {
             let circumstances_label: &'static str = circumstances.into();
-            self.0
-                .with_label_values(&["retry", circumstances_label])
-                .inc();
+            self.0.with_label_values(&["retry", circumstances_label]);
             OngoingCalculationGuard {
                 inc_drop_calculation: Some(DROP_CALCULATION.retry.clone()),
             }
@@ -688,54 +683,14 @@ pub static STARTUP_IS_LOADING: Lazy<UIntGauge> = Lazy::new(|| {
     .expect("Failed to register pageserver_startup_is_loading")
 });
 
-/// Metrics related to the lifecycle of a [`crate::tenant::Tenant`] object: things
-/// like how long it took to load.
-///
-/// Note that these are process-global metrics, _not_ per-tenant metrics.  Per-tenant
-/// metrics are rather expensive, and usually fine grained stuff makes more sense
-/// at a timeline level than tenant level.
-pub(crate) struct TenantMetrics {
-    /// How long did tenants take to go from construction to active state?
-    pub(crate) activation: Histogram,
-    pub(crate) preload: Histogram,
-    pub(crate) attach: Histogram,
-
-    /// How many tenants are included in the initial startup of the pagesrever?
-    pub(crate) startup_scheduled: IntCounter,
-    pub(crate) startup_complete: IntCounter,
-}
-
-pub(crate) static TENANT: Lazy<TenantMetrics> = Lazy::new(|| {
-    TenantMetrics {
-    activation: register_histogram!(
+/// How long did tenants take to go from construction to active state?
+pub(crate) static TENANT_ACTIVATION: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
         "pageserver_tenant_activation_seconds",
         "Time taken by tenants to activate, in seconds",
         CRITICAL_OP_BUCKETS.into()
     )
-    .expect("Failed to register metric"),
-    preload: register_histogram!(
-        "pageserver_tenant_preload_seconds",
-        "Time taken by tenants to load remote metadata on startup/attach, in seconds",
-        CRITICAL_OP_BUCKETS.into()
-    )
-    .expect("Failed to register metric"),
-    attach: register_histogram!(
-        "pageserver_tenant_attach_seconds",
-        "Time taken by tenants to intialize, after remote metadata is already loaded",
-        CRITICAL_OP_BUCKETS.into()
-    )
-    .expect("Failed to register metric"),
-    startup_scheduled: register_int_counter!(
-        "pageserver_tenant_startup_scheduled",
-        "Number of tenants included in pageserver startup (doesn't count tenants attached later)"
-    ).expect("Failed to register metric"),
-    startup_complete: register_int_counter!(
-        "pageserver_tenant_startup_complete",
-        "Number of tenants that have completed warm-up, or activated on-demand during initial startup: \
-         should eventually reach `pageserver_tenant_startup_scheduled_total`.  Does not include broken \
-         tenants: such cases will lead to this metric never reaching the scheduled count."
-    ).expect("Failed to register metric"),
-}
+    .expect("Failed to register pageserver_tenant_activation_seconds metric")
 });
 
 /// Each `Timeline`'s  [`EVICTIONS_WITH_LOW_RESIDENCE_DURATION`] metric.
@@ -1023,62 +978,12 @@ static SMGR_QUERY_TIME_PER_TENANT_TIMELINE: Lazy<HistogramVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-static SMGR_QUERY_TIME_GLOBAL_BUCKETS: Lazy<Vec<f64>> = Lazy::new(|| {
-    [
-        1,
-        10,
-        20,
-        40,
-        60,
-        80,
-        100,
-        200,
-        300,
-        400,
-        500,
-        600,
-        700,
-        800,
-        900,
-        1_000, // 1ms
-        2_000,
-        4_000,
-        6_000,
-        8_000,
-        10_000, // 10ms
-        20_000,
-        40_000,
-        60_000,
-        80_000,
-        100_000,
-        200_000,
-        400_000,
-        600_000,
-        800_000,
-        1_000_000, // 1s
-        2_000_000,
-        4_000_000,
-        6_000_000,
-        8_000_000,
-        10_000_000, // 10s
-        20_000_000,
-        50_000_000,
-        100_000_000,
-        200_000_000,
-        1_000_000_000, // 1000s
-    ]
-    .into_iter()
-    .map(Duration::from_micros)
-    .map(|d| d.as_secs_f64())
-    .collect()
-});
-
 static SMGR_QUERY_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
     register_histogram_vec!(
         "pageserver_smgr_query_seconds_global",
         "Time spent on smgr query handling, aggregated by query type.",
         &["smgr_query_type"],
-        SMGR_QUERY_TIME_GLOBAL_BUCKETS.clone(),
+        CRITICAL_OP_BUCKETS.into(),
     )
     .expect("failed to define a metric")
 });
@@ -1365,28 +1270,6 @@ pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMet
     )
     .expect("failed to define a metric"),
 });
-pub(crate) struct SecondaryModeMetrics {
-    pub(crate) upload_heatmap: IntCounter,
-    pub(crate) upload_heatmap_errors: IntCounter,
-    pub(crate) upload_heatmap_duration: Histogram,
-}
-pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| SecondaryModeMetrics {
-    upload_heatmap: register_int_counter!(
-        "pageserver_secondary_upload_heatmap",
-        "Number of heatmaps written to remote storage by attached tenants"
-    )
-    .expect("failed to define a metric"),
-    upload_heatmap_errors: register_int_counter!(
-        "pageserver_secondary_upload_heatmap_errors",
-        "Failures writing heatmap to remote storage"
-    )
-    .expect("failed to define a metric"),
-    upload_heatmap_duration: register_histogram!(
-        "pageserver_secondary_upload_heatmap_duration",
-        "Time to build and upload a heatmap, including any waiting inside the S3 client"
-    )
-    .expect("failed to define a metric"),
-});
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 pub enum RemoteOpKind {
@@ -1438,16 +1321,25 @@ pub(crate) static TENANT_TASK_EVENTS: Lazy<IntCounterVec> = Lazy::new(|| {
     .expect("Failed to register tenant_task_events metric")
 });
 
-pub(crate) static BACKGROUND_LOOP_SEMAPHORE_WAIT_GAUGE: Lazy<IntCounterPairVec> = Lazy::new(|| {
-    register_int_counter_pair_vec!(
-        "pageserver_background_loop_semaphore_wait_start_count",
-        "Counter for background loop concurrency-limiting semaphore acquire calls started",
-        "pageserver_background_loop_semaphore_wait_finish_count",
-        "Counter for background loop concurrency-limiting semaphore acquire calls finished",
-        &["task"],
-    )
-    .unwrap()
-});
+pub(crate) static BACKGROUND_LOOP_SEMAPHORE_WAIT_START_COUNT: Lazy<IntCounterVec> =
+    Lazy::new(|| {
+        register_int_counter_vec!(
+            "pageserver_background_loop_semaphore_wait_start_count",
+            "Counter for background loop concurrency-limiting semaphore acquire calls started",
+            &["task"],
+        )
+        .unwrap()
+    });
+
+pub(crate) static BACKGROUND_LOOP_SEMAPHORE_WAIT_FINISH_COUNT: Lazy<IntCounterVec> =
+    Lazy::new(|| {
+        register_int_counter_vec!(
+            "pageserver_background_loop_semaphore_wait_finish_count",
+            "Counter for background loop concurrency-limiting semaphore acquire calls finished",
+            &["task"],
+        )
+        .unwrap()
+    });
 
 pub(crate) static BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT: Lazy<IntCounterVec> = Lazy::new(|| {
     register_int_counter_vec!(
@@ -2307,9 +2199,6 @@ pub fn preinitialize_metrics() {
     // Deletion queue stats
     Lazy::force(&DELETION_QUEUE);
 
-    // Tenant stats
-    Lazy::force(&TENANT);
-
     // Tenant manager stats
     Lazy::force(&TENANT_MANAGER);
 

pageserver/src/repository.rs

@@ -2,11 +2,38 @@ use crate::walrecord::NeonWalRecord;
 use anyhow::Result;
 use bytes::Bytes;
 use serde::{Deserialize, Serialize};
-use std::ops::AddAssign;
+use std::ops::{AddAssign, Range};
 use std::time::Duration;
 
 pub use pageserver_api::key::{Key, KEY_SIZE};
 
+pub fn key_range_size(key_range: &Range<Key>) -> u32 {
+    let start = key_range.start;
+    let end = key_range.end;
+
+    if end.field1 != start.field1
+        || end.field2 != start.field2
+        || end.field3 != start.field3
+        || end.field4 != start.field4
+    {
+        return u32::MAX;
+    }
+
+    let start = (start.field5 as u64) << 32 | start.field6 as u64;
+    let end = (end.field5 as u64) << 32 | end.field6 as u64;
+
+    let diff = end - start;
+    if diff > u32::MAX as u64 {
+        u32::MAX
+    } else {
+        diff as u32
+    }
+}
+
+pub fn singleton_range(key: Key) -> Range<Key> {
+    key..key.next()
+}
+
 /// A 'value' stored for a one Key.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 #[cfg_attr(test, derive(PartialEq))]

pageserver/src/task_mgr.rs

@@ -258,9 +258,6 @@ pub enum TaskKind {
     /// See [`crate::disk_usage_eviction_task`].
     DiskUsageEviction,
 
-    /// See [`crate::tenant::secondary`].
-    SecondaryUploads,
-
     // Initial logical size calculation
     InitialLogicalSizeCalculation,
 
@@ -561,14 +558,9 @@ pub async fn shutdown_watcher() {
 /// cancelled. It can however be moved to other tasks, such as `tokio::task::spawn_blocking` or
 /// `tokio::task::JoinSet::spawn`.
 pub fn shutdown_token() -> CancellationToken {
-    let res = SHUTDOWN_TOKEN.try_with(|t| t.clone());
-
-    if cfg!(test) {
-        // in tests this method is called from non-taskmgr spawned tasks, and that is all ok.
-        res.unwrap_or_default()
-    } else {
-        res.expect("shutdown_token() called in an unexpected task or thread")
-    }
+    SHUTDOWN_TOKEN
+        .try_with(|t| t.clone())
+        .expect("shutdown_token() called in an unexpected task or thread")
 }
 
 /// Has the current task been requested to shut down?

pageserver/src/tenant.rs

@@ -36,8 +36,6 @@ use utils::crashsafe::path_with_suffix_extension;
 use utils::fs_ext;
 use utils::sync::gate::Gate;
 use utils::sync::gate::GateGuard;
-use utils::timeout::timeout_cancellable;
-use utils::timeout::TimeoutCancellableError;
 
 use self::config::AttachedLocationConfig;
 use self::config::AttachmentMode;
@@ -50,7 +48,6 @@ use self::mgr::GetActiveTenantError;
 use self::mgr::GetTenantError;
 use self::mgr::TenantsMap;
 use self::remote_timeline_client::RemoteTimelineClient;
-use self::timeline::uninit::TimelineExclusionError;
 use self::timeline::uninit::TimelineUninitMark;
 use self::timeline::uninit::UninitializedTimeline;
 use self::timeline::EvictionTaskTenantState;
@@ -61,7 +58,7 @@ use crate::deletion_queue::DeletionQueueClient;
 use crate::deletion_queue::DeletionQueueError;
 use crate::import_datadir;
 use crate::is_uninit_mark;
-use crate::metrics::TENANT;
+use crate::metrics::TENANT_ACTIVATION;
 use crate::metrics::{remove_tenant_metrics, TENANT_STATE_METRIC, TENANT_SYNTHETIC_SIZE_METRIC};
 use crate::repository::GcResult;
 use crate::task_mgr;
@@ -90,6 +87,7 @@ use std::process::Stdio;
 use std::sync::atomic::AtomicU64;
 use std::sync::atomic::Ordering;
 use std::sync::Arc;
+use std::sync::MutexGuard;
 use std::sync::{Mutex, RwLock};
 use std::time::{Duration, Instant};
 
@@ -146,7 +144,6 @@ pub mod storage_layer;
 pub mod config;
 pub mod delete;
 pub mod mgr;
-pub mod secondary;
 pub mod tasks;
 pub mod upload_queue;
 
@@ -228,7 +225,7 @@ pub struct Tenant {
 
     /// The value creation timestamp, used to measure activation delay, see:
     /// <https://github.com/neondatabase/neon/issues/4025>
-    constructed_at: Instant,
+    loading_started_at: Instant,
 
     state: watch::Sender<TenantState>,
 
@@ -251,12 +248,6 @@ pub struct Tenant {
     generation: Generation,
 
     timelines: Mutex<HashMap<TimelineId, Arc<Timeline>>>,
-
-    /// During timeline creation, we first insert the TimelineId to the
-    /// creating map, then `timelines`, then remove it from the creating map.
-    /// **Lock order**: if acquring both, acquire`timelines` before `timelines_creating`
-    timelines_creating: std::sync::Mutex<HashSet<TimelineId>>,
-
     // This mutex prevents creation of new timelines during GC.
     // Adding yet another mutex (in addition to `timelines`) is needed because holding
     // `timelines` mutex during all GC iteration
@@ -278,11 +269,6 @@ pub struct Tenant {
 
     eviction_task_tenant_state: tokio::sync::Mutex<EvictionTaskTenantState>,
 
-    /// If the tenant is in Activating state, notify this to encourage it
-    /// to proceed to Active as soon as possible, rather than waiting for lazy
-    /// background warmup.
-    pub(crate) activate_now_sem: tokio::sync::Semaphore,
-
     pub(crate) delete_progress: Arc<tokio::sync::Mutex<DeleteTenantFlow>>,
 
     // Cancellation token fires when we have entered shutdown().  This is a parent of
@@ -420,10 +406,8 @@ impl Debug for SetStoppingError {
 
 #[derive(thiserror::Error, Debug)]
 pub enum CreateTimelineError {
-    #[error("creation of timeline with the given ID is in progress")]
-    AlreadyCreating,
-    #[error("timeline already exists with different parameters")]
-    Conflict,
+    #[error("a timeline with the given ID already exists")]
+    AlreadyExists,
     #[error(transparent)]
     AncestorLsn(anyhow::Error),
     #[error("ancestor timeline is not active")]
@@ -629,14 +613,6 @@ impl Tenant {
             "attach tenant",
             false,
             async move {
-                // Is this tenant being spawned as part of process startup?
-                let starting_up = init_order.is_some();
-                scopeguard::defer! {
-                    if starting_up {
-                        TENANT.startup_complete.inc();
-                    }
-                }
-
                 // Ideally we should use Tenant::set_broken_no_wait, but it is not supposed to be used when tenant is in loading state.
                 let make_broken =
                     |t: &Tenant, err: anyhow::Error| {
@@ -663,62 +639,8 @@ impl Tenant {
                     .as_mut()
                     .and_then(|x| x.initial_tenant_load_remote.take());
 
-                enum AttachType<'a> {
-                    // During pageserver startup, we are attaching this tenant lazily in the background
-                    Warmup(tokio::sync::SemaphorePermit<'a>),
-                    // During pageserver startup, we are attaching this tenant as soon as we can,
-                    // because a client tried to access it.
-                    OnDemand,
-                    // During normal operations after startup, we are attaching a tenant.
-                    Normal,
-                }
-
-                // Before doing any I/O, wait for either or:
-                // - A client to attempt to access to this tenant (on-demand loading)
-                // - A permit to become available in the warmup semaphore (background warmup)
-                //
-                // Some-ness of init_order is how we know if we're attaching during startup or later
-                // in process lifetime.
-                let attach_type = if init_order.is_some() {
-                    tokio::select!(
-                        _ = tenant_clone.activate_now_sem.acquire() => {
-                            tracing::info!("Activating tenant (on-demand)");
-                            AttachType::OnDemand
-                        },
-                        permit_result = conf.concurrent_tenant_warmup.inner().acquire() => {
-                            match permit_result {
-                                Ok(p) => {
-                                    tracing::info!("Activating tenant (warmup)");
-                                    AttachType::Warmup(p)
-                                }
-                                Err(_) => {
-                                    // This is unexpected: the warmup semaphore should stay alive
-                                    // for the lifetime of init_order.  Log a warning and proceed.
-                                    tracing::warn!("warmup_limit semaphore unexpectedly closed");
-                                    AttachType::Normal
-                                }
-                            }
-
-                        }
-                        _ = tenant_clone.cancel.cancelled() => {
-                            // This is safe, but should be pretty rare: it is interesting if a tenant
-                            // stayed in Activating for such a long time that shutdown found it in
-                            // that state.
-                            tracing::info!(state=%tenant_clone.current_state(), "Tenant shut down before activation");
-                            return Ok(());
-                        },
-                    )
-                } else {
-                    AttachType::Normal
-                };
-
-                let preload_timer = TENANT.preload.start_timer();
                 let preload = match mode {
-                    SpawnMode::Create => {
-                        // Don't count the skipped preload into the histogram of preload durations
-                        preload_timer.stop_and_discard();
-                        None
-                    },
+                    SpawnMode::Create => {None},
                     SpawnMode::Normal => {
                         match &remote_storage {
                             Some(remote_storage) => Some(
@@ -728,11 +650,7 @@ impl Tenant {
                                         tracing::info_span!(parent: None, "attach_preload", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()),
                                     )
                                     .await {
-                                        Ok(p) => {
-                                            preload_timer.observe_duration();
-                                            p
-                                        }
-                                            ,
+                                        Ok(p) => p,
                                         Err(e) => {
                                             make_broken(&tenant_clone, anyhow::anyhow!(e));
                                                 return Ok(());
@@ -794,43 +712,15 @@ impl Tenant {
                     }
                 }
 
-                // We will time the duration of the attach phase unless this is a creation (attach will do no work)
-                let attach_timer = match mode {
-                    SpawnMode::Create => None,
-                    SpawnMode::Normal => {Some(TENANT.attach.start_timer())}
-                };
                 match tenant_clone.attach(preload, &ctx).await {
                     Ok(()) => {
                         info!("attach finished, activating");
-                        if let Some(t)=  attach_timer {t.observe_duration();}
                         tenant_clone.activate(broker_client, None, &ctx);
                     }
                     Err(e) => {
-                        if let Some(t)=  attach_timer {t.observe_duration();}
                         make_broken(&tenant_clone, anyhow::anyhow!(e));
                     }
                 }
-
-                // If we are doing an opportunistic warmup attachment at startup, initialize
-                // logical size at the same time.  This is better than starting a bunch of idle tenants
-                // with cold caches and then coming back later to initialize their logical sizes.
-                //
-                // It also prevents the warmup proccess competing with the concurrency limit on
-                // logical size calculations: if logical size calculation semaphore is saturated,
-                // then warmup will wait for that before proceeding to the next tenant.
-                if let AttachType::Warmup(_permit) = attach_type {
-                    let mut futs = FuturesUnordered::new();
-                    let timelines: Vec<_> = tenant_clone.timelines.lock().unwrap().values().cloned().collect();
-                    for t in timelines {
-                        futs.push(t.await_initial_logical_size())
-                    }
-                    tracing::info!("Waiting for initial logical sizes while warming up...");
-                    while futs.next().await.is_some() {
-
-                    }
-                    tracing::info!("Warm-up complete");
-                }
-
                 Ok(())
             }
             .instrument({
@@ -1567,7 +1457,7 @@ impl Tenant {
     /// For tests, use `DatadirModification::init_empty_test_timeline` + `commit` to setup the
     /// minimum amount of keys required to get a writable timeline.
     /// (Without it, `put` might fail due to `repartition` failing.)
-    pub(crate) async fn create_empty_timeline(
+    pub async fn create_empty_timeline(
         &self,
         new_timeline_id: TimelineId,
         initdb_lsn: Lsn,
@@ -1579,7 +1469,10 @@ impl Tenant {
             "Cannot create empty timelines on inactive tenant"
         );
 
-        let timeline_uninit_mark = self.create_timeline_uninit_mark(new_timeline_id)?;
+        let timeline_uninit_mark = {
+            let timelines = self.timelines.lock().unwrap();
+            self.create_timeline_uninit_mark(new_timeline_id, &timelines)?
+        };
         let new_metadata = TimelineMetadata::new(
             // Initialize disk_consistent LSN to 0, The caller must import some data to
             // make it valid, before calling finish_creation()
@@ -1656,7 +1549,7 @@ impl Tenant {
     /// If the caller specified the timeline ID to use (`new_timeline_id`), and timeline with
     /// the same timeline ID already exists, returns CreateTimelineError::AlreadyExists.
     #[allow(clippy::too_many_arguments)]
-    pub(crate) async fn create_timeline(
+    pub async fn create_timeline(
         &self,
         new_timeline_id: TimelineId,
         ancestor_timeline_id: Option<TimelineId>,
@@ -1677,51 +1570,26 @@ impl Tenant {
             .enter()
             .map_err(|_| CreateTimelineError::ShuttingDown)?;
 
-        // Get exclusive access to the timeline ID: this ensures that it does not already exist,
-        // and that no other creation attempts will be allowed in while we are working.  The
-        // uninit_mark is a guard.
-        let uninit_mark = match self.create_timeline_uninit_mark(new_timeline_id) {
-            Ok(m) => m,
-            Err(TimelineExclusionError::AlreadyCreating) => {
-                // Creation is in progress, we cannot create it again, and we cannot
-                // check if this request matches the existing one, so caller must try
-                // again later.
-                return Err(CreateTimelineError::AlreadyCreating);
-            }
-            Err(TimelineExclusionError::Other(e)) => {
-                return Err(CreateTimelineError::Other(e));
-            }
-            Err(TimelineExclusionError::AlreadyExists(existing)) => {
-                debug!("timeline {new_timeline_id} already exists");
+        if let Ok(existing) = self.get_timeline(new_timeline_id, false) {
+            debug!("timeline {new_timeline_id} already exists");
 
-                // Idempotency: creating the same timeline twice is not an error, unless
-                // the second creation has different parameters.
-                if existing.get_ancestor_timeline_id() != ancestor_timeline_id
-                    || existing.pg_version != pg_version
-                    || (ancestor_start_lsn.is_some()
-                        && ancestor_start_lsn != Some(existing.get_ancestor_lsn()))
-                {
-                    return Err(CreateTimelineError::Conflict);
-                }
-
-                if let Some(remote_client) = existing.remote_client.as_ref() {
-                    // Wait for uploads to complete, so that when we return Ok, the timeline
-                    // is known to be durable on remote storage. Just like we do at the end of
-                    // this function, after we have created the timeline ourselves.
-                    //
-                    // We only really care that the initial version of `index_part.json` has
-                    // been uploaded. That's enough to remember that the timeline
-                    // exists. However, there is no function to wait specifically for that so
-                    // we just wait for all in-progress uploads to finish.
-                    remote_client
-                        .wait_completion()
-                        .await
-                        .context("wait for timeline uploads to complete")?;
-                }
-
-                return Ok(existing);
+            if let Some(remote_client) = existing.remote_client.as_ref() {
+                // Wait for uploads to complete, so that when we return Ok, the timeline
+                // is known to be durable on remote storage. Just like we do at the end of
+                // this function, after we have created the timeline ourselves.
+                //
+                // We only really care that the initial version of `index_part.json` has
+                // been uploaded. That's enough to remember that the timeline
+                // exists. However, there is no function to wait specifically for that so
+                // we just wait for all in-progress uploads to finish.
+                remote_client
+                    .wait_completion()
+                    .await
+                    .context("wait for timeline uploads to complete")?;
             }
-        };
+
+            return Err(CreateTimelineError::AlreadyExists);
+        }
 
         let loaded_timeline = match ancestor_timeline_id {
             Some(ancestor_timeline_id) => {
@@ -1758,32 +1626,18 @@ impl Tenant {
                     ancestor_timeline.wait_lsn(*lsn, ctx).await?;
                 }
 
-                self.branch_timeline(
-                    &ancestor_timeline,
-                    new_timeline_id,
-                    ancestor_start_lsn,
-                    uninit_mark,
-                    ctx,
-                )
-                .await?
+                self.branch_timeline(&ancestor_timeline, new_timeline_id, ancestor_start_lsn, ctx)
+                    .await?
             }
             None => {
-                self.bootstrap_timeline(
-                    new_timeline_id,
-                    pg_version,
-                    load_existing_initdb,
-                    uninit_mark,
-                    ctx,
-                )
-                .await?
+                self.bootstrap_timeline(new_timeline_id, pg_version, load_existing_initdb, ctx)
+                    .await?
             }
         };
 
-        // At this point we have dropped our guard on [`Self::timelines_creating`], and
-        // the timeline is visible in [`Self::timelines`], but it is _not_ durable yet.  We must
-        // not send a success to the caller until it is.  The same applies to handling retries,
-        // see the handling of [`TimelineExclusionError::AlreadyExists`] above.
         if let Some(remote_client) = loaded_timeline.remote_client.as_ref() {
+            // Wait for the upload of the 'index_part.json` file to finish, so that when we return
+            // Ok, the timeline is durable in remote storage.
             let kind = ancestor_timeline_id
                 .map(|_| "branched")
                 .unwrap_or("bootstrapped");
@@ -1797,15 +1651,6 @@ impl Tenant {
         Ok(loaded_timeline)
     }
 
-    pub(crate) async fn delete_timeline(
-        self: Arc<Self>,
-        timeline_id: TimelineId,
-    ) -> Result<(), DeleteTimelineError> {
-        DeleteTimelineFlow::run(&self, timeline_id, false).await?;
-
-        Ok(())
-    }
-
     /// perform one garbage collection iteration, removing old data files from disk.
     /// this function is periodically called by gc task.
     /// also it can be explicitly requested through page server api 'do_gc' command.
@@ -1967,7 +1812,7 @@ impl Tenant {
                 );
                 *current_state = TenantState::Active;
 
-                let elapsed = self.constructed_at.elapsed();
+                let elapsed = self.loading_started_at.elapsed();
                 let total_timelines = timelines_accessor.len();
 
                 // log a lot of stuff, because some tenants sometimes suffer from user-visible
@@ -1982,7 +1827,7 @@ impl Tenant {
                     "activation attempt finished"
                 );
 
-                TENANT.activation.observe(elapsed.as_secs_f64());
+                TENANT_ACTIVATION.observe(elapsed.as_secs_f64());
             });
         }
     }
@@ -2237,41 +2082,18 @@ impl Tenant {
         self.state.subscribe()
     }
 
-    /// The activate_now semaphore is initialized with zero units.  As soon as
-    /// we add a unit, waiters will be able to acquire a unit and proceed.
-    pub(crate) fn activate_now(&self) {
-        self.activate_now_sem.add_permits(1);
-    }
-
-    pub(crate) async fn wait_to_become_active(
-        &self,
-        timeout: Duration,
-    ) -> Result<(), GetActiveTenantError> {
+    pub(crate) async fn wait_to_become_active(&self) -> Result<(), GetActiveTenantError> {
         let mut receiver = self.state.subscribe();
         loop {
             let current_state = receiver.borrow_and_update().clone();
             match current_state {
                 TenantState::Loading | TenantState::Attaching | TenantState::Activating(_) => {
                     // in these states, there's a chance that we can reach ::Active
-                    self.activate_now();
-                    match timeout_cancellable(timeout, &self.cancel, receiver.changed()).await {
-                        Ok(r) => {
-                            r.map_err(
-                            |_e: tokio::sync::watch::error::RecvError|
-                                // Tenant existed but was dropped: report it as non-existent
-                                GetActiveTenantError::NotFound(GetTenantError::NotFound(self.tenant_shard_id.tenant_id))
-                        )?
-                        }
-                        Err(TimeoutCancellableError::Cancelled) => {
-                            return Err(GetActiveTenantError::Cancelled);
-                        }
-                        Err(TimeoutCancellableError::Timeout) => {
-                            return Err(GetActiveTenantError::WaitForActiveTimeout {
-                                latest_state: Some(self.current_state()),
-                                wait_time: timeout,
-                            });
-                        }
-                    }
+                    receiver.changed().await.map_err(
+                        |_e: tokio::sync::watch::error::RecvError|
+                            // Tenant existed but was dropped: report it as non-existent
+                            GetActiveTenantError::NotFound(GetTenantError::NotFound(self.tenant_shard_id.tenant_id))
+                    )?;
                 }
                 TenantState::Active { .. } => {
                     return Ok(());
@@ -2292,14 +2114,6 @@ impl Tenant {
             .attach_mode
             .clone()
     }
-
-    pub(crate) fn get_tenant_shard_id(&self) -> &TenantShardId {
-        &self.tenant_shard_id
-    }
-
-    pub(crate) fn get_generation(&self) -> Generation {
-        self.generation
-    }
 }
 
 /// Given a Vec of timelines and their ancestors (timeline_id, ancestor_id),
@@ -2438,18 +2252,6 @@ impl Tenant {
             .or(self.conf.default_tenant_conf.min_resident_size_override)
     }
 
-    pub fn get_heatmap_period(&self) -> Option<Duration> {
-        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
-        let heatmap_period = tenant_conf
-            .heatmap_period
-            .unwrap_or(self.conf.default_tenant_conf.heatmap_period);
-        if heatmap_period.is_zero() {
-            None
-        } else {
-            Some(heatmap_period)
-        }
-    }
-
     pub fn set_new_tenant_config(&self, new_tenant_conf: TenantConfOpt) {
         self.tenant_conf.write().unwrap().tenant_conf = new_tenant_conf;
         // Don't hold self.timelines.lock() during the notifies.
@@ -2596,10 +2398,9 @@ impl Tenant {
             conf,
             // using now here is good enough approximation to catch tenants with really long
             // activation times.
-            constructed_at: Instant::now(),
+            loading_started_at: Instant::now(),
             tenant_conf: Arc::new(RwLock::new(attached_conf)),
             timelines: Mutex::new(HashMap::new()),
-            timelines_creating: Mutex::new(HashSet::new()),
             gc_cs: tokio::sync::Mutex::new(()),
             walredo_mgr,
             remote_storage,
@@ -2608,7 +2409,6 @@ impl Tenant {
             cached_logical_sizes: tokio::sync::Mutex::new(HashMap::new()),
             cached_synthetic_tenant_size: Arc::new(AtomicU64::new(0)),
             eviction_task_tenant_state: tokio::sync::Mutex::new(EvictionTaskTenantState::default()),
-            activate_now_sem: tokio::sync::Semaphore::new(0),
             delete_progress: Arc::new(tokio::sync::Mutex::new(DeleteTenantFlow::default())),
             cancel: CancellationToken::default(),
             gate: Gate::new(format!("Tenant<{tenant_shard_id}>")),
@@ -2992,9 +2792,8 @@ impl Tenant {
         start_lsn: Option<Lsn>,
         ctx: &RequestContext,
     ) -> Result<Arc<Timeline>, CreateTimelineError> {
-        let uninit_mark = self.create_timeline_uninit_mark(dst_id).unwrap();
         let tl = self
-            .branch_timeline_impl(src_timeline, dst_id, start_lsn, uninit_mark, ctx)
+            .branch_timeline_impl(src_timeline, dst_id, start_lsn, ctx)
             .await?;
         tl.set_state(TimelineState::Active);
         Ok(tl)
@@ -3008,10 +2807,9 @@ impl Tenant {
         src_timeline: &Arc<Timeline>,
         dst_id: TimelineId,
         start_lsn: Option<Lsn>,
-        timeline_uninit_mark: TimelineUninitMark<'_>,
         ctx: &RequestContext,
     ) -> Result<Arc<Timeline>, CreateTimelineError> {
-        self.branch_timeline_impl(src_timeline, dst_id, start_lsn, timeline_uninit_mark, ctx)
+        self.branch_timeline_impl(src_timeline, dst_id, start_lsn, ctx)
             .await
     }
 
@@ -3020,14 +2818,13 @@ impl Tenant {
         src_timeline: &Arc<Timeline>,
         dst_id: TimelineId,
         start_lsn: Option<Lsn>,
-        timeline_uninit_mark: TimelineUninitMark<'_>,
         _ctx: &RequestContext,
     ) -> Result<Arc<Timeline>, CreateTimelineError> {
         let src_id = src_timeline.timeline_id;
 
-        // We will validate our ancestor LSN in this function.  Acquire the GC lock so that
-        // this check cannot race with GC, and the ancestor LSN is guaranteed to remain
-        // valid while we are creating the branch.
+        // First acquire the GC lock so that another task cannot advance the GC
+        // cutoff in 'gc_info', and make 'start_lsn' invalid, while we are
+        // creating the branch.
         let _gc_cs = self.gc_cs.lock().await;
 
         // If no start LSN is specified, we branch the new timeline from the source timeline's last record LSN
@@ -3037,6 +2834,13 @@ impl Tenant {
             lsn
         });
 
+        // Create a placeholder for the new branch. This will error
+        // out if the new timeline ID is already in use.
+        let timeline_uninit_mark = {
+            let timelines = self.timelines.lock().unwrap();
+            self.create_timeline_uninit_mark(dst_id, &timelines)?
+        };
+
         // Ensure that `start_lsn` is valid, i.e. the LSN is within the PITR
         // horizon on the source timeline
         //
@@ -3128,38 +2932,21 @@ impl Tenant {
         Ok(new_timeline)
     }
 
-    /// For unit tests, make this visible so that other modules can directly create timelines
-    #[cfg(test)]
-    pub(crate) async fn bootstrap_timeline_test(
-        &self,
-        timeline_id: TimelineId,
-        pg_version: u32,
-        load_existing_initdb: Option<TimelineId>,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<Arc<Timeline>> {
-        let uninit_mark = self.create_timeline_uninit_mark(timeline_id).unwrap();
-        self.bootstrap_timeline(
-            timeline_id,
-            pg_version,
-            load_existing_initdb,
-            uninit_mark,
-            ctx,
-        )
-        .await
-    }
-
     /// - run initdb to init temporary instance and get bootstrap data
     /// - after initialization completes, tar up the temp dir and upload it to S3.
     ///
     /// The caller is responsible for activating the returned timeline.
-    async fn bootstrap_timeline(
+    pub(crate) async fn bootstrap_timeline(
         &self,
         timeline_id: TimelineId,
         pg_version: u32,
         load_existing_initdb: Option<TimelineId>,
-        timeline_uninit_mark: TimelineUninitMark<'_>,
         ctx: &RequestContext,
     ) -> anyhow::Result<Arc<Timeline>> {
+        let timeline_uninit_mark = {
+            let timelines = self.timelines.lock().unwrap();
+            self.create_timeline_uninit_mark(timeline_id, &timelines)?
+        };
         // create a `tenant/{tenant_id}/timelines/basebackup-{timeline_id}.{TEMP_FILE_SUFFIX}/`
         // temporary directory for basebackup files for the given timeline.
 
@@ -3193,7 +2980,6 @@ impl Tenant {
                     storage,
                     &self.tenant_shard_id,
                     &existing_initdb_timeline_id,
-                    &self.cancel,
                 )
                 .await
                 .context("download initdb tar")?;
@@ -3234,7 +3020,6 @@ impl Tenant {
                             &timeline_id,
                             pgdata_zstd.try_clone().await?,
                             tar_zst_size,
-                            &self.cancel,
                         )
                         .await
                     },
@@ -3242,7 +3027,8 @@ impl Tenant {
                     3,
                     u32::MAX,
                     "persist_initdb_tar_zst",
-                    backoff::Cancel::new(self.cancel.clone(), || anyhow::anyhow!("Cancelled")),
+                    // TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
+                    backoff::Cancel::new(CancellationToken::new(), || unreachable!()),
                 )
                 .await?;
 
@@ -3357,11 +3143,11 @@ impl Tenant {
     /// at 'disk_consistent_lsn'. After any initial data has been imported, call
     /// `finish_creation` to insert the Timeline into the timelines map and to remove the
     /// uninit mark file.
-    async fn prepare_new_timeline<'a>(
-        &'a self,
+    async fn prepare_new_timeline(
+        &self,
         new_timeline_id: TimelineId,
         new_metadata: &TimelineMetadata,
-        uninit_mark: TimelineUninitMark<'a>,
+        uninit_mark: TimelineUninitMark,
         start_lsn: Lsn,
         ancestor: Option<Arc<Timeline>>,
     ) -> anyhow::Result<UninitializedTimeline> {
@@ -3434,38 +3220,23 @@ impl Tenant {
     fn create_timeline_uninit_mark(
         &self,
         timeline_id: TimelineId,
-    ) -> Result<TimelineUninitMark, TimelineExclusionError> {
+        timelines: &MutexGuard<HashMap<TimelineId, Arc<Timeline>>>,
+    ) -> anyhow::Result<TimelineUninitMark> {
         let tenant_shard_id = self.tenant_shard_id;
 
+        anyhow::ensure!(
+            timelines.get(&timeline_id).is_none(),
+            "Timeline {tenant_shard_id}/{timeline_id} already exists in pageserver's memory"
+        );
+        let timeline_path = self.conf.timeline_path(&tenant_shard_id, &timeline_id);
+        anyhow::ensure!(
+            !timeline_path.exists(),
+            "Timeline {timeline_path} already exists, cannot create its uninit mark file",
+        );
+
         let uninit_mark_path = self
             .conf
             .timeline_uninit_mark_file_path(tenant_shard_id, timeline_id);
-        let timeline_path = self.conf.timeline_path(&tenant_shard_id, &timeline_id);
-
-        let uninit_mark = TimelineUninitMark::new(
-            self,
-            timeline_id,
-            uninit_mark_path.clone(),
-            timeline_path.clone(),
-        )?;
-
-        // At this stage, we have got exclusive access to in-memory state for this timeline ID
-        // for creation.
-        // A timeline directory should never exist on disk already:
-        // - a previous failed creation would have cleaned up after itself
-        // - a pageserver restart would clean up timeline directories that don't have valid remote state
-        //
-        // Therefore it is an unexpected internal error to encounter a timeline directory already existing here,
-        // this error may indicate a bug in cleanup on failed creations.
-        if timeline_path.exists() {
-            return Err(TimelineExclusionError::Other(anyhow::anyhow!(
-                "Timeline directory already exists! This is a bug."
-            )));
-        }
-
-        // Create the on-disk uninit mark _after_ the in-memory acquisition of the tenant ID: guarantees
-        // that during process runtime, colliding creations will be caught in-memory without getting
-        // as far as failing to write a file.
         fs::OpenOptions::new()
             .write(true)
             .create_new(true)
@@ -3479,6 +3250,8 @@ impl Tenant {
                 format!("Failed to crate uninit mark for timeline {tenant_shard_id}/{timeline_id}")
             })?;
 
+        let uninit_mark = TimelineUninitMark::new(uninit_mark_path, timeline_path);
+
         Ok(uninit_mark)
     }
 
@@ -3921,7 +3694,6 @@ pub(crate) mod harness {
                     tenant_conf.evictions_low_residence_duration_metric_threshold,
                 ),
                 gc_feedback: Some(tenant_conf.gc_feedback),
-                heatmap_period: Some(tenant_conf.heatmap_period),
             }
         }
     }
@@ -4228,7 +4000,13 @@ mod tests {
             .await
         {
             Ok(_) => panic!("duplicate timeline creation should fail"),
-            Err(e) => assert_eq!(e.to_string(), "Already exists".to_string()),
+            Err(e) => assert_eq!(
+                e.to_string(),
+                format!(
+                    "Timeline {}/{} already exists in pageserver's memory",
+                    tenant.tenant_shard_id, TIMELINE_ID
+                )
+            ),
         }
 
         Ok(())

pageserver/src/tenant/config.rs

@@ -334,11 +334,6 @@ pub struct TenantConf {
     #[serde(with = "humantime_serde")]
     pub evictions_low_residence_duration_metric_threshold: Duration,
     pub gc_feedback: bool,
-
-    /// If non-zero, the period between uploads of a heatmap from attached tenants.  This
-    /// may be disabled if a Tenant will not have secondary locations: only secondary
-    /// locations will use the heatmap uploaded by attached locations.
-    pub heatmap_period: Duration,
 }
 
 /// Same as TenantConf, but this struct preserves the information about
@@ -419,11 +414,6 @@ pub struct TenantConfOpt {
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(default)]
     pub gc_feedback: Option<bool>,
-
-    #[serde(skip_serializing_if = "Option::is_none")]
-    #[serde(with = "humantime_serde")]
-    #[serde(default)]
-    pub heatmap_period: Option<Duration>,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
@@ -492,7 +482,6 @@ impl TenantConfOpt {
                 .evictions_low_residence_duration_metric_threshold
                 .unwrap_or(global_conf.evictions_low_residence_duration_metric_threshold),
             gc_feedback: self.gc_feedback.unwrap_or(global_conf.gc_feedback),
-            heatmap_period: self.heatmap_period.unwrap_or(global_conf.heatmap_period),
         }
     }
 }
@@ -530,7 +519,6 @@ impl Default for TenantConf {
             )
             .expect("cannot parse default evictions_low_residence_duration_metric_threshold"),
             gc_feedback: false,
-            heatmap_period: Duration::ZERO,
         }
     }
 }

pageserver/src/tenant/delete.rs

@@ -71,7 +71,6 @@ async fn create_remote_delete_mark(
     conf: &PageServerConf,
     remote_storage: &GenericRemoteStorage,
     tenant_shard_id: &TenantShardId,
-    cancel: &CancellationToken,
 ) -> Result<(), DeleteTenantError> {
     let remote_mark_path = remote_tenant_delete_mark_path(conf, tenant_shard_id)?;
 
@@ -88,7 +87,8 @@ async fn create_remote_delete_mark(
         FAILED_UPLOAD_WARN_THRESHOLD,
         FAILED_REMOTE_OP_RETRIES,
         "mark_upload",
-        backoff::Cancel::new(cancel.clone(), || anyhow::anyhow!("Cancelled")),
+        // TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
+        backoff::Cancel::new(CancellationToken::new(), || unreachable!()),
     )
     .await
     .context("mark_upload")?;
@@ -170,7 +170,6 @@ async fn remove_tenant_remote_delete_mark(
     conf: &PageServerConf,
     remote_storage: Option<&GenericRemoteStorage>,
     tenant_shard_id: &TenantShardId,
-    cancel: &CancellationToken,
 ) -> Result<(), DeleteTenantError> {
     if let Some(remote_storage) = remote_storage {
         let path = remote_tenant_delete_mark_path(conf, tenant_shard_id)?;
@@ -180,7 +179,8 @@ async fn remove_tenant_remote_delete_mark(
             FAILED_UPLOAD_WARN_THRESHOLD,
             FAILED_REMOTE_OP_RETRIES,
             "remove_tenant_remote_delete_mark",
-            backoff::Cancel::new(cancel.clone(), || anyhow::anyhow!("Cancelled")),
+            // TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
+            backoff::Cancel::new(CancellationToken::new(), || unreachable!()),
         )
         .await
         .context("remove_tenant_remote_delete_mark")?;
@@ -322,15 +322,9 @@ impl DeleteTenantFlow {
         // Though sounds scary, different mark name?
         // Detach currently uses remove_dir_all so in case of a crash we can end up in a weird state.
         if let Some(remote_storage) = &remote_storage {
-            create_remote_delete_mark(
-                conf,
-                remote_storage,
-                &tenant.tenant_shard_id,
-                // Can't use tenant.cancel, it's already shut down.  TODO: wire in an appropriate token
-                &CancellationToken::new(),
-            )
-            .await
-            .context("remote_mark")?
+            create_remote_delete_mark(conf, remote_storage, &tenant.tenant_shard_id)
+                .await
+                .context("remote_mark")?
         }
 
         fail::fail_point!("tenant-delete-before-create-local-mark", |_| {
@@ -530,14 +524,8 @@ impl DeleteTenantFlow {
                 .context("timelines dir not empty")?;
         }
 
-        remove_tenant_remote_delete_mark(
-            conf,
-            remote_storage.as_ref(),
-            &tenant.tenant_shard_id,
-            // Can't use tenant.cancel, it's already shut down.  TODO: wire in an appropriate token
-            &CancellationToken::new(),
-        )
-        .await?;
+        remove_tenant_remote_delete_mark(conf, remote_storage.as_ref(), &tenant.tenant_shard_id)
+            .await?;
 
         fail::fail_point!("tenant-delete-before-cleanup-remaining-fs-traces", |_| {
             Err(anyhow::anyhow!(

pageserver/src/tenant/mgr.rs

@@ -28,7 +28,7 @@ use crate::control_plane_client::{
     ControlPlaneClient, ControlPlaneGenerationsApi, RetryForeverError,
 };
 use crate::deletion_queue::DeletionQueueClient;
-use crate::metrics::{TENANT, TENANT_MANAGER as METRICS};
+use crate::metrics::TENANT_MANAGER as METRICS;
 use crate::task_mgr::{self, TaskKind};
 use crate::tenant::config::{
     AttachedLocationConfig, AttachmentMode, LocationConf, LocationMode, TenantConfOpt,
@@ -44,6 +44,7 @@ use utils::generation::Generation;
 use utils::id::{TenantId, TimelineId};
 
 use super::delete::DeleteTenantError;
+use super::timeline::delete::DeleteTimelineFlow;
 use super::TenantSharedResources;
 
 /// For a tenant that appears in TenantsMap, it may either be
@@ -429,13 +430,6 @@ pub async fn init_tenant_mgr(
     let tenant_generations =
         init_load_generations(conf, &tenant_configs, &resources, &cancel).await?;
 
-    tracing::info!(
-        "Attaching {} tenants at startup, warming up {} at a time",
-        tenant_configs.len(),
-        conf.concurrent_tenant_warmup.initial_permits()
-    );
-    TENANT.startup_scheduled.inc_by(tenant_configs.len() as u64);
-
     // Construct `Tenant` objects and start them running
     for (tenant_shard_id, location_conf) in tenant_configs {
         let tenant_dir_path = conf.tenant_path(&tenant_shard_id);
@@ -813,12 +807,6 @@ pub(crate) async fn set_new_tenant_config(
 }
 
 impl TenantManager {
-    /// Convenience function so that anyone with a TenantManager can get at the global configuration, without
-    /// having to pass it around everywhere as a separate object.
-    pub(crate) fn get_conf(&self) -> &'static PageServerConf {
-        self.conf
-    }
-
     /// Gets the attached tenant from the in-memory data, erroring if it's absent, in secondary mode, or is not fitting to the query.
     /// `active_only = true` allows to query only tenants that are ready for operations, erroring on other kinds of tenants.
     pub(crate) fn get_attached_tenant_shard(
@@ -854,6 +842,17 @@ impl TenantManager {
         }
     }
 
+    pub(crate) async fn delete_timeline(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        _ctx: &RequestContext,
+    ) -> Result<(), DeleteTimelineError> {
+        let tenant = self.get_attached_tenant_shard(tenant_shard_id, true)?;
+        DeleteTimelineFlow::run(&tenant, timeline_id, false).await?;
+        Ok(())
+    }
+
     #[instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()))]
     pub(crate) async fn upsert_location(
         &self,
@@ -1088,20 +1087,6 @@ impl TenantManager {
 
         Ok(())
     }
-
-    pub(crate) fn get_attached_active_tenant_shards(&self) -> Vec<Arc<Tenant>> {
-        let locked = self.tenants.read().unwrap();
-        match &*locked {
-            TenantsMap::Initializing => Vec::new(),
-            TenantsMap::Open(map) | TenantsMap::ShuttingDown(map) => map
-                .values()
-                .filter_map(|slot| {
-                    slot.get_attached()
-                        .and_then(|t| if t.is_active() { Some(t.clone()) } else { None })
-                })
-                .collect(),
-        }
-    }
 }
 
 #[derive(Debug, thiserror::Error)]
@@ -1216,10 +1201,7 @@ pub(crate) async fn get_active_tenant_with_timeout(
                         // Fast path: we don't need to do any async waiting.
                         return Ok(tenant.clone());
                     }
-                    _ => {
-                        tenant.activate_now();
-                        (WaitFor::Tenant(tenant.clone()), tenant_shard_id)
-                    }
+                    _ => (WaitFor::Tenant(tenant.clone()), tenant_shard_id),
                 }
             }
             Some(TenantSlot::Secondary) => {
@@ -1273,10 +1255,28 @@ pub(crate) async fn get_active_tenant_with_timeout(
     };
 
     tracing::debug!("Waiting for tenant to enter active state...");
-    tenant
-        .wait_to_become_active(deadline.duration_since(Instant::now()))
-        .await?;
-    Ok(tenant)
+    match timeout_cancellable(
+        deadline.duration_since(Instant::now()),
+        cancel,
+        tenant.wait_to_become_active(),
+    )
+    .await
+    {
+        Ok(Ok(())) => Ok(tenant),
+        Ok(Err(e)) => Err(e),
+        Err(TimeoutCancellableError::Timeout) => {
+            let latest_state = tenant.current_state();
+            if latest_state == TenantState::Active {
+                Ok(tenant)
+            } else {
+                Err(GetActiveTenantError::WaitForActiveTimeout {
+                    latest_state: Some(latest_state),
+                    wait_time: timeout,
+                })
+            }
+        }
+        Err(TimeoutCancellableError::Cancelled) => Err(GetActiveTenantError::Cancelled),
+    }
 }
 
 pub(crate) async fn delete_tenant(

pageserver/src/tenant/remote_timeline_client.rs

@@ -180,7 +180,7 @@
 //! [`Tenant::timeline_init_and_sync`]: super::Tenant::timeline_init_and_sync
 //! [`Timeline::load_layer_map`]: super::Timeline::load_layer_map
 
-pub(crate) mod download;
+mod download;
 pub mod index;
 mod upload;
 
@@ -196,12 +196,10 @@ pub(crate) use upload::upload_initdb_dir;
 use utils::backoff::{
     self, exponential_backoff, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS,
 };
-use utils::timeout::{timeout_cancellable, TimeoutCancellableError};
 
 use std::collections::{HashMap, VecDeque};
 use std::sync::atomic::{AtomicU32, Ordering};
 use std::sync::{Arc, Mutex};
-use std::time::Duration;
 
 use remote_storage::{DownloadError, GenericRemoteStorage, RemotePath};
 use std::ops::DerefMut;
@@ -318,47 +316,6 @@ pub struct RemoteTimelineClient {
     storage_impl: GenericRemoteStorage,
 
     deletion_queue_client: DeletionQueueClient,
-
-    cancel: CancellationToken,
-}
-
-/// This timeout is intended to deal with hangs in lower layers, e.g. stuck TCP flows.  It is not
-/// intended to be snappy enough for prompt shutdown, as we have a CancellationToken for that.
-const UPLOAD_TIMEOUT: Duration = Duration::from_secs(120);
-const DOWNLOAD_TIMEOUT: Duration = Duration::from_secs(120);
-
-/// Wrapper for timeout_cancellable that flattens result and converts TimeoutCancellableError to anyhow.
-///
-/// This is a convenience for the various upload functions.  In future
-/// the anyhow::Error result should be replaced with a more structured type that
-/// enables callers to avoid handling shutdown as an error.
-async fn upload_cancellable<F>(cancel: &CancellationToken, future: F) -> anyhow::Result<()>
-where
-    F: std::future::Future<Output = anyhow::Result<()>>,
-{
-    match timeout_cancellable(UPLOAD_TIMEOUT, cancel, future).await {
-        Ok(Ok(())) => Ok(()),
-        Ok(Err(e)) => Err(e),
-        Err(TimeoutCancellableError::Timeout) => Err(anyhow::anyhow!("Timeout")),
-        Err(TimeoutCancellableError::Cancelled) => Err(anyhow::anyhow!("Shutting down")),
-    }
-}
-/// Wrapper for timeout_cancellable that flattens result and converts TimeoutCancellableError to DownloaDError.
-async fn download_cancellable<F, R>(
-    cancel: &CancellationToken,
-    future: F,
-) -> Result<R, DownloadError>
-where
-    F: std::future::Future<Output = Result<R, DownloadError>>,
-{
-    match timeout_cancellable(DOWNLOAD_TIMEOUT, cancel, future).await {
-        Ok(Ok(r)) => Ok(r),
-        Ok(Err(e)) => Err(e),
-        Err(TimeoutCancellableError::Timeout) => {
-            Err(DownloadError::Other(anyhow::anyhow!("Timed out")))
-        }
-        Err(TimeoutCancellableError::Cancelled) => Err(DownloadError::Cancelled),
-    }
 }
 
 impl RemoteTimelineClient {
@@ -394,7 +351,6 @@ impl RemoteTimelineClient {
                 &tenant_shard_id,
                 &timeline_id,
             )),
-            cancel: CancellationToken::new(),
         }
     }
 
@@ -545,7 +501,6 @@ impl RemoteTimelineClient {
         &self,
         layer_file_name: &LayerFileName,
         layer_metadata: &LayerFileMetadata,
-        cancel: &CancellationToken,
     ) -> anyhow::Result<u64> {
         let downloaded_size = {
             let _unfinished_gauge_guard = self.metrics.call_begin(
@@ -562,7 +517,6 @@ impl RemoteTimelineClient {
                 self.timeline_id,
                 layer_file_name,
                 layer_metadata,
-                cancel,
             )
             .measure_remote_op(
                 self.tenant_shard_id.tenant_id,
@@ -1017,7 +971,6 @@ impl RemoteTimelineClient {
                     &self.timeline_id,
                     self.generation,
                     &index_part_with_deleted_at,
-                    &self.cancel,
                 )
             },
             |_e| false,
@@ -1027,7 +980,8 @@ impl RemoteTimelineClient {
             // when executed as part of tenant deletion this happens in the background
             2,
             "persist_index_part_with_deleted_flag",
-            backoff::Cancel::new(self.cancel.clone(), || anyhow::anyhow!("Cancelled")),
+            // TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
+            backoff::Cancel::new(CancellationToken::new(), || unreachable!()),
         )
         .await?;
 
@@ -1327,7 +1281,6 @@ impl RemoteTimelineClient {
                         path,
                         layer_metadata,
                         self.generation,
-                        &self.cancel,
                     )
                     .measure_remote_op(
                         self.tenant_shard_id.tenant_id,
@@ -1354,7 +1307,6 @@ impl RemoteTimelineClient {
                         &self.timeline_id,
                         self.generation,
                         index_part,
-                        &self.cancel,
                     )
                     .measure_remote_op(
                         self.tenant_shard_id.tenant_id,
@@ -1652,23 +1604,6 @@ impl RemoteTimelineClient {
             }
         }
     }
-
-    pub(crate) fn get_layers_metadata(
-        &self,
-        layers: Vec<LayerFileName>,
-    ) -> anyhow::Result<Vec<Option<LayerFileMetadata>>> {
-        let q = self.upload_queue.lock().unwrap();
-        let q = match &*q {
-            UploadQueue::Stopped(_) | UploadQueue::Uninitialized => {
-                anyhow::bail!("queue is in state {}", q.as_str())
-            }
-            UploadQueue::Initialized(inner) => inner,
-        };
-
-        let decorated = layers.into_iter().map(|l| q.latest_files.get(&l).cloned());
-
-        Ok(decorated.collect())
-    }
 }
 
 pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
@@ -1724,13 +1659,6 @@ pub fn remote_index_path(
     .expect("Failed to construct path")
 }
 
-pub const HEATMAP_BASENAME: &str = "heatmap-v1.json";
-
-pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
-    RemotePath::from_string(&format!("tenants/{tenant_shard_id}/{HEATMAP_BASENAME}"))
-        .expect("Failed to construct path")
-}
-
 /// Given the key of an index, parse out the generation part of the name
 pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
     let file_name = match path.get_path().file_name() {
@@ -1876,7 +1804,6 @@ mod tests {
                     &self.harness.tenant_shard_id,
                     &TIMELINE_ID,
                 )),
-                cancel: CancellationToken::new(),
             })
         }
 

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -5,6 +5,7 @@
 
 use std::collections::HashSet;
 use std::future::Future;
+use std::time::Duration;
 
 use anyhow::{anyhow, Context};
 use camino::{Utf8Path, Utf8PathBuf};
@@ -13,17 +14,13 @@ use tokio::fs::{self, File, OpenOptions};
 use tokio::io::{AsyncSeekExt, AsyncWriteExt};
 use tokio_util::sync::CancellationToken;
 use tracing::warn;
-use utils::timeout::timeout_cancellable;
 use utils::{backoff, crashsafe};
 
 use crate::config::PageServerConf;
-use crate::tenant::remote_timeline_client::{
-    download_cancellable, remote_layer_path, remote_timelines_path, DOWNLOAD_TIMEOUT,
-};
+use crate::tenant::remote_timeline_client::{remote_layer_path, remote_timelines_path};
 use crate::tenant::storage_layer::LayerFileName;
 use crate::tenant::timeline::span::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::tenant::Generation;
-use crate::virtual_file::on_fatal_io_error;
 use crate::TEMP_FILE_SUFFIX;
 use remote_storage::{DownloadError, GenericRemoteStorage, ListingMode};
 use utils::crashsafe::path_with_suffix_extension;
@@ -35,6 +32,8 @@ use super::{
     FAILED_DOWNLOAD_WARN_THRESHOLD, FAILED_REMOTE_OP_RETRIES, INITDB_PATH,
 };
 
+static MAX_DOWNLOAD_DURATION: Duration = Duration::from_secs(120);
+
 ///
 /// If 'metadata' is given, we will validate that the downloaded file's size matches that
 /// in the metadata. (In the future, we might do more cross-checks, like CRC validation)
@@ -47,7 +46,6 @@ pub async fn download_layer_file<'a>(
     timeline_id: TimelineId,
     layer_file_name: &'a LayerFileName,
     layer_metadata: &'a LayerFileMetadata,
-    cancel: &CancellationToken,
 ) -> Result<u64, DownloadError> {
     debug_assert_current_span_has_tenant_and_timeline_id();
 
@@ -75,18 +73,14 @@ pub async fn download_layer_file<'a>(
     // If pageserver crashes the temp file will be deleted on startup and re-downloaded.
     let temp_file_path = path_with_suffix_extension(&local_path, TEMP_DOWNLOAD_EXTENSION);
 
-    let cancel_inner = cancel.clone();
     let (mut destination_file, bytes_amount) = download_retry(
         || async {
             let destination_file = tokio::fs::File::create(&temp_file_path)
                 .await
                 .with_context(|| format!("create a destination file for layer '{temp_file_path}'"))
                 .map_err(DownloadError::Other)?;
-
-            // Cancellation safety: it is safe to cancel this future, because it isn't writing to a local
-            // file: the write to local file doesn't start until after the request header is returned
-            // and we start draining the body stream below
-            let download = download_cancellable(&cancel_inner, storage.download(&remote_path))
+            let download = storage
+                .download(&remote_path)
                 .await
                 .with_context(|| {
                     format!(
@@ -100,33 +94,12 @@ pub async fn download_layer_file<'a>(
 
             let mut reader = tokio_util::io::StreamReader::new(download.download_stream);
 
-            // Cancellation safety: it is safe to cancel this future because it is writing into a temporary file,
-            // and we will unlink the temporary file if there is an error.  This unlink is important because we
-            // are in a retry loop, and we wouldn't want to leave behind a rogue write I/O to a file that
-            // we will imminiently try and write to again.
-            let bytes_amount: u64 = match timeout_cancellable(
-                DOWNLOAD_TIMEOUT,
-                &cancel_inner,
+            let bytes_amount = tokio::time::timeout(
+                MAX_DOWNLOAD_DURATION,
                 tokio::io::copy_buf(&mut reader, &mut destination_file),
             )
             .await
-            .with_context(|| {
-                format!(
-                    "download layer at remote path '{remote_path:?}' into file {temp_file_path:?}"
-                )
-            })
-            .map_err(DownloadError::Other)?
-            {
-                Ok(b) => Ok(b),
-                Err(e) => {
-                    // Remove incomplete files: on restart Timeline would do this anyway, but we must
-                    // do it here for the retry case.
-                    if let Err(e) = tokio::fs::remove_file(&temp_file_path).await {
-                        on_fatal_io_error(&e, &format!("Removing temporary file {temp_file_path}"));
-                    }
-                    Err(e)
-                }
-            }
+            .map_err(|e| DownloadError::Other(anyhow::anyhow!("Timed out  {:?}", e)))?
             .with_context(|| {
                 format!(
                     "download layer at remote path '{remote_path:?}' into file {temp_file_path:?}"
@@ -139,7 +112,6 @@ pub async fn download_layer_file<'a>(
             Ok((destination_file, bytes_amount))
         },
         &format!("download {remote_path:?}"),
-        cancel,
     )
     .await?;
 
@@ -216,14 +188,8 @@ pub async fn list_remote_timelines(
         anyhow::bail!("storage-sync-list-remote-timelines");
     });
 
-    let cancel_inner = cancel.clone();
     let listing = download_retry_forever(
-        || {
-            download_cancellable(
-                &cancel_inner,
-                storage.list(Some(&remote_path), ListingMode::WithDelimiter),
-            )
-        },
+        || storage.list(Some(&remote_path), ListingMode::WithDelimiter),
         &format!("list timelines for {tenant_shard_id}"),
         cancel,
     )
@@ -264,13 +230,9 @@ async fn do_download_index_part(
 
     let remote_path = remote_index_path(tenant_shard_id, timeline_id, index_generation);
 
-    let cancel_inner = cancel.clone();
     let index_part_bytes = download_retry_forever(
         || async {
-            // Cancellation: if is safe to cancel this future because we're just downloading into
-            // a memory buffer, not touching local disk.
-            let index_part_download =
-                download_cancellable(&cancel_inner, storage.download(&remote_path)).await?;
+            let index_part_download = storage.download(&remote_path).await?;
 
             let mut index_part_bytes = Vec::new();
             let mut stream = std::pin::pin!(index_part_download.download_stream);
@@ -385,7 +347,10 @@ pub(super) async fn download_index_part(
         FAILED_DOWNLOAD_WARN_THRESHOLD,
         FAILED_REMOTE_OP_RETRIES,
         "listing index_part files",
-        backoff::Cancel::new(cancel.clone(), || anyhow::anyhow!("Cancelled")),
+        // TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
+        backoff::Cancel::new(CancellationToken::new(), || -> anyhow::Error {
+            unreachable!()
+        }),
     )
     .await
     .map_err(DownloadError::Other)?;
@@ -424,7 +389,6 @@ pub(crate) async fn download_initdb_tar_zst(
     storage: &GenericRemoteStorage,
     tenant_shard_id: &TenantShardId,
     timeline_id: &TimelineId,
-    cancel: &CancellationToken,
 ) -> Result<(Utf8PathBuf, File), DownloadError> {
     debug_assert_current_span_has_tenant_and_timeline_id();
 
@@ -442,8 +406,6 @@ pub(crate) async fn download_initdb_tar_zst(
         "{INITDB_PATH}.download-{timeline_id}.{TEMP_FILE_SUFFIX}"
     ));
 
-    let cancel_inner = cancel.clone();
-
     let file = download_retry(
         || async {
             let file = OpenOptions::new()
@@ -456,14 +418,10 @@ pub(crate) async fn download_initdb_tar_zst(
                 .with_context(|| format!("tempfile creation {temp_path}"))
                 .map_err(DownloadError::Other)?;
 
-            let download =
-                download_cancellable(&cancel_inner, storage.download(&remote_path)).await?;
+            let download = storage.download(&remote_path).await?;
             let mut download = tokio_util::io::StreamReader::new(download.download_stream);
             let mut writer = tokio::io::BufWriter::with_capacity(8 * 1024, file);
 
-            // TODO: this consumption of the response body should be subject to timeout + cancellation, but
-            // not without thinking carefully about how to recover safely from cancelling a write to
-            // local storage (e.g. by writing into a temp file as we do in download_layer)
             tokio::io::copy_buf(&mut download, &mut writer)
                 .await
                 .with_context(|| format!("download initdb.tar.zst at {remote_path:?}"))
@@ -479,7 +437,6 @@ pub(crate) async fn download_initdb_tar_zst(
             Ok(file)
         },
         &format!("download {remote_path}"),
-        cancel,
     )
     .await
     .map_err(|e| {
@@ -503,11 +460,7 @@ pub(crate) async fn download_initdb_tar_zst(
 /// with backoff.
 ///
 /// (See similar logic for uploads in `perform_upload_task`)
-async fn download_retry<T, O, F>(
-    op: O,
-    description: &str,
-    cancel: &CancellationToken,
-) -> Result<T, DownloadError>
+async fn download_retry<T, O, F>(op: O, description: &str) -> Result<T, DownloadError>
 where
     O: FnMut() -> F,
     F: Future<Output = Result<T, DownloadError>>,
@@ -518,7 +471,10 @@ where
         FAILED_DOWNLOAD_WARN_THRESHOLD,
         FAILED_REMOTE_OP_RETRIES,
         description,
-        backoff::Cancel::new(cancel.clone(), || DownloadError::Cancelled),
+        // TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
+        backoff::Cancel::new(CancellationToken::new(), || -> DownloadError {
+            unreachable!()
+        }),
     )
     .await
 }

pageserver/src/tenant/remote_timeline_client/upload.rs

@@ -4,17 +4,14 @@ use anyhow::{bail, Context};
 use camino::Utf8Path;
 use fail::fail_point;
 use pageserver_api::shard::TenantShardId;
-use std::io::{ErrorKind, SeekFrom};
+use std::io::ErrorKind;
 use tokio::fs::{self, File};
-use tokio::io::AsyncSeekExt;
-use tokio_util::sync::CancellationToken;
 
 use super::Generation;
 use crate::{
     config::PageServerConf,
     tenant::remote_timeline_client::{
         index::IndexPart, remote_index_path, remote_initdb_archive_path, remote_path,
-        upload_cancellable,
     },
 };
 use remote_storage::GenericRemoteStorage;
@@ -31,7 +28,6 @@ pub(super) async fn upload_index_part<'a>(
     timeline_id: &TimelineId,
     generation: Generation,
     index_part: &'a IndexPart,
-    cancel: &CancellationToken,
 ) -> anyhow::Result<()> {
     tracing::trace!("uploading new index part");
 
@@ -47,16 +43,14 @@ pub(super) async fn upload_index_part<'a>(
     let index_part_bytes = bytes::Bytes::from(index_part_bytes);
 
     let remote_path = remote_index_path(tenant_shard_id, timeline_id, generation);
-    upload_cancellable(
-        cancel,
-        storage.upload_storage_object(
+    storage
+        .upload_storage_object(
             futures::stream::once(futures::future::ready(Ok(index_part_bytes))),
             index_part_size,
             &remote_path,
-        ),
-    )
-    .await
-    .with_context(|| format!("upload index part for '{tenant_shard_id} / {timeline_id}'"))
+        )
+        .await
+        .with_context(|| format!("upload index part for '{tenant_shard_id} / {timeline_id}'"))
 }
 
 /// Attempts to upload given layer files.
@@ -69,7 +63,6 @@ pub(super) async fn upload_timeline_layer<'a>(
     source_path: &'a Utf8Path,
     known_metadata: &'a LayerFileMetadata,
     generation: Generation,
-    cancel: &CancellationToken,
 ) -> anyhow::Result<()> {
     fail_point!("before-upload-layer", |_| {
         bail!("failpoint before-upload-layer")
@@ -113,7 +106,8 @@ pub(super) async fn upload_timeline_layer<'a>(
 
     let reader = tokio_util::io::ReaderStream::with_capacity(source_file, super::BUFFER_SIZE);
 
-    upload_cancellable(cancel, storage.upload(reader, fs_size, &storage_path, None))
+    storage
+        .upload(reader, fs_size, &storage_path, None)
         .await
         .with_context(|| format!("upload layer from local path '{source_path}'"))?;
 
@@ -125,22 +119,16 @@ pub(crate) async fn upload_initdb_dir(
     storage: &GenericRemoteStorage,
     tenant_id: &TenantId,
     timeline_id: &TimelineId,
-    mut initdb_tar_zst: File,
+    initdb_tar_zst: File,
     size: u64,
-    cancel: &CancellationToken,
 ) -> anyhow::Result<()> {
     tracing::trace!("uploading initdb dir");
 
-    // We might have read somewhat into the file already in the prior retry attempt
-    initdb_tar_zst.seek(SeekFrom::Start(0)).await?;
-
     let file = tokio_util::io::ReaderStream::with_capacity(initdb_tar_zst, super::BUFFER_SIZE);
 
     let remote_path = remote_initdb_archive_path(tenant_id, timeline_id);
-    upload_cancellable(
-        cancel,
-        storage.upload_storage_object(file, size as usize, &remote_path),
-    )
-    .await
-    .with_context(|| format!("upload initdb dir for '{tenant_id} / {timeline_id}'"))
+    storage
+        .upload_storage_object(file, size as usize, &remote_path)
+        .await
+        .with_context(|| format!("upload initdb dir for '{tenant_id} / {timeline_id}'"))
 }

pageserver/src/tenant/secondary.rs

@@ -1,104 +0,0 @@
-pub mod heatmap;
-mod heatmap_uploader;
-
-use std::sync::Arc;
-
-use crate::task_mgr::{self, TaskKind, BACKGROUND_RUNTIME};
-
-use self::heatmap_uploader::heatmap_uploader_task;
-
-use super::mgr::TenantManager;
-
-use pageserver_api::shard::TenantShardId;
-use remote_storage::GenericRemoteStorage;
-
-use tokio_util::sync::CancellationToken;
-use utils::completion::Barrier;
-
-enum UploadCommand {
-    Upload(TenantShardId),
-}
-
-struct CommandRequest<T> {
-    payload: T,
-    response_tx: tokio::sync::oneshot::Sender<CommandResponse>,
-}
-
-struct CommandResponse {
-    result: anyhow::Result<()>,
-}
-
-/// The SecondaryController is a pseudo-rpc client for administrative control of secondary mode downloads,
-/// and heatmap uploads.  This is not a hot data path: it's primarily a hook for tests,
-/// where we want to immediately upload/download for a particular tenant.  In normal operation
-/// uploads & downloads are autonomous and not driven by this interface.
-pub struct SecondaryController {
-    upload_req_tx: tokio::sync::mpsc::Sender<CommandRequest<UploadCommand>>,
-}
-
-impl SecondaryController {
-    async fn dispatch<T>(
-        &self,
-        queue: &tokio::sync::mpsc::Sender<CommandRequest<T>>,
-        payload: T,
-    ) -> anyhow::Result<()> {
-        let (response_tx, response_rx) = tokio::sync::oneshot::channel();
-
-        queue
-            .send(CommandRequest {
-                payload,
-                response_tx,
-            })
-            .await
-            .map_err(|_| anyhow::anyhow!("Receiver shut down"))?;
-
-        let response = response_rx
-            .await
-            .map_err(|_| anyhow::anyhow!("Request dropped"))?;
-
-        response.result
-    }
-
-    pub async fn upload_tenant(&self, tenant_shard_id: TenantShardId) -> anyhow::Result<()> {
-        self.dispatch(&self.upload_req_tx, UploadCommand::Upload(tenant_shard_id))
-            .await
-    }
-}
-
-pub fn spawn_tasks(
-    tenant_manager: Arc<TenantManager>,
-    remote_storage: GenericRemoteStorage,
-    background_jobs_can_start: Barrier,
-    cancel: CancellationToken,
-) -> SecondaryController {
-    let (upload_req_tx, upload_req_rx) =
-        tokio::sync::mpsc::channel::<CommandRequest<UploadCommand>>(16);
-
-    task_mgr::spawn(
-        BACKGROUND_RUNTIME.handle(),
-        TaskKind::SecondaryUploads,
-        None,
-        None,
-        "heatmap uploads",
-        false,
-        async move {
-            heatmap_uploader_task(
-                tenant_manager,
-                remote_storage,
-                upload_req_rx,
-                background_jobs_can_start,
-                cancel,
-            )
-            .await
-        },
-    );
-
-    SecondaryController { upload_req_tx }
-}
-
-/// For running with remote storage disabled: a SecondaryController that is connected to nothing.
-pub fn null_controller() -> SecondaryController {
-    let (upload_req_tx, _upload_req_rx) =
-        tokio::sync::mpsc::channel::<CommandRequest<UploadCommand>>(16);
-    SecondaryController { upload_req_tx }
-}

pageserver/src/tenant/secondary/heatmap.rs

@@ -1,64 +0,0 @@
-use std::time::SystemTime;
-
-use crate::tenant::{
-    remote_timeline_client::index::IndexLayerMetadata, storage_layer::LayerFileName,
-};
-
-use serde::{Deserialize, Serialize};
-use serde_with::{serde_as, DisplayFromStr, TimestampSeconds};
-
-use utils::{generation::Generation, id::TimelineId};
-
-#[derive(Serialize, Deserialize)]
-pub(super) struct HeatMapTenant {
-    /// Generation of the attached location that uploaded the heatmap: this is not required
-    /// for correctness, but acts as a hint to secondary locations in order to detect thrashing
-    /// in the unlikely event that two attached locations are both uploading conflicting heatmaps.
-    pub(super) generation: Generation,
-
-    pub(super) timelines: Vec<HeatMapTimeline>,
-}
-
-#[serde_as]
-#[derive(Serialize, Deserialize)]
-pub(crate) struct HeatMapTimeline {
-    #[serde_as(as = "DisplayFromStr")]
-    pub(super) timeline_id: TimelineId,
-
-    pub(super) layers: Vec<HeatMapLayer>,
-}
-
-#[serde_as]
-#[derive(Serialize, Deserialize)]
-pub(crate) struct HeatMapLayer {
-    pub(super) name: LayerFileName,
-    pub(super) metadata: IndexLayerMetadata,
-
-    #[serde_as(as = "TimestampSeconds<i64>")]
-    pub(super) access_time: SystemTime,
-    // TODO: an actual 'heat' score that would let secondary locations prioritize downloading
-    // the hottest layers, rather than trying to simply mirror whatever layers are on-disk on the primary.
-}
-
-impl HeatMapLayer {
-    pub(crate) fn new(
-        name: LayerFileName,
-        metadata: IndexLayerMetadata,
-        access_time: SystemTime,
-    ) -> Self {
-        Self {
-            name,
-            metadata,
-            access_time,
-        }
-    }
-}
-
-impl HeatMapTimeline {
-    pub(crate) fn new(timeline_id: TimelineId, layers: Vec<HeatMapLayer>) -> Self {
-        Self {
-            timeline_id,
-            layers,
-        }
-    }
-}

pageserver/src/tenant/secondary/heatmap_uploader.rs

@@ -1,582 +0,0 @@
-use std::{
-    collections::HashMap,
-    sync::{Arc, Weak},
-    time::{Duration, Instant},
-};
-
-use crate::{
-    metrics::SECONDARY_MODE,
-    tenant::{
-        config::AttachmentMode, mgr::TenantManager, remote_timeline_client::remote_heatmap_path,
-        secondary::CommandResponse, span::debug_assert_current_span_has_tenant_id, Tenant,
-    },
-};
-
-use md5;
-use pageserver_api::shard::TenantShardId;
-use remote_storage::GenericRemoteStorage;
-
-use tokio::task::JoinSet;
-use tokio_util::sync::CancellationToken;
-use tracing::instrument;
-use utils::{backoff, completion::Barrier};
-
-use super::{heatmap::HeatMapTenant, CommandRequest, UploadCommand};
-
-/// Period between heatmap uploader walking Tenants to look for work to do.
-/// If any tenants have a heatmap upload period lower than this, it will be adjusted
-/// downward to match.
-const DEFAULT_SCHEDULING_INTERVAL: Duration = Duration::from_millis(60000);
-const MIN_SCHEDULING_INTERVAL: Duration = Duration::from_millis(1000);
-
-struct WriteInProgress {
-    barrier: Barrier,
-}
-
-struct UploadPending {
-    tenant: Arc<Tenant>,
-    last_digest: Option<md5::Digest>,
-}
-
-struct WriteComplete {
-    tenant_shard_id: TenantShardId,
-    completed_at: Instant,
-    digest: Option<md5::Digest>,
-    next_upload: Option<Instant>,
-}
-
-/// The heatmap uploader keeps a little bit of per-tenant state, mainly to remember
-/// when we last did a write.  We only populate this after doing at least one
-/// write for a tenant -- this avoids holding state for tenants that have
-/// uploads disabled.
-
-struct UploaderTenantState {
-    // This Weak only exists to enable culling idle instances of this type
-    // when the Tenant has been deallocated.
-    tenant: Weak<Tenant>,
-
-    /// Digest of the serialized heatmap that we last successfully uploaded
-    ///
-    /// md5 is generally a bad hash.  We use it because it's convenient for interop with AWS S3's ETag,
-    /// which is also an md5sum.
-    last_digest: Option<md5::Digest>,
-
-    /// When the last upload attempt completed (may have been successful or failed)
-    last_upload: Option<Instant>,
-
-    /// When should we next do an upload?  None means never.
-    next_upload: Option<Instant>,
-}
-
-/// This type is owned by a single task ([`heatmap_uploader_task`]) which runs an event
-/// handling loop and mutates it as needed: there are no locks here, because that event loop
-/// can hold &mut references to this type throughout.
-struct HeatmapUploader {
-    tenant_manager: Arc<TenantManager>,
-    remote_storage: GenericRemoteStorage,
-    cancel: CancellationToken,
-
-    tenants: HashMap<TenantShardId, UploaderTenantState>,
-
-    /// Tenants with work to do, for which tasks should be spawned as soon as concurrency
-    /// limits permit it.
-    tenants_pending: std::collections::VecDeque<UploadPending>,
-
-    /// Tenants for which a task in `tasks` has been spawned.
-    tenants_uploading: HashMap<TenantShardId, WriteInProgress>,
-
-    tasks: JoinSet<()>,
-
-    /// Channel for our child tasks to send results to: we use a channel for results rather than
-    /// just getting task results via JoinSet because we need the channel's recv() "sleep until something
-    /// is available" semantic, rather than JoinSet::join_next()'s "sleep until next thing is available _or_ I'm empty"
-    /// behavior.
-    task_result_tx: tokio::sync::mpsc::UnboundedSender<WriteComplete>,
-    task_result_rx: tokio::sync::mpsc::UnboundedReceiver<WriteComplete>,
-
-    concurrent_uploads: usize,
-
-    scheduling_interval: Duration,
-}
-
-/// The uploader task runs a loop that periodically wakes up and schedules tasks for
-/// tenants that require an upload, or handles any commands that have been sent into
-/// `command_queue`.  No I/O is done in this loop: that all happens in the tasks we
-/// spawn.
-///
-/// Scheduling iterations are somewhat infrequent.  However, each one will enqueue
-/// all tenants that require an upload, and in between scheduling iterations we will
-/// continue to spawn new tasks for pending tenants, as our concurrency limit permits.
-///
-/// While we take a CancellationToken here, it is subordinate to the CancellationTokens
-/// of tenants: i.e. we expect all Tenants to have been shut down before we are shut down, otherwise
-/// we might block waiting on a Tenant.
-pub(super) async fn heatmap_uploader_task(
-    tenant_manager: Arc<TenantManager>,
-    remote_storage: GenericRemoteStorage,
-    mut command_queue: tokio::sync::mpsc::Receiver<CommandRequest<UploadCommand>>,
-    background_jobs_can_start: Barrier,
-    cancel: CancellationToken,
-) -> anyhow::Result<()> {
-    let concurrent_uploads = tenant_manager.get_conf().heatmap_upload_concurrency;
-
-    let (result_tx, result_rx) = tokio::sync::mpsc::unbounded_channel();
-
-    let mut uploader = HeatmapUploader {
-        tenant_manager,
-        remote_storage,
-        cancel: cancel.clone(),
-        tasks: JoinSet::new(),
-        tenants: HashMap::new(),
-        tenants_pending: std::collections::VecDeque::new(),
-        tenants_uploading: HashMap::new(),
-        task_result_tx: result_tx,
-        task_result_rx: result_rx,
-        concurrent_uploads,
-        scheduling_interval: DEFAULT_SCHEDULING_INTERVAL,
-    };
-
-    tracing::info!("Waiting for background_jobs_can start...");
-    background_jobs_can_start.wait().await;
-    tracing::info!("background_jobs_can is ready, proceeding.");
-
-    while !cancel.is_cancelled() {
-        // Look for new work: this is relatively expensive because we have to go acquire the lock on
-        // the tenant manager to retrieve tenants, and then iterate over them to figure out which ones
-        // require an upload.
-        uploader.schedule_iteration().await?;
-
-        // Between scheduling iterations, we will:
-        //  - Drain any complete tasks and spawn pending tasks
-        //  - Handle incoming administrative commands
-        //  - Check our cancellation token
-        let next_scheduling_iteration = Instant::now()
-            .checked_add(uploader.scheduling_interval)
-            .unwrap_or_else(|| {
-                tracing::warn!(
-                    "Scheduling interval invalid ({}s), running immediately!",
-                    uploader.scheduling_interval.as_secs_f64()
-                );
-                Instant::now()
-            });
-        loop {
-            tokio::select! {
-                _ = cancel.cancelled() => {
-                    // We do not simply drop the JoinSet, in order to have an orderly shutdown without cancellation.
-                    tracing::info!("Heatmap uploader joining tasks");
-                    while let Some(_r) = uploader.tasks.join_next().await {};
-                    tracing::info!("Heatmap uploader terminating");
-
-                    break;
-                },
-                _ = tokio::time::sleep(next_scheduling_iteration.duration_since(Instant::now())) => {
-                    tracing::debug!("heatmap_uploader_task: woke for scheduling interval");
-                    break;},
-                cmd = command_queue.recv() => {
-                    tracing::debug!("heatmap_uploader_task: woke for command queue");
-                    let cmd = match cmd {
-                        Some(c) =>c,
-                        None => {
-                            // SecondaryController was destroyed, and this has raced with
-                            // our CancellationToken
-                            tracing::info!("Heatmap uploader terminating");
-                            cancel.cancel();
-                            break;
-                        }
-                    };
-
-                    let CommandRequest{
-                        response_tx,
-                        payload
-                    } = cmd;
-                    uploader.handle_command(payload, response_tx);
-                },
-                _ = uploader.process_next_completion() => {
-                    if !cancel.is_cancelled() {
-                        uploader.spawn_pending();
-                    }
-                }
-            }
-        }
-    }
-
-    Ok(())
-}
-
-impl HeatmapUploader {
-    /// Periodic execution phase: inspect all attached tenants and schedule any work they require.
-    async fn schedule_iteration(&mut self) -> anyhow::Result<()> {
-        // Cull any entries in self.tenants whose Arc<Tenant> is gone
-        self.tenants
-            .retain(|_k, v| v.tenant.upgrade().is_some() && v.next_upload.is_some());
-
-        // The priority order of previously scheduled work may be invalidated by current state: drop
-        // all pending work (it will be re-scheduled if still needed)
-        self.tenants_pending.clear();
-
-        // Used a fixed 'now' through the following loop, for efficiency and fairness.
-        let now = Instant::now();
-
-        // While iterating over the potentially-long list of tenants, we will periodically yield
-        // to avoid blocking executor.
-        const YIELD_ITERATIONS: usize = 1000;
-
-        // Iterate over tenants looking for work to do.
-        let tenants = self.tenant_manager.get_attached_active_tenant_shards();
-        for (i, tenant) in tenants.into_iter().enumerate() {
-            // Process is shutting down, drop out
-            if self.cancel.is_cancelled() {
-                return Ok(());
-            }
-
-            // Skip tenants that already have a write in flight
-            if self
-                .tenants_uploading
-                .contains_key(tenant.get_tenant_shard_id())
-            {
-                continue;
-            }
-
-            self.maybe_schedule_upload(&now, tenant);
-
-            if i + 1 % YIELD_ITERATIONS == 0 {
-                tokio::task::yield_now().await;
-            }
-        }
-
-        // Spawn tasks for as many of our pending tenants as we can.
-        self.spawn_pending();
-
-        Ok(())
-    }
-
-    ///
-    /// Cancellation: this method is cancel-safe.
-    async fn process_next_completion(&mut self) {
-        match self.task_result_rx.recv().await {
-            Some(r) => {
-                self.on_completion(r);
-            }
-            None => {
-                unreachable!("Result sender is stored on Self");
-            }
-        }
-    }
-
-    /// The 'maybe' refers to the tenant's state: whether it is configured
-    /// for heatmap uploads at all, and whether sufficient time has passed
-    /// since the last upload.
-    fn maybe_schedule_upload(&mut self, now: &Instant, tenant: Arc<Tenant>) {
-        match tenant.get_heatmap_period() {
-            None => {
-                // Heatmaps are disabled for this tenant
-                return;
-            }
-            Some(period) => {
-                // If any tenant has asked for uploads more frequent than our scheduling interval,
-                // reduce it to match so that we can keep up.  This is mainly useful in testing, where
-                // we may set rather short intervals.
-                if period < self.scheduling_interval {
-                    self.scheduling_interval = std::cmp::max(period, MIN_SCHEDULING_INTERVAL);
-                }
-            }
-        }
-
-        // Stale attachments do not upload anything: if we are in this state, there is probably some
-        // other attachment in mode Single or Multi running on another pageserver, and we don't
-        // want to thrash and overwrite their heatmap uploads.
-        if tenant.get_attach_mode() == AttachmentMode::Stale {
-            return;
-        }
-
-        // Create an entry in self.tenants if one doesn't already exist: this will later be updated
-        // with the completion time in on_completion.
-        let state = self
-            .tenants
-            .entry(*tenant.get_tenant_shard_id())
-            .or_insert_with(|| UploaderTenantState {
-                tenant: Arc::downgrade(&tenant),
-                last_upload: None,
-                next_upload: Some(Instant::now()),
-                last_digest: None,
-            });
-
-        // Decline to do the upload if insufficient time has passed
-        if state.next_upload.map(|nu| &nu > now).unwrap_or(false) {
-            return;
-        }
-
-        let last_digest = state.last_digest;
-        self.tenants_pending.push_back(UploadPending {
-            tenant,
-            last_digest,
-        })
-    }
-
-    fn spawn_pending(&mut self) {
-        while !self.tenants_pending.is_empty()
-            && self.tenants_uploading.len() < self.concurrent_uploads
-        {
-            // unwrap: loop condition includes !is_empty()
-            let pending = self.tenants_pending.pop_front().unwrap();
-            self.spawn_upload(pending.tenant, pending.last_digest);
-        }
-    }
-
-    fn spawn_upload(&mut self, tenant: Arc<Tenant>, last_digest: Option<md5::Digest>) {
-        let remote_storage = self.remote_storage.clone();
-        let tenant_shard_id = *tenant.get_tenant_shard_id();
-        let (completion, barrier) = utils::completion::channel();
-        let result_tx = self.task_result_tx.clone();
-        self.tasks.spawn(async move {
-            // Guard for the barrier in [`WriteInProgress`]
-            let _completion = completion;
-
-            let started_at = Instant::now();
-            let digest = match upload_tenant_heatmap(remote_storage, &tenant, last_digest).await {
-                Ok(UploadHeatmapOutcome::Uploaded(digest)) => {
-                    let duration = Instant::now().duration_since(started_at);
-                    SECONDARY_MODE
-                        .upload_heatmap_duration
-                        .observe(duration.as_secs_f64());
-                    SECONDARY_MODE.upload_heatmap.inc();
-                    Some(digest)
-                }
-                Ok(UploadHeatmapOutcome::NoChange | UploadHeatmapOutcome::Skipped) => last_digest,
-                Err(UploadHeatmapError::Upload(e)) => {
-                    tracing::warn!(
-                        "Failed to upload heatmap for tenant {}: {e:#}",
-                        tenant.get_tenant_shard_id(),
-                    );
-                    let duration = Instant::now().duration_since(started_at);
-                    SECONDARY_MODE
-                        .upload_heatmap_duration
-                        .observe(duration.as_secs_f64());
-                    SECONDARY_MODE.upload_heatmap_errors.inc();
-                    last_digest
-                }
-                Err(UploadHeatmapError::Cancelled) => {
-                    tracing::info!("Cancelled heatmap upload, shutting down");
-                    last_digest
-                }
-            };
-
-            let now = Instant::now();
-            let next_upload = tenant
-                .get_heatmap_period()
-                .and_then(|period| now.checked_add(period));
-
-            result_tx
-                .send(WriteComplete {
-                    tenant_shard_id: *tenant.get_tenant_shard_id(),
-                    completed_at: now,
-                    digest,
-                    next_upload,
-                })
-                .ok();
-        });
-
-        self.tenants_uploading
-            .insert(tenant_shard_id, WriteInProgress { barrier });
-    }
-
-    #[instrument(skip_all, fields(tenant_id=%completion.tenant_shard_id.tenant_id, shard_id=%completion.tenant_shard_id.shard_slug()))]
-    fn on_completion(&mut self, completion: WriteComplete) {
-        tracing::debug!("Heatmap upload completed");
-        let WriteComplete {
-            tenant_shard_id,
-            completed_at,
-            digest,
-            next_upload,
-        } = completion;
-        self.tenants_uploading.remove(&tenant_shard_id);
-        use std::collections::hash_map::Entry;
-        match self.tenants.entry(tenant_shard_id) {
-            Entry::Vacant(_) => {
-                // Tenant state was dropped, nothing to update.
-            }
-            Entry::Occupied(mut entry) => {
-                entry.get_mut().last_upload = Some(completed_at);
-                entry.get_mut().last_digest = digest;
-                entry.get_mut().next_upload = next_upload
-            }
-        }
-    }
-
-    fn handle_command(
-        &mut self,
-        command: UploadCommand,
-        response_tx: tokio::sync::oneshot::Sender<CommandResponse>,
-    ) {
-        match command {
-            UploadCommand::Upload(tenant_shard_id) => {
-                // If an upload was ongoing for this tenant, let it finish first.
-                let barrier = if let Some(writing_state) =
-                    self.tenants_uploading.get(&tenant_shard_id)
-                {
-                    tracing::info!(
-                        tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(),
-                        "Waiting for heatmap write to complete");
-                    writing_state.barrier.clone()
-                } else {
-                    // Spawn the upload then immediately wait for it.  This will block processing of other commands and
-                    // starting of other background work.
-                    tracing::info!(
-                        tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(),
-                        "Starting heatmap write on command");
-                    let tenant = match self
-                        .tenant_manager
-                        .get_attached_tenant_shard(tenant_shard_id, true)
-                    {
-                        Ok(t) => t,
-                        Err(e) => {
-                            // Drop result of send: we don't care if caller dropped their receiver
-                            drop(response_tx.send(CommandResponse {
-                                result: Err(e.into()),
-                            }));
-                            return;
-                        }
-                    };
-                    self.spawn_upload(tenant, None);
-                    let writing_state = self
-                        .tenants_uploading
-                        .get(&tenant_shard_id)
-                        .expect("We just inserted this");
-                    tracing::info!(
-                        tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(),
-                        "Waiting for heatmap upload to complete");
-
-                    writing_state.barrier.clone()
-                };
-
-                // This task does no I/O: it only listens for a barrier's completion and then
-                // sends to the command response channel.  It is therefore safe to spawn this without
-                // any gates/task_mgr hooks.
-                tokio::task::spawn(async move {
-                    barrier.wait().await;
-
-                    tracing::info!(
-                        tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(),
-                        "Heatmap upload complete");
-
-                    // Drop result of send: we don't care if caller dropped their receiver
-                    drop(response_tx.send(CommandResponse { result: Ok(()) }))
-                });
-            }
-        }
-    }
-}
-
-enum UploadHeatmapOutcome {
-    /// We successfully wrote to remote storage, with this digest.
-    Uploaded(md5::Digest),
-    /// We did not upload because the heatmap digest was unchanged since the last upload
-    NoChange,
-    /// We skipped the upload for some reason, such as tenant/timeline not ready
-    Skipped,
-}
-
-#[derive(thiserror::Error, Debug)]
-enum UploadHeatmapError {
-    #[error("Cancelled")]
-    Cancelled,
-
-    #[error(transparent)]
-    Upload(#[from] anyhow::Error),
-}
-
-/// The inner upload operation.  This will skip if `last_digest` is Some and matches the digest
-/// of the object we would have uploaded.
-#[instrument(skip_all, fields(tenant_id = %tenant.get_tenant_shard_id().tenant_id, shard_id = %tenant.get_tenant_shard_id().shard_slug()))]
-async fn upload_tenant_heatmap(
-    remote_storage: GenericRemoteStorage,
-    tenant: &Arc<Tenant>,
-    last_digest: Option<md5::Digest>,
-) -> Result<UploadHeatmapOutcome, UploadHeatmapError> {
-    debug_assert_current_span_has_tenant_id();
-
-    let generation = tenant.get_generation();
-    if generation.is_none() {
-        // We do not expect this: generations were implemented before heatmap uploads.  However,
-        // handle it so that we don't have to make the generation in the heatmap an Option<>
-        // (Generation::none is not serializable)
-        tracing::warn!("Skipping heatmap upload for tenant with generation==None");
-        return Ok(UploadHeatmapOutcome::Skipped);
-    }
-
-    let mut heatmap = HeatMapTenant {
-        timelines: Vec::new(),
-        generation,
-    };
-    let timelines = tenant.timelines.lock().unwrap().clone();
-
-    let tenant_cancel = tenant.cancel.clone();
-
-    // Ensure that Tenant::shutdown waits for any upload in flight: this is needed because otherwise
-    // when we delete a tenant, we might race with an upload in flight and end up leaving a heatmap behind
-    // in remote storage.
-    let _guard = match tenant.gate.enter() {
-        Ok(g) => g,
-        Err(_) => {
-            tracing::info!("Skipping heatmap upload for tenant which is shutting down");
-            return Err(UploadHeatmapError::Cancelled);
-        }
-    };
-
-    for (timeline_id, timeline) in timelines {
-        let heatmap_timeline = timeline.generate_heatmap().await;
-        match heatmap_timeline {
-            None => {
-                tracing::debug!(
-                    "Skipping heatmap upload because timeline {timeline_id} is not ready"
-                );
-                return Ok(UploadHeatmapOutcome::Skipped);
-            }
-            Some(heatmap_timeline) => {
-                heatmap.timelines.push(heatmap_timeline);
-            }
-        }
-    }
-
-    // Serialize the heatmap
-    let bytes = serde_json::to_vec(&heatmap).map_err(|e| anyhow::anyhow!(e))?;
-    let size = bytes.len();
-
-    // Drop out early if nothing changed since our last upload
-    let digest = md5::compute(&bytes);
-    if Some(digest) == last_digest {
-        return Ok(UploadHeatmapOutcome::NoChange);
-    }
-
-    let path = remote_heatmap_path(tenant.get_tenant_shard_id());
-
-    // Write the heatmap.
-    tracing::debug!("Uploading {size} byte heatmap to {path}");
-    if let Err(e) = backoff::retry(
-        || async {
-            let bytes = futures::stream::once(futures::future::ready(Ok(bytes::Bytes::from(
-                bytes.clone(),
-            ))));
-            remote_storage
-                .upload_storage_object(bytes, size, &path)
-                .await
-        },
-        |_| false,
-        3,
-        u32::MAX,
-        "Uploading heatmap",
-        backoff::Cancel::new(tenant_cancel.clone(), || anyhow::anyhow!("Shutting down")),
-    )
-    .await
-    {
-        if tenant_cancel.is_cancelled() {
-            return Err(UploadHeatmapError::Cancelled);
-        } else {
-            return Err(e.into());
-        }
-    }
-
-    tracing::info!("Successfully uploaded {size} byte heatmap to {path}");
-
-    Ok(UploadHeatmapOutcome::Uploaded(digest))
-}

pageserver/src/tenant/storage_layer/layer.rs

@@ -259,9 +259,8 @@ impl Layer {
 
         layer
             .get_value_reconstruct_data(key, lsn_range, reconstruct_data, &self.0, ctx)
-            .instrument(tracing::debug_span!("get_value_reconstruct_data", layer=%self))
+            .instrument(tracing::info_span!("get_value_reconstruct_data", layer=%self))
             .await
-            .with_context(|| format!("get_value_reconstruct_data for layer {self}"))
     }
 
     /// Download the layer if evicted.
@@ -458,8 +457,6 @@ struct LayerInner {
     /// For loaded layers, this may be some other value if the tenant has undergone
     /// a shard split since the layer was originally written.
     shard: ShardIndex,
-
-    last_evicted_at: std::sync::Mutex<Option<std::time::Instant>>,
 }
 
 impl std::fmt::Display for LayerInner {
@@ -590,7 +587,6 @@ impl LayerInner {
             consecutive_failures: AtomicUsize::new(0),
             generation,
             shard,
-            last_evicted_at: std::sync::Mutex::default(),
         }
     }
 
@@ -655,6 +651,7 @@ impl LayerInner {
     }
 
     /// Cancellation safe.
+    #[tracing::instrument(skip_all, fields(layer=%self))]
     async fn get_or_maybe_download(
         self: &Arc<Self>,
         allow_download: bool,
@@ -663,101 +660,87 @@ impl LayerInner {
         let mut init_permit = None;
 
         loop {
-            let download = move |permit| {
-                async move {
-                    // disable any scheduled but not yet running eviction deletions for this
-                    let next_version = 1 + self.version.fetch_add(1, Ordering::Relaxed);
+            let download = move |permit| async move {
+                // disable any scheduled but not yet running eviction deletions for this
+                let next_version = 1 + self.version.fetch_add(1, Ordering::Relaxed);
 
-                    // count cancellations, which currently remain largely unexpected
-                    let init_cancelled =
-                        scopeguard::guard((), |_| LAYER_IMPL_METRICS.inc_init_cancelled());
+                // count cancellations, which currently remain largely unexpected
+                let init_cancelled =
+                    scopeguard::guard((), |_| LAYER_IMPL_METRICS.inc_init_cancelled());
 
-                    // no need to make the evict_and_wait wait for the actual download to complete
-                    drop(self.status.send(Status::Downloaded));
+                // no need to make the evict_and_wait wait for the actual download to complete
+                drop(self.status.send(Status::Downloaded));
 
-                    let timeline = self
-                        .timeline
-                        .upgrade()
-                        .ok_or_else(|| DownloadError::TimelineShutdown)?;
+                let timeline = self
+                    .timeline
+                    .upgrade()
+                    .ok_or_else(|| DownloadError::TimelineShutdown)?;
 
-                    // FIXME: grab a gate
+                // FIXME: grab a gate
 
-                    let can_ever_evict = timeline.remote_client.as_ref().is_some();
+                let can_ever_evict = timeline.remote_client.as_ref().is_some();
 
-                    // check if we really need to be downloaded; could have been already downloaded by a
-                    // cancelled previous attempt.
-                    let needs_download = self
-                        .needs_download()
-                        .await
-                        .map_err(DownloadError::PreStatFailed)?;
+                // check if we really need to be downloaded; could have been already downloaded by a
+                // cancelled previous attempt.
+                let needs_download = self
+                    .needs_download()
+                    .await
+                    .map_err(DownloadError::PreStatFailed)?;
 
-                    let permit = if let Some(reason) = needs_download {
-                        if let NeedsDownload::NotFile(ft) = reason {
-                            return Err(DownloadError::NotFile(ft));
-                        }
-
-                        // only reset this after we've decided we really need to download. otherwise it'd
-                        // be impossible to mark cancelled downloads for eviction, like one could imagine
-                        // we would like to do for prefetching which was not needed.
-                        self.wanted_evicted.store(false, Ordering::Release);
-
-                        if !can_ever_evict {
-                            return Err(DownloadError::NoRemoteStorage);
-                        }
-
-                        if let Some(ctx) = ctx {
-                            self.check_expected_download(ctx)?;
-                        }
-
-                        if !allow_download {
-                            // this does look weird, but for LayerInner the "downloading" means also changing
-                            // internal once related state ...
-                            return Err(DownloadError::DownloadRequired);
-                        }
-
-                        tracing::info!(%reason, "downloading on-demand");
-
-                        self.spawn_download_and_wait(timeline, permit).await?
-                    } else {
-                        // the file is present locally, probably by a previous but cancelled call to
-                        // get_or_maybe_download. alternatively we might be running without remote storage.
-                        LAYER_IMPL_METRICS.inc_init_needed_no_download();
-
-                        permit
-                    };
-
-                    let since_last_eviction =
-                        self.last_evicted_at.lock().unwrap().map(|ts| ts.elapsed());
-                    if let Some(since_last_eviction) = since_last_eviction {
-                        // FIXME: this will not always be recorded correctly until #6028 (the no
-                        // download needed branch above)
-                        LAYER_IMPL_METRICS.record_redownloaded_after(since_last_eviction);
+                let permit = if let Some(reason) = needs_download {
+                    if let NeedsDownload::NotFile(ft) = reason {
+                        return Err(DownloadError::NotFile(ft));
                     }
 
-                    let res = Arc::new(DownloadedLayer {
-                        owner: Arc::downgrade(self),
-                        kind: tokio::sync::OnceCell::default(),
-                        version: next_version,
-                    });
+                    // only reset this after we've decided we really need to download. otherwise it'd
+                    // be impossible to mark cancelled downloads for eviction, like one could imagine
+                    // we would like to do for prefetching which was not needed.
+                    self.wanted_evicted.store(false, Ordering::Release);
 
-                    self.access_stats.record_residence_event(
-                        LayerResidenceStatus::Resident,
-                        LayerResidenceEventReason::ResidenceChange,
-                    );
-
-                    let waiters = self.inner.initializer_count();
-                    if waiters > 0 {
-                        tracing::info!(
-                            waiters,
-                            "completing the on-demand download for other tasks"
-                        );
+                    if !can_ever_evict {
+                        return Err(DownloadError::NoRemoteStorage);
                     }
 
-                    scopeguard::ScopeGuard::into_inner(init_cancelled);
+                    if let Some(ctx) = ctx {
+                        self.check_expected_download(ctx)?;
+                    }
 
-                    Ok((ResidentOrWantedEvicted::Resident(res), permit))
+                    if !allow_download {
+                        // this does look weird, but for LayerInner the "downloading" means also changing
+                        // internal once related state ...
+                        return Err(DownloadError::DownloadRequired);
+                    }
+
+                    tracing::info!(%reason, "downloading on-demand");
+
+                    self.spawn_download_and_wait(timeline, permit).await?
+                } else {
+                    // the file is present locally, probably by a previous but cancelled call to
+                    // get_or_maybe_download. alternatively we might be running without remote storage.
+                    LAYER_IMPL_METRICS.inc_init_needed_no_download();
+
+                    permit
+                };
+
+                let res = Arc::new(DownloadedLayer {
+                    owner: Arc::downgrade(self),
+                    kind: tokio::sync::OnceCell::default(),
+                    version: next_version,
+                });
+
+                self.access_stats.record_residence_event(
+                    LayerResidenceStatus::Resident,
+                    LayerResidenceEventReason::ResidenceChange,
+                );
+
+                let waiters = self.inner.initializer_count();
+                if waiters > 0 {
+                    tracing::info!(waiters, "completing the on-demand download for other tasks");
                 }
-                .instrument(tracing::info_span!("get_or_maybe_download", layer=%self))
+
+                scopeguard::ScopeGuard::into_inner(init_cancelled);
+
+                Ok((ResidentOrWantedEvicted::Resident(res), permit))
             };
 
             if let Some(init_permit) = init_permit.take() {
@@ -868,7 +851,6 @@ impl LayerInner {
                 let result = client.download_layer_file(
                     &this.desc.filename(),
                     &this.metadata(),
-                    &crate::task_mgr::shutdown_token()
                 )
                 .await;
 
@@ -1135,8 +1117,6 @@ impl LayerInner {
         // we are still holding the permit, so no new spawn_download_and_wait can happen
         drop(self.status.send(Status::Evicted));
 
-        *self.last_evicted_at.lock().unwrap() = Some(std::time::Instant::now());
-
         res
     }
 
@@ -1441,7 +1421,6 @@ pub(crate) struct LayerImplMetrics {
 
     rare_counters: enum_map::EnumMap<RareEvent, IntCounter>,
     inits_cancelled: metrics::core::GenericCounter<metrics::core::AtomicU64>,
-    redownload_after: metrics::Histogram,
 }
 
 impl Default for LayerImplMetrics {
@@ -1517,26 +1496,6 @@ impl Default for LayerImplMetrics {
         )
         .unwrap();
 
-        let redownload_after = {
-            let minute = 60.0;
-            let hour = 60.0 * minute;
-            metrics::register_histogram!(
-                "pageserver_layer_redownloaded_after",
-                "Time between evicting and re-downloading.",
-                vec![
-                    10.0,
-                    30.0,
-                    minute,
-                    5.0 * minute,
-                    15.0 * minute,
-                    30.0 * minute,
-                    hour,
-                    12.0 * hour,
-                ]
-            )
-            .unwrap()
-        };
-
         Self {
             started_evictions,
             completed_evictions,
@@ -1548,7 +1507,6 @@ impl Default for LayerImplMetrics {
 
             rare_counters,
             inits_cancelled,
-            redownload_after,
         }
     }
 }
@@ -1616,10 +1574,6 @@ impl LayerImplMetrics {
     fn inc_init_cancelled(&self) {
         self.inits_cancelled.inc()
     }
-
-    fn record_redownloaded_after(&self, duration: std::time::Duration) {
-        self.redownload_after.observe(duration.as_secs_f64())
-    }
 }
 
 #[derive(enum_map::Enum)]

pageserver/src/tenant/tasks.rs

@@ -54,18 +54,31 @@ impl BackgroundLoopKind {
     }
 }
 
-/// Cancellation safe.
-pub(crate) async fn concurrent_background_tasks_rate_limit_permit(
+pub(crate) enum RateLimitError {
+    Cancelled,
+}
+
+pub(crate) async fn concurrent_background_tasks_rate_limit(
     loop_kind: BackgroundLoopKind,
     _ctx: &RequestContext,
-) -> impl Drop {
-    let _guard = crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_GAUGE
+    cancel: &CancellationToken,
+) -> Result<impl Drop, RateLimitError> {
+    crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_START_COUNT
         .with_label_values(&[loop_kind.as_static_str()])
-        .guard();
-
-    match CONCURRENT_BACKGROUND_TASKS.acquire().await {
-        Ok(permit) => permit,
-        Err(_closed) => unreachable!("we never close the semaphore"),
+        .inc();
+    scopeguard::defer!(
+        crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_FINISH_COUNT.with_label_values(&[loop_kind.as_static_str()]).inc();
+    );
+    tokio::select! {
+        permit = CONCURRENT_BACKGROUND_TASKS.acquire() => {
+            match permit {
+                Ok(permit) => Ok(permit),
+                Err(_closed) => unreachable!("we never close the semaphore"),
+            }
+        },
+        _ = cancel.cancelled() => {
+            Err(RateLimitError::Cancelled)
+        }
     }
 }
 

pageserver/src/tenant/timeline.rs

@@ -51,7 +51,7 @@ use crate::tenant::storage_layer::{
     LayerAccessStatsReset, LayerFileName, ResidentLayer, ValueReconstructResult,
     ValueReconstructState,
 };
-use crate::tenant::tasks::BackgroundLoopKind;
+use crate::tenant::tasks::{BackgroundLoopKind, RateLimitError};
 use crate::tenant::timeline::logical_size::CurrentLogicalSize;
 use crate::tenant::{
     layer_map::{LayerMap, SearchResult},
@@ -98,9 +98,8 @@ use self::logical_size::LogicalSize;
 use self::walreceiver::{WalReceiver, WalReceiverConf};
 
 use super::config::TenantConf;
-use super::remote_timeline_client::index::{IndexLayerMetadata, IndexPart};
+use super::remote_timeline_client::index::IndexPart;
 use super::remote_timeline_client::RemoteTimelineClient;
-use super::secondary::heatmap::{HeatMapLayer, HeatMapTimeline};
 use super::{debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf};
 
 #[derive(Debug, PartialEq, Eq, Clone, Copy)]
@@ -446,12 +445,6 @@ pub(crate) enum CompactFlags {
     ForceRepartition,
 }
 
-impl std::fmt::Debug for Timeline {
-    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
-        write!(f, "Timeline<{}>", self.timeline_id)
-    }
-}
-
 /// Public interface functions
 impl Timeline {
     /// Get the LSN where this branch was created
@@ -715,27 +708,19 @@ impl Timeline {
         flags: EnumSet<CompactFlags>,
         ctx: &RequestContext,
     ) -> Result<(), CompactionError> {
-        // most likely the cancellation token is from background task, but in tests it could be the
-        // request task as well.
-
-        let prepare = async move {
-            let guard = self.compaction_lock.lock().await;
-
-            let permit = super::tasks::concurrent_background_tasks_rate_limit_permit(
-                BackgroundLoopKind::Compaction,
-                ctx,
-            )
-            .await;
-
-            (guard, permit)
-        };
+        let _g = self.compaction_lock.lock().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! {
-            tuple = prepare => { tuple },
-            _ = self.cancel.cancelled() => return Ok(()),
-            _ = cancel.cancelled() => return Ok(()),
+        let _permit = match super::tasks::concurrent_background_tasks_rate_limit(
+            BackgroundLoopKind::Compaction,
+            ctx,
+            cancel,
+        )
+        .await
+        {
+            Ok(permit) => permit,
+            Err(RateLimitError::Cancelled) => return Ok(()),
         };
 
         let last_record_lsn = self.get_last_record_lsn();
@@ -1734,7 +1719,6 @@ impl Timeline {
                 self.current_logical_size.current_size().accuracy(),
                 logical_size::Accuracy::Exact,
             );
-            self.current_logical_size.initialized.add_permits(1);
             return;
         };
 
@@ -1780,11 +1764,6 @@ impl Timeline {
         cancel: CancellationToken,
         background_ctx: RequestContext,
     ) {
-        scopeguard::defer! {
-            // Irrespective of the outcome of this operation, we should unblock anyone waiting for it.
-            self.current_logical_size.initialized.add_permits(1);
-        }
-
         enum BackgroundCalculationError {
             Cancelled,
             Other(anyhow::Error),
@@ -1796,22 +1775,22 @@ impl Timeline {
             let skip_concurrency_limiter = &skip_concurrency_limiter;
             async move {
                 let cancel = task_mgr::shutdown_token();
-                let wait_for_permit = super::tasks::concurrent_background_tasks_rate_limit_permit(
+                let wait_for_permit = super::tasks::concurrent_background_tasks_rate_limit(
                     BackgroundLoopKind::InitialLogicalSizeCalculation,
                     background_ctx,
+                    &cancel,
                 );
 
                 use crate::metrics::initial_logical_size::StartCircumstances;
                 let (_maybe_permit, circumstances) = tokio::select! {
-                    permit = wait_for_permit => {
-                        (Some(permit), StartCircumstances::AfterBackgroundTasksRateLimit)
+                    res = wait_for_permit => {
+                        match res {
+                            Ok(permit) => (Some(permit), StartCircumstances::AfterBackgroundTasksRateLimit),
+                            Err(RateLimitError::Cancelled) => {
+                                return Err(BackgroundCalculationError::Cancelled);
+                            }
+                        }
                     }
-                    _ = self_ref.cancel.cancelled() => {
-                        return Err(BackgroundCalculationError::Cancelled);
-                    }
-                    _ = cancel.cancelled() => {
-                        return Err(BackgroundCalculationError::Cancelled);
-                    },
                     () = skip_concurrency_limiter.cancelled() => {
                         // Some action that is part of a end user interaction requested logical size
                         // => break out of the rate limit
@@ -2076,55 +2055,6 @@ impl Timeline {
 
         None
     }
-
-    /// The timeline heatmap is a hint to secondary locations from the primary location,
-    /// indicating which layers are currently on-disk on the primary.
-    ///
-    /// None is returned if the Timeline is in a state where uploading a heatmap
-    /// doesn't make sense, such as shutting down or initializing.  The caller
-    /// should treat this as a cue to simply skip doing any heatmap uploading
-    /// for this timeline.
-    pub(crate) async fn generate_heatmap(&self) -> Option<HeatMapTimeline> {
-        let eviction_info = self.get_local_layers_for_disk_usage_eviction().await;
-
-        let remote_client = match &self.remote_client {
-            Some(c) => c,
-            None => return None,
-        };
-
-        let layer_file_names = eviction_info
-            .resident_layers
-            .iter()
-            .map(|l| l.layer.layer_desc().filename())
-            .collect::<Vec<_>>();
-
-        let decorated = match remote_client.get_layers_metadata(layer_file_names) {
-            Ok(d) => d,
-            Err(_) => {
-                // Getting metadata only fails on Timeline in bad state.
-                return None;
-            }
-        };
-
-        let heatmap_layers = std::iter::zip(
-            eviction_info.resident_layers.into_iter(),
-            decorated.into_iter(),
-        )
-        .filter_map(|(layer, remote_info)| {
-            remote_info.map(|remote_info| {
-                HeatMapLayer::new(
-                    layer.layer.layer_desc().filename(),
-                    IndexLayerMetadata::from(remote_info),
-                    layer.last_activity_ts,
-                )
-            })
-        });
-
-        Some(HeatMapTimeline::new(
-            self.timeline_id,
-            heatmap_layers.collect(),
-        ))
-    }
 }
 
 type TraversalId = String;
@@ -3110,32 +3040,6 @@ impl Timeline {
 
         Ok(image_layers)
     }
-
-    /// Wait until the background initial logical size calculation is complete, or
-    /// this Timeline is shut down.  Calling this function will cause the initial
-    /// logical size calculation to skip waiting for the background jobs barrier.
-    pub(crate) async fn await_initial_logical_size(self: Arc<Self>) {
-        if let Some(await_bg_cancel) = self
-            .current_logical_size
-            .cancel_wait_for_background_loop_concurrency_limit_semaphore
-            .get()
-        {
-            await_bg_cancel.cancel();
-        } else {
-            // We should not wait if we were not able to explicitly instruct
-            // the logical size cancellation to skip the concurrency limit semaphore.
-            // TODO: this is an unexpected case.  We should restructure so that it
-            // can't happen.
-            tracing::info!(
-                "await_initial_logical_size: can't get semaphore cancel token, skipping"
-            );
-        }
-
-        tokio::select!(
-            _ = self.current_logical_size.initialized.acquire() => {},
-            _ = self.cancel.cancelled() => {}
-        )
-    }
 }
 
 #[derive(Default)]
@@ -3892,14 +3796,7 @@ impl Timeline {
     /// within a layer file. We can only remove the whole file if it's fully
     /// obsolete.
     pub(super) async fn gc(&self) -> anyhow::Result<GcResult> {
-        // this is most likely the background tasks, but it might be the spawned task from
-        // immediate_gc
-        let cancel = crate::task_mgr::shutdown_token();
-        let _g = tokio::select! {
-            guard = self.gc_lock.lock() => guard,
-            _ = self.cancel.cancelled() => return Ok(GcResult::default()),
-            _ = cancel.cancelled() => return Ok(GcResult::default()),
-        };
+        let _g = self.gc_lock.lock().await;
         let timer = self.metrics.garbage_collect_histo.start_timer();
 
         fail_point!("before-timeline-gc");

pageserver/src/tenant/timeline/eviction_task.rs

@@ -30,7 +30,7 @@ use crate::{
     task_mgr::{self, TaskKind, BACKGROUND_RUNTIME},
     tenant::{
         config::{EvictionPolicy, EvictionPolicyLayerAccessThreshold},
-        tasks::BackgroundLoopKind,
+        tasks::{BackgroundLoopKind, RateLimitError},
         timeline::EvictionError,
         LogicalSizeCalculationCause, Tenant,
     },
@@ -158,15 +158,15 @@ impl Timeline {
     ) -> ControlFlow<()> {
         let now = SystemTime::now();
 
-        let acquire_permit = crate::tenant::tasks::concurrent_background_tasks_rate_limit_permit(
+        let _permit = match crate::tenant::tasks::concurrent_background_tasks_rate_limit(
             BackgroundLoopKind::Eviction,
             ctx,
-        );
-
-        let _permit = tokio::select! {
-            permit = acquire_permit => permit,
-            _ = cancel.cancelled() => return ControlFlow::Break(()),
-            _ = self.cancel.cancelled() => return ControlFlow::Break(()),
+            cancel,
+        )
+        .await
+        {
+            Ok(permit) => permit,
+            Err(RateLimitError::Cancelled) => return ControlFlow::Break(()),
         };
 
         // If we evict layers but keep cached values derived from those layers, then

pageserver/src/tenant/timeline/logical_size.rs

@@ -34,9 +34,6 @@ pub(super) struct LogicalSize {
     pub(crate) cancel_wait_for_background_loop_concurrency_limit_semaphore:
         OnceCell<CancellationToken>,
 
-    /// Once the initial logical size is initialized, this is notified.
-    pub(crate) initialized: tokio::sync::Semaphore,
-
     /// Latest Lsn that has its size uncalculated, could be absent for freshly created timelines.
     pub initial_part_end: Option<Lsn>,
 
@@ -128,7 +125,6 @@ impl LogicalSize {
             initial_part_end: None,
             size_added_after_initial: AtomicI64::new(0),
             did_return_approximate_to_walreceiver: AtomicBool::new(false),
-            initialized: tokio::sync::Semaphore::new(0),
         }
     }
 
@@ -139,7 +135,6 @@ impl LogicalSize {
             initial_part_end: Some(compute_to),
             size_added_after_initial: AtomicI64::new(0),
             did_return_approximate_to_walreceiver: AtomicBool::new(false),
-            initialized: tokio::sync::Semaphore::new(0),
         }
     }
 

pageserver/src/tenant/timeline/uninit.rs

@@ -19,14 +19,14 @@ use super::Timeline;
 pub struct UninitializedTimeline<'t> {
     pub(crate) owning_tenant: &'t Tenant,
     timeline_id: TimelineId,
-    raw_timeline: Option<(Arc<Timeline>, TimelineUninitMark<'t>)>,
+    raw_timeline: Option<(Arc<Timeline>, TimelineUninitMark)>,
 }
 
 impl<'t> UninitializedTimeline<'t> {
     pub(crate) fn new(
         owning_tenant: &'t Tenant,
         timeline_id: TimelineId,
-        raw_timeline: Option<(Arc<Timeline>, TimelineUninitMark<'t>)>,
+        raw_timeline: Option<(Arc<Timeline>, TimelineUninitMark)>,
     ) -> Self {
         Self {
             owning_tenant,
@@ -169,55 +169,18 @@ pub(crate) fn cleanup_timeline_directory(uninit_mark: TimelineUninitMark) {
 ///
 /// XXX: it's important to create it near the timeline dir, not inside it to ensure timeline dir gets removed first.
 #[must_use]
-pub(crate) struct TimelineUninitMark<'t> {
-    owning_tenant: &'t Tenant,
-    timeline_id: TimelineId,
+pub(crate) struct TimelineUninitMark {
     uninit_mark_deleted: bool,
     uninit_mark_path: Utf8PathBuf,
     pub(crate) timeline_path: Utf8PathBuf,
 }
 
-/// Errors when acquiring exclusive access to a timeline ID for creation
-#[derive(thiserror::Error, Debug)]
-pub(crate) enum TimelineExclusionError {
-    #[error("Already exists")]
-    AlreadyExists(Arc<Timeline>),
-    #[error("Already creating")]
-    AlreadyCreating,
-
-    // e.g. I/O errors, or some failure deep in postgres initdb
-    #[error(transparent)]
-    Other(#[from] anyhow::Error),
-}
-
-impl<'t> TimelineUninitMark<'t> {
-    pub(crate) fn new(
-        owning_tenant: &'t Tenant,
-        timeline_id: TimelineId,
-        uninit_mark_path: Utf8PathBuf,
-        timeline_path: Utf8PathBuf,
-    ) -> Result<Self, TimelineExclusionError> {
-        // Lock order: this is the only place we take both locks.  During drop() we only
-        // lock creating_timelines
-        let timelines = owning_tenant.timelines.lock().unwrap();
-        let mut creating_timelines: std::sync::MutexGuard<
-            '_,
-            std::collections::HashSet<TimelineId>,
-        > = owning_tenant.timelines_creating.lock().unwrap();
-
-        if let Some(existing) = timelines.get(&timeline_id) {
-            Err(TimelineExclusionError::AlreadyExists(existing.clone()))
-        } else if creating_timelines.contains(&timeline_id) {
-            Err(TimelineExclusionError::AlreadyCreating)
-        } else {
-            creating_timelines.insert(timeline_id);
-            Ok(Self {
-                owning_tenant,
-                timeline_id,
-                uninit_mark_deleted: false,
-                uninit_mark_path,
-                timeline_path,
-            })
+impl TimelineUninitMark {
+    pub(crate) fn new(uninit_mark_path: Utf8PathBuf, timeline_path: Utf8PathBuf) -> Self {
+        Self {
+            uninit_mark_deleted: false,
+            uninit_mark_path,
+            timeline_path,
         }
     }
 
@@ -244,7 +207,7 @@ impl<'t> TimelineUninitMark<'t> {
     }
 }
 
-impl Drop for TimelineUninitMark<'_> {
+impl Drop for TimelineUninitMark {
     fn drop(&mut self) {
         if !self.uninit_mark_deleted {
             if self.timeline_path.exists() {
@@ -263,11 +226,5 @@ impl Drop for TimelineUninitMark<'_> {
                 }
             }
         }
-
-        self.owning_tenant
-            .timelines_creating
-            .lock()
-            .unwrap()
-            .remove(&self.timeline_id);
     }
 }

pageserver/src/walingest.rs

@@ -2191,7 +2191,7 @@ mod tests {
             .load()
             .await;
         let tline = tenant
-            .bootstrap_timeline_test(TIMELINE_ID, pg_version, None, &ctx)
+            .bootstrap_timeline(TIMELINE_ID, pg_version, None, &ctx)
             .await
             .unwrap();
 

pgxn/neon/neon_utils.c

@@ -1,6 +1,3 @@
-
-#include <sys/resource.h>
-
 #include "postgres.h"
 
 #include "access/timeline.h"
@@ -117,25 +114,3 @@ pq_sendint64_le(StringInfo buf, uint64 i)
 	memcpy(buf->data + buf->len, &i, sizeof(uint64));
 	buf->len += sizeof(uint64);
 }
-
-/*
- * Disables core dump for the current process.
- */
-void
-disable_core_dump()
-{
-	struct rlimit rlim;
-
-#ifdef WALPROPOSER_LIB			/* skip in simulation mode */
-	return;
-#endif
-
-	rlim.rlim_cur = 0;
-	rlim.rlim_max = 0;
-	if (setrlimit(RLIMIT_CORE, &rlim))
-	{
-		int			save_errno = errno;
-
-		fprintf(stderr, "WARNING: disable cores setrlimit failed: %s", strerror(save_errno));
-	}
-}

pgxn/neon/neon_utils.h

@@ -8,6 +8,5 @@ uint32		pq_getmsgint32_le(StringInfo msg);
 uint64		pq_getmsgint64_le(StringInfo msg);
 void		pq_sendint32_le(StringInfo buf, uint32 i);
 void		pq_sendint64_le(StringInfo buf, uint64 i);
-extern void disable_core_dump();
 
 #endif							/* __NEON_UTILS_H__ */

pgxn/neon/walproposer.c

@@ -35,8 +35,6 @@
  *
  *-------------------------------------------------------------------------
  */
-#include <sys/resource.h>
-
 #include "postgres.h"
 #include "libpq/pqformat.h"
 #include "neon.h"
@@ -1071,12 +1069,6 @@ DetermineEpochStartLsn(WalProposer *wp)
 			if (!((dth->n_entries >= 1) && (dth->entries[dth->n_entries - 1].term ==
 											walprop_shared->mineLastElectedTerm)))
 			{
-				/*
-				 * Panic to restart PG as we need to retake basebackup.
-				 * However, don't dump core as this is kinda expected
-				 * scenario.
-				 */
-				disable_core_dump();
 				walprop_log(PANIC,
 							"collected propEpochStartLsn %X/%X, but basebackup LSN %X/%X",
 							LSN_FORMAT_ARGS(wp->propEpochStartLsn),
@@ -1453,12 +1445,7 @@ RecvAppendResponses(Safekeeper *sk)
 
 		if (sk->appendResponse.term > wp->propTerm)
 		{
-			/*
-			 * Another compute with higher term is running. Panic to restart
-			 * PG as we likely need to retake basebackup. However, don't dump
-			 * core as this is kinda expected scenario.
-			 */
-			disable_core_dump();
+			/* Another compute with higher term is running. */
 			walprop_log(PANIC, "WAL acceptor %s:%s with term " INT64_FORMAT " rejected our request, our term " INT64_FORMAT "",
 						sk->host, sk->port,
 						sk->appendResponse.term, wp->propTerm);

proxy/src/auth.rs

@@ -4,7 +4,7 @@ pub mod backend;
 pub use backend::BackendType;
 
 mod credentials;
-pub use credentials::{check_peer_addr_is_in_list, ClientCredentials};
+pub use credentials::{check_peer_addr_is_in_list, ClientCredentials, IpPattern};
 
 mod password_hack;
 pub use password_hack::parse_endpoint_param;

proxy/src/auth/backend.rs

@@ -11,8 +11,7 @@ use crate::auth::validate_password_and_exchange;
 use crate::console::errors::GetAuthInfoError;
 use crate::console::provider::AuthInfo;
 use crate::console::AuthSecret;
-use crate::proxy::connect_compute::handle_try_wake;
-use crate::proxy::retry::retry_after;
+use crate::proxy::{handle_try_wake, retry_after, LatencyTimer};
 use crate::scram;
 use crate::stream::Stream;
 use crate::{
@@ -23,7 +22,6 @@ use crate::{
         provider::{CachedNodeInfo, ConsoleReqExtra},
         Api,
     },
-    metrics::LatencyTimer,
     stream, url,
 };
 use futures::TryFutureExt;
@@ -34,6 +32,8 @@ use std::sync::Arc;
 use tokio::io::{AsyncRead, AsyncWrite};
 use tracing::{error, info, warn};
 
+use super::IpPattern;
+
 /// This type serves two purposes:
 ///
 /// * When `T` is `()`, it's just a regular auth backend selector
@@ -57,7 +57,7 @@ pub enum BackendType<'a, T> {
 
 pub trait TestBackend: Send + Sync + 'static {
     fn wake_compute(&self) -> Result<CachedNodeInfo, console::errors::WakeComputeError>;
-    fn get_allowed_ips(&self) -> Result<Arc<Vec<String>>, console::errors::GetAuthInfoError>;
+    fn get_allowed_ips(&self) -> Result<Arc<Vec<IpPattern>>, console::errors::GetAuthInfoError>;
 }
 
 impl std::fmt::Display for BackendType<'_, ()> {
@@ -390,7 +390,7 @@ impl BackendType<'_, ComputeUserInfo> {
     pub async fn get_allowed_ips(
         &self,
         extra: &ConsoleReqExtra,
-    ) -> Result<Arc<Vec<String>>, GetAuthInfoError> {
+    ) -> Result<Arc<Vec<IpPattern>>, GetAuthInfoError> {
         use BackendType::*;
         match self {
             Console(api, creds) => api.get_allowed_ips(extra, creds).await,

proxy/src/auth/backend/classic.rs

@@ -4,7 +4,7 @@ use crate::{
     compute,
     config::AuthenticationConfig,
     console::AuthSecret,
-    metrics::LatencyTimer,
+    proxy::LatencyTimer,
     sasl,
     stream::{PqStream, Stream},
 };

proxy/src/auth/backend/hacks.rs

@@ -4,7 +4,7 @@ use super::{
 use crate::{
     auth::{self, AuthFlow},
     console::AuthSecret,
-    metrics::LatencyTimer,
+    proxy::LatencyTimer,
     sasl,
     stream::{self, Stream},
 };

proxy/src/auth/credentials.rs

@@ -1,8 +1,9 @@
 //! User credentials used in authentication.
 
 use crate::{
-    auth::password_hack::parse_endpoint_param, error::UserFacingError,
-    metrics::NUM_CONNECTION_ACCEPTED_BY_SNI, proxy::neon_options_str,
+    auth::password_hack::parse_endpoint_param,
+    error::UserFacingError,
+    proxy::{neon_options_str, NUM_CONNECTION_ACCEPTED_BY_SNI},
 };
 use itertools::Itertools;
 use pq_proto::StartupMessageParams;
@@ -152,44 +153,65 @@ impl ClientCredentials {
     }
 }
 
-pub fn check_peer_addr_is_in_list(peer_addr: &IpAddr, ip_list: &Vec<String>) -> bool {
+pub fn check_peer_addr_is_in_list(peer_addr: &IpAddr, ip_list: &[IpPattern]) -> bool {
     if ip_list.is_empty() {
         return true;
     }
     for ip in ip_list {
-        // We expect that all ip addresses from control plane are correct.
-        // However, if some of them are broken, we still can check the others.
-        match parse_ip_pattern(ip) {
-            Ok(pattern) => {
-                if check_ip(peer_addr, &pattern) {
-                    return true;
-                }
-            }
-            Err(err) => warn!("Cannot parse ip: {}; err: {}", ip, err),
+        if check_ip(peer_addr, ip) {
+            return true;
         }
     }
     false
 }
 
 #[derive(Debug, Clone, Eq, PartialEq)]
-enum IpPattern {
+pub enum IpPattern {
     Subnet(ipnet::IpNet),
     Range(IpAddr, IpAddr),
     Single(IpAddr),
 }
 
-fn parse_ip_pattern(pattern: &str) -> anyhow::Result<IpPattern> {
-    if pattern.contains('/') {
-        let subnet: ipnet::IpNet = pattern.parse()?;
-        return Ok(IpPattern::Subnet(subnet));
+impl<'de> serde::Deserialize<'de> for IpPattern {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        let ip = <&'de str>::deserialize(deserializer)?;
+        Ok(Self::from_str_lossy(ip))
     }
-    if let Some((start, end)) = pattern.split_once('-') {
-        let start: IpAddr = start.parse()?;
-        let end: IpAddr = end.parse()?;
-        return Ok(IpPattern::Range(start, end));
+}
+
+impl std::str::FromStr for IpPattern {
+    type Err = anyhow::Error;
+
+    fn from_str(pattern: &str) -> Result<Self, Self::Err> {
+        if pattern.contains('/') {
+            let subnet: ipnet::IpNet = pattern.parse()?;
+            return Ok(IpPattern::Subnet(subnet));
+        }
+        if let Some((start, end)) = pattern.split_once('-') {
+            let start: IpAddr = start.parse()?;
+            let end: IpAddr = end.parse()?;
+            return Ok(IpPattern::Range(start, end));
+        }
+        let addr: IpAddr = pattern.parse()?;
+        Ok(IpPattern::Single(addr))
+    }
+}
+
+impl IpPattern {
+    pub fn from_str_lossy(pattern: &str) -> Self {
+        match pattern.parse() {
+            Ok(pattern) => pattern,
+            Err(err) => {
+                warn!("Cannot parse ip: {}; err: {}", pattern, err);
+                // We expect that all ip addresses from control plane are correct.
+                // However, if some of them are broken, we still can check the others.
+                Self::Single([0, 0, 0, 0].into())
+            }
+        }
     }
-    let addr: IpAddr = pattern.parse()?;
-    Ok(IpPattern::Single(addr))
 }
 
 fn check_ip(ip: &IpAddr, pattern: &IpPattern) -> bool {
@@ -212,6 +234,8 @@ fn subdomain_from_sni(sni: &str, common_name: &str) -> Option<SmolStr> {
 
 #[cfg(test)]
 mod tests {
+    use std::str::FromStr;
+
     use super::*;
     use ClientCredsParseError::*;
 
@@ -413,41 +437,47 @@ mod tests {
     #[test]
     fn test_check_peer_addr_is_in_list() {
         let peer_addr = IpAddr::from([127, 0, 0, 1]);
-        assert!(check_peer_addr_is_in_list(&peer_addr, &vec![]));
+        assert!(check_peer_addr_is_in_list(&peer_addr, &[]));
         assert!(check_peer_addr_is_in_list(
             &peer_addr,
-            &vec!["127.0.0.1".into()]
+            &[IpPattern::from_str_lossy("127.0.0.1")]
         ));
         assert!(!check_peer_addr_is_in_list(
             &peer_addr,
-            &vec!["8.8.8.8".into()]
+            &[IpPattern::from_str_lossy("8.8.8.8")]
         ));
         // If there is an incorrect address, it will be skipped.
         assert!(check_peer_addr_is_in_list(
             &peer_addr,
-            &vec!["88.8.8".into(), "127.0.0.1".into()]
+            &[
+                IpPattern::from_str_lossy("88.8.8"),
+                IpPattern::from_str_lossy("127.0.0.1")
+            ]
         ));
     }
     #[test]
     fn test_parse_ip_v4() -> anyhow::Result<()> {
         let peer_addr = IpAddr::from([127, 0, 0, 1]);
         // Ok
-        assert_eq!(parse_ip_pattern("127.0.0.1")?, IpPattern::Single(peer_addr));
         assert_eq!(
-            parse_ip_pattern("127.0.0.1/31")?,
+            IpPattern::from_str("127.0.0.1")?,
+            IpPattern::Single(peer_addr)
+        );
+        assert_eq!(
+            IpPattern::from_str("127.0.0.1/31")?,
             IpPattern::Subnet(ipnet::IpNet::new(peer_addr, 31)?)
         );
         assert_eq!(
-            parse_ip_pattern("0.0.0.0-200.0.1.2")?,
+            IpPattern::from_str("0.0.0.0-200.0.1.2")?,
             IpPattern::Range(IpAddr::from([0, 0, 0, 0]), IpAddr::from([200, 0, 1, 2]))
         );
 
         // Error
-        assert!(parse_ip_pattern("300.0.1.2").is_err());
-        assert!(parse_ip_pattern("30.1.2").is_err());
-        assert!(parse_ip_pattern("127.0.0.1/33").is_err());
-        assert!(parse_ip_pattern("127.0.0.1-127.0.3").is_err());
-        assert!(parse_ip_pattern("1234.0.0.1-127.0.3.0").is_err());
+        assert!(IpPattern::from_str("300.0.1.2").is_err());
+        assert!(IpPattern::from_str("30.1.2").is_err());
+        assert!(IpPattern::from_str("127.0.0.1/33").is_err());
+        assert!(IpPattern::from_str("127.0.0.1-127.0.3").is_err());
+        assert!(IpPattern::from_str("1234.0.0.1-127.0.3.0").is_err());
         Ok(())
     }
 

proxy/src/compute.rs

@@ -1,10 +1,9 @@
 use crate::{
     auth::parse_endpoint_param, cancellation::CancelClosure, console::errors::WakeComputeError,
-    error::UserFacingError, metrics::NUM_DB_CONNECTIONS_GAUGE, proxy::neon_option,
+    error::UserFacingError, proxy::neon_option,
 };
 use futures::{FutureExt, TryFutureExt};
 use itertools::Itertools;
-use metrics::IntCounterPairGuard;
 use pq_proto::StartupMessageParams;
 use std::{io, net::SocketAddr, time::Duration};
 use thiserror::Error;
@@ -224,8 +223,6 @@ pub struct PostgresConnection {
     pub params: std::collections::HashMap<String, String>,
     /// Query cancellation token.
     pub cancel_closure: CancelClosure,
-
-    _guage: IntCounterPairGuard,
 }
 
 impl ConnCfg {
@@ -234,7 +231,6 @@ impl ConnCfg {
         &self,
         allow_self_signed_compute: bool,
         timeout: Duration,
-        proto: &'static str,
     ) -> Result<PostgresConnection, ConnectionError> {
         let (socket_addr, stream, host) = self.connect_raw(timeout).await?;
 
@@ -268,7 +264,6 @@ impl ConnCfg {
             stream,
             params,
             cancel_closure,
-            _guage: NUM_DB_CONNECTIONS_GAUGE.with_label_values(&[proto]).guard(),
         };
 
         Ok(connection)

proxy/src/console/messages.rs

@@ -2,6 +2,8 @@ use serde::Deserialize;
 use smol_str::SmolStr;
 use std::fmt;
 
+use crate::auth::IpPattern;
+
 /// Generic error response with human-readable description.
 /// Note that we can't always present it to user as is.
 #[derive(Debug, Deserialize)]
@@ -14,7 +16,7 @@ pub struct ConsoleError {
 #[derive(Deserialize)]
 pub struct GetRoleSecret {
     pub role_secret: Box<str>,
-    pub allowed_ips: Option<Vec<Box<str>>>,
+    pub allowed_ips: Option<Vec<IpPattern>>,
 }
 
 // Manually implement debug to omit sensitive info.

proxy/src/console/provider.rs

@@ -4,12 +4,13 @@ pub mod neon;
 
 use super::messages::MetricsAuxInfo;
 use crate::{
-    auth::backend::ComputeUserInfo,
+    auth::{backend::ComputeUserInfo, IpPattern},
     cache::{timed_lru, TimedLru},
     compute, scram,
 };
 use async_trait::async_trait;
 use dashmap::DashMap;
+use smol_str::SmolStr;
 use std::{sync::Arc, time::Duration};
 use tokio::{
     sync::{OwnedSemaphorePermit, Semaphore},
@@ -21,7 +22,7 @@ pub mod errors {
     use crate::{
         error::{io_error, UserFacingError},
         http,
-        proxy::retry::ShouldRetry,
+        proxy::ShouldRetry,
     };
     use thiserror::Error;
 
@@ -229,7 +230,7 @@ pub enum AuthSecret {
 pub struct AuthInfo {
     pub secret: Option<AuthSecret>,
     /// List of IP addresses allowed for the autorization.
-    pub allowed_ips: Vec<String>,
+    pub allowed_ips: Vec<IpPattern>,
 }
 
 /// Info for establishing a connection to a compute node.
@@ -250,7 +251,7 @@ pub struct NodeInfo {
 
 pub type NodeInfoCache = TimedLru<Arc<str>, NodeInfo>;
 pub type CachedNodeInfo = timed_lru::Cached<&'static NodeInfoCache>;
-pub type AllowedIpsCache = TimedLru<Arc<str>, Arc<Vec<String>>>;
+pub type AllowedIpsCache = TimedLru<Arc<str>, Arc<Vec<IpPattern>>>;
 
 /// This will allocate per each call, but the http requests alone
 /// already require a few allocations, so it should be fine.
@@ -267,7 +268,7 @@ pub trait Api {
         &self,
         extra: &ConsoleReqExtra,
         creds: &ComputeUserInfo,
-    ) -> Result<Arc<Vec<String>>, errors::GetAuthInfoError>;
+    ) -> Result<Arc<Vec<IpPattern>>, errors::GetAuthInfoError>;
 
     /// Wake up the compute node and return the corresponding connection info.
     async fn wake_compute(
@@ -282,13 +283,13 @@ pub struct ApiCaches {
     /// Cache for the `wake_compute` API method.
     pub node_info: NodeInfoCache,
     /// Cache for the `get_allowed_ips`. TODO(anna): use notifications listener instead.
-    pub allowed_ips: TimedLru<Arc<str>, Arc<Vec<String>>>,
+    pub allowed_ips: TimedLru<Arc<str>, Arc<Vec<IpPattern>>>,
 }
 
-/// Various caches for [`console`](super).
+/// Per-endpoint semaphore
 pub struct ApiLocks {
     name: &'static str,
-    node_locks: DashMap<Arc<str>, Arc<Semaphore>>,
+    node_locks: DashMap<SmolStr, Arc<Semaphore>>,
     permits: usize,
     timeout: Duration,
     registered: prometheus::IntCounter,
@@ -354,9 +355,9 @@ impl ApiLocks {
         })
     }
 
-    pub async fn get_wake_compute_permit(
+    pub async fn get_permit(
         &self,
-        key: &Arc<str>,
+        key: &SmolStr,
     ) -> Result<WakeComputePermit, errors::WakeComputeError> {
         if self.permits == 0 {
             return Ok(WakeComputePermit { permit: None });

proxy/src/console/provider/mock.rs

@@ -6,7 +6,13 @@ use super::{
     errors::{ApiError, GetAuthInfoError, WakeComputeError},
     AuthInfo, AuthSecret, CachedNodeInfo, ConsoleReqExtra, NodeInfo,
 };
-use crate::{auth::backend::ComputeUserInfo, compute, error::io_error, scram, url::ApiUrl};
+use crate::{
+    auth::{backend::ComputeUserInfo, IpPattern},
+    compute,
+    error::io_error,
+    scram,
+    url::ApiUrl,
+};
 use async_trait::async_trait;
 use futures::TryFutureExt;
 use thiserror::Error;
@@ -85,7 +91,7 @@ impl Api {
             {
                 Some(s) => {
                     info!("got allowed_ips: {s}");
-                    s.split(',').map(String::from).collect()
+                    s.split(',').map(IpPattern::from_str_lossy).collect()
                 }
                 None => vec![],
             };
@@ -154,7 +160,7 @@ impl super::Api for Api {
         &self,
         _extra: &ConsoleReqExtra,
         creds: &ComputeUserInfo,
-    ) -> Result<Arc<Vec<String>>, GetAuthInfoError> {
+    ) -> Result<Arc<Vec<IpPattern>>, GetAuthInfoError> {
         Ok(Arc::new(self.do_get_auth_info(creds).await?.allowed_ips))
     }
 

proxy/src/console/provider/neon.rs

@@ -5,11 +5,13 @@ use super::{
     errors::{ApiError, GetAuthInfoError, WakeComputeError},
     ApiCaches, ApiLocks, AuthInfo, AuthSecret, CachedNodeInfo, ConsoleReqExtra, NodeInfo,
 };
-use crate::metrics::{ALLOWED_IPS_BY_CACHE_OUTCOME, ALLOWED_IPS_NUMBER};
 use crate::{auth::backend::ComputeUserInfo, compute, http, scram};
+use crate::{
+    auth::IpPattern,
+    proxy::{ALLOWED_IPS_BY_CACHE_OUTCOME, ALLOWED_IPS_NUMBER},
+};
 use async_trait::async_trait;
 use futures::TryFutureExt;
-use itertools::Itertools;
 use std::{net::SocketAddr, sync::Arc};
 use tokio::time::Instant;
 use tokio_postgres::config::SslMode;
@@ -82,12 +84,7 @@ impl Api {
             let secret = scram::ServerSecret::parse(&body.role_secret)
                 .map(AuthSecret::Scram)
                 .ok_or(GetAuthInfoError::BadSecret)?;
-            let allowed_ips = body
-                .allowed_ips
-                .into_iter()
-                .flatten()
-                .map(String::from)
-                .collect_vec();
+            let allowed_ips = body.allowed_ips.unwrap_or_default();
             ALLOWED_IPS_NUMBER.observe(allowed_ips.len() as f64);
             Ok(AuthInfo {
                 secret: Some(secret),
@@ -171,7 +168,7 @@ impl super::Api for Api {
         &self,
         extra: &ConsoleReqExtra,
         creds: &ComputeUserInfo,
-    ) -> Result<Arc<Vec<String>>, GetAuthInfoError> {
+    ) -> Result<Arc<Vec<IpPattern>>, GetAuthInfoError> {
         let key: &str = &creds.endpoint;
         if let Some(allowed_ips) = self.caches.allowed_ips.get(key) {
             ALLOWED_IPS_BY_CACHE_OUTCOME
@@ -195,33 +192,31 @@ impl super::Api for Api {
         extra: &ConsoleReqExtra,
         creds: &ComputeUserInfo,
     ) -> Result<CachedNodeInfo, WakeComputeError> {
-        let key: &str = &creds.inner.cache_key;
+        let key = &creds.inner.cache_key;
 
         // Every time we do a wakeup http request, the compute node will stay up
         // for some time (highly depends on the console's scale-to-zero policy);
         // The connection info remains the same during that period of time,
         // which means that we might cache it to reduce the load and latency.
-        if let Some(cached) = self.caches.node_info.get(key) {
-            info!(key = key, "found cached compute node info");
+        if let Some(cached) = self.caches.node_info.get(&**key) {
+            info!(key = &**key, "found cached compute node info");
             return Ok(cached);
         }
 
-        let key: Arc<str> = key.into();
-
-        let permit = self.locks.get_wake_compute_permit(&key).await?;
+        let permit = self.locks.get_permit(key).await?;
 
         // after getting back a permit - it's possible the cache was filled
         // double check
         if permit.should_check_cache() {
-            if let Some(cached) = self.caches.node_info.get(&key) {
-                info!(key = &*key, "found cached compute node info");
+            if let Some(cached) = self.caches.node_info.get(&**key) {
+                info!(key = &**key, "found cached compute node info");
                 return Ok(cached);
             }
         }
 
         let node = self.do_wake_compute(extra, creds).await?;
-        let (_, cached) = self.caches.node_info.insert(key.clone(), node);
-        info!(key = &*key, "created a cache entry for compute node info");
+        let (_, cached) = self.caches.node_info.insert(key.as_str().into(), node);
+        info!(key = &**key, "created a cache entry for compute node info");
 
         Ok(cached)
     }

proxy/src/http.rs

@@ -13,7 +13,7 @@ pub use reqwest_retry::{policies::ExponentialBackoff, RetryTransientMiddleware};
 use tokio::time::Instant;
 use tracing::trace;
 
-use crate::{metrics::CONSOLE_REQUEST_LATENCY, rate_limiter, url::ApiUrl};
+use crate::{proxy::CONSOLE_REQUEST_LATENCY, rate_limiter, url::ApiUrl};
 use reqwest_middleware::RequestBuilder;
 
 /// This is the preferred way to create new http clients,

proxy/src/lib.rs

@@ -16,7 +16,6 @@ pub mod console;
 pub mod error;
 pub mod http;
 pub mod logging;
-pub mod metrics;
 pub mod parse;
 pub mod protocol2;
 pub mod proxy;

proxy/src/metrics.rs

@@ -1,232 +0,0 @@
-use ::metrics::{
-    exponential_buckets, register_int_counter_pair_vec, register_int_counter_vec,
-    IntCounterPairVec, IntCounterVec,
-};
-use prometheus::{
-    register_histogram, register_histogram_vec, register_int_gauge_vec, Histogram, HistogramVec,
-    IntGaugeVec,
-};
-
-use once_cell::sync::Lazy;
-use tokio::time;
-
-pub static NUM_DB_CONNECTIONS_GAUGE: Lazy<IntCounterPairVec> = Lazy::new(|| {
-    register_int_counter_pair_vec!(
-        "proxy_opened_db_connections_total",
-        "Number of opened connections to a database.",
-        "proxy_closed_db_connections_total",
-        "Number of closed connections to a database.",
-        &["protocol"],
-    )
-    .unwrap()
-});
-
-pub static NUM_CLIENT_CONNECTION_GAUGE: Lazy<IntCounterPairVec> = Lazy::new(|| {
-    register_int_counter_pair_vec!(
-        "proxy_opened_client_connections_total",
-        "Number of opened connections from a client.",
-        "proxy_closed_client_connections_total",
-        "Number of closed connections from a client.",
-        &["protocol"],
-    )
-    .unwrap()
-});
-
-pub static NUM_CONNECTION_REQUESTS_GAUGE: Lazy<IntCounterPairVec> = Lazy::new(|| {
-    register_int_counter_pair_vec!(
-        "proxy_accepted_connections_total",
-        "Number of client connections accepted.",
-        "proxy_closed_connections_total",
-        "Number of client connections closed.",
-        &["protocol"],
-    )
-    .unwrap()
-});
-
-pub static COMPUTE_CONNECTION_LATENCY: Lazy<HistogramVec> = Lazy::new(|| {
-    register_histogram_vec!(
-        "proxy_compute_connection_latency_seconds",
-        "Time it took for proxy to establish a connection to the compute endpoint",
-        // http/ws/tcp, true/false, true/false, success/failure
-        // 3 * 2 * 2 * 2 = 24 counters
-        &["protocol", "cache_miss", "pool_miss", "outcome"],
-        // largest bucket = 2^16 * 0.5ms = 32s
-        exponential_buckets(0.0005, 2.0, 16).unwrap(),
-    )
-    .unwrap()
-});
-
-pub static CONSOLE_REQUEST_LATENCY: Lazy<HistogramVec> = Lazy::new(|| {
-    register_histogram_vec!(
-        "proxy_console_request_latency",
-        "Time it took for proxy to establish a connection to the compute endpoint",
-        // proxy_wake_compute/proxy_get_role_info
-        &["request"],
-        // largest bucket = 2^16 * 0.2ms = 13s
-        exponential_buckets(0.0002, 2.0, 16).unwrap(),
-    )
-    .unwrap()
-});
-
-pub static ALLOWED_IPS_BY_CACHE_OUTCOME: Lazy<IntCounterVec> = Lazy::new(|| {
-    register_int_counter_vec!(
-        "proxy_allowed_ips_cache_misses",
-        "Number of cache hits/misses for allowed ips",
-        // hit/miss
-        &["outcome"],
-    )
-    .unwrap()
-});
-
-pub static RATE_LIMITER_ACQUIRE_LATENCY: Lazy<Histogram> = Lazy::new(|| {
-    register_histogram!(
-        "proxy_control_plane_token_acquire_seconds",
-        "Time it took for proxy to establish a connection to the compute endpoint",
-        // largest bucket = 3^16 * 0.05ms = 2.15s
-        exponential_buckets(0.00005, 3.0, 16).unwrap(),
-    )
-    .unwrap()
-});
-
-pub static RATE_LIMITER_LIMIT: Lazy<IntGaugeVec> = Lazy::new(|| {
-    register_int_gauge_vec!(
-        "semaphore_control_plane_limit",
-        "Current limit of the semaphore control plane",
-        &["limit"], // 2 counters
-    )
-    .unwrap()
-});
-
-pub static NUM_CONNECTION_ACCEPTED_BY_SNI: Lazy<IntCounterVec> = Lazy::new(|| {
-    register_int_counter_vec!(
-        "proxy_accepted_connections_by_sni",
-        "Number of connections (per sni).",
-        &["kind"],
-    )
-    .unwrap()
-});
-
-pub static ALLOWED_IPS_NUMBER: Lazy<Histogram> = Lazy::new(|| {
-    register_histogram!(
-        "proxy_allowed_ips_number",
-        "Number of allowed ips",
-        vec![0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 10.0, 20.0, 50.0, 100.0],
-    )
-    .unwrap()
-});
-
-pub struct LatencyTimer {
-    // time since the stopwatch was started
-    start: Option<time::Instant>,
-    // accumulated time on the stopwatch
-    accumulated: std::time::Duration,
-    // label data
-    protocol: &'static str,
-    cache_miss: bool,
-    pool_miss: bool,
-    outcome: &'static str,
-}
-
-pub struct LatencyTimerPause<'a> {
-    timer: &'a mut LatencyTimer,
-}
-
-impl LatencyTimer {
-    pub fn new(protocol: &'static str) -> Self {
-        Self {
-            start: Some(time::Instant::now()),
-            accumulated: std::time::Duration::ZERO,
-            protocol,
-            cache_miss: false,
-            // by default we don't do pooling
-            pool_miss: true,
-            // assume failed unless otherwise specified
-            outcome: "failed",
-        }
-    }
-
-    pub fn pause(&mut self) -> LatencyTimerPause<'_> {
-        // stop the stopwatch and record the time that we have accumulated
-        let start = self.start.take().expect("latency timer should be started");
-        self.accumulated += start.elapsed();
-        LatencyTimerPause { timer: self }
-    }
-
-    pub fn cache_miss(&mut self) {
-        self.cache_miss = true;
-    }
-
-    pub fn pool_hit(&mut self) {
-        self.pool_miss = false;
-    }
-
-    pub fn success(mut self) {
-        self.outcome = "success";
-    }
-}
-
-impl Drop for LatencyTimerPause<'_> {
-    fn drop(&mut self) {
-        // start the stopwatch again
-        self.timer.start = Some(time::Instant::now());
-    }
-}
-
-impl Drop for LatencyTimer {
-    fn drop(&mut self) {
-        let duration =
-            self.start.map(|start| start.elapsed()).unwrap_or_default() + self.accumulated;
-        COMPUTE_CONNECTION_LATENCY
-            .with_label_values(&[
-                self.protocol,
-                bool_to_str(self.cache_miss),
-                bool_to_str(self.pool_miss),
-                self.outcome,
-            ])
-            .observe(duration.as_secs_f64())
-    }
-}
-
-pub static NUM_CONNECTION_FAILURES: Lazy<IntCounterVec> = Lazy::new(|| {
-    register_int_counter_vec!(
-        "proxy_connection_failures_total",
-        "Number of connection failures (per kind).",
-        &["kind"],
-    )
-    .unwrap()
-});
-
-pub static NUM_WAKEUP_FAILURES: Lazy<IntCounterVec> = Lazy::new(|| {
-    register_int_counter_vec!(
-        "proxy_connection_failures_breakdown",
-        "Number of wake-up failures (per kind).",
-        &["retry", "kind"],
-    )
-    .unwrap()
-});
-
-pub static NUM_BYTES_PROXIED_PER_CLIENT_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
-    register_int_counter_vec!(
-        "proxy_io_bytes_per_client",
-        "Number of bytes sent/received between client and backend.",
-        crate::console::messages::MetricsAuxInfo::TRAFFIC_LABELS,
-    )
-    .unwrap()
-});
-
-pub static NUM_BYTES_PROXIED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
-    register_int_counter_vec!(
-        "proxy_io_bytes",
-        "Number of bytes sent/received between all clients and backends.",
-        &["direction"],
-    )
-    .unwrap()
-});
-
-pub const fn bool_to_str(x: bool) -> &'static str {
-    if x {
-        "true"
-    } else {
-        "false"
-    }
-}

proxy/src/proxy.rs

@@ -1,41 +1,283 @@
 #[cfg(test)]
 mod tests;
 
-pub mod connect_compute;
-pub mod retry;
-
 use crate::{
     auth,
     cancellation::{self, CancelMap},
-    compute,
+    compute::{self, PostgresConnection},
     config::{AuthenticationConfig, ProxyConfig, TlsConfig},
-    console::{self, messages::MetricsAuxInfo},
-    metrics::{
-        LatencyTimer, NUM_BYTES_PROXIED_COUNTER, NUM_BYTES_PROXIED_PER_CLIENT_COUNTER,
-        NUM_CLIENT_CONNECTION_GAUGE, NUM_CONNECTION_REQUESTS_GAUGE,
-    },
+    console::{self, errors::WakeComputeError, messages::MetricsAuxInfo, Api},
+    http::StatusCode,
     protocol2::WithClientIp,
     rate_limiter::EndpointRateLimiter,
     stream::{PqStream, Stream},
     usage_metrics::{Ids, USAGE_METRICS},
 };
 use anyhow::{bail, Context};
+use async_trait::async_trait;
 use futures::TryFutureExt;
 use itertools::Itertools;
-use once_cell::sync::OnceCell;
+use metrics::{exponential_buckets, register_int_counter_vec, IntCounterVec};
+use once_cell::sync::{Lazy, OnceCell};
 use pq_proto::{BeMessage as Be, FeStartupPacket, StartupMessageParams};
+use prometheus::{
+    register_histogram, register_histogram_vec, register_int_gauge_vec, Histogram, HistogramVec,
+    IntGaugeVec,
+};
 use regex::Regex;
-use std::{net::IpAddr, sync::Arc};
-use tokio::io::{AsyncRead, AsyncWrite, AsyncWriteExt};
+use std::{error::Error, io, net::IpAddr, ops::ControlFlow, sync::Arc, time::Instant};
+use tokio::{
+    io::{AsyncRead, AsyncWrite, AsyncWriteExt},
+    time,
+};
 use tokio_util::sync::CancellationToken;
-use tracing::{error, info, info_span, Instrument};
+use tracing::{error, info, info_span, warn, Instrument};
 use utils::measured_stream::MeasuredStream;
 
-use self::connect_compute::{connect_to_compute, TcpMechanism};
+/// Number of times we should retry the `/proxy_wake_compute` http request.
+/// Retry duration is BASE_RETRY_WAIT_DURATION * RETRY_WAIT_EXPONENT_BASE ^ n, where n starts at 0
+pub const NUM_RETRIES_CONNECT: u32 = 16;
+const CONNECT_TIMEOUT: time::Duration = time::Duration::from_secs(2);
+const BASE_RETRY_WAIT_DURATION: time::Duration = time::Duration::from_millis(25);
+const RETRY_WAIT_EXPONENT_BASE: f64 = std::f64::consts::SQRT_2;
 
 const ERR_INSECURE_CONNECTION: &str = "connection is insecure (try using `sslmode=require`)";
 const ERR_PROTO_VIOLATION: &str = "protocol violation";
 
+pub static NUM_DB_CONNECTIONS_OPENED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_opened_db_connections_total",
+        "Number of opened connections to a database.",
+        &["protocol"],
+    )
+    .unwrap()
+});
+
+pub static NUM_DB_CONNECTIONS_CLOSED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_closed_db_connections_total",
+        "Number of closed connections to a database.",
+        &["protocol"],
+    )
+    .unwrap()
+});
+
+pub static NUM_CLIENT_CONNECTION_OPENED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_opened_client_connections_total",
+        "Number of opened connections from a client.",
+        &["protocol"],
+    )
+    .unwrap()
+});
+
+pub static NUM_CLIENT_CONNECTION_CLOSED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_closed_client_connections_total",
+        "Number of closed connections from a client.",
+        &["protocol"],
+    )
+    .unwrap()
+});
+
+pub static NUM_CONNECTIONS_ACCEPTED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_accepted_connections_total",
+        "Number of client connections accepted.",
+        &["protocol"],
+    )
+    .unwrap()
+});
+
+pub static NUM_CONNECTIONS_CLOSED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_closed_connections_total",
+        "Number of client connections closed.",
+        &["protocol"],
+    )
+    .unwrap()
+});
+
+static COMPUTE_CONNECTION_LATENCY: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
+        "proxy_compute_connection_latency_seconds",
+        "Time it took for proxy to establish a connection to the compute endpoint",
+        // http/ws/tcp, true/false, true/false, success/failure
+        // 3 * 2 * 2 * 2 = 24 counters
+        &["protocol", "cache_miss", "pool_miss", "outcome"],
+        // largest bucket = 2^16 * 0.5ms = 32s
+        exponential_buckets(0.0005, 2.0, 16).unwrap(),
+    )
+    .unwrap()
+});
+
+pub static CONSOLE_REQUEST_LATENCY: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
+        "proxy_console_request_latency",
+        "Time it took for proxy to establish a connection to the compute endpoint",
+        // proxy_wake_compute/proxy_get_role_info
+        &["request"],
+        // largest bucket = 2^16 * 0.2ms = 13s
+        exponential_buckets(0.0002, 2.0, 16).unwrap(),
+    )
+    .unwrap()
+});
+
+pub static ALLOWED_IPS_BY_CACHE_OUTCOME: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_allowed_ips_cache_misses",
+        "Number of cache hits/misses for allowed ips",
+        // hit/miss
+        &["outcome"],
+    )
+    .unwrap()
+});
+
+pub static RATE_LIMITER_ACQUIRE_LATENCY: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
+        "proxy_control_plane_token_acquire_seconds",
+        "Time it took for proxy to establish a connection to the compute endpoint",
+        // largest bucket = 3^16 * 0.05ms = 2.15s
+        exponential_buckets(0.00005, 3.0, 16).unwrap(),
+    )
+    .unwrap()
+});
+
+pub static RATE_LIMITER_LIMIT: Lazy<IntGaugeVec> = Lazy::new(|| {
+    register_int_gauge_vec!(
+        "semaphore_control_plane_limit",
+        "Current limit of the semaphore control plane",
+        &["limit"], // 2 counters
+    )
+    .unwrap()
+});
+
+pub static NUM_CONNECTION_ACCEPTED_BY_SNI: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_accepted_connections_by_sni",
+        "Number of connections (per sni).",
+        &["kind"],
+    )
+    .unwrap()
+});
+
+pub static ALLOWED_IPS_NUMBER: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
+        "proxy_allowed_ips_number",
+        "Number of allowed ips",
+        vec![0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 10.0, 20.0, 50.0, 100.0],
+    )
+    .unwrap()
+});
+
+pub struct LatencyTimer {
+    // time since the stopwatch was started
+    start: Option<Instant>,
+    // accumulated time on the stopwatch
+    accumulated: std::time::Duration,
+    // label data
+    protocol: &'static str,
+    cache_miss: bool,
+    pool_miss: bool,
+    outcome: &'static str,
+}
+
+pub struct LatencyTimerPause<'a> {
+    timer: &'a mut LatencyTimer,
+}
+
+impl LatencyTimer {
+    pub fn new(protocol: &'static str) -> Self {
+        Self {
+            start: Some(Instant::now()),
+            accumulated: std::time::Duration::ZERO,
+            protocol,
+            cache_miss: false,
+            // by default we don't do pooling
+            pool_miss: true,
+            // assume failed unless otherwise specified
+            outcome: "failed",
+        }
+    }
+
+    pub fn pause(&mut self) -> LatencyTimerPause<'_> {
+        // stop the stopwatch and record the time that we have accumulated
+        let start = self.start.take().expect("latency timer should be started");
+        self.accumulated += start.elapsed();
+        LatencyTimerPause { timer: self }
+    }
+
+    pub fn cache_miss(&mut self) {
+        self.cache_miss = true;
+    }
+
+    pub fn pool_hit(&mut self) {
+        self.pool_miss = false;
+    }
+
+    pub fn success(mut self) {
+        self.outcome = "success";
+    }
+}
+
+impl Drop for LatencyTimerPause<'_> {
+    fn drop(&mut self) {
+        // start the stopwatch again
+        self.timer.start = Some(Instant::now());
+    }
+}
+
+impl Drop for LatencyTimer {
+    fn drop(&mut self) {
+        let duration =
+            self.start.map(|start| start.elapsed()).unwrap_or_default() + self.accumulated;
+        COMPUTE_CONNECTION_LATENCY
+            .with_label_values(&[
+                self.protocol,
+                bool_to_str(self.cache_miss),
+                bool_to_str(self.pool_miss),
+                self.outcome,
+            ])
+            .observe(duration.as_secs_f64())
+    }
+}
+
+static NUM_CONNECTION_FAILURES: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_connection_failures_total",
+        "Number of connection failures (per kind).",
+        &["kind"],
+    )
+    .unwrap()
+});
+
+static NUM_WAKEUP_FAILURES: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_connection_failures_breakdown",
+        "Number of wake-up failures (per kind).",
+        &["retry", "kind"],
+    )
+    .unwrap()
+});
+
+static NUM_BYTES_PROXIED_PER_CLIENT_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_io_bytes_per_client",
+        "Number of bytes sent/received between client and backend.",
+        crate::console::messages::MetricsAuxInfo::TRAFFIC_LABELS,
+    )
+    .unwrap()
+});
+
+static NUM_BYTES_PROXIED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "proxy_io_bytes",
+        "Number of bytes sent/received between all clients and backends.",
+        &["direction"],
+    )
+    .unwrap()
+});
+
 pub async fn run_until_cancelled<F: std::future::Future>(
     f: F,
     cancellation_token: &CancellationToken,
@@ -186,12 +428,16 @@ pub async fn handle_client<S: AsyncRead + AsyncWrite + Unpin>(
     );
 
     let proto = mode.protocol_label();
-    let _client_gauge = NUM_CLIENT_CONNECTION_GAUGE
+    NUM_CLIENT_CONNECTION_OPENED_COUNTER
         .with_label_values(&[proto])
-        .guard();
-    let _request_gauge = NUM_CONNECTION_REQUESTS_GAUGE
+        .inc();
+    NUM_CONNECTIONS_ACCEPTED_COUNTER
         .with_label_values(&[proto])
-        .guard();
+        .inc();
+    scopeguard::defer! {
+        NUM_CLIENT_CONNECTION_CLOSED_COUNTER.with_label_values(&[proto]).inc();
+        NUM_CONNECTIONS_CLOSED_COUNTER.with_label_values(&[proto]).inc();
+    }
 
     let tls = config.tls_config.as_ref();
 
@@ -315,6 +561,294 @@ async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
     }
 }
 
+/// If we couldn't connect, a cached connection info might be to blame
+/// (e.g. the compute node's address might've changed at the wrong time).
+/// Invalidate the cache entry (if any) to prevent subsequent errors.
+#[tracing::instrument(name = "invalidate_cache", skip_all)]
+pub fn invalidate_cache(node_info: console::CachedNodeInfo) -> compute::ConnCfg {
+    let is_cached = node_info.cached();
+    if is_cached {
+        warn!("invalidating stalled compute node info cache entry");
+    }
+    let label = match is_cached {
+        true => "compute_cached",
+        false => "compute_uncached",
+    };
+    NUM_CONNECTION_FAILURES.with_label_values(&[label]).inc();
+
+    node_info.invalidate().config
+}
+
+/// Try to connect to the compute node once.
+#[tracing::instrument(name = "connect_once", fields(pid = tracing::field::Empty), skip_all)]
+async fn connect_to_compute_once(
+    node_info: &console::CachedNodeInfo,
+    timeout: time::Duration,
+) -> Result<PostgresConnection, compute::ConnectionError> {
+    let allow_self_signed_compute = node_info.allow_self_signed_compute;
+
+    node_info
+        .config
+        .connect(allow_self_signed_compute, timeout)
+        .await
+}
+
+#[async_trait]
+pub trait ConnectMechanism {
+    type Connection;
+    type ConnectError;
+    type Error: From<Self::ConnectError>;
+    async fn connect_once(
+        &self,
+        node_info: &console::CachedNodeInfo,
+        timeout: time::Duration,
+    ) -> Result<Self::Connection, Self::ConnectError>;
+
+    fn update_connect_config(&self, conf: &mut compute::ConnCfg);
+}
+
+pub struct TcpMechanism<'a> {
+    /// KV-dictionary with PostgreSQL connection params.
+    pub params: &'a StartupMessageParams,
+}
+
+#[async_trait]
+impl ConnectMechanism for TcpMechanism<'_> {
+    type Connection = PostgresConnection;
+    type ConnectError = compute::ConnectionError;
+    type Error = compute::ConnectionError;
+
+    async fn connect_once(
+        &self,
+        node_info: &console::CachedNodeInfo,
+        timeout: time::Duration,
+    ) -> Result<PostgresConnection, Self::Error> {
+        connect_to_compute_once(node_info, timeout).await
+    }
+
+    fn update_connect_config(&self, config: &mut compute::ConnCfg) {
+        config.set_startup_params(self.params);
+    }
+}
+
+const fn bool_to_str(x: bool) -> &'static str {
+    if x {
+        "true"
+    } else {
+        "false"
+    }
+}
+
+fn report_error(e: &WakeComputeError, retry: bool) {
+    use crate::console::errors::ApiError;
+    let retry = bool_to_str(retry);
+    let kind = match e {
+        WakeComputeError::BadComputeAddress(_) => "bad_compute_address",
+        WakeComputeError::ApiError(ApiError::Transport(_)) => "api_transport_error",
+        WakeComputeError::ApiError(ApiError::Console {
+            status: StatusCode::LOCKED,
+            ref text,
+        }) if text.contains("written data quota exceeded")
+            || text.contains("the limit for current plan reached") =>
+        {
+            "quota_exceeded"
+        }
+        WakeComputeError::ApiError(ApiError::Console {
+            status: StatusCode::LOCKED,
+            ..
+        }) => "api_console_locked",
+        WakeComputeError::ApiError(ApiError::Console {
+            status: StatusCode::BAD_REQUEST,
+            ..
+        }) => "api_console_bad_request",
+        WakeComputeError::ApiError(ApiError::Console { status, .. })
+            if status.is_server_error() =>
+        {
+            "api_console_other_server_error"
+        }
+        WakeComputeError::ApiError(ApiError::Console { .. }) => "api_console_other_error",
+        WakeComputeError::TimeoutError => "timeout_error",
+    };
+    NUM_WAKEUP_FAILURES.with_label_values(&[retry, kind]).inc();
+}
+
+/// Try to connect to the compute node, retrying if necessary.
+/// This function might update `node_info`, so we take it by `&mut`.
+#[tracing::instrument(skip_all)]
+pub async fn connect_to_compute<M: ConnectMechanism>(
+    mechanism: &M,
+    mut node_info: console::CachedNodeInfo,
+    extra: &console::ConsoleReqExtra,
+    creds: &auth::BackendType<'_, auth::backend::ComputeUserInfo>,
+    mut latency_timer: LatencyTimer,
+) -> Result<M::Connection, M::Error>
+where
+    M::ConnectError: ShouldRetry + std::fmt::Debug,
+    M::Error: From<WakeComputeError>,
+{
+    mechanism.update_connect_config(&mut node_info.config);
+
+    // try once
+    let (config, err) = match mechanism.connect_once(&node_info, CONNECT_TIMEOUT).await {
+        Ok(res) => {
+            latency_timer.success();
+            return Ok(res);
+        }
+        Err(e) => {
+            error!(error = ?e, "could not connect to compute node");
+            (invalidate_cache(node_info), e)
+        }
+    };
+
+    latency_timer.cache_miss();
+
+    let mut num_retries = 1;
+
+    // if we failed to connect, it's likely that the compute node was suspended, wake a new compute node
+    info!("compute node's state has likely changed; requesting a wake-up");
+    let node_info = loop {
+        let wake_res = match creds {
+            auth::BackendType::Console(api, creds) => api.wake_compute(extra, creds).await,
+            #[cfg(feature = "testing")]
+            auth::BackendType::Postgres(api, creds) => api.wake_compute(extra, creds).await,
+            // nothing to do?
+            auth::BackendType::Link(_) => return Err(err.into()),
+            // test backend
+            #[cfg(test)]
+            auth::BackendType::Test(x) => x.wake_compute(),
+        };
+
+        match handle_try_wake(wake_res, num_retries) {
+            Err(e) => {
+                error!(error = ?e, num_retries, retriable = false, "couldn't wake compute node");
+                report_error(&e, false);
+                return Err(e.into());
+            }
+            // failed to wake up but we can continue to retry
+            Ok(ControlFlow::Continue(e)) => {
+                report_error(&e, true);
+                warn!(error = ?e, num_retries, retriable = true, "couldn't wake compute node");
+            }
+            // successfully woke up a compute node and can break the wakeup loop
+            Ok(ControlFlow::Break(mut node_info)) => {
+                node_info.config.reuse_password(&config);
+                mechanism.update_connect_config(&mut node_info.config);
+                break node_info;
+            }
+        }
+
+        let wait_duration = retry_after(num_retries);
+        num_retries += 1;
+
+        time::sleep(wait_duration).await;
+    };
+
+    // now that we have a new node, try connect to it repeatedly.
+    // this can error for a few reasons, for instance:
+    // * DNS connection settings haven't quite propagated yet
+    info!("wake_compute success. attempting to connect");
+    loop {
+        match mechanism.connect_once(&node_info, CONNECT_TIMEOUT).await {
+            Ok(res) => {
+                latency_timer.success();
+                return Ok(res);
+            }
+            Err(e) => {
+                let retriable = e.should_retry(num_retries);
+                if !retriable {
+                    error!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
+                    return Err(e.into());
+                }
+                warn!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
+            }
+        }
+
+        let wait_duration = retry_after(num_retries);
+        num_retries += 1;
+
+        time::sleep(wait_duration).await;
+    }
+}
+
+/// Attempts to wake up the compute node.
+/// * Returns Ok(Continue(e)) if there was an error waking but retries are acceptable
+/// * Returns Ok(Break(node)) if the wakeup succeeded
+/// * Returns Err(e) if there was an error
+pub fn handle_try_wake(
+    result: Result<console::CachedNodeInfo, WakeComputeError>,
+    num_retries: u32,
+) -> Result<ControlFlow<console::CachedNodeInfo, WakeComputeError>, WakeComputeError> {
+    match result {
+        Err(err) => match &err {
+            WakeComputeError::ApiError(api) if api.should_retry(num_retries) => {
+                Ok(ControlFlow::Continue(err))
+            }
+            _ => Err(err),
+        },
+        // Ready to try again.
+        Ok(new) => Ok(ControlFlow::Break(new)),
+    }
+}
+
+pub trait ShouldRetry {
+    fn could_retry(&self) -> bool;
+    fn should_retry(&self, num_retries: u32) -> bool {
+        match self {
+            _ if num_retries >= NUM_RETRIES_CONNECT => false,
+            err => err.could_retry(),
+        }
+    }
+}
+
+impl ShouldRetry for io::Error {
+    fn could_retry(&self) -> bool {
+        use std::io::ErrorKind;
+        matches!(
+            self.kind(),
+            ErrorKind::ConnectionRefused | ErrorKind::AddrNotAvailable | ErrorKind::TimedOut
+        )
+    }
+}
+
+impl ShouldRetry for tokio_postgres::error::DbError {
+    fn could_retry(&self) -> bool {
+        use tokio_postgres::error::SqlState;
+        matches!(
+            self.code(),
+            &SqlState::CONNECTION_FAILURE
+                | &SqlState::CONNECTION_EXCEPTION
+                | &SqlState::CONNECTION_DOES_NOT_EXIST
+                | &SqlState::SQLCLIENT_UNABLE_TO_ESTABLISH_SQLCONNECTION,
+        )
+    }
+}
+
+impl ShouldRetry for tokio_postgres::Error {
+    fn could_retry(&self) -> bool {
+        if let Some(io_err) = self.source().and_then(|x| x.downcast_ref()) {
+            io::Error::could_retry(io_err)
+        } else if let Some(db_err) = self.source().and_then(|x| x.downcast_ref()) {
+            tokio_postgres::error::DbError::could_retry(db_err)
+        } else {
+            false
+        }
+    }
+}
+
+impl ShouldRetry for compute::ConnectionError {
+    fn could_retry(&self) -> bool {
+        match self {
+            compute::ConnectionError::Postgres(err) => err.could_retry(),
+            compute::ConnectionError::CouldNotConnect(err) => err.could_retry(),
+            _ => false,
+        }
+    }
+}
+
+pub fn retry_after(num_retries: u32) -> time::Duration {
+    BASE_RETRY_WAIT_DURATION.mul_f64(RETRY_WAIT_EXPONENT_BASE.powi((num_retries as i32) - 1))
+}
+
 /// Finish client connection initialization: confirm auth success, send params, etc.
 #[tracing::instrument(skip_all)]
 async fn prepare_client_connection(
@@ -494,7 +1028,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> Client<'_, S> {
 
         let aux = node_info.aux.clone();
         let mut node = connect_to_compute(
-            &TcpMechanism { params, proto },
+            &TcpMechanism { params },
             node_info,
             &extra,
             &creds,
@@ -503,6 +1037,13 @@ impl<S: AsyncRead + AsyncWrite + Unpin> Client<'_, S> {
         .or_else(|e| stream.throw_error(e))
         .await?;
 
+        NUM_DB_CONNECTIONS_OPENED_COUNTER
+            .with_label_values(&[proto])
+            .inc();
+        scopeguard::defer! {
+            NUM_DB_CONNECTIONS_CLOSED_COUNTER.with_label_values(&[proto]).inc();
+        }
+
         prepare_client_connection(&node, session, &mut stream).await?;
         // Before proxy passing, forward to compute whatever data is left in the
         // PqStream input buffer. Normally there is none, but our serverless npm

proxy/src/proxy/connect_compute.rs

@@ -1,238 +0,0 @@
-use crate::{
-    auth,
-    compute::{self, PostgresConnection},
-    console::{self, errors::WakeComputeError, Api},
-    metrics::{bool_to_str, LatencyTimer, NUM_CONNECTION_FAILURES, NUM_WAKEUP_FAILURES},
-    proxy::retry::{retry_after, ShouldRetry},
-};
-use async_trait::async_trait;
-use hyper::StatusCode;
-use pq_proto::StartupMessageParams;
-use std::ops::ControlFlow;
-use tokio::time;
-use tracing::{error, info, warn};
-
-const CONNECT_TIMEOUT: time::Duration = time::Duration::from_secs(2);
-
-/// If we couldn't connect, a cached connection info might be to blame
-/// (e.g. the compute node's address might've changed at the wrong time).
-/// Invalidate the cache entry (if any) to prevent subsequent errors.
-#[tracing::instrument(name = "invalidate_cache", skip_all)]
-pub fn invalidate_cache(node_info: console::CachedNodeInfo) -> compute::ConnCfg {
-    let is_cached = node_info.cached();
-    if is_cached {
-        warn!("invalidating stalled compute node info cache entry");
-    }
-    let label = match is_cached {
-        true => "compute_cached",
-        false => "compute_uncached",
-    };
-    NUM_CONNECTION_FAILURES.with_label_values(&[label]).inc();
-
-    node_info.invalidate().config
-}
-
-/// Try to connect to the compute node once.
-#[tracing::instrument(name = "connect_once", fields(pid = tracing::field::Empty), skip_all)]
-async fn connect_to_compute_once(
-    node_info: &console::CachedNodeInfo,
-    timeout: time::Duration,
-    proto: &'static str,
-) -> Result<PostgresConnection, compute::ConnectionError> {
-    let allow_self_signed_compute = node_info.allow_self_signed_compute;
-
-    node_info
-        .config
-        .connect(allow_self_signed_compute, timeout, proto)
-        .await
-}
-
-#[async_trait]
-pub trait ConnectMechanism {
-    type Connection;
-    type ConnectError;
-    type Error: From<Self::ConnectError>;
-    async fn connect_once(
-        &self,
-        node_info: &console::CachedNodeInfo,
-        timeout: time::Duration,
-    ) -> Result<Self::Connection, Self::ConnectError>;
-
-    fn update_connect_config(&self, conf: &mut compute::ConnCfg);
-}
-
-pub struct TcpMechanism<'a> {
-    /// KV-dictionary with PostgreSQL connection params.
-    pub params: &'a StartupMessageParams,
-    pub proto: &'static str,
-}
-
-#[async_trait]
-impl ConnectMechanism for TcpMechanism<'_> {
-    type Connection = PostgresConnection;
-    type ConnectError = compute::ConnectionError;
-    type Error = compute::ConnectionError;
-
-    async fn connect_once(
-        &self,
-        node_info: &console::CachedNodeInfo,
-        timeout: time::Duration,
-    ) -> Result<PostgresConnection, Self::Error> {
-        connect_to_compute_once(node_info, timeout, self.proto).await
-    }
-
-    fn update_connect_config(&self, config: &mut compute::ConnCfg) {
-        config.set_startup_params(self.params);
-    }
-}
-
-fn report_error(e: &WakeComputeError, retry: bool) {
-    use crate::console::errors::ApiError;
-    let retry = bool_to_str(retry);
-    let kind = match e {
-        WakeComputeError::BadComputeAddress(_) => "bad_compute_address",
-        WakeComputeError::ApiError(ApiError::Transport(_)) => "api_transport_error",
-        WakeComputeError::ApiError(ApiError::Console {
-            status: StatusCode::LOCKED,
-            ref text,
-        }) if text.contains("written data quota exceeded")
-            || text.contains("the limit for current plan reached") =>
-        {
-            "quota_exceeded"
-        }
-        WakeComputeError::ApiError(ApiError::Console {
-            status: StatusCode::LOCKED,
-            ..
-        }) => "api_console_locked",
-        WakeComputeError::ApiError(ApiError::Console {
-            status: StatusCode::BAD_REQUEST,
-            ..
-        }) => "api_console_bad_request",
-        WakeComputeError::ApiError(ApiError::Console { status, .. })
-            if status.is_server_error() =>
-        {
-            "api_console_other_server_error"
-        }
-        WakeComputeError::ApiError(ApiError::Console { .. }) => "api_console_other_error",
-        WakeComputeError::TimeoutError => "timeout_error",
-    };
-    NUM_WAKEUP_FAILURES.with_label_values(&[retry, kind]).inc();
-}
-
-/// Try to connect to the compute node, retrying if necessary.
-/// This function might update `node_info`, so we take it by `&mut`.
-#[tracing::instrument(skip_all)]
-pub async fn connect_to_compute<M: ConnectMechanism>(
-    mechanism: &M,
-    mut node_info: console::CachedNodeInfo,
-    extra: &console::ConsoleReqExtra,
-    creds: &auth::BackendType<'_, auth::backend::ComputeUserInfo>,
-    mut latency_timer: LatencyTimer,
-) -> Result<M::Connection, M::Error>
-where
-    M::ConnectError: ShouldRetry + std::fmt::Debug,
-    M::Error: From<WakeComputeError>,
-{
-    mechanism.update_connect_config(&mut node_info.config);
-
-    // try once
-    let (config, err) = match mechanism.connect_once(&node_info, CONNECT_TIMEOUT).await {
-        Ok(res) => {
-            latency_timer.success();
-            return Ok(res);
-        }
-        Err(e) => {
-            error!(error = ?e, "could not connect to compute node");
-            (invalidate_cache(node_info), e)
-        }
-    };
-
-    latency_timer.cache_miss();
-
-    let mut num_retries = 1;
-
-    // if we failed to connect, it's likely that the compute node was suspended, wake a new compute node
-    info!("compute node's state has likely changed; requesting a wake-up");
-    let node_info = loop {
-        let wake_res = match creds {
-            auth::BackendType::Console(api, creds) => api.wake_compute(extra, creds).await,
-            #[cfg(feature = "testing")]
-            auth::BackendType::Postgres(api, creds) => api.wake_compute(extra, creds).await,
-            // nothing to do?
-            auth::BackendType::Link(_) => return Err(err.into()),
-            // test backend
-            #[cfg(test)]
-            auth::BackendType::Test(x) => x.wake_compute(),
-        };
-
-        match handle_try_wake(wake_res, num_retries) {
-            Err(e) => {
-                error!(error = ?e, num_retries, retriable = false, "couldn't wake compute node");
-                report_error(&e, false);
-                return Err(e.into());
-            }
-            // failed to wake up but we can continue to retry
-            Ok(ControlFlow::Continue(e)) => {
-                report_error(&e, true);
-                warn!(error = ?e, num_retries, retriable = true, "couldn't wake compute node");
-            }
-            // successfully woke up a compute node and can break the wakeup loop
-            Ok(ControlFlow::Break(mut node_info)) => {
-                node_info.config.reuse_password(&config);
-                mechanism.update_connect_config(&mut node_info.config);
-                break node_info;
-            }
-        }
-
-        let wait_duration = retry_after(num_retries);
-        num_retries += 1;
-
-        time::sleep(wait_duration).await;
-    };
-
-    // now that we have a new node, try connect to it repeatedly.
-    // this can error for a few reasons, for instance:
-    // * DNS connection settings haven't quite propagated yet
-    info!("wake_compute success. attempting to connect");
-    loop {
-        match mechanism.connect_once(&node_info, CONNECT_TIMEOUT).await {
-            Ok(res) => {
-                latency_timer.success();
-                return Ok(res);
-            }
-            Err(e) => {
-                let retriable = e.should_retry(num_retries);
-                if !retriable {
-                    error!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
-                    return Err(e.into());
-                }
-                warn!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
-            }
-        }
-
-        let wait_duration = retry_after(num_retries);
-        num_retries += 1;
-
-        time::sleep(wait_duration).await;
-    }
-}
-
-/// Attempts to wake up the compute node.
-/// * Returns Ok(Continue(e)) if there was an error waking but retries are acceptable
-/// * Returns Ok(Break(node)) if the wakeup succeeded
-/// * Returns Err(e) if there was an error
-pub fn handle_try_wake(
-    result: Result<console::CachedNodeInfo, WakeComputeError>,
-    num_retries: u32,
-) -> Result<ControlFlow<console::CachedNodeInfo, WakeComputeError>, WakeComputeError> {
-    match result {
-        Err(err) => match &err {
-            WakeComputeError::ApiError(api) if api.should_retry(num_retries) => {
-                Ok(ControlFlow::Continue(err))
-            }
-            _ => Err(err),
-        },
-        // Ready to try again.
-        Ok(new) => Ok(ControlFlow::Break(new)),
-    }
-}

proxy/src/proxy/retry.rs

@@ -1,68 +0,0 @@
-use crate::compute;
-use std::{error::Error, io};
-use tokio::time;
-
-/// Number of times we should retry the `/proxy_wake_compute` http request.
-/// Retry duration is BASE_RETRY_WAIT_DURATION * RETRY_WAIT_EXPONENT_BASE ^ n, where n starts at 0
-pub const NUM_RETRIES_CONNECT: u32 = 16;
-const BASE_RETRY_WAIT_DURATION: time::Duration = time::Duration::from_millis(25);
-const RETRY_WAIT_EXPONENT_BASE: f64 = std::f64::consts::SQRT_2;
-
-pub trait ShouldRetry {
-    fn could_retry(&self) -> bool;
-    fn should_retry(&self, num_retries: u32) -> bool {
-        match self {
-            _ if num_retries >= NUM_RETRIES_CONNECT => false,
-            err => err.could_retry(),
-        }
-    }
-}
-
-impl ShouldRetry for io::Error {
-    fn could_retry(&self) -> bool {
-        use std::io::ErrorKind;
-        matches!(
-            self.kind(),
-            ErrorKind::ConnectionRefused | ErrorKind::AddrNotAvailable | ErrorKind::TimedOut
-        )
-    }
-}
-
-impl ShouldRetry for tokio_postgres::error::DbError {
-    fn could_retry(&self) -> bool {
-        use tokio_postgres::error::SqlState;
-        matches!(
-            self.code(),
-            &SqlState::CONNECTION_FAILURE
-                | &SqlState::CONNECTION_EXCEPTION
-                | &SqlState::CONNECTION_DOES_NOT_EXIST
-                | &SqlState::SQLCLIENT_UNABLE_TO_ESTABLISH_SQLCONNECTION,
-        )
-    }
-}
-
-impl ShouldRetry for tokio_postgres::Error {
-    fn could_retry(&self) -> bool {
-        if let Some(io_err) = self.source().and_then(|x| x.downcast_ref()) {
-            io::Error::could_retry(io_err)
-        } else if let Some(db_err) = self.source().and_then(|x| x.downcast_ref()) {
-            tokio_postgres::error::DbError::could_retry(db_err)
-        } else {
-            false
-        }
-    }
-}
-
-impl ShouldRetry for compute::ConnectionError {
-    fn could_retry(&self) -> bool {
-        match self {
-            compute::ConnectionError::Postgres(err) => err.could_retry(),
-            compute::ConnectionError::CouldNotConnect(err) => err.could_retry(),
-            _ => false,
-        }
-    }
-}
-
-pub fn retry_after(num_retries: u32) -> time::Duration {
-    BASE_RETRY_WAIT_DURATION.mul_f64(RETRY_WAIT_EXPONENT_BASE.powi((num_retries as i32) - 1))
-}

proxy/src/proxy/tests.rs

@@ -2,13 +2,11 @@
 
 mod mitm;
 
-use super::connect_compute::ConnectMechanism;
-use super::retry::ShouldRetry;
 use super::*;
 use crate::auth::backend::{ComputeUserInfo, TestBackend};
+use crate::auth::IpPattern;
 use crate::config::CertResolver;
 use crate::console::{CachedNodeInfo, NodeInfo};
-use crate::proxy::retry::{retry_after, NUM_RETRIES_CONNECT};
 use crate::{auth, http, sasl, scram};
 use async_trait::async_trait;
 use rstest::rstest;
@@ -426,7 +424,7 @@ impl ConnectMechanism for TestConnectMechanism {
     async fn connect_once(
         &self,
         _node_info: &console::CachedNodeInfo,
-        _timeout: std::time::Duration,
+        _timeout: time::Duration,
     ) -> Result<Self::Connection, Self::ConnectError> {
         let mut counter = self.counter.lock().unwrap();
         let action = self.sequence[*counter];
@@ -469,7 +467,7 @@ impl TestBackend for TestConnectMechanism {
         }
     }
 
-    fn get_allowed_ips(&self) -> Result<Arc<Vec<String>>, console::errors::GetAuthInfoError> {
+    fn get_allowed_ips(&self) -> Result<Arc<Vec<IpPattern>>, console::errors::GetAuthInfoError> {
         unimplemented!("not used in tests")
     }
 }

proxy/src/proxy/tests/mitm.rs

@@ -120,7 +120,7 @@ where
 struct PgFrame;
 impl Decoder for PgFrame {
     type Item = Bytes;
-    type Error = std::io::Error;
+    type Error = io::Error;
 
     fn decode(&mut self, src: &mut BytesMut) -> Result<Option<Self::Item>, Self::Error> {
         if src.len() < 5 {
@@ -136,7 +136,7 @@ impl Decoder for PgFrame {
     }
 }
 impl Encoder<Bytes> for PgFrame {
-    type Error = std::io::Error;
+    type Error = io::Error;
 
     fn encode(&mut self, item: Bytes, dst: &mut BytesMut) -> Result<(), Self::Error> {
         dst.extend_from_slice(&item);

proxy/src/rate_limiter/aimd.rs

@@ -33,6 +33,39 @@ impl Aimd {
             min_utilisation_threshold: config.aimd_min_utilisation_threshold,
         }
     }
+
+    pub fn decrease_factor(self, factor: f32) -> Self {
+        assert!((0.5..1.0).contains(&factor));
+        Self {
+            decrease_factor: factor,
+            ..self
+        }
+    }
+
+    pub fn increase_by(self, increase: usize) -> Self {
+        assert!(increase > 0);
+        Self {
+            increase_by: increase,
+            ..self
+        }
+    }
+
+    pub fn with_max_limit(self, max: usize) -> Self {
+        assert!(max > 0);
+        Self {
+            max_limit: max,
+            ..self
+        }
+    }
+
+    /// A threshold below which the limit won't be increased. 0.5 = 50%.
+    pub fn with_min_utilisation_threshold(self, min_util: f32) -> Self {
+        assert!(min_util > 0. && min_util < 1.);
+        Self {
+            min_utilisation_threshold: min_util,
+            ..self
+        }
+    }
 }
 
 #[async_trait]

proxy/src/rate_limiter/limiter.rs

@@ -1,16 +1,12 @@
-use std::{
-    collections::hash_map::RandomState,
-    hash::BuildHasher,
-    sync::{
-        atomic::{AtomicUsize, Ordering},
-        Arc, Mutex,
-    },
+use std::sync::{
+    atomic::{AtomicUsize, Ordering},
+    Arc,
 };
 
 use anyhow::bail;
 use dashmap::DashMap;
 use itertools::Itertools;
-use rand::{rngs::StdRng, Rng, SeedableRng};
+use rand::{thread_rng, Rng};
 use smol_str::SmolStr;
 use tokio::sync::{Mutex as AsyncMutex, Semaphore, SemaphorePermit};
 use tokio::time::{timeout, Duration, Instant};
@@ -32,11 +28,10 @@ use super::{
 // saw SNI, before doing TLS handshake. User-side error messages in that case
 // does not look very nice (`SSL SYSCALL error: Undefined error: 0`), so for now
 // I went with a more expensive way that yields user-friendlier error messages.
-pub struct EndpointRateLimiter<Rand = StdRng, Hasher = RandomState> {
-    map: DashMap<SmolStr, Vec<RateBucket>, Hasher>,
+pub struct EndpointRateLimiter {
+    map: DashMap<SmolStr, Vec<RateBucket>>,
     info: &'static [RateBucketInfo],
     access_count: AtomicUsize,
-    rand: Mutex<Rand>,
 }
 
 #[derive(Clone, Copy)]
@@ -130,18 +125,11 @@ impl RateBucketInfo {
 
 impl EndpointRateLimiter {
     pub fn new(info: &'static [RateBucketInfo]) -> Self {
-        Self::new_with_rand_and_hasher(info, StdRng::from_entropy(), RandomState::new())
-    }
-}
-
-impl<R: Rng, S: BuildHasher + Clone> EndpointRateLimiter<R, S> {
-    fn new_with_rand_and_hasher(info: &'static [RateBucketInfo], rand: R, hasher: S) -> Self {
         info!(buckets = ?info, "endpoint rate limiter");
         Self {
             info,
-            map: DashMap::with_hasher_and_shard_amount(hasher, 64),
+            map: DashMap::with_shard_amount(64),
             access_count: AtomicUsize::new(1), // start from 1 to avoid GC on the first request
-            rand: Mutex::new(rand),
         }
     }
 
@@ -188,9 +176,7 @@ impl<R: Rng, S: BuildHasher + Clone> EndpointRateLimiter<R, S> {
             self.map.len()
         );
         let n = self.map.shards().len();
-        // this lock is ok as the periodic cycle of do_gc makes this very unlikely to collide
-        // (impossible, infact, unless we have 2048 threads)
-        let shard = self.rand.lock().unwrap().gen_range(0..n);
+        let shard = thread_rng().gen_range(0..n);
         self.map.shards()[shard].write().clear();
     }
 }
@@ -233,6 +219,7 @@ pub struct Token<'t> {
 #[derive(Debug, Clone, Copy)]
 pub struct LimiterState {
     limit: usize,
+    available: usize,
     in_flight: usize,
 }
 
@@ -393,10 +380,10 @@ impl Limiter {
             }
             new_limit
         };
-        crate::metrics::RATE_LIMITER_LIMIT
+        crate::proxy::RATE_LIMITER_LIMIT
             .with_label_values(&["expected"])
             .set(new_limit as i64);
-        crate::metrics::RATE_LIMITER_LIMIT
+        crate::proxy::RATE_LIMITER_LIMIT
             .with_label_values(&["actual"])
             .set(actual_limit as i64);
         self.limits.store(new_limit, Ordering::Release);
@@ -410,7 +397,11 @@ impl Limiter {
     pub fn state(&self) -> LimiterState {
         let limit = self.limits.load(Ordering::Relaxed);
         let in_flight = self.in_flight.load(Ordering::Relaxed);
-        LimiterState { limit, in_flight }
+        LimiterState {
+            limit,
+            available: limit.saturating_sub(in_flight),
+            in_flight,
+        }
     }
 }
 
@@ -423,6 +414,13 @@ impl<'t> Token<'t> {
         }
     }
 
+    #[cfg(test)]
+    pub fn set_latency(&mut self, latency: Duration) {
+        use std::ops::Sub;
+
+        self.start = Instant::now().sub(latency);
+    }
+
     pub fn forget(&mut self) {
         if let Some(permit) = self.permit.take() {
             permit.forget();
@@ -441,6 +439,10 @@ impl LimiterState {
     pub fn limit(&self) -> usize {
         self.limit
     }
+    /// The amount of concurrency available to use.
+    pub fn available(&self) -> usize {
+        self.available
+    }
     /// The number of jobs in flight.
     pub fn in_flight(&self) -> usize {
         self.in_flight
@@ -470,7 +472,7 @@ impl reqwest_middleware::Middleware for Limiter {
                 )
             })?;
         info!(duration = ?start.elapsed(), "waiting for token to connect to the control plane");
-        crate::metrics::RATE_LIMITER_ACQUIRE_LATENCY.observe(start.elapsed().as_secs_f64());
+        crate::proxy::RATE_LIMITER_ACQUIRE_LATENCY.observe(start.elapsed().as_secs_f64());
         match next.run(req, extensions).await {
             Ok(response) => {
                 self.release(token, Some(Outcome::from_reqwest_response(&response)))
@@ -488,11 +490,9 @@ impl reqwest_middleware::Middleware for Limiter {
 
 #[cfg(test)]
 mod tests {
-    use std::{hash::BuildHasherDefault, pin::pin, task::Context, time::Duration};
+    use std::{pin::pin, task::Context, time::Duration};
 
     use futures::{task::noop_waker_ref, Future};
-    use rand::SeedableRng;
-    use rustc_hash::FxHasher;
     use smol_str::SmolStr;
     use tokio::time;
 
@@ -690,21 +690,4 @@ mod tests {
             assert!(limiter.check(endpoint.clone()));
         }
     }
-
-    #[tokio::test]
-    async fn test_rate_limits_gc() {
-        // fixed seeded random/hasher to ensure that the test is not flaky
-        let rand = rand::rngs::StdRng::from_seed([1; 32]);
-        let hasher = BuildHasherDefault::<FxHasher>::default();
-
-        let limiter = EndpointRateLimiter::new_with_rand_and_hasher(
-            &RateBucketInfo::DEFAULT_SET,
-            rand,
-            hasher,
-        );
-        for i in 0..1_000_000 {
-            limiter.check(format!("{i}").into());
-        }
-        assert!(limiter.map.len() < 150_000);
-    }
 }

proxy/src/serverless.rs

@@ -8,13 +8,12 @@ mod websocket;
 
 use anyhow::bail;
 use hyper::StatusCode;
-use metrics::IntCounterPairGuard;
 pub use reqwest_middleware::{ClientWithMiddleware, Error};
 pub use reqwest_retry::{policies::ExponentialBackoff, RetryTransientMiddleware};
 use tokio_util::task::TaskTracker;
 
-use crate::metrics::NUM_CLIENT_CONNECTION_GAUGE;
 use crate::protocol2::{ProxyProtocolAccept, WithClientIp};
+use crate::proxy::{NUM_CLIENT_CONNECTION_CLOSED_COUNTER, NUM_CLIENT_CONNECTION_OPENED_COUNTER};
 use crate::rate_limiter::EndpointRateLimiter;
 use crate::{cancellation::CancelMap, config::ProxyConfig};
 use futures::StreamExt;
@@ -150,17 +149,22 @@ pub async fn task_main(
 
 struct MetricService<S> {
     inner: S,
-    _gauge: IntCounterPairGuard,
 }
 
 impl<S> MetricService<S> {
     fn new(inner: S) -> MetricService<S> {
-        MetricService {
-            inner,
-            _gauge: NUM_CLIENT_CONNECTION_GAUGE
-                .with_label_values(&["http"])
-                .guard(),
-        }
+        NUM_CLIENT_CONNECTION_OPENED_COUNTER
+            .with_label_values(&["http"])
+            .inc();
+        MetricService { inner }
+    }
+}
+
+impl<S> Drop for MetricService<S> {
+    fn drop(&mut self) {
+        NUM_CLIENT_CONNECTION_CLOSED_COUNTER
+            .with_label_values(&["http"])
+            .inc();
     }
 }
 

proxy/src/serverless/conn_pool.rs

@@ -9,7 +9,7 @@ use pbkdf2::{
 };
 use pq_proto::StartupMessageParams;
 use smol_str::SmolStr;
-use std::{collections::HashMap, net::IpAddr, sync::Arc};
+use std::{collections::HashMap, net::IpAddr, sync::Arc, time::Duration};
 use std::{
     fmt,
     task::{ready, Poll},
@@ -23,13 +23,17 @@ use tokio_postgres::{AsyncMessage, ReadyForQueryStatus};
 
 use crate::{
     auth::{self, backend::ComputeUserInfo, check_peer_addr_is_in_list},
-    console,
-    metrics::{LatencyTimer, NUM_DB_CONNECTIONS_GAUGE},
-    proxy::{connect_compute::ConnectMechanism, neon_options},
+    console::{self, locks::ApiLocks},
+    proxy::{
+        neon_options, LatencyTimer, NUM_DB_CONNECTIONS_CLOSED_COUNTER,
+        NUM_DB_CONNECTIONS_OPENED_COUNTER,
+    },
     usage_metrics::{Ids, MetricCounter, USAGE_METRICS},
 };
 use crate::{compute, config};
 
+use crate::proxy::ConnectMechanism;
+
 use tracing::{error, warn, Span};
 use tracing::{info, info_span, Instrument};
 
@@ -110,17 +114,32 @@ pub struct GlobalConnPool {
     // Using a lock to remove any race conditions.
     // Eg cleaning up connections while a new connection is returned
     closed: RwLock<bool>,
+
+    // semaphore guarding unauthenticated postgres connections
+    connection_lock: ApiLocks,
 }
 
 impl GlobalConnPool {
     pub fn new(config: &'static crate::config::ProxyConfig) -> Arc<Self> {
-        Arc::new(Self {
+        let connection_lock =
+            ApiLocks::new("http_connect_lock", 2, 32, Duration::from_secs(10)).unwrap();
+        let this = Arc::new(Self {
             global_pool: DashMap::new(),
             global_pool_size: AtomicUsize::new(0),
             max_conns_per_endpoint: MAX_CONNS_PER_ENDPOINT,
             proxy_config: config,
             closed: RwLock::new(false),
-        })
+            connection_lock,
+        });
+
+        let this2 = this.clone();
+        tokio::spawn(async move {
+            this2
+                .connection_lock
+                .garbage_collect_worker(Duration::from_secs(600))
+                .await
+        });
+        this
     }
 
     pub fn shutdown(&self) {
@@ -217,6 +236,11 @@ impl GlobalConnPool {
                 return Ok(Client::new(client, pool).await);
             }
         } else {
+            // acquire a permit for un-authenticated access to the compute.
+            // to be clear, postgres will authenticate, but we want to limit the connections
+            // that have potential to be unauthenticated.
+            let _permit = self.connection_lock.get_permit(&conn_info.hostname).await?;
+
             let conn_id = uuid::Uuid::new_v4();
             info!(%conn_id, "pool: opening a new connection '{conn_info}'");
             connect_to_compute(
@@ -443,7 +467,7 @@ async fn connect_to_compute(
         .await?
         .context("missing cache entry from wake_compute")?;
 
-    crate::proxy::connect_compute::connect_to_compute(
+    crate::proxy::connect_to_compute(
         &TokioMechanism {
             conn_id,
             conn_info,
@@ -473,11 +497,6 @@ async fn connect_to_compute_once(
         .connect_timeout(timeout)
         .connect(tokio_postgres::NoTls)
         .await?;
-
-    let conn_gauge = NUM_DB_CONNECTIONS_GAUGE
-        .with_label_values(&["http"])
-        .guard();
-
     tracing::Span::current().record("pid", &tracing::field::display(client.get_process_id()));
 
     let (tx, mut rx) = tokio::sync::watch::channel(session);
@@ -493,7 +512,10 @@ async fn connect_to_compute_once(
 
     tokio::spawn(
         async move {
-            let _conn_gauge = conn_gauge;
+            NUM_DB_CONNECTIONS_OPENED_COUNTER.with_label_values(&["http"]).inc();
+            scopeguard::defer! {
+                NUM_DB_CONNECTIONS_CLOSED_COUNTER.with_label_values(&["http"]).inc();
+            }
             poll_fn(move |cx| {
                 if matches!(rx.has_changed(), Ok(true)) {
                     session = *rx.borrow_and_update();

proxy/src/serverless/sql_over_http.rs

@@ -29,7 +29,7 @@ use utils::http::error::ApiError;
 use utils::http::json::json_response;
 
 use crate::config::HttpConfig;
-use crate::metrics::NUM_CONNECTION_REQUESTS_GAUGE;
+use crate::proxy::{NUM_CONNECTIONS_ACCEPTED_COUNTER, NUM_CONNECTIONS_CLOSED_COUNTER};
 
 use super::conn_pool::ConnInfo;
 use super::conn_pool::GlobalConnPool;
@@ -303,9 +303,12 @@ async fn handle_inner(
     session_id: uuid::Uuid,
     peer_addr: IpAddr,
 ) -> anyhow::Result<Response<Body>> {
-    let _request_gauge = NUM_CONNECTION_REQUESTS_GAUGE
+    NUM_CONNECTIONS_ACCEPTED_COUNTER
         .with_label_values(&["http"])
-        .guard();
+        .inc();
+    scopeguard::defer! {
+        NUM_CONNECTIONS_CLOSED_COUNTER.with_label_values(&["http"]).inc();
+    }
 
     //
     // Determine the destination and connection params

proxy/src/serverless/websocket.rs

@@ -27,15 +27,15 @@ use sync_wrapper::SyncWrapper;
 pin_project! {
     /// This is a wrapper around a [`WebSocketStream`] that
     /// implements [`AsyncRead`] and [`AsyncWrite`].
-    pub struct WebSocketRw<S = Upgraded> {
+    pub struct WebSocketRw {
         #[pin]
-        stream: SyncWrapper<WebSocketStream<S>>,
+        stream: SyncWrapper<WebSocketStream<Upgraded>>,
         bytes: Bytes,
     }
 }
 
-impl<S> WebSocketRw<S> {
-    pub fn new(stream: WebSocketStream<S>) -> Self {
+impl WebSocketRw {
+    pub fn new(stream: WebSocketStream<Upgraded>) -> Self {
         Self {
             stream: stream.into(),
             bytes: Bytes::new(),
@@ -43,7 +43,7 @@ impl<S> WebSocketRw<S> {
     }
 }
 
-impl<S: AsyncRead + AsyncWrite + Unpin> AsyncWrite for WebSocketRw<S> {
+impl AsyncWrite for WebSocketRw {
     fn poll_write(
         self: Pin<&mut Self>,
         cx: &mut Context<'_>,
@@ -69,7 +69,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AsyncWrite for WebSocketRw<S> {
     }
 }
 
-impl<S: AsyncRead + AsyncWrite + Unpin> AsyncRead for WebSocketRw<S> {
+impl AsyncRead for WebSocketRw {
     fn poll_read(
         mut self: Pin<&mut Self>,
         cx: &mut Context<'_>,
@@ -86,7 +86,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AsyncRead for WebSocketRw<S> {
     }
 }
 
-impl<S: AsyncRead + AsyncWrite + Unpin> AsyncBufRead for WebSocketRw<S> {
+impl AsyncBufRead for WebSocketRw {
     fn poll_fill_buf(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<&[u8]>> {
         // Please refer to poll_fill_buf's documentation.
         const EOF: Poll<io::Result<&[u8]>> = Poll::Ready(Ok(&[]));
@@ -151,60 +151,3 @@ pub async fn serve_websocket(
     .await?;
     Ok(())
 }
-
-#[cfg(test)]
-mod tests {
-    use std::pin::pin;
-
-    use futures::{SinkExt, StreamExt};
-    use hyper_tungstenite::{
-        tungstenite::{protocol::Role, Message},
-        WebSocketStream,
-    };
-    use tokio::{
-        io::{duplex, AsyncReadExt, AsyncWriteExt},
-        task::JoinSet,
-    };
-
-    use super::WebSocketRw;
-
-    #[tokio::test]
-    async fn websocket_stream_wrapper_happy_path() {
-        let (stream1, stream2) = duplex(1024);
-
-        let mut js = JoinSet::new();
-
-        js.spawn(async move {
-            let mut client = WebSocketStream::from_raw_socket(stream1, Role::Client, None).await;
-
-            client
-                .send(Message::Binary(b"hello world".to_vec()))
-                .await
-                .unwrap();
-
-            let message = client.next().await.unwrap().unwrap();
-            assert_eq!(message, Message::Binary(b"websockets are cool".to_vec()));
-
-            client.close(None).await.unwrap();
-        });
-
-        js.spawn(async move {
-            let mut rw = pin!(WebSocketRw::new(
-                WebSocketStream::from_raw_socket(stream2, Role::Server, None).await
-            ));
-
-            let mut buf = vec![0; 1024];
-            let n = rw.read(&mut buf).await.unwrap();
-            assert_eq!(&buf[..n], b"hello world");
-
-            rw.write_all(b"websockets are cool").await.unwrap();
-            rw.flush().await.unwrap();
-
-            let n = rw.read_to_end(&mut buf).await.unwrap();
-            assert_eq!(n, 0);
-        });
-
-        js.join_next().await.unwrap().unwrap();
-        js.join_next().await.unwrap().unwrap();
-    }
-}

s3_scrubber/Cargo.toml

@@ -31,7 +31,6 @@ reqwest = { workspace = true, default-features = false, features = ["rustls-tls"
 aws-config = { workspace = true, default-features = false, features = ["rustls", "sso"] }
 
 pageserver = { path = "../pageserver" }
-pageserver_api = { path = "../libs/pageserver_api" }
 remote_storage = { path = "../libs/remote_storage" }
 
 tracing.workspace = true

s3_scrubber/src/checks.rs

@@ -7,12 +7,13 @@ use utils::generation::Generation;
 
 use crate::cloud_admin_api::BranchData;
 use crate::metadata_stream::stream_listing;
-use crate::{download_object_with_retries, RootTarget, TenantShardTimelineId};
+use crate::{download_object_with_retries, RootTarget};
 use futures_util::{pin_mut, StreamExt};
 use pageserver::tenant::remote_timeline_client::parse_remote_index_path;
 use pageserver::tenant::storage_layer::LayerFileName;
 use pageserver::tenant::IndexPart;
 use remote_storage::RemotePath;
+use utils::id::TenantTimelineId;
 
 pub(crate) struct TimelineAnalysis {
     /// Anomalies detected
@@ -38,8 +39,8 @@ impl TimelineAnalysis {
     }
 }
 
-pub(crate) fn branch_cleanup_and_check_errors(
-    id: &TenantShardTimelineId,
+pub(crate) async fn branch_cleanup_and_check_errors(
+    id: &TenantTimelineId,
     s3_root: &RootTarget,
     s3_active_branch: Option<&BranchData>,
     console_branch: Option<BranchData>,
@@ -237,7 +238,7 @@ fn parse_layer_object_name(name: &str) -> Result<(LayerFileName, Generation), St
 
 pub(crate) async fn list_timeline_blobs(
     s3_client: &Client,
-    id: TenantShardTimelineId,
+    id: TenantTimelineId,
     s3_root: &RootTarget,
 ) -> anyhow::Result<S3TimelineBlobData> {
     let mut s3_layers = HashSet::new();

s3_scrubber/src/garbage.rs

@@ -10,16 +10,15 @@ use aws_sdk_s3::{
     Client,
 };
 use futures_util::{pin_mut, TryStreamExt};
-use pageserver_api::shard::TenantShardId;
 use serde::{Deserialize, Serialize};
 use tokio_stream::StreamExt;
-use utils::id::TenantId;
+use utils::id::{TenantId, TenantTimelineId};
 
 use crate::{
     cloud_admin_api::{CloudAdminApiClient, MaybeDeleted, ProjectData},
     init_remote,
     metadata_stream::{stream_listing, stream_tenant_timelines, stream_tenants},
-    BucketConfig, ConsoleConfig, NodeKind, RootTarget, TenantShardTimelineId, TraversingDepth,
+    BucketConfig, ConsoleConfig, NodeKind, RootTarget, TraversingDepth,
 };
 
 #[derive(Serialize, Deserialize, Debug)]
@@ -30,8 +29,8 @@ enum GarbageReason {
 
 #[derive(Serialize, Deserialize, Debug)]
 enum GarbageEntity {
-    Tenant(TenantShardId),
-    Timeline(TenantShardTimelineId),
+    Tenant(TenantId),
+    Timeline(TenantTimelineId),
 }
 
 #[derive(Serialize, Deserialize, Debug)]
@@ -143,9 +142,6 @@ async fn find_garbage_inner(
         console_projects.len()
     );
 
-    // TODO(sharding): batch calls into Console so that we only call once for each TenantId,
-    // rather than checking the same TenantId for multiple TenantShardId
-
     // Enumerate Tenants in S3, and check if each one exists in Console
     tracing::info!("Finding all tenants in bucket {}...", bucket_config.bucket);
     let tenants = stream_tenants(&s3_client, &target);
@@ -153,10 +149,10 @@ async fn find_garbage_inner(
         let api_client = cloud_admin_api_client.clone();
         let console_projects = &console_projects;
         async move {
-            match console_projects.get(&t.tenant_id) {
+            match console_projects.get(&t) {
                 Some(project_data) => Ok((t, Some(project_data.clone()))),
                 None => api_client
-                    .find_tenant_project(t.tenant_id)
+                    .find_tenant_project(t)
                     .await
                     .map_err(|e| anyhow::anyhow!(e))
                     .map(|r| (t, r)),
@@ -170,21 +166,21 @@ async fn find_garbage_inner(
     // checks if they are enabled by the `depth` parameter.
     pin_mut!(tenants_checked);
     let mut garbage = GarbageList::new(node_kind, bucket_config);
-    let mut active_tenants: Vec<TenantShardId> = vec![];
+    let mut active_tenants: Vec<TenantId> = vec![];
     let mut counter = 0;
     while let Some(result) = tenants_checked.next().await {
-        let (tenant_shard_id, console_result) = result?;
+        let (tenant_id, console_result) = result?;
 
         // Paranoia check
         if let Some(project) = &console_result {
-            assert!(project.tenant == tenant_shard_id.tenant_id);
+            assert!(project.tenant == tenant_id);
         }
 
-        if garbage.maybe_append(GarbageEntity::Tenant(tenant_shard_id), console_result) {
-            tracing::debug!("Tenant {tenant_shard_id} is garbage");
+        if garbage.maybe_append(GarbageEntity::Tenant(tenant_id), console_result) {
+            tracing::debug!("Tenant {tenant_id} is garbage");
         } else {
-            tracing::debug!("Tenant {tenant_shard_id} is active");
-            active_tenants.push(tenant_shard_id);
+            tracing::debug!("Tenant {tenant_id} is active");
+            active_tenants.push(tenant_id);
         }
 
         counter += 1;
@@ -270,13 +266,13 @@ impl std::fmt::Display for PurgeMode {
 pub async fn get_tenant_objects(
     s3_client: &Arc<Client>,
     target: RootTarget,
-    tenant_shard_id: TenantShardId,
+    tenant_id: TenantId,
 ) -> anyhow::Result<Vec<ObjectIdentifier>> {
-    tracing::debug!("Listing objects in tenant {tenant_shard_id}");
+    tracing::debug!("Listing objects in tenant {tenant_id}");
     // TODO: apply extra validation based on object modification time.  Don't purge
     // tenants where any timeline's index_part.json has been touched recently.
 
-    let mut tenant_root = target.tenant_root(&tenant_shard_id);
+    let mut tenant_root = target.tenant_root(&tenant_id);
 
     // Remove delimiter, so that object listing lists all keys in the prefix and not just
     // common prefixes.
@@ -289,7 +285,7 @@ pub async fn get_tenant_objects(
 pub async fn get_timeline_objects(
     s3_client: &Arc<Client>,
     target: RootTarget,
-    ttid: TenantShardTimelineId,
+    ttid: TenantTimelineId,
 ) -> anyhow::Result<Vec<ObjectIdentifier>> {
     tracing::debug!("Listing objects in timeline {ttid}");
     let mut timeline_root = target.timeline_root(&ttid);

s3_scrubber/src/lib.rs

@@ -22,7 +22,6 @@ use aws_sdk_s3::{Client, Config};
 
 use clap::ValueEnum;
 use pageserver::tenant::TENANTS_SEGMENT_NAME;
-use pageserver_api::shard::TenantShardId;
 use reqwest::Url;
 use serde::{Deserialize, Serialize};
 use std::io::IsTerminal;
@@ -30,7 +29,7 @@ use tokio::io::AsyncReadExt;
 use tracing::error;
 use tracing_appender::non_blocking::WorkerGuard;
 use tracing_subscriber::{fmt, prelude::*, EnvFilter};
-use utils::id::TimelineId;
+use utils::id::{TenantId, TenantTimelineId};
 
 const MAX_RETRIES: usize = 20;
 const CLOUD_ADMIN_API_TOKEN_ENV_VAR: &str = "CLOUD_ADMIN_API_TOKEN";
@@ -45,35 +44,6 @@ pub struct S3Target {
     pub delimiter: String,
 }
 
-/// Convenience for referring to timelines within a particular shard: more ergonomic
-/// than using a 2-tuple.
-///
-/// This is the shard-aware equivalent of TenantTimelineId.  It's defined here rather
-/// than somewhere more broadly exposed, because this kind of thing is rarely needed
-/// in the pageserver, as all timeline objects existing in the scope of a particular
-/// tenant: the scrubber is different in that it handles collections of data referring to many
-/// TenantShardTimelineIds in on place.
-#[derive(Serialize, Deserialize, Debug, Clone, Copy, Hash, PartialEq, Eq)]
-pub struct TenantShardTimelineId {
-    tenant_shard_id: TenantShardId,
-    timeline_id: TimelineId,
-}
-
-impl TenantShardTimelineId {
-    fn new(tenant_shard_id: TenantShardId, timeline_id: TimelineId) -> Self {
-        Self {
-            tenant_shard_id,
-            timeline_id,
-        }
-    }
-}
-
-impl Display for TenantShardTimelineId {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{}/{}", self.tenant_shard_id, self.timeline_id)
-    }
-}
-
 #[derive(clap::ValueEnum, Debug, Clone, Copy, PartialEq, Eq)]
 pub enum TraversingDepth {
     Tenant,
@@ -140,19 +110,19 @@ impl RootTarget {
         }
     }
 
-    pub fn tenant_root(&self, tenant_id: &TenantShardId) -> S3Target {
+    pub fn tenant_root(&self, tenant_id: &TenantId) -> S3Target {
         self.tenants_root().with_sub_segment(&tenant_id.to_string())
     }
 
-    pub fn timelines_root(&self, tenant_id: &TenantShardId) -> S3Target {
+    pub fn timelines_root(&self, tenant_id: &TenantId) -> S3Target {
         match self {
             Self::Pageserver(_) => self.tenant_root(tenant_id).with_sub_segment("timelines"),
             Self::Safekeeper(_) => self.tenant_root(tenant_id),
         }
     }
 
-    pub fn timeline_root(&self, id: &TenantShardTimelineId) -> S3Target {
-        self.timelines_root(&id.tenant_shard_id)
+    pub fn timeline_root(&self, id: &TenantTimelineId) -> S3Target {
+        self.timelines_root(&id.tenant_id)
             .with_sub_segment(&id.timeline_id.to_string())
     }
 

s3_scrubber/src/metadata_stream.rs

@@ -3,15 +3,14 @@ use async_stream::{stream, try_stream};
 use aws_sdk_s3::{types::ObjectIdentifier, Client};
 use tokio_stream::Stream;
 
-use crate::{list_objects_with_retries, RootTarget, S3Target, TenantShardTimelineId};
-use pageserver_api::shard::TenantShardId;
-use utils::id::TimelineId;
+use crate::{list_objects_with_retries, RootTarget, S3Target, TenantId};
+use utils::id::{TenantTimelineId, TimelineId};
 
 /// Given an S3 bucket, output a stream of TenantIds discovered via ListObjectsv2
 pub fn stream_tenants<'a>(
     s3_client: &'a Client,
     target: &'a RootTarget,
-) -> impl Stream<Item = anyhow::Result<TenantShardId>> + 'a {
+) -> impl Stream<Item = anyhow::Result<TenantId>> + 'a {
     try_stream! {
         let mut continuation_token = None;
         let tenants_target = target.tenants_root();
@@ -45,14 +44,14 @@ pub fn stream_tenants<'a>(
     }
 }
 
-/// Given a TenantShardId, output a stream of the timelines within that tenant, discovered
+/// Given a TenantId, output a stream of the timelines within that tenant, discovered
 /// using ListObjectsv2.  The listing is done before the stream is built, so that this
 /// function can be used to generate concurrency on a stream using buffer_unordered.
 pub async fn stream_tenant_timelines<'a>(
     s3_client: &'a Client,
     target: &'a RootTarget,
-    tenant: TenantShardId,
-) -> anyhow::Result<impl Stream<Item = Result<TenantShardTimelineId, anyhow::Error>> + 'a> {
+    tenant: TenantId,
+) -> anyhow::Result<impl Stream<Item = Result<TenantTimelineId, anyhow::Error>> + 'a> {
     let mut timeline_ids: Vec<Result<TimelineId, anyhow::Error>> = Vec::new();
     let mut continuation_token = None;
     let timelines_target = target.timelines_root(&tenant);
@@ -99,7 +98,7 @@ pub async fn stream_tenant_timelines<'a>(
     Ok(stream! {
         for i in timeline_ids {
             let id = i?;
-            yield Ok(TenantShardTimelineId::new(tenant, id));
+            yield Ok(TenantTimelineId::new(tenant, id));
         }
     })
 }

s3_scrubber/src/scan_metadata.rs

@@ -5,19 +5,20 @@ use crate::checks::{
     TimelineAnalysis,
 };
 use crate::metadata_stream::{stream_tenant_timelines, stream_tenants};
-use crate::{init_remote, BucketConfig, NodeKind, RootTarget, TenantShardTimelineId};
+use crate::{init_remote, BucketConfig, NodeKind, RootTarget};
 use aws_sdk_s3::Client;
 use futures_util::{pin_mut, StreamExt, TryStreamExt};
 use histogram::Histogram;
 use pageserver::tenant::IndexPart;
 use serde::Serialize;
+use utils::id::TenantTimelineId;
 
 #[derive(Serialize)]
 pub struct MetadataSummary {
     count: usize,
-    with_errors: HashSet<TenantShardTimelineId>,
-    with_warnings: HashSet<TenantShardTimelineId>,
-    with_garbage: HashSet<TenantShardTimelineId>,
+    with_errors: HashSet<TenantTimelineId>,
+    with_warnings: HashSet<TenantTimelineId>,
+    with_garbage: HashSet<TenantTimelineId>,
     indices_by_version: HashMap<usize, usize>,
 
     layer_count: MinMaxHisto,
@@ -131,7 +132,7 @@ impl MetadataSummary {
         }
     }
 
-    fn update_analysis(&mut self, id: &TenantShardTimelineId, analysis: &TimelineAnalysis) {
+    fn update_analysis(&mut self, id: &TenantTimelineId, analysis: &TimelineAnalysis) {
         if !analysis.errors.is_empty() {
             self.with_errors.insert(*id);
         }
@@ -198,8 +199,8 @@ pub async fn scan_metadata(bucket_config: BucketConfig) -> anyhow::Result<Metada
     async fn report_on_timeline(
         s3_client: &Client,
         target: &RootTarget,
-        ttid: TenantShardTimelineId,
-    ) -> anyhow::Result<(TenantShardTimelineId, S3TimelineBlobData)> {
+        ttid: TenantTimelineId,
+    ) -> anyhow::Result<(TenantTimelineId, S3TimelineBlobData)> {
         let data = list_timeline_blobs(s3_client, ttid, target).await?;
         Ok((ttid, data))
     }
@@ -212,7 +213,8 @@ pub async fn scan_metadata(bucket_config: BucketConfig) -> anyhow::Result<Metada
         let (ttid, data) = i?;
         summary.update_data(&data);
 
-        let analysis = branch_cleanup_and_check_errors(&ttid, &target, None, None, Some(data));
+        let analysis =
+            branch_cleanup_and_check_errors(&ttid, &target, None, None, Some(data)).await;
 
         summary.update_analysis(&ttid, &analysis);
     }

safekeeper/src/handler.rs

@@ -11,7 +11,7 @@ use tracing::{debug, info, info_span, Instrument};
 use crate::auth::check_permission;
 use crate::json_ctrl::{handle_json_ctrl, AppendLogicalMessage};
 
-use crate::metrics::{TrafficMetrics, PG_QUERIES_GAUGE};
+use crate::metrics::{TrafficMetrics, PG_QUERIES_FINISHED, PG_QUERIES_RECEIVED};
 use crate::safekeeper::Term;
 use crate::timeline::TimelineError;
 use crate::wal_service::ConnectionId;
@@ -210,7 +210,10 @@ impl<IO: AsyncRead + AsyncWrite + Unpin + Send> postgres_backend::Handler<IO>
         let cmd = parse_cmd(query_string)?;
         let cmd_str = cmd_to_string(&cmd);
 
-        let _guard = PG_QUERIES_GAUGE.with_label_values(&[cmd_str]).guard();
+        PG_QUERIES_RECEIVED.with_label_values(&[cmd_str]).inc();
+        scopeguard::defer! {
+            PG_QUERIES_FINISHED.with_label_values(&[cmd_str]).inc();
+        }
 
         info!("got query {:?}", query_string);
 

safekeeper/src/metrics.rs

@@ -11,8 +11,7 @@ use futures::Future;
 use metrics::{
     core::{AtomicU64, Collector, Desc, GenericCounter, GenericGaugeVec, Opts},
     proto::MetricFamily,
-    register_int_counter, register_int_counter_pair_vec, register_int_counter_vec, Gauge,
-    IntCounter, IntCounterPairVec, IntCounterVec, IntGaugeVec,
+    register_int_counter, register_int_counter_vec, Gauge, IntCounter, IntCounterVec, IntGaugeVec,
 };
 use once_cell::sync::Lazy;
 
@@ -90,10 +89,16 @@ pub static BROKER_PULLED_UPDATES: Lazy<IntCounterVec> = Lazy::new(|| {
     )
     .expect("Failed to register safekeeper_broker_pulled_updates_total counter")
 });
-pub static PG_QUERIES_GAUGE: Lazy<IntCounterPairVec> = Lazy::new(|| {
-    register_int_counter_pair_vec!(
+pub static PG_QUERIES_RECEIVED: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
         "safekeeper_pg_queries_received_total",
         "Number of queries received through pg protocol",
+        &["query"]
+    )
+    .expect("Failed to register safekeeper_pg_queries_received_total counter")
+});
+pub static PG_QUERIES_FINISHED: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
         "safekeeper_pg_queries_finished_total",
         "Number of queries finished through pg protocol",
         &["query"]

test_runner/fixtures/neon_fixtures.py

@@ -2945,7 +2945,7 @@ class Safekeeper:
         tli_dir = self.timeline_dir(tenant_id, timeline_id)
         segments = []
         for _, _, filenames in os.walk(tli_dir):
-            segments.extend([f for f in filenames if not f.startswith("safekeeper.control")])
+            segments.extend([f for f in filenames if f != "safekeeper.control"])
         segments.sort()
         return segments
 

test_runner/fixtures/pageserver/allowed_errors.py

@@ -79,9 +79,6 @@ DEFAULT_PAGESERVER_ALLOWED_ERRORS = (
     # AWS S3 may emit 500 errors for keys in a DeleteObjects response: we retry these
     # and it is not a failure of our code when it happens.
     ".*DeleteObjects.*We encountered an internal error. Please try again.*",
-    # During shutdown, DownloadError::Cancelled may be logged as an error.  Cleaning this
-    # up is tracked in https://github.com/neondatabase/neon/issues/6096
-    ".*Cancelled, shutting down.*",
 )
 
 

test_runner/fixtures/pageserver/http.py

@@ -322,10 +322,6 @@ class PageserverHttpClient(requests.Session):
         self.verbose_error(res)
         return TenantConfig.from_json(res.json())
 
-    def tenant_heatmap_upload(self, tenant_id: TenantId):
-        res = self.post(f"http://localhost:{self.port}/v1/tenant/{tenant_id}/heatmap_upload")
-        self.verbose_error(res)
-
     def set_tenant_config(self, tenant_id: TenantId, config: dict[str, Any]):
         assert "tenant_id" not in config.keys()
         res = self.put(

test_runner/fixtures/remote_storage.py

@@ -16,7 +16,6 @@ from fixtures.log_helper import log
 from fixtures.types import TenantId, TimelineId
 
 TIMELINE_INDEX_PART_FILE_NAME = "index_part.json"
-TENANT_HEATMAP_FILE_NAME = "heatmap-v1.json"
 
 
 @enum.unique
@@ -134,13 +133,6 @@ class LocalFsStorage:
         with self.index_path(tenant_id, timeline_id).open("r") as f:
             return json.load(f)
 
-    def heatmap_path(self, tenant_id: TenantId) -> Path:
-        return self.tenant_path(tenant_id) / TENANT_HEATMAP_FILE_NAME
-
-    def heatmap_content(self, tenant_id):
-        with self.heatmap_path(tenant_id).open("r") as f:
-            return json.load(f)
-
     def to_toml_inline_table(self) -> str:
         rv = {
             "local_path": str(self.root),

test_runner/regress/test_attach_tenant_config.py

@@ -52,16 +52,7 @@ def negative_env(neon_env_builder: NeonEnvBuilder) -> Generator[NegativeTests, N
         TenantId(t["id"]) for t in ps_http.tenant_list()
     ], "tenant should not be attached after negative test"
 
-    env.pageserver.allowed_errors.extend(
-        [
-            # This fixture detaches the tenant, and tests using it will tend to re-attach it
-            # shortly after. There may be un-processed deletion_queue validations from the
-            # initial attachment
-            ".*Dropped remote consistent LSN updates.*",
-            # This fixture is for tests that will intentionally generate 400 responses
-            ".*Error processing HTTP request: Bad request",
-        ]
-    )
+    env.pageserver.allowed_errors.append(".*Error processing HTTP request: Bad request")
 
     def log_contains_bad_request():
         env.pageserver.log_contains(".*Error processing HTTP request: Bad request")
@@ -172,7 +163,6 @@ def test_fully_custom_config(positive_env: NeonEnv):
         "gc_feedback": True,
         "gc_horizon": 23 * (1024 * 1024),
         "gc_period": "2h 13m",
-        "heatmap_period": "10m",
         "image_creation_threshold": 7,
         "pitr_interval": "1m",
         "lagging_wal_timeout": "23m",

test_runner/regress/test_branching.py

@@ -1,7 +1,8 @@
 import random
 import threading
 import time
-from typing import List
+from queue import SimpleQueue
+from typing import Any, Dict, List, Union
 
 import pytest
 from fixtures.log_helper import log
@@ -238,6 +239,92 @@ def test_cannot_branch_from_non_uploaded_branch(neon_env_builder: NeonEnvBuilder
         t.join()
 
 
+def test_competing_branchings_from_loading_race_to_ok_or_err(neon_env_builder: NeonEnvBuilder):
+    """
+    If the activate only after upload is used, then retries could become competing.
+    """
+
+    env = neon_env_builder.init_configs()
+    env.start()
+
+    env.pageserver.allowed_errors.extend(
+        [
+            ".*request{method=POST path=/v1/tenant/.*/timeline request_id=.*}: request was dropped before completing.*",
+            ".*Error processing HTTP request: InternalServerError\\(Timeline .*/.* already exists in pageserver's memory",
+        ]
+    )
+    ps_http = env.pageserver.http_client()
+
+    # pause all uploads
+    ps_http.configure_failpoints(("before-upload-index-pausable", "pause"))
+    env.pageserver.tenant_create(env.initial_tenant)
+
+    def start_creating_timeline():
+        ps_http.timeline_create(
+            env.pg_version, env.initial_tenant, env.initial_timeline, timeout=60
+        )
+
+    create_root = threading.Thread(target=start_creating_timeline)
+
+    branch_id = TimelineId.generate()
+
+    queue: SimpleQueue[Union[Dict[Any, Any], Exception]] = SimpleQueue()
+    barrier = threading.Barrier(3)
+
+    def try_branch():
+        barrier.wait()
+        barrier.wait()
+        try:
+            ret = ps_http.timeline_create(
+                env.pg_version,
+                env.initial_tenant,
+                branch_id,
+                ancestor_timeline_id=env.initial_timeline,
+                timeout=5,
+            )
+            queue.put(ret)
+        except Exception as e:
+            queue.put(e)
+
+    threads = [threading.Thread(target=try_branch) for _ in range(2)]
+
+    try:
+        create_root.start()
+
+        for t in threads:
+            t.start()
+
+        wait_until_paused(env, "before-upload-index-pausable")
+
+        barrier.wait()
+        ps_http.configure_failpoints(("before-upload-index-pausable", "off"))
+        barrier.wait()
+
+        # now both requests race to branch, only one can win because they take gc_cs, Tenant::timelines or marker files
+        first = queue.get()
+        second = queue.get()
+
+        log.info(first)
+        log.info(second)
+
+        (succeeded, failed) = (first, second) if isinstance(second, Exception) else (second, first)
+        assert isinstance(failed, Exception)
+        assert isinstance(succeeded, Dict)
+
+        # there's multiple valid status codes:
+        # - Timeline x/y already exists
+        # - whatever 409 response says, but that is a subclass of PageserverApiException
+        assert isinstance(failed, PageserverApiException)
+        assert succeeded["state"] == "Active"
+    finally:
+        # we might still have the failpoint active
+        env.pageserver.stop(immediate=True)
+
+        for t in threads:
+            t.join()
+        create_root.join()
+
+
 def test_non_uploaded_root_timeline_is_deleted_after_restart(neon_env_builder: NeonEnvBuilder):
     """
     Check that a timeline is deleted locally on subsequent restart if it never successfully uploaded during creation.

test_runner/regress/test_broken_timeline.py

@@ -20,7 +20,7 @@ def test_local_corruption(neon_env_builder: NeonEnvBuilder):
 
     env.pageserver.allowed_errors.extend(
         [
-            ".*get_value_reconstruct_data for layer .*",
+            ".*layer loading failed:.*",
             ".*could not find data for key.*",
             ".*is not active. Current state: Broken.*",
             ".*will not become active. Current state: Broken.*",
@@ -83,7 +83,7 @@ def test_local_corruption(neon_env_builder: NeonEnvBuilder):
     # (We don't check layer file contents on startup, when loading the timeline)
     #
     # This will change when we implement checksums for layers
-    with pytest.raises(Exception, match="get_value_reconstruct_data for layer ") as err:
+    with pytest.raises(Exception, match="layer loading failed:") as err:
         pg2.start()
     log.info(
         f"As expected, compute startup failed for timeline {tenant2}/{timeline2} with corrupt layers: {err}"

test_runner/regress/test_pageserver_secondary.py

@@ -4,7 +4,7 @@ from typing import Any, Dict, Optional
 import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnvBuilder, NeonPageserver
-from fixtures.remote_storage import LocalFsStorage, RemoteStorageKind
+from fixtures.remote_storage import RemoteStorageKind
 from fixtures.types import TenantId, TimelineId
 from fixtures.utils import wait_until
 from fixtures.workload import Workload
@@ -330,46 +330,3 @@ def test_live_migration(neon_env_builder: NeonEnvBuilder):
 
     workload.churn_rows(64, pageserver_b.id)
     workload.validate(pageserver_b.id)
-
-
-def test_heatmap_uploads(neon_env_builder: NeonEnvBuilder):
-    """
-    Test the sequence of location states that are used in a live migration.
-    """
-    env = neon_env_builder.init_start()  # initial_tenant_conf=TENANT_CONF)
-    assert isinstance(env.pageserver_remote_storage, LocalFsStorage)
-
-    tenant_id = env.initial_tenant
-    timeline_id = env.initial_timeline
-
-    # Write some data so that we have some layers
-    workload = Workload(env, tenant_id, timeline_id)
-    workload.init(env.pageservers[0].id)
-
-    # Write some layers and upload a heatmap
-    workload.write_rows(256, env.pageservers[0].id)
-    env.pageserver.http_client().tenant_heatmap_upload(tenant_id)
-
-    def validate_heatmap(heatmap):
-        assert len(heatmap["timelines"]) == 1
-        assert heatmap["timelines"][0]["timeline_id"] == str(timeline_id)
-        assert len(heatmap["timelines"][0]["layers"]) > 0
-        layers = heatmap["timelines"][0]["layers"]
-
-        # Each layer appears at most once
-        assert len(set(layer["name"] for layer in layers)) == len(layers)
-
-    # Download and inspect the heatmap that the pageserver uploaded
-    heatmap_first = env.pageserver_remote_storage.heatmap_content(tenant_id)
-    log.info(f"Read back heatmap: {heatmap_first}")
-    validate_heatmap(heatmap_first)
-
-    # Do some more I/O to generate more layers
-    workload.churn_rows(64, env.pageservers[0].id)
-    env.pageserver.http_client().tenant_heatmap_upload(tenant_id)
-
-    # Ensure that another heatmap upload includes the new layers
-    heatmap_second = env.pageserver_remote_storage.heatmap_content(tenant_id)
-    log.info(f"Read back heatmap: {heatmap_second}")
-    assert heatmap_second != heatmap_first
-    validate_heatmap(heatmap_second)

test_runner/regress/test_timeline_size.py

@@ -300,8 +300,7 @@ def test_timeline_initial_logical_size_calculation_cancellation(
     env = neon_env_builder.init_start()
     client = env.pageserver.http_client()
 
-    tenant_id = env.initial_tenant
-    timeline_id = env.initial_timeline
+    tenant_id, timeline_id = env.neon_cli.create_tenant()
 
     # load in some data
     endpoint = env.endpoints.create_start("main", tenant_id=tenant_id)
@@ -733,142 +732,3 @@ def wait_for_timeline_size_init(
     raise Exception(
         f"timed out while waiting for current_logical_size of a timeline to reach its non-incremental value, details: {timeline_details}"
     )
-
-
-def test_ondemand_activation(neon_env_builder: NeonEnvBuilder):
-    """
-    Tenants warmuping up opportunistically will wait for one another's logical size calculations to complete
-    before proceeding.  However, they skip this if a client is actively trying to access them.
-
-    This test is not purely about logical sizes, but logical size calculation is the phase that we
-    use as a proxy for "warming up" in this test: it happens within the semaphore guard used
-    to limit concurrent tenant warm-up.
-    """
-
-    # We will run with the limit set to 1, so that once we have one tenant stuck
-    # in a pausable failpoint, the rest are prevented from proceeding through warmup.
-    neon_env_builder.pageserver_config_override = "concurrent_tenant_warmup = '1'"
-
-    env = neon_env_builder.init_start()
-    pageserver_http = env.pageserver.http_client()
-
-    # Create some tenants
-    n_tenants = 10
-    tenant_ids = {env.initial_tenant}
-    for _i in range(0, n_tenants - 1):
-        tenant_id = TenantId.generate()
-        env.pageserver.tenant_create(tenant_id)
-
-        # Empty tenants are not subject to waiting for logical size calculations, because
-        # those hapen on timeline level
-        timeline_id = TimelineId.generate()
-        env.neon_cli.create_timeline(
-            new_branch_name="main", tenant_id=tenant_id, timeline_id=timeline_id
-        )
-
-        tenant_ids.add(tenant_id)
-
-    # Restart pageserver with logical size calculations paused
-    env.pageserver.stop()
-    env.pageserver.start(
-        extra_env_vars={"FAILPOINTS": "timeline-calculate-logical-size-pause=pause"}
-    )
-
-    def get_tenant_states():
-        states = {}
-        for tenant_id in tenant_ids:
-            tenant = pageserver_http.tenant_status(tenant_id=tenant_id)
-            states[tenant_id] = tenant["state"]["slug"]
-        log.info(f"Tenant states: {states}")
-        return states
-
-    def at_least_one_active():
-        assert "Active" in set(get_tenant_states().values())
-
-    # One tenant should activate, then get stuck in their logical size calculation
-    wait_until(10, 1, at_least_one_active)
-
-    # Wait some walltime to gain confidence that other tenants really are stuck and not proceeding to activate
-    time.sleep(5)
-
-    # We should see one tenant win the activation race, and enter logical size calculation.  The rest
-    # will stay in Attaching state, waiting for the "warmup_limit" semaphore
-    expect_activated = 1
-    states = get_tenant_states()
-    assert len([s for s in states.values() if s == "Active"]) == expect_activated
-    assert len([s for s in states.values() if s == "Attaching"]) == n_tenants - expect_activated
-
-    assert (
-        pageserver_http.get_metric_value("pageserver_tenant_startup_scheduled_total") == n_tenants
-    )
-
-    # This is zero, and subsequent checks are expect_activated - 1, because this counter does not
-    # count how may tenants are Active, it counts how many have finished warmup.  The first tenant
-    # that reached Active is still stuck in its local size calculation, and has therefore not finished warmup.
-    assert pageserver_http.get_metric_value("pageserver_tenant_startup_complete_total") == 0
-
-    # If a client accesses one of the blocked tenants, it should skip waiting for warmup and
-    # go active as fast as it can.
-    stuck_tenant_id = list(
-        [(tid, s) for (tid, s) in get_tenant_states().items() if s == "Attaching"]
-    )[0][0]
-
-    endpoint = env.endpoints.create_start(branch_name="main", tenant_id=stuck_tenant_id)
-    endpoint.safe_psql_many(
-        [
-            "CREATE TABLE foo (x INTEGER)",
-            "INSERT INTO foo SELECT g FROM generate_series(1, 10) g",
-        ]
-    )
-    endpoint.stop()
-
-    # That one that we successfully accessed is now Active
-    expect_activated += 1
-    assert pageserver_http.tenant_status(tenant_id=stuck_tenant_id)["state"]["slug"] == "Active"
-    assert (
-        pageserver_http.get_metric_value("pageserver_tenant_startup_complete_total")
-        == expect_activated - 1
-    )
-
-    # The ones we didn't touch are still in Attaching
-    assert (
-        len([s for s in get_tenant_states().values() if s == "Attaching"])
-        == n_tenants - expect_activated
-    )
-
-    # Timeline creation operations also wake up Attaching tenants
-    stuck_tenant_id = list(
-        [(tid, s) for (tid, s) in get_tenant_states().items() if s == "Attaching"]
-    )[0][0]
-    pageserver_http.timeline_create(env.pg_version, stuck_tenant_id, TimelineId.generate())
-    expect_activated += 1
-    assert pageserver_http.tenant_status(tenant_id=stuck_tenant_id)["state"]["slug"] == "Active"
-    assert (
-        len([s for s in get_tenant_states().values() if s == "Attaching"])
-        == n_tenants - expect_activated
-    )
-
-    assert (
-        pageserver_http.get_metric_value("pageserver_tenant_startup_complete_total")
-        == expect_activated - 1
-    )
-
-    # When we unblock logical size calculation, all tenants should proceed to active state via
-    # the warmup route.
-    pageserver_http.configure_failpoints(("timeline-calculate-logical-size-pause", "off"))
-
-    def all_active():
-        assert all(s == "Active" for s in get_tenant_states().values())
-
-    wait_until(10, 1, all_active)
-
-    # Final control check: restarting with no failpoints at all results in all tenants coming active
-    # without being prompted by client I/O
-    env.pageserver.stop()
-    env.pageserver.start()
-    wait_until(10, 1, all_active)
-
-    assert (
-        pageserver_http.get_metric_value("pageserver_tenant_startup_scheduled_total") == n_tenants
-    )
-    assert pageserver_http.get_metric_value("pageserver_tenant_startup_complete_total") == n_tenants