Increase parallel workers to trigger more errors

.circleci/config.yml

@@ -24,12 +24,6 @@ jobs:
   # A job to build postgres
   build-postgres:
     executor: zenith-build-executor
-    parameters:
-      build_type:
-        type: enum
-        enum: ["debug", "release"]
-    environment:
-      BUILD_TYPE: << parameters.build_type >>
     steps:
         # Checkout the git repo (circleci doesn't have a flag to enable submodules here)
       - checkout
@@ -45,7 +39,7 @@ jobs:
           name: Restore postgres cache
           keys:
             # Restore ONLY if the rev key matches exactly
-            - v04-postgres-cache-<< parameters.build_type >>-{{ checksum "/tmp/cache-key-postgres" }}
+            - v03-postgres-cache-{{ checksum "/tmp/cache-key-postgres" }}
 
         # FIXME We could cache our own docker container, instead of installing packages every time.
       - run:
@@ -65,12 +59,12 @@ jobs:
             if [ ! -e tmp_install/bin/postgres ]; then
               # "depth 1" saves some time by not cloning the whole repo
               git submodule update --init --depth 1
-              make postgres -j8
+              make postgres
             fi
 
       - save_cache:
           name: Save postgres cache
-          key: v04-postgres-cache-<< parameters.build_type >>-{{ checksum "/tmp/cache-key-postgres" }}
+          key: v03-postgres-cache-{{ checksum "/tmp/cache-key-postgres" }}
           paths:
             - tmp_install
 
@@ -102,7 +96,7 @@ jobs:
           name: Restore postgres cache
           keys:
             # Restore ONLY if the rev key matches exactly
-            - v04-postgres-cache-<< parameters.build_type >>-{{ checksum "/tmp/cache-key-postgres" }}
+            - v03-postgres-cache-{{ checksum "/tmp/cache-key-postgres" }}
 
       - restore_cache:
           name: Restore rust cache
@@ -239,7 +233,7 @@ jobs:
               exit 1
             fi
             if << parameters.run_in_parallel >>; then
-              EXTRA_PARAMS="-n4 $EXTRA_PARAMS"
+              EXTRA_PARAMS="-n16 $EXTRA_PARAMS"
             fi;
             # Run the tests.
             #
@@ -251,7 +245,7 @@ jobs:
             # multiple tests in one file)
             # -rA prints summary in the end
             # -n4 uses four processes to run tests via pytest-xdist
-            pipenv run pytest --junitxml=$TEST_OUTPUT/junit.xml --tb=short -s --verbose -rA $TEST_SELECTION $EXTRA_PARAMS
+            pipenv run pytest --junitxml=$TEST_OUTPUT/junit.xml --tb=short --verbose -rA $TEST_SELECTION $EXTRA_PARAMS
       - run:
           # CircleCI artifacts are preserved one file at a time, so skipping
           # this step isn't a good idea. If you want to extract the
@@ -334,18 +328,14 @@ workflows:
   build_and_test:
     jobs:
       - check-codestyle
-      - build-postgres:
-          name: build-postgres-<< matrix.build_type >>
-          matrix:
-            parameters:
-              build_type: ["debug", "release"]
+      - build-postgres
       - build-zenith:
           name: build-zenith-<< matrix.build_type >>
           matrix:
             parameters:
               build_type: ["debug", "release"]
           requires:
-            - build-postgres-<< matrix.build_type >>
+            - build-postgres
       - run-pytest:
           name: pg_regress-tests-<< matrix.build_type >>
           matrix:

.dockerignore

@@ -11,8 +11,4 @@ test_output
 .vscode
 .zenith
 integration_tests/.zenith
-.mypy_cache
-
-Dockerfile
-.dockerignore
-
+.mypy_cache

Cargo.lock

@@ -1204,7 +1204,6 @@ dependencies = [
  "daemonize",
  "futures",
  "hex",
- "hex-literal",
  "humantime",
  "hyper",
  "lazy_static",
@@ -2339,7 +2338,6 @@ dependencies = [
  "byteorder",
  "bytes",
  "clap",
- "const_format",
  "crc32c",
  "daemonize",
  "fs2",
@@ -2359,7 +2357,6 @@ dependencies = [
  "tokio-stream",
  "walkdir",
  "workspace_hack",
- "zenith_metrics",
  "zenith_utils",
 ]
 
@@ -2580,7 +2577,6 @@ version = "0.1.0"
 dependencies = [
  "lazy_static",
  "libc",
- "once_cell",
  "prometheus",
 ]
 
@@ -2592,7 +2588,6 @@ dependencies = [
  "bincode",
  "byteorder",
  "bytes",
- "crossbeam-utils",
  "hex",
  "hex-literal",
  "hyper",
@@ -2611,7 +2606,6 @@ dependencies = [
  "slog-scope",
  "slog-stdlog",
  "slog-term",
- "tempfile",
  "thiserror",
  "tokio",
  "webpki",

Dockerfile

@@ -10,7 +10,6 @@ FROM zenithdb/build:buster AS pg-build
 WORKDIR /zenith
 COPY ./vendor/postgres vendor/postgres
 COPY ./Makefile Makefile
-ENV BUILD_TYPE release
 RUN make -j $(getconf _NPROCESSORS_ONLN) -s postgres
 RUN rm -rf postgres_install/build
 

Makefile

@@ -6,18 +6,6 @@ else
 	SECCOMP =
 endif
 
-#
-# We differentiate between release / debug build types using the BUILD_TYPE
-# environment variable.
-#
-ifeq ($(BUILD_TYPE),release)
-	PG_CONFIGURE_OPTS = --enable-debug
-	PG_CFLAGS = -O2 -g3 ${CFLAGS}
-else
-	PG_CONFIGURE_OPTS = --enable-debug --enable-cassert --enable-depend
-	PG_CFLAGS = -O0 -g3 ${CFLAGS}
-endif
-
 #
 # Top level Makefile to build Zenith and PostgreSQL
 #
@@ -42,8 +30,10 @@ tmp_install/build/config.status:
 	+@echo "Configuring postgres build"
 	mkdir -p tmp_install/build
 	(cd tmp_install/build && \
-	../../vendor/postgres/configure CFLAGS='$(PG_CFLAGS)' \
-		$(PG_CONFIGURE_OPTS) \
+	../../vendor/postgres/configure CFLAGS='-O0 -g3 $(CFLAGS)' \
+		--enable-cassert \
+		--enable-debug \
+		--enable-depend \
 		$(SECCOMP) \
 		--prefix=$(abspath tmp_install) > configure.log)
 

README.md

@@ -25,7 +25,7 @@ Pageserver consists of:
 On Ubuntu or Debian this set of packages should be sufficient to build the code:
 ```text
 apt install build-essential libtool libreadline-dev zlib1g-dev flex bison libseccomp-dev \
-libssl-dev clang pkg-config libpq-dev
+libssl-dev clang
 ```
 
 [Rust] 1.52 or later is also required.
@@ -108,13 +108,6 @@ postgres=# insert into t values(2,2);
 INSERT 0 1
 ```
 
-6. If you want to run tests afterwards (see below), you have to stop pageserver and all postgres instances you have just started:
-```sh
-> ./target/debug/zenith pg stop migration_check
-> ./target/debug/zenith pg stop main
-> ./target/debug/zenith stop
-```
-
 ## Running tests
 
 ```sh

control_plane/src/compute.rs

@@ -452,7 +452,9 @@ impl PostgresNode {
             .output()
             .expect("failed to execute whoami");
 
-        assert!(output.status.success(), "whoami failed");
+        if !output.status.success() {
+            panic!("whoami failed");
+        }
 
         String::from_utf8(output.stdout).unwrap().trim().to_string()
     }

docs/README.md

@@ -10,5 +10,5 @@
 - [pageserver/README](/pageserver/README) — pageserver overview.
 - [postgres_ffi/README](/postgres_ffi/README) — Postgres FFI overview.
 - [test_runner/README.md](/test_runner/README.md) — tests infrastructure overview.
-- [walkeeper/README](/walkeeper/README) — WAL service overview.
+- [walkeeper/README](/walkeeper/README.md) — WAL service overview.
 - [core_changes.md](core_changes.md) - Description of Zenith changes in Postgres core

pageserver/Cargo.toml

@@ -40,6 +40,3 @@ postgres_ffi = { path = "../postgres_ffi" }
 zenith_metrics = { path = "../zenith_metrics" }
 zenith_utils = { path = "../zenith_utils" }
 workspace_hack = { path = "../workspace_hack" }
-
-[dev-dependencies]
-hex-literal = "0.3"

pageserver/README

@@ -7,9 +7,8 @@ The Page Server has a few different duties:
 - Replay WAL that's applicable to the chunks that the Page Server maintains
 - Backup to S3
 
-S3 is the main fault-tolerant storage of all data, as there are no Page Server
-replicas. We use a separate fault-tolerant WAL service to reduce latency. It
-keeps track of WAL records which are not syncted to S3 yet.
+
+
 
 The Page Server consists of multiple threads that operate on a shared
 repository of page versions:

pageserver/src/bin/pageserver.rs

@@ -20,12 +20,8 @@ use daemonize::Daemonize;
 
 use pageserver::{
     branches,
-    defaults::{
-        DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_PG_LISTEN_ADDR,
-        DEFAULT_RELISH_STORAGE_MAX_CONCURRENT_SYNC_LIMITS,
-    },
-    http, page_service, relish_storage, tenant_mgr, PageServerConf, RelishStorageConfig,
-    RelishStorageKind, S3Config, LOG_FILE_NAME,
+    defaults::{DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_PG_LISTEN_ADDR},
+    http, page_service, tenant_mgr, PageServerConf, RelishStorageConfig, S3Config, LOG_FILE_NAME,
 };
 use zenith_utils::http::endpoint;
 
@@ -45,7 +41,6 @@ struct CfgFileParams {
     auth_type: Option<String>,
     // see https://github.com/alexcrichton/toml-rs/blob/6c162e6562c3e432bf04c82a3d1d789d80761a86/examples/enum_external.rs for enum deserialisation examples
     relish_storage: Option<RelishStorage>,
-    relish_storage_max_concurrent_sync: Option<String>,
 }
 
 #[derive(Serialize, Deserialize, Clone)]
@@ -96,7 +91,6 @@ impl CfgFileParams {
             auth_validation_public_key_path: get_arg("auth-validation-public-key-path"),
             auth_type: get_arg("auth-type"),
             relish_storage,
-            relish_storage_max_concurrent_sync: get_arg("relish-storage-max-concurrent-sync"),
         }
     }
 
@@ -116,9 +110,6 @@ impl CfgFileParams {
                 .or(other.auth_validation_public_key_path),
             auth_type: self.auth_type.or(other.auth_type),
             relish_storage: self.relish_storage.or(other.relish_storage),
-            relish_storage_max_concurrent_sync: self
-                .relish_storage_max_concurrent_sync
-                .or(other.relish_storage_max_concurrent_sync),
         }
     }
 
@@ -187,34 +178,25 @@ impl CfgFileParams {
             );
         }
 
-        let max_concurrent_sync = match self.relish_storage_max_concurrent_sync.as_deref() {
-            Some(relish_storage_max_concurrent_sync) => {
-                relish_storage_max_concurrent_sync.parse()?
-            }
-            None => DEFAULT_RELISH_STORAGE_MAX_CONCURRENT_SYNC_LIMITS,
-        };
-        let relish_storage_config = self.relish_storage.as_ref().map(|storage_params| {
-            let storage = match storage_params.clone() {
-                RelishStorage::Local { local_path } => {
-                    RelishStorageKind::LocalFs(PathBuf::from(local_path))
-                }
-                RelishStorage::AwsS3 {
-                    bucket_name,
-                    bucket_region,
-                    access_key_id,
-                    secret_access_key,
-                } => RelishStorageKind::AwsS3(S3Config {
-                    bucket_name,
-                    bucket_region,
-                    access_key_id,
-                    secret_access_key,
-                }),
-            };
-            RelishStorageConfig {
-                max_concurrent_sync,
-                storage,
-            }
-        });
+        let relish_storage_config =
+            self.relish_storage
+                .as_ref()
+                .map(|storage_params| match storage_params.clone() {
+                    RelishStorage::Local { local_path } => {
+                        RelishStorageConfig::LocalFs(PathBuf::from(local_path))
+                    }
+                    RelishStorage::AwsS3 {
+                        bucket_name,
+                        bucket_region,
+                        access_key_id,
+                        secret_access_key,
+                    } => RelishStorageConfig::AwsS3(S3Config {
+                        bucket_name,
+                        bucket_region,
+                        access_key_id,
+                        secret_access_key,
+                    }),
+                });
 
         Ok(PageServerConf {
             daemonize: false,
@@ -240,7 +222,6 @@ impl CfgFileParams {
 }
 
 fn main() -> Result<()> {
-    zenith_metrics::set_common_metrics_prefix("pageserver");
     let arg_matches = App::new("Zenith page server")
         .about("Materializes WAL stream to pages and serves them to the postgres")
         .arg(
@@ -364,12 +345,6 @@ fn main() -> Result<()> {
                 .takes_value(true)
                 .help("Credentials to access the AWS S3 bucket"),
         )
-        .arg(
-            Arg::with_name("relish-storage-max-concurrent-sync")
-                .long("relish-storage-max-concurrent-sync")
-                .takes_value(true)
-                .help("Maximum allowed concurrent synchronisations with storage"),
-        )
         .get_matches();
 
     let workdir = Path::new(arg_matches.value_of("workdir").unwrap_or(".zenith"));
@@ -480,20 +455,16 @@ fn start_pageserver(conf: &'static PageServerConf) -> Result<()> {
 
         match daemonize.start() {
             Ok(_) => info!("Success, daemonized"),
-            Err(e) => error!("could not daemonize: {:#}", e),
+            Err(e) => error!("Error, {}", e),
         }
     }
 
-    // keep join handles for spawned threads
-    // don't spawn threads before daemonizing
-    let mut join_handles = Vec::new();
-
-    if let Some(handle) = relish_storage::run_storage_sync_thread(conf)? {
-        join_handles.push(handle);
-    }
     // Initialize tenant manager.
     tenant_mgr::init(conf);
 
+    // keep join handles for spawned threads
+    let mut join_handles = vec![];
+
     // initialize authentication for incoming connections
     let auth = match &conf.auth_type {
         AuthType::Trust | AuthType::MD5 => None,

pageserver/src/branches.rs

@@ -17,7 +17,6 @@ use std::{
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 use log::*;
-use zenith_utils::crashsafe_dir;
 use zenith_utils::logging;
 use zenith_utils::lsn::Lsn;
 
@@ -119,7 +118,7 @@ pub fn init_pageserver(conf: &'static PageServerConf, create_tenant: Option<&str
         println!("initializing tenantid {}", tenantid);
         create_repo(conf, tenantid, dummy_redo_mgr).with_context(|| "failed to create repo")?;
     }
-    crashsafe_dir::create_dir_all(conf.tenants_path())?;
+    fs::create_dir_all(conf.tenants_path())?;
 
     println!("pageserver init succeeded");
     Ok(())
@@ -136,12 +135,12 @@ pub fn create_repo(
     }
 
     // top-level dir may exist if we are creating it through CLI
-    crashsafe_dir::create_dir_all(&repo_dir)
+    fs::create_dir_all(&repo_dir)
         .with_context(|| format!("could not create directory {}", repo_dir.display()))?;
 
-    crashsafe_dir::create_dir(conf.timelines_path(&tenantid))?;
-    crashsafe_dir::create_dir_all(conf.branches_path(&tenantid))?;
-    crashsafe_dir::create_dir_all(conf.tags_path(&tenantid))?;
+    fs::create_dir(conf.timelines_path(&tenantid))?;
+    fs::create_dir_all(conf.branches_path(&tenantid))?;
+    fs::create_dir_all(conf.tags_path(&tenantid))?;
 
     info!("created directory structure in {}", repo_dir.display());
 
@@ -151,13 +150,12 @@ pub fn create_repo(
         conf,
         wal_redo_manager,
         tenantid,
-        false,
     ));
 
     // Load data into pageserver
     // TODO To implement zenith import we need to
     //      move data loading out of create_repo()
-    bootstrap_timeline(conf, tenantid, tli, repo.as_ref())?;
+    bootstrap_timeline(conf, tenantid, tli, &*repo)?;
 
     Ok(repo)
 }
@@ -223,11 +221,7 @@ fn bootstrap_timeline(
     // Import the contents of the data directory at the initial checkpoint
     // LSN, and any WAL after that.
     let timeline = repo.create_empty_timeline(tli)?;
-    restore_local_repo::import_timeline_from_postgres_datadir(
-        &pgdata_path,
-        timeline.as_ref(),
-        lsn,
-    )?;
+    restore_local_repo::import_timeline_from_postgres_datadir(&pgdata_path, &*timeline, lsn)?;
     timeline.checkpoint()?;
 
     println!(
@@ -419,6 +413,7 @@ fn create_timeline(
     let timelinedir = conf.timeline_path(&timelineid, tenantid);
 
     fs::create_dir(&timelinedir)?;
+    fs::create_dir(&timelinedir.join("wal"))?;
 
     if let Some(ancestor) = ancestor {
         let data = format!("{}@{}", ancestor.timelineid, ancestor.lsn);

pageserver/src/layered_repository.rs

@@ -18,26 +18,23 @@ use lazy_static::lazy_static;
 use log::*;
 use postgres_ffi::pg_constants::BLCKSZ;
 use serde::{Deserialize, Serialize};
-use zenith_utils::batch_fsync::batch_fsync;
 
 use std::collections::hash_map::Entry;
 use std::collections::HashMap;
 use std::collections::{BTreeSet, HashSet};
-use std::convert::TryInto;
-use std::fs;
-use std::fs::{File, OpenOptions};
+use std::fs::File;
 use std::io::Write;
 use std::ops::Bound::Included;
 use std::path::{Path, PathBuf};
 use std::sync::atomic::{AtomicUsize, Ordering};
 use std::sync::{Arc, Mutex, MutexGuard};
 use std::time::{Duration, Instant};
+use std::{fs, thread};
 
 use crate::layered_repository::inmemory_layer::FreezeLayers;
 use crate::relish::*;
-use crate::relish_storage::schedule_timeline_upload;
+use crate::relish_storage::storage_uploader::QueueBasedRelishUploader;
 use crate::repository::{GcResult, Repository, Timeline, WALRecord};
-use crate::walreceiver::IS_WAL_RECEIVER;
 use crate::walredo::WalRedoManager;
 use crate::PageServerConf;
 use crate::{ZTenantId, ZTimelineId};
@@ -47,7 +44,6 @@ use zenith_metrics::{
 };
 use zenith_metrics::{register_histogram_vec, HistogramVec};
 use zenith_utils::bin_ser::BeSer;
-use zenith_utils::crashsafe_dir;
 use zenith_utils::lsn::{AtomicLsn, Lsn, RecordLsn};
 use zenith_utils::seqwait::SeqWait;
 
@@ -58,7 +54,6 @@ mod image_layer;
 mod inmemory_layer;
 mod interval_tree;
 mod layer_map;
-mod page_versions;
 mod storage_layer;
 
 use delta_layer::DeltaLayer;
@@ -77,11 +72,6 @@ static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; 8192]);
 // Timeout when waiting for WAL receiver to catch up to an LSN given in a GetPage@LSN call.
 static TIMEOUT: Duration = Duration::from_secs(60);
 
-// Taken from PG_CONTROL_MAX_SAFE_SIZE
-const METADATA_MAX_SAFE_SIZE: usize = 512;
-const METADATA_CHECKSUM_SIZE: usize = std::mem::size_of::<u32>();
-const METADATA_MAX_DATA_SIZE: usize = METADATA_MAX_SAFE_SIZE - METADATA_CHECKSUM_SIZE;
-
 // Metrics collected on operations on the storage repository.
 lazy_static! {
     static ref STORAGE_TIME: HistogramVec = register_histogram_vec!(
@@ -112,9 +102,6 @@ lazy_static! {
     .expect("failed to define a metric");
 }
 
-/// The name of the metadata file pageserver creates per timeline.
-pub const METADATA_FILE_NAME: &str = "metadata";
-
 ///
 /// Repository consists of multiple timelines. Keep them in a hash table.
 ///
@@ -124,9 +111,7 @@ pub struct LayeredRepository {
     timelines: Mutex<HashMap<ZTimelineId, Arc<LayeredTimeline>>>,
 
     walredo_mgr: Arc<dyn WalRedoManager + Send + Sync>,
-    /// Makes evey repo's timelines to backup their files to remote storage,
-    /// when they get frozen.
-    upload_relishes: bool,
+    relish_uploader: Option<Arc<QueueBasedRelishUploader>>,
 }
 
 /// Public interface
@@ -141,7 +126,7 @@ impl Repository for LayeredRepository {
         let mut timelines = self.timelines.lock().unwrap();
 
         // Create the timeline directory, and write initial metadata to file.
-        crashsafe_dir::create_dir_all(self.conf.timeline_path(&timelineid, &self.tenantid))?;
+        std::fs::create_dir_all(self.conf.timeline_path(&timelineid, &self.tenantid))?;
 
         let metadata = TimelineMetadata {
             disk_consistent_lsn: Lsn(0),
@@ -149,7 +134,7 @@ impl Repository for LayeredRepository {
             ancestor_timeline: None,
             ancestor_lsn: Lsn(0),
         };
-        Self::save_metadata(self.conf, timelineid, self.tenantid, &metadata, true)?;
+        Self::save_metadata(self.conf, timelineid, self.tenantid, &metadata)?;
 
         let timeline = LayeredTimeline::new(
             self.conf,
@@ -158,8 +143,8 @@ impl Repository for LayeredRepository {
             timelineid,
             self.tenantid,
             Arc::clone(&self.walredo_mgr),
+            self.relish_uploader.as_ref().map(Arc::clone),
             0,
-            false,
         )?;
 
         let timeline_rc = Arc::new(timeline);
@@ -193,8 +178,8 @@ impl Repository for LayeredRepository {
             ancestor_timeline: Some(src),
             ancestor_lsn: start_lsn,
         };
-        crashsafe_dir::create_dir_all(self.conf.timeline_path(&dst, &self.tenantid))?;
-        Self::save_metadata(self.conf, dst, self.tenantid, &metadata, true)?;
+        std::fs::create_dir_all(self.conf.timeline_path(&dst, &self.tenantid))?;
+        Self::save_metadata(self.conf, dst, self.tenantid, &metadata)?;
 
         info!("branched timeline {} from {} at {}", dst, src, start_lsn);
 
@@ -231,7 +216,6 @@ impl LayeredRepository {
             Some(timeline) => Ok(timeline.clone()),
             None => {
                 let metadata = Self::load_metadata(self.conf, timelineid, self.tenantid)?;
-                let disk_consistent_lsn = metadata.disk_consistent_lsn;
 
                 // Recurse to look up the ancestor timeline.
                 //
@@ -251,26 +235,30 @@ impl LayeredRepository {
                     timelineid,
                     self.tenantid,
                     Arc::clone(&self.walredo_mgr),
-                    0, // init with 0 and update after layers are loaded,
-                    self.upload_relishes,
+                    self.relish_uploader.as_ref().map(Arc::clone),
+                    0, // init with 0 and update after layers are loaded
                 )?;
 
                 // List the layers on disk, and load them into the layer map
-                let _loaded_layers = timeline.load_layer_map(disk_consistent_lsn)?;
-                if self.upload_relishes {
-                    schedule_timeline_upload(());
-                    // schedule_timeline_upload(
-                    //     self.tenantid,
-                    //     timelineid,
-                    //     loaded_layers,
-                    //     disk_consistent_lsn,
-                    // );
-                }
+                timeline.load_layer_map()?;
 
                 // needs to be after load_layer_map
                 timeline.init_current_logical_size()?;
 
                 let timeline = Arc::new(timeline);
+
+                // Load any new WAL after the last checkpoint into memory.
+                info!(
+                    "Loading WAL for timeline {} starting at {}",
+                    timelineid,
+                    timeline.get_last_record_lsn()
+                );
+
+                if cfg!(debug_assertions) {
+                    // check again after wal loading
+                    Self::assert_size_calculation_matches_offloaded(Arc::clone(&timeline));
+                }
+
                 timelines.insert(timelineid, timeline.clone());
                 Ok(timeline)
             }
@@ -281,14 +269,15 @@ impl LayeredRepository {
         conf: &'static PageServerConf,
         walredo_mgr: Arc<dyn WalRedoManager + Send + Sync>,
         tenantid: ZTenantId,
-        upload_relishes: bool,
     ) -> LayeredRepository {
         LayeredRepository {
             tenantid,
             conf,
             timelines: Mutex::new(HashMap::new()),
             walredo_mgr,
-            upload_relishes,
+            relish_uploader: conf.relish_storage_config.as_ref().map(|config| {
+                Arc::new(QueueBasedRelishUploader::new(config, &conf.workdir).unwrap())
+            }),
         }
     }
 
@@ -363,41 +352,15 @@ impl LayeredRepository {
         timelineid: ZTimelineId,
         tenantid: ZTenantId,
         data: &TimelineMetadata,
-        first_save: bool,
-    ) -> Result<()> {
-        let path = metadata_path(conf, timelineid, tenantid);
-        // use OpenOptions to ensure file presence is consistent with first_save
-        let mut file = OpenOptions::new()
-            .write(true)
-            .create_new(first_save)
-            .open(&path)?;
+    ) -> Result<PathBuf> {
+        let path = conf.timeline_path(&timelineid, &tenantid).join("metadata");
+        let mut file = File::create(&path)?;
 
         info!("saving metadata {}", path.display());
 
-        let mut metadata_bytes = TimelineMetadata::ser(data)?;
+        file.write_all(&TimelineMetadata::ser(data)?)?;
 
-        assert!(metadata_bytes.len() <= METADATA_MAX_DATA_SIZE);
-        metadata_bytes.resize(METADATA_MAX_SAFE_SIZE, 0u8);
-
-        let checksum = crc32c::crc32c(&metadata_bytes[..METADATA_MAX_DATA_SIZE]);
-        metadata_bytes[METADATA_MAX_DATA_SIZE..].copy_from_slice(&u32::to_le_bytes(checksum));
-
-        if file.write(&metadata_bytes)? != metadata_bytes.len() {
-            bail!("Could not write all the metadata bytes in a single call");
-        }
-        file.sync_all()?;
-
-        // fsync the parent directory to ensure the directory entry is durable
-        if first_save {
-            let timeline_dir = File::open(
-                &path
-                    .parent()
-                    .expect("Metadata should always have a parent dir"),
-            )?;
-            timeline_dir.sync_all()?;
-        }
-
-        Ok(())
+        Ok(path)
     }
 
     fn load_metadata(
@@ -405,19 +368,10 @@ impl LayeredRepository {
         timelineid: ZTimelineId,
         tenantid: ZTenantId,
     ) -> Result<TimelineMetadata> {
-        let path = metadata_path(conf, timelineid, tenantid);
-        let metadata_bytes = std::fs::read(&path)?;
-        ensure!(metadata_bytes.len() == METADATA_MAX_SAFE_SIZE);
+        let path = conf.timeline_path(&timelineid, &tenantid).join("metadata");
+        let data = std::fs::read(&path)?;
 
-        let data = &metadata_bytes[..METADATA_MAX_DATA_SIZE];
-        let calculated_checksum = crc32c::crc32c(data);
-
-        let checksum_bytes: &[u8; METADATA_CHECKSUM_SIZE] =
-            metadata_bytes[METADATA_MAX_DATA_SIZE..].try_into()?;
-        let expected_checksum = u32::from_le_bytes(*checksum_bytes);
-        ensure!(calculated_checksum == expected_checksum);
-
-        let data = TimelineMetadata::des_prefix(data)?;
+        let data = TimelineMetadata::des(&data)?;
         assert!(data.disk_consistent_lsn.is_aligned());
 
         Ok(data)
@@ -485,7 +439,7 @@ impl LayeredRepository {
             let timeline = self.get_timeline_locked(*timelineid, &mut *timelines)?;
 
             if let Some(ancestor_timeline) = &timeline.ancestor_timeline {
-                // If target_timeline is specified, we only need to know branchpoints of its children
+                // If target_timeline is specified, we only need to know branchpoints of its childs
                 if let Some(timelineid) = target_timelineid {
                     if ancestor_timeline.timelineid == timelineid {
                         all_branchpoints
@@ -539,6 +493,24 @@ impl LayeredRepository {
         totals.elapsed = now.elapsed();
         Ok(totals)
     }
+
+    fn assert_size_calculation_matches(incremental: usize, timeline: &LayeredTimeline) {
+        match timeline.get_current_logical_size_non_incremental(timeline.get_last_record_lsn()) {
+            Ok(non_incremental) => {
+                if incremental != non_incremental {
+                    error!("timeline size calculation diverged, incremental doesn't match non incremental. incremental={} non_incremental={}", incremental, non_incremental);
+                }
+            }
+            Err(e) => error!("failed to calculate non incremental timeline size: {}", e),
+        }
+    }
+
+    fn assert_size_calculation_matches_offloaded(timeline: Arc<LayeredTimeline>) {
+        let incremental = timeline.get_current_logical_size();
+        thread::spawn(move || {
+            Self::assert_size_calculation_matches(incremental, &timeline);
+        });
+    }
 }
 
 /// Metadata stored on disk for each timeline
@@ -575,6 +547,8 @@ pub struct LayeredTimeline {
     // WAL redo manager
     walredo_mgr: Arc<dyn WalRedoManager + Sync + Send>,
 
+    relish_uploader: Option<Arc<QueueBasedRelishUploader>>,
+
     // What page versions do we hold in the repository? If we get a
     // request > last_record_lsn, we need to wait until we receive all
     // the WAL up to the request. The SeqWait provides functions for
@@ -622,9 +596,6 @@ pub struct LayeredTimeline {
     // TODO: it is possible to combine these two fields into single one using custom metric which uses SeqCst
     // ordering for its operations, but involves private modules, and macro trickery
     current_logical_size_gauge: IntGauge,
-
-    /// If `true`, will backup its timeline files to remote storage after freezing.
-    upload_relishes: bool,
 }
 
 /// Public interface functions
@@ -632,11 +603,8 @@ impl Timeline for LayeredTimeline {
     /// Wait until WAL has been received up to the given LSN.
     fn wait_lsn(&self, lsn: Lsn) -> Result<()> {
         // This should never be called from the WAL receiver thread, because that could lead
-        // to a deadlock.
-        assert!(
-            !IS_WAL_RECEIVER.with(|c| c.get()),
-            "wait_lsn called by WAL receiver thread"
-        );
+        // to a deadlock. FIXME: Is there a less hacky way to check that?
+        assert_ne!(thread::current().name(), Some("WAL receiver thread"));
 
         self.last_record_lsn
             .wait_for_timeout(lsn, TIMEOUT)
@@ -1008,8 +976,8 @@ impl LayeredTimeline {
         timelineid: ZTimelineId,
         tenantid: ZTenantId,
         walredo_mgr: Arc<dyn WalRedoManager + Send + Sync>,
+        relish_uploader: Option<Arc<QueueBasedRelishUploader>>,
         current_logical_size: usize,
-        upload_relishes: bool,
     ) -> Result<LayeredTimeline> {
         let current_logical_size_gauge = LOGICAL_TIMELINE_SIZE
             .get_metric_with_label_values(&[&tenantid.to_string(), &timelineid.to_string()])
@@ -1021,6 +989,7 @@ impl LayeredTimeline {
             layers: Mutex::new(LayerMap::default()),
 
             walredo_mgr,
+            relish_uploader,
 
             // initialize in-memory 'last_record_lsn' from 'disk_consistent_lsn'.
             last_record_lsn: SeqWait::new(RecordLsn {
@@ -1033,38 +1002,24 @@ impl LayeredTimeline {
             ancestor_lsn: metadata.ancestor_lsn,
             current_logical_size: AtomicUsize::new(current_logical_size),
             current_logical_size_gauge,
-            upload_relishes,
         };
         Ok(timeline)
     }
 
     ///
-    /// Scan the timeline directory to populate the layer map.
-    /// Returns all timeline-related files that were found and loaded.
+    /// Scan the timeline directory to populate the layer map
     ///
-    fn load_layer_map(&self, disk_consistent_lsn: Lsn) -> anyhow::Result<Vec<PathBuf>> {
+    fn load_layer_map(&self) -> anyhow::Result<()> {
         info!(
             "loading layer map for timeline {} into memory",
             self.timelineid
         );
         let mut layers = self.layers.lock().unwrap();
-        let (imgfilenames, deltafilenames) =
+        let (imgfilenames, mut deltafilenames) =
             filename::list_files(self.conf, self.timelineid, self.tenantid)?;
 
-        let timeline_path = self.conf.timeline_path(&self.timelineid, &self.tenantid);
-        let mut local_layers = Vec::with_capacity(imgfilenames.len() + deltafilenames.len());
         // First create ImageLayer structs for each image file.
-        for filename in &imgfilenames {
-            if filename.lsn > disk_consistent_lsn {
-                warn!(
-                    "found future image layer {} on timeline {}",
-                    filename, self.timelineid
-                );
-
-                rename_to_backup(timeline_path.join(filename.to_string()))?;
-                continue;
-            }
-
+        for filename in imgfilenames.iter() {
             let layer = ImageLayer::new(self.conf, self.timelineid, self.tenantid, filename);
 
             info!(
@@ -1073,34 +1028,41 @@ impl LayeredTimeline {
                 layer.get_start_lsn(),
                 self.timelineid
             );
-            local_layers.push(layer.path());
             layers.insert_historic(Arc::new(layer));
         }
 
-        for filename in &deltafilenames {
-            ensure!(filename.start_lsn < filename.end_lsn);
-            if filename.end_lsn > disk_consistent_lsn {
-                warn!(
-                    "found future delta layer {} on timeline {}",
-                    filename, self.timelineid
-                );
+        // Then for the Delta files. The delta files are created in order starting
+        // from the oldest file, because each DeltaLayer needs a reference to its
+        // predecessor.
+        deltafilenames.sort();
 
-                rename_to_backup(timeline_path.join(filename.to_string()))?;
-                continue;
-            }
+        for filename in deltafilenames.iter() {
+            let predecessor = layers.get(&filename.seg, filename.start_lsn);
 
-            let layer = DeltaLayer::new(self.conf, self.timelineid, self.tenantid, filename);
+            let predecessor_str: String = if let Some(prec) = &predecessor {
+                prec.filename().display().to_string()
+            } else {
+                "none".to_string()
+            };
+
+            let layer = DeltaLayer::new(
+                self.conf,
+                self.timelineid,
+                self.tenantid,
+                filename,
+                predecessor,
+            );
 
             info!(
-                "found layer {} on timeline {}",
+                "found layer {} on timeline {}, predecessor: {}",
                 layer.filename().display(),
                 self.timelineid,
+                predecessor_str,
             );
-            local_layers.push(layer.path());
             layers.insert_historic(Arc::new(layer));
         }
 
-        Ok(local_layers)
+        Ok(())
     }
 
     ///
@@ -1222,12 +1184,13 @@ impl LayeredTimeline {
         assert!(lsn.is_aligned());
 
         let last_record_lsn = self.get_last_record_lsn();
-        assert!(
-            lsn > last_record_lsn,
-            "cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})",
-            lsn,
-            last_record_lsn,
-        );
+        if lsn <= last_record_lsn {
+            panic!(
+                "cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})",
+                lsn,
+                last_record_lsn
+            );
+        }
 
         // Do we have a layer open for writing already?
         let layer;
@@ -1270,7 +1233,7 @@ impl LayeredTimeline {
             let start_lsn;
             if prev_layer.get_timeline_id() != self.timelineid {
                 // First modification on this timeline
-                start_lsn = self.ancestor_lsn + 1;
+                start_lsn = self.ancestor_lsn;
                 trace!(
                     "creating layer for write for {} at branch point {}/{}",
                     seg,
@@ -1358,8 +1321,6 @@ impl LayeredTimeline {
         // a lot of memory and/or aren't receiving much updates anymore.
         let mut disk_consistent_lsn = last_record_lsn;
 
-        let mut created_historics = false;
-        let mut layer_uploads = Vec::new();
         while let Some((oldest_layer, oldest_generation)) = layers.peek_oldest_open() {
             let oldest_pending_lsn = oldest_layer.get_oldest_pending_lsn();
 
@@ -1412,22 +1373,31 @@ impl LayeredTimeline {
             drop(layers);
             let new_historics = frozen.write_to_disk(self)?;
             layers = self.layers.lock().unwrap();
-
-            if !new_historics.is_empty() {
-                created_historics = true;
+            if let Some(relish_uploader) = &self.relish_uploader {
+                for label_path in new_historics.iter().filter_map(|layer| layer.path()) {
+                    relish_uploader.schedule_upload(self.timelineid, label_path);
+                }
             }
 
             // Finally, replace the frozen in-memory layer with the new on-disk layers
             layers.remove_historic(frozen.clone());
 
-            // Add the historics to the LayerMap
-            for delta_layer in new_historics.delta_layers {
-                layer_uploads.push(delta_layer.path());
-                layers.insert_historic(Arc::new(delta_layer));
+            // If we created a successor InMemoryLayer, its predecessor is
+            // currently the frozen layer. We need to update the predecessor
+            // to be the latest on-disk layer.
+            if let Some(last_historic) = new_historics.last() {
+                if let Some(new_open) = &maybe_new_open {
+                    let maybe_old_predecessor =
+                        new_open.update_predecessor(Arc::clone(last_historic));
+                    let old_predecessor = maybe_old_predecessor
+                        .expect("new_open should always be a successor to frozen");
+                    assert!(layer_ptr_eq(frozen.as_ref(), old_predecessor.as_ref()));
+                }
             }
-            for image_layer in new_historics.image_layers {
-                layer_uploads.push(image_layer.path());
-                layers.insert_historic(Arc::new(image_layer));
+
+            // Add the historics to the LayerMap
+            for n in new_historics {
+                layers.insert_historic(n);
             }
         }
 
@@ -1438,16 +1408,6 @@ impl LayeredTimeline {
             layer.unload()?;
         }
 
-        drop(layers);
-
-        if created_historics {
-            // We must fsync the timeline dir to ensure the directory entries for
-            // new layer files are durable
-            layer_uploads.push(self.conf.timeline_path(&self.timelineid, &self.tenantid));
-            batch_fsync(&layer_uploads)?;
-            layer_uploads.pop().unwrap();
-        }
-
         // Save the metadata, with updated 'disk_consistent_lsn', to a
         // file in the timeline dir. After crash, we will restart WAL
         // streaming and processing from that point.
@@ -1472,21 +1432,10 @@ impl LayeredTimeline {
             ancestor_timeline: ancestor_timelineid,
             ancestor_lsn: self.ancestor_lsn,
         };
-        LayeredRepository::save_metadata(
-            self.conf,
-            self.timelineid,
-            self.tenantid,
-            &metadata,
-            false,
-        )?;
-        if self.upload_relishes {
-            schedule_timeline_upload(())
-            // schedule_timeline_upload(
-            //     self.tenantid,
-            //     self.timelineid,
-            //     layer_uploads,
-            //     disk_consistent_lsn,
-            // });
+        let metadata_path =
+            LayeredRepository::save_metadata(self.conf, self.timelineid, self.tenantid, &metadata)?;
+        if let Some(relish_uploader) = &self.relish_uploader {
+            relish_uploader.schedule_upload(self.timelineid, metadata_path);
         }
 
         // Also update the in-memory copy
@@ -1731,20 +1680,12 @@ impl LayeredTimeline {
         loop {
             match layer_ref.get_page_reconstruct_data(blknum, curr_lsn, &mut data)? {
                 PageReconstructResult::Complete => break,
-                PageReconstructResult::Continue(cont_lsn) => {
+                PageReconstructResult::Continue(cont_lsn, cont_layer) => {
                     // Fetch base image / more WAL from the returned predecessor layer
-                    if let Some((cont_layer, cont_lsn)) = self.get_layer_for_read(seg, cont_lsn)? {
-                        layer_arc = cont_layer;
-                        layer_ref = &*layer_arc;
-                        curr_lsn = cont_lsn;
-                        continue;
-                    } else {
-                        bail!(
-                            "could not find predecessor layer of segment {} at {}",
-                            seg.rel,
-                            cont_lsn
-                        );
-                    }
+                    layer_arc = cont_layer;
+                    layer_ref = &*layer_arc;
+                    curr_lsn = cont_lsn;
+                    continue;
                 }
                 PageReconstructResult::Missing(lsn) => {
                     // Oops, we could not reconstruct the page.
@@ -1917,31 +1858,13 @@ pub fn dump_layerfile_from_path(path: &Path) -> Result<()> {
     Ok(())
 }
 
-fn metadata_path(
-    conf: &'static PageServerConf,
-    timelineid: ZTimelineId,
-    tenantid: ZTenantId,
-) -> PathBuf {
-    conf.timeline_path(&timelineid, &tenantid)
-        .join(METADATA_FILE_NAME)
-}
-
-/// Add a suffix to a layer file's name: .{num}.old
-/// Uses the first available num (starts at 0)
-fn rename_to_backup(path: PathBuf) -> anyhow::Result<()> {
-    let filename = path.file_name().unwrap().to_str().unwrap();
-    let mut new_path = path.clone();
-
-    for i in 0u32.. {
-        new_path.set_file_name(format!("{}.{}.old", filename, i));
-        if !new_path.exists() {
-            std::fs::rename(&path, &new_path)?;
-            return Ok(());
-        }
-    }
-
-    Err(anyhow!(
-        "couldn't find an unused backup number for {:?}",
-        path
-    ))
+/// Check for equality of Layer memory addresses
+fn layer_ptr_eq(l1: &dyn Layer, l2: &dyn Layer) -> bool {
+    let l1_ptr = l1 as *const dyn Layer;
+    let l2_ptr = l2 as *const dyn Layer;
+    // comparing *const dyn Layer will not only check for data address equality,
+    // but also for vtable address equality.
+    // to avoid this, we compare *const ().
+    // see here for more https://github.com/rust-lang/rust/issues/46139
+    std::ptr::eq(l1_ptr as *const (), l2_ptr as *const ())
 }

pageserver/src/layered_repository/delta_layer.rs

@@ -42,13 +42,15 @@ use crate::layered_repository::filename::{DeltaFileName, PathOrConf};
 use crate::layered_repository::storage_layer::{
     Layer, PageReconstructData, PageReconstructResult, PageVersion, SegmentTag,
 };
+use crate::repository::WALRecord;
 use crate::waldecoder;
 use crate::PageServerConf;
 use crate::{ZTenantId, ZTimelineId};
-use anyhow::{bail, ensure, Result};
+use anyhow::{bail, Result};
+use bytes::Bytes;
 use log::*;
 use serde::{Deserialize, Serialize};
-use zenith_utils::vec_map::VecMap;
+use std::collections::BTreeMap;
 // avoid binding to Write (conflicts with std::io::Write)
 // while being able to use std::fmt::Write's methods
 use std::fmt::Write as _;
@@ -57,7 +59,7 @@ use std::fs::File;
 use std::io::{BufWriter, Write};
 use std::ops::Bound::Included;
 use std::path::{Path, PathBuf};
-use std::sync::{Mutex, MutexGuard};
+use std::sync::{Arc, Mutex, MutexGuard};
 
 use bookfile::{Book, BookWriter};
 
@@ -107,6 +109,12 @@ impl From<&DeltaLayer> for Summary {
     }
 }
 
+#[derive(Serialize, Deserialize)]
+struct PageVersionMeta {
+    page_image_range: Option<BlobRange>,
+    record_range: Option<BlobRange>,
+}
+
 ///
 /// DeltaLayer is the in-memory data structure associated with an
 /// on-disk delta file.  We keep a DeltaLayer in memory for each
@@ -131,6 +139,9 @@ pub struct DeltaLayer {
 
     dropped: bool,
 
+    /// Predecessor layer
+    predecessor: Option<Arc<dyn Layer>>,
+
     inner: Mutex<DeltaLayerInner>,
 }
 
@@ -141,10 +152,10 @@ pub struct DeltaLayerInner {
 
     /// All versions of all pages in the file are are kept here.
     /// Indexed by block number and LSN.
-    page_version_metas: VecMap<(u32, Lsn), BlobRange>,
+    page_version_metas: BTreeMap<(u32, Lsn), PageVersionMeta>,
 
     /// `relsizes` tracks the size of the relation at different points in time.
-    relsizes: VecMap<Lsn, u32>,
+    relsizes: BTreeMap<Lsn, u32>,
 }
 
 impl Layer for DeltaLayer {
@@ -169,7 +180,29 @@ impl Layer for DeltaLayer {
     }
 
     fn filename(&self) -> PathBuf {
-        PathBuf::from(self.layer_name().to_string())
+        PathBuf::from(
+            DeltaFileName {
+                seg: self.seg,
+                start_lsn: self.start_lsn,
+                end_lsn: self.end_lsn,
+                dropped: self.dropped,
+            }
+            .to_string(),
+        )
+    }
+
+    fn path(&self) -> Option<PathBuf> {
+        Some(Self::path_for(
+            &self.path_or_conf,
+            self.timelineid,
+            self.tenantid,
+            &DeltaFileName {
+                seg: self.seg,
+                start_lsn: self.start_lsn,
+                end_lsn: self.end_lsn,
+                dropped: self.dropped,
+            },
+        ))
     }
 
     /// Look up given page in the cache.
@@ -193,20 +226,18 @@ impl Layer for DeltaLayer {
             // Scan the metadata BTreeMap backwards, starting from the given entry.
             let minkey = (blknum, Lsn(0));
             let maxkey = (blknum, lsn);
-            let iter = inner
+            let mut iter = inner
                 .page_version_metas
-                .slice_range((Included(&minkey), Included(&maxkey)))
-                .iter()
-                .rev();
-            for ((_blknum, _lsn), blob_range) in iter {
-                let pv = PageVersion::des(&read_blob(&page_version_reader, blob_range)?)?;
-
-                if let Some(img) = pv.page_image {
+                .range((Included(&minkey), Included(&maxkey)));
+            while let Some(((_blknum, _entry_lsn), entry)) = iter.next_back() {
+                if let Some(img_range) = &entry.page_image_range {
                     // Found a page image, return it
+                    let img = Bytes::from(read_blob(&page_version_reader, img_range)?);
                     reconstruct_data.page_img = Some(img);
                     need_image = false;
                     break;
-                } else if let Some(rec) = pv.record {
+                } else if let Some(rec_range) = &entry.record_range {
+                    let rec = WALRecord::des(&read_blob(&page_version_reader, rec_range)?)?;
                     let will_init = rec.will_init;
                     reconstruct_data.records.push(rec);
                     if will_init {
@@ -224,9 +255,16 @@ impl Layer for DeltaLayer {
         }
 
         // If an older page image is needed to reconstruct the page, let the
-        // caller know.
+        // caller know about the predecessor layer.
         if need_image {
-            Ok(PageReconstructResult::Continue(self.start_lsn))
+            if let Some(cont_layer) = &self.predecessor {
+                Ok(PageReconstructResult::Continue(
+                    self.start_lsn,
+                    Arc::clone(cont_layer),
+                ))
+            } else {
+                Ok(PageReconstructResult::Missing(self.start_lsn))
+            }
         } else {
             Ok(PageReconstructResult::Complete)
         }
@@ -235,22 +273,21 @@ impl Layer for DeltaLayer {
     /// Get size of the relation at given LSN
     fn get_seg_size(&self, lsn: Lsn) -> Result<u32> {
         assert!(lsn >= self.start_lsn);
-        ensure!(
-            self.seg.rel.is_blocky(),
-            "get_seg_size() called on a non-blocky rel"
-        );
 
         // Scan the BTreeMap backwards, starting from the given entry.
         let inner = self.load()?;
-        let slice = inner
-            .relsizes
-            .slice_range((Included(&Lsn(0)), Included(&lsn)));
+        let mut iter = inner.relsizes.range((Included(&Lsn(0)), Included(&lsn)));
 
-        if let Some((_entry_lsn, entry)) = slice.last() {
-            Ok(*entry)
+        let result;
+        if let Some((_entry_lsn, entry)) = iter.next_back() {
+            result = *entry;
+        // Use the base image if needed
+        } else if let Some(predecessor) = &self.predecessor {
+            result = predecessor.get_seg_size(lsn)?;
         } else {
-            Err(anyhow::anyhow!("could not find seg size in delta layer"))
+            result = 0;
         }
+        Ok(result)
     }
 
     /// Does this segment exist at given LSN?
@@ -270,15 +307,17 @@ impl Layer for DeltaLayer {
     ///
     fn unload(&self) -> Result<()> {
         let mut inner = self.inner.lock().unwrap();
-        inner.page_version_metas = VecMap::default();
-        inner.relsizes = VecMap::default();
+        inner.page_version_metas = BTreeMap::new();
+        inner.relsizes = BTreeMap::new();
         inner.loaded = false;
         Ok(())
     }
 
     fn delete(&self) -> Result<()> {
         // delete underlying file
-        fs::remove_file(self.path())?;
+        if let Some(path) = self.path() {
+            fs::remove_file(path)?;
+        }
         Ok(())
     }
 
@@ -295,22 +334,22 @@ impl Layer for DeltaLayer {
 
         println!("--- relsizes ---");
         let inner = self.load()?;
-        for (k, v) in inner.relsizes.as_slice() {
+        for (k, v) in inner.relsizes.iter() {
             println!("  {}: {}", k, v);
         }
         println!("--- page versions ---");
         let (_path, book) = self.open_book()?;
         let chapter = book.chapter_reader(PAGE_VERSIONS_CHAPTER)?;
-        for ((blk, lsn), blob_range) in inner.page_version_metas.as_slice() {
+        for (k, v) in inner.page_version_metas.iter() {
             let mut desc = String::new();
 
-            let buf = read_blob(&chapter, blob_range)?;
-            let pv = PageVersion::des(&buf)?;
-
-            if let Some(img) = pv.page_image.as_ref() {
-                write!(&mut desc, " img {} bytes", img.len())?;
+            if let Some(page_image_range) = v.page_image_range.as_ref() {
+                let image = read_blob(&chapter, page_image_range)?;
+                write!(&mut desc, " img {} bytes", image.len())?;
             }
-            if let Some(rec) = pv.record.as_ref() {
+            if let Some(record_range) = v.record_range.as_ref() {
+                let record_bytes = read_blob(&chapter, record_range)?;
+                let rec = WALRecord::des(&record_bytes)?;
                 let wal_desc = waldecoder::describe_wal_record(&rec.rec);
                 write!(
                     &mut desc,
@@ -320,7 +359,7 @@ impl Layer for DeltaLayer {
                     wal_desc
                 )?;
             }
-            println!("  blk {} at {}: {}", blk, lsn, desc);
+            println!("  blk {} at {}: {}", k.0, k.1, desc);
         }
 
         Ok(())
@@ -358,13 +397,10 @@ impl DeltaLayer {
         start_lsn: Lsn,
         end_lsn: Lsn,
         dropped: bool,
-        page_versions: impl Iterator<Item = (u32, Lsn, &'a PageVersion)>,
-        relsizes: VecMap<Lsn, u32>,
+        predecessor: Option<Arc<dyn Layer>>,
+        page_versions: impl Iterator<Item = (&'a (u32, Lsn), &'a PageVersion)>,
+        relsizes: BTreeMap<Lsn, u32>,
     ) -> Result<DeltaLayer> {
-        if seg.rel.is_blocky() {
-            assert!(!relsizes.is_empty());
-        }
-
         let delta_layer = DeltaLayer {
             path_or_conf: PathOrConf::Conf(conf),
             timelineid,
@@ -375,14 +411,17 @@ impl DeltaLayer {
             dropped,
             inner: Mutex::new(DeltaLayerInner {
                 loaded: true,
-                page_version_metas: VecMap::default(),
+                page_version_metas: BTreeMap::new(),
                 relsizes,
             }),
+            predecessor,
         };
         let mut inner = delta_layer.inner.lock().unwrap();
 
         // Write the in-memory btreemaps into a file
-        let path = delta_layer.path();
+        let path = delta_layer
+            .path()
+            .expect("DeltaLayer is supposed to have a layer path on disk");
 
         // Note: This overwrites any existing file. There shouldn't be any.
         // FIXME: throw an error instead?
@@ -392,27 +431,44 @@ impl DeltaLayer {
 
         let mut page_version_writer = BlobWriter::new(book, PAGE_VERSIONS_CHAPTER);
 
-        for (blknum, lsn, page_version) in page_versions {
-            let buf = PageVersion::ser(page_version)?;
-            let blob_range = page_version_writer.write_blob(&buf)?;
+        for (key, page_version) in page_versions {
+            let page_image_range = page_version
+                .page_image
+                .as_ref()
+                .map(|page_image| page_version_writer.write_blob(page_image))
+                .transpose()?;
 
-            inner
-                .page_version_metas
-                .append((blknum, lsn), blob_range)
-                .unwrap();
+            let record_range = page_version
+                .record
+                .as_ref()
+                .map(|record| {
+                    let buf = WALRecord::ser(record)?;
+                    page_version_writer.write_blob(&buf)
+                })
+                .transpose()?;
+
+            let old = inner.page_version_metas.insert(
+                *key,
+                PageVersionMeta {
+                    page_image_range,
+                    record_range,
+                },
+            );
+
+            assert!(old.is_none());
         }
 
         let book = page_version_writer.close()?;
 
         // Write out page versions
         let mut chapter = book.new_chapter(PAGE_VERSION_METAS_CHAPTER);
-        let buf = VecMap::ser(&inner.page_version_metas)?;
+        let buf = BTreeMap::ser(&inner.page_version_metas)?;
         chapter.write_all(&buf)?;
         let book = chapter.close()?;
 
         // and relsizes to separate chapter
         let mut chapter = book.new_chapter(REL_SIZES_CHAPTER);
-        let buf = VecMap::ser(&inner.relsizes)?;
+        let buf = BTreeMap::ser(&inner.relsizes)?;
         chapter.write_all(&buf)?;
         let book = chapter.close()?;
 
@@ -445,7 +501,12 @@ impl DeltaLayer {
             &self.path_or_conf,
             self.timelineid,
             self.tenantid,
-            &self.layer_name(),
+            &DeltaFileName {
+                seg: self.seg,
+                start_lsn: self.start_lsn,
+                end_lsn: self.end_lsn,
+                dropped: self.dropped,
+            },
         );
 
         let file = File::open(&path)?;
@@ -493,10 +554,10 @@ impl DeltaLayer {
         }
 
         let chapter = book.read_chapter(PAGE_VERSION_METAS_CHAPTER)?;
-        let page_version_metas = VecMap::des(&chapter)?;
+        let page_version_metas = BTreeMap::des(&chapter)?;
 
         let chapter = book.read_chapter(REL_SIZES_CHAPTER)?;
-        let relsizes = VecMap::des(&chapter)?;
+        let relsizes = BTreeMap::des(&chapter)?;
 
         debug!("loaded from {}", &path.display());
 
@@ -515,6 +576,7 @@ impl DeltaLayer {
         timelineid: ZTimelineId,
         tenantid: ZTenantId,
         filename: &DeltaFileName,
+        predecessor: Option<Arc<dyn Layer>>,
     ) -> DeltaLayer {
         DeltaLayer {
             path_or_conf: PathOrConf::Conf(conf),
@@ -526,9 +588,10 @@ impl DeltaLayer {
             dropped: filename.dropped,
             inner: Mutex::new(DeltaLayerInner {
                 loaded: false,
-                page_version_metas: VecMap::default(),
-                relsizes: VecMap::default(),
+                page_version_metas: BTreeMap::new(),
+                relsizes: BTreeMap::new(),
             }),
+            predecessor,
         }
     }
 
@@ -549,28 +612,10 @@ impl DeltaLayer {
             dropped: summary.dropped,
             inner: Mutex::new(DeltaLayerInner {
                 loaded: false,
-                page_version_metas: VecMap::default(),
-                relsizes: VecMap::default(),
+                page_version_metas: BTreeMap::new(),
+                relsizes: BTreeMap::new(),
             }),
+            predecessor: None,
         })
     }
-
-    fn layer_name(&self) -> DeltaFileName {
-        DeltaFileName {
-            seg: self.seg,
-            start_lsn: self.start_lsn,
-            end_lsn: self.end_lsn,
-            dropped: self.dropped,
-        }
-    }
-
-    /// Path to the layer file in pageserver workdir.
-    pub fn path(&self) -> PathBuf {
-        Self::path_for(
-            &self.path_or_conf,
-            self.timelineid,
-            self.tenantid,
-            &self.layer_name(),
-        )
-    }
 }

pageserver/src/layered_repository/filename.rs

@@ -13,8 +13,6 @@ use anyhow::Result;
 use log::*;
 use zenith_utils::lsn::Lsn;
 
-use super::METADATA_FILE_NAME;
-
 // Note: LayeredTimeline::load_layer_map() relies on this sort order
 #[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone)]
 pub struct DeltaFileName {
@@ -37,7 +35,7 @@ impl DeltaFileName {
     /// Parse a string as a delta file name. Returns None if the filename does not
     /// match the expected pattern.
     ///
-    pub fn parse_str(fname: &str) -> Option<Self> {
+    pub fn from_str(fname: &str) -> Option<Self> {
         let rel;
         let mut parts;
         if let Some(rest) = fname.strip_prefix("rel_") {
@@ -170,7 +168,7 @@ impl ImageFileName {
     /// Parse a string as an image file name. Returns None if the filename does not
     /// match the expected pattern.
     ///
-    pub fn parse_str(fname: &str) -> Option<Self> {
+    pub fn from_str(fname: &str) -> Option<Self> {
         let rel;
         let mut parts;
         if let Some(rest) = fname.strip_prefix("rel_") {
@@ -288,11 +286,11 @@ pub fn list_files(
         let fname = direntry?.file_name();
         let fname = fname.to_str().unwrap();
 
-        if let Some(deltafilename) = DeltaFileName::parse_str(fname) {
+        if let Some(deltafilename) = DeltaFileName::from_str(fname) {
             deltafiles.push(deltafilename);
-        } else if let Some(imgfilename) = ImageFileName::parse_str(fname) {
+        } else if let Some(imgfilename) = ImageFileName::from_str(fname) {
             imgfiles.push(imgfilename);
-        } else if fname == METADATA_FILE_NAME || fname == "ancestor" || fname.ends_with(".old") {
+        } else if fname == "wal" || fname == "metadata" || fname == "ancestor" {
             // ignore these
         } else {
             warn!("unrecognized filename in timeline dir: {}", fname);

pageserver/src/layered_repository/image_layer.rs

@@ -114,7 +114,25 @@ pub struct ImageLayerInner {
 
 impl Layer for ImageLayer {
     fn filename(&self) -> PathBuf {
-        PathBuf::from(self.layer_name().to_string())
+        PathBuf::from(
+            ImageFileName {
+                seg: self.seg,
+                lsn: self.lsn,
+            }
+            .to_string(),
+        )
+    }
+
+    fn path(&self) -> Option<PathBuf> {
+        Some(Self::path_for(
+            &self.path_or_conf,
+            self.timelineid,
+            self.tenantid,
+            &ImageFileName {
+                seg: self.seg,
+                lsn: self.lsn,
+            },
+        ))
     }
 
     fn get_timeline_id(&self) -> ZTimelineId {
@@ -204,7 +222,9 @@ impl Layer for ImageLayer {
 
     fn delete(&self) -> Result<()> {
         // delete underlying file
-        fs::remove_file(self.path())?;
+        if let Some(path) = self.path() {
+            fs::remove_file(path)?;
+        }
         Ok(())
     }
 
@@ -280,7 +300,9 @@ impl ImageLayer {
         let inner = layer.inner.lock().unwrap();
 
         // Write the images into a file
-        let path = layer.path();
+        let path = layer
+            .path()
+            .expect("ImageLayer is supposed to have a layer path on disk");
         // Note: This overwrites any existing file. There shouldn't be any.
         // FIXME: throw an error instead?
         let file = File::create(&path)?;
@@ -317,7 +339,7 @@ impl ImageLayer {
         // This flushes the underlying 'buf_writer'.
         book.close()?;
 
-        trace!("saved {}", path.display());
+        trace!("saved {}", &path.display());
 
         drop(inner);
 
@@ -422,7 +444,15 @@ impl ImageLayer {
     }
 
     fn open_book(&self) -> Result<(PathBuf, Book<File>)> {
-        let path = self.path();
+        let path = Self::path_for(
+            &self.path_or_conf,
+            self.timelineid,
+            self.tenantid,
+            &ImageFileName {
+                seg: self.seg,
+                lsn: self.lsn,
+            },
+        );
 
         let file = File::open(&path)?;
         let book = Book::new(file)?;
@@ -469,21 +499,4 @@ impl ImageLayer {
             }),
         })
     }
-
-    fn layer_name(&self) -> ImageFileName {
-        ImageFileName {
-            seg: self.seg,
-            lsn: self.lsn,
-        }
-    }
-
-    /// Path to the layer file in pageserver workdir.
-    pub fn path(&self) -> PathBuf {
-        Self::path_for(
-            &self.path_or_conf,
-            self.timelineid,
-            self.tenantid,
-            &self.layer_name(),
-        )
-    }
 }

pageserver/src/layered_repository/inmemory_layer.rs

@@ -12,19 +12,18 @@ use crate::layered_repository::{DeltaLayer, ImageLayer};
 use crate::repository::WALRecord;
 use crate::PageServerConf;
 use crate::{ZTenantId, ZTimelineId};
-use anyhow::{bail, ensure, Result};
+use anyhow::{bail, Result};
 use bytes::Bytes;
 use log::*;
 use std::cmp::min;
+use std::collections::BTreeMap;
+use std::ops::Bound::Included;
 use std::path::PathBuf;
 use std::sync::{Arc, RwLock};
-use zenith_utils::vec_map::VecMap;
 
 use zenith_utils::accum::Accum;
 use zenith_utils::lsn::Lsn;
 
-use super::page_versions::PageVersions;
-
 pub struct InMemoryLayer {
     conf: &'static PageServerConf,
     tenantid: ZTenantId,
@@ -46,9 +45,6 @@ pub struct InMemoryLayer {
     /// The above fields never change. The parts that do change are in 'inner',
     /// and protected by mutex.
     inner: RwLock<InMemoryLayerInner>,
-
-    /// Predecessor layer might be needed?
-    incremental: bool,
 }
 
 pub struct InMemoryLayerInner {
@@ -59,20 +55,19 @@ pub struct InMemoryLayerInner {
     /// All versions of all pages in the layer are are kept here.
     /// Indexed by block number and LSN.
     ///
-    page_versions: PageVersions,
+    page_versions: BTreeMap<(u32, Lsn), PageVersion>,
 
     ///
     /// `segsizes` tracks the size of the segment at different points in time.
     ///
-    /// For a blocky rel, there is always one entry, at the layer's start_lsn,
-    /// so that determining the size never depends on the predecessor layer. For
-    /// a non-blocky rel, 'segsizes' is not used and is always empty.
-    ///
-    segsizes: VecMap<Lsn, u32>,
+    segsizes: BTreeMap<Lsn, u32>,
 
     /// Writes are only allowed when true.
     /// Set to false when this layer is in the process of being replaced.
     writeable: bool,
+
+    /// Predecessor layer
+    predecessor: Option<Arc<dyn Layer>>,
 }
 
 impl InMemoryLayerInner {
@@ -86,13 +81,12 @@ impl InMemoryLayerInner {
 
     fn get_seg_size(&self, lsn: Lsn) -> u32 {
         // Scan the BTreeMap backwards, starting from the given entry.
-        let slice = self.segsizes.slice_range(..=lsn);
+        let mut iter = self.segsizes.range((Included(&Lsn(0)), Included(&lsn)));
 
-        // We make sure there is always at least one entry
-        if let Some((_entry_lsn, entry)) = slice.last() {
+        if let Some((_entry_lsn, entry)) = iter.next_back() {
             *entry
         } else {
-            panic!("could not find seg size in in-memory layer");
+            0
         }
     }
 }
@@ -124,6 +118,10 @@ impl Layer for InMemoryLayer {
         PathBuf::from(format!("inmem-{}", delta_filename))
     }
 
+    fn path(&self) -> Option<PathBuf> {
+        None
+    }
+
     fn get_timeline_id(&self) -> ZTimelineId {
         self.timelineid
     }
@@ -166,16 +164,18 @@ impl Layer for InMemoryLayer {
 
         assert!(self.seg.blknum_in_seg(blknum));
 
+        let predecessor: Option<Arc<dyn Layer>>;
+
         {
             let inner = self.inner.read().unwrap();
 
-            // Scan the page versions backwards, starting from `lsn`.
-            let iter = inner
+            // Scan the BTreeMap backwards, starting from reconstruct_data.lsn.
+            let minkey = (blknum, Lsn(0));
+            let maxkey = (blknum, lsn);
+            let mut iter = inner
                 .page_versions
-                .get_block_lsn_range(blknum, ..=lsn)
-                .iter()
-                .rev();
-            for (_entry_lsn, entry) in iter {
+                .range((Included(&minkey), Included(&maxkey)));
+            while let Some(((_blknum, _entry_lsn), entry)) = iter.next_back() {
                 if let Some(img) = &entry.page_image {
                     reconstruct_data.page_img = Some(img.clone());
                     need_image = false;
@@ -192,14 +192,16 @@ impl Layer for InMemoryLayer {
                     bail!("no page image or WAL record for requested page");
                 }
             }
+
+            predecessor = inner.predecessor.clone();
             // release lock on 'inner'
         }
 
         // If an older page image is needed to reconstruct the page, let the
-        // caller know
+        // caller know about the predecessor layer.
         if need_image {
-            if self.incremental {
-                Ok(PageReconstructResult::Continue(Lsn(self.start_lsn.0 - 1)))
+            if let Some(cont_layer) = predecessor {
+                Ok(PageReconstructResult::Continue(self.start_lsn, cont_layer))
             } else {
                 Ok(PageReconstructResult::Missing(self.start_lsn))
             }
@@ -211,10 +213,6 @@ impl Layer for InMemoryLayer {
     /// Get size of the relation at given LSN
     fn get_seg_size(&self, lsn: Lsn) -> Result<u32> {
         assert!(lsn >= self.start_lsn);
-        ensure!(
-            self.seg.rel.is_blocky(),
-            "get_seg_size() called on a non-blocky rel"
-        );
 
         let inner = self.inner.read().unwrap();
         Ok(inner.get_seg_size(lsn))
@@ -254,7 +252,8 @@ impl Layer for InMemoryLayer {
     }
 
     fn is_incremental(&self) -> bool {
-        self.incremental
+        let inner = self.inner.read().unwrap();
+        inner.predecessor.is_some()
     }
 
     /// debugging function to print out the contents of the layer
@@ -272,17 +271,17 @@ impl Layer for InMemoryLayer {
             self.timelineid, self.seg, self.start_lsn, end_str
         );
 
-        for (k, v) in inner.segsizes.as_slice() {
+        for (k, v) in inner.segsizes.iter() {
             println!("segsizes {}: {}", k, v);
         }
 
-        for (blknum, lsn, pv) in inner.page_versions.ordered_page_version_iter(None) {
+        for (k, v) in inner.page_versions.iter() {
             println!(
                 "blk {} at {}: {}/{}\n",
-                blknum,
-                lsn,
-                pv.page_image.is_some(),
-                pv.record.is_some()
+                k.0,
+                k.1,
+                v.page_image.is_some(),
+                v.record.is_some()
             );
         }
 
@@ -305,18 +304,6 @@ pub struct FreezeLayers {
     pub open: Option<Arc<InMemoryLayer>>,
 }
 
-/// A result of an inmemory layer data being written to disk.
-pub struct LayersOnDisk {
-    pub delta_layers: Vec<DeltaLayer>,
-    pub image_layers: Vec<ImageLayer>,
-}
-
-impl LayersOnDisk {
-    pub fn is_empty(&self) -> bool {
-        self.delta_layers.is_empty() && self.image_layers.is_empty()
-    }
-}
-
 impl InMemoryLayer {
     fn assert_not_frozen(&self) {
         assert!(self.end_lsn.is_none());
@@ -345,12 +332,6 @@ impl InMemoryLayer {
             start_lsn
         );
 
-        // The segment is initially empty, so initialize 'segsizes' with 0.
-        let mut segsizes = VecMap::default();
-        if seg.rel.is_blocky() {
-            segsizes.append(start_lsn, 0).unwrap();
-        }
-
         Ok(InMemoryLayer {
             conf,
             timelineid,
@@ -359,12 +340,12 @@ impl InMemoryLayer {
             start_lsn,
             end_lsn: None,
             oldest_pending_lsn,
-            incremental: false,
             inner: RwLock::new(InMemoryLayerInner {
                 drop_lsn: None,
-                page_versions: PageVersions::default(),
-                segsizes,
+                page_versions: BTreeMap::new(),
+                segsizes: BTreeMap::new(),
                 writeable: true,
+                predecessor: None,
             }),
         })
     }
@@ -412,7 +393,7 @@ impl InMemoryLayer {
 
         inner.check_writeable()?;
 
-        let old = inner.page_versions.append_or_update_last(blknum, lsn, pv);
+        let old = inner.page_versions.insert((blknum, lsn), pv);
 
         if old.is_some() {
             // We already had an entry for this LSN. That's odd..
@@ -457,9 +438,7 @@ impl InMemoryLayer {
                         gapblknum,
                         blknum
                     );
-                    let old = inner
-                        .page_versions
-                        .append_or_update_last(gapblknum, lsn, zeropv);
+                    let old = inner.page_versions.insert((gapblknum, lsn), zeropv);
                     // We already had an entry for this LSN. That's odd..
 
                     if old.is_some() {
@@ -470,7 +449,7 @@ impl InMemoryLayer {
                     }
                 }
 
-                inner.segsizes.append_or_update_last(lsn, newsize).unwrap();
+                inner.segsizes.insert(lsn, newsize);
                 return Ok(newsize - oldsize);
             }
         }
@@ -479,10 +458,6 @@ impl InMemoryLayer {
 
     /// Remember that the relation was truncated at given LSN
     pub fn put_truncation(&self, lsn: Lsn, segsize: u32) -> WriteResult<()> {
-        assert!(
-            self.seg.rel.is_blocky(),
-            "put_truncation() called on a non-blocky rel"
-        );
         self.assert_not_frozen();
 
         let mut inner = self.inner.write().unwrap();
@@ -492,7 +467,7 @@ impl InMemoryLayer {
         let oldsize = inner.get_seg_size(lsn);
         assert!(segsize < oldsize);
 
-        let old = inner.segsizes.append_or_update_last(lsn, segsize).unwrap();
+        let old = inner.segsizes.insert(lsn, segsize);
 
         if old.is_some() {
             // We already had an entry for this LSN. That's odd..
@@ -543,11 +518,11 @@ impl InMemoryLayer {
             start_lsn,
         );
 
-        // Copy the segment size at the start LSN from the predecessor layer.
-        let mut segsizes = VecMap::default();
+        // For convenience, copy the segment size from the predecessor layer
+        let mut segsizes = BTreeMap::new();
         if seg.rel.is_blocky() {
             let size = src.get_seg_size(start_lsn)?;
-            segsizes.append(start_lsn, size).unwrap();
+            segsizes.insert(start_lsn, size);
         }
 
         Ok(InMemoryLayer {
@@ -558,12 +533,12 @@ impl InMemoryLayer {
             start_lsn,
             end_lsn: None,
             oldest_pending_lsn,
-            incremental: true,
             inner: RwLock::new(InMemoryLayerInner {
                 drop_lsn: None,
-                page_versions: PageVersions::default(),
+                page_versions: BTreeMap::new(),
                 segsizes,
                 writeable: true,
+                predecessor: Some(src),
             }),
         })
     }
@@ -614,18 +589,28 @@ impl InMemoryLayer {
 
         // Divide all the page versions into old and new
         // at the 'cutoff_lsn' point.
+        let mut before_segsizes = BTreeMap::new();
+        let mut after_segsizes = BTreeMap::new();
         let mut after_oldest_lsn: Accum<Lsn> = Accum(None);
-
-        let cutoff_lsn_exclusive = Lsn(cutoff_lsn.0 + 1);
-
-        let (before_segsizes, mut after_segsizes) = inner.segsizes.split_at(&cutoff_lsn_exclusive);
-        if let Some((lsn, _size)) = after_segsizes.as_slice().first() {
-            after_oldest_lsn.accum(min, *lsn);
+        for (lsn, size) in inner.segsizes.iter() {
+            if *lsn > cutoff_lsn {
+                after_segsizes.insert(*lsn, *size);
+                after_oldest_lsn.accum(min, *lsn);
+            } else {
+                before_segsizes.insert(*lsn, *size);
+            }
         }
 
-        let (before_page_versions, after_page_versions) = inner
-            .page_versions
-            .split_at(cutoff_lsn_exclusive, &mut after_oldest_lsn);
+        let mut before_page_versions = BTreeMap::new();
+        let mut after_page_versions = BTreeMap::new();
+        for ((blknum, lsn), pv) in inner.page_versions.iter() {
+            if *lsn > cutoff_lsn {
+                after_page_versions.insert((*blknum, *lsn), pv.clone());
+                after_oldest_lsn.accum(min, *lsn);
+            } else {
+                before_page_versions.insert((*blknum, *lsn), pv.clone());
+            }
+        }
 
         let frozen = Arc::new(InMemoryLayer {
             conf: self.conf,
@@ -635,12 +620,12 @@ impl InMemoryLayer {
             start_lsn: self.start_lsn,
             end_lsn: Some(cutoff_lsn),
             oldest_pending_lsn: self.start_lsn,
-            incremental: self.incremental,
             inner: RwLock::new(InMemoryLayerInner {
                 drop_lsn: inner.drop_lsn,
                 page_versions: before_page_versions,
                 segsizes: before_segsizes,
                 writeable: false,
+                predecessor: inner.predecessor.clone(),
             }),
         });
 
@@ -655,11 +640,8 @@ impl InMemoryLayer {
             )?;
 
             let new_inner = new_open.inner.get_mut().unwrap();
-            // Ensure page_versions doesn't contain anything
-            // so we can just replace it
-            assert!(new_inner.page_versions.is_empty());
-            new_inner.page_versions = after_page_versions;
-            new_inner.segsizes.extend(&mut after_segsizes).unwrap();
+            new_inner.page_versions.append(&mut after_page_versions);
+            new_inner.segsizes.append(&mut after_segsizes);
 
             Some(Arc::new(new_open))
         } else {
@@ -677,7 +659,7 @@ impl InMemoryLayer {
     /// WAL records between start and end LSN. (The delta layer is not needed
     /// when a new relish is created with a single LSN, so that the start and
     /// end LSN are the same.)
-    pub fn write_to_disk(&self, timeline: &LayeredTimeline) -> Result<LayersOnDisk> {
+    pub fn write_to_disk(&self, timeline: &LayeredTimeline) -> Result<Vec<Arc<dyn Layer>>> {
         trace!(
             "write_to_disk {} end_lsn is {} get_end_lsn is {}",
             self.filename().display(),
@@ -686,7 +668,7 @@ impl InMemoryLayer {
         );
 
         // Grab the lock in read-mode. We hold it over the I/O, but because this
-        // layer is not writeable anymore, no one should be trying to acquire the
+        // layer is not writeable anymore, no one should be trying to aquire the
         // write lock on it, so we shouldn't block anyone. There's one exception
         // though: another thread might have grabbed a reference to this layer
         // in `get_layer_for_write' just before the checkpointer called
@@ -697,6 +679,8 @@ impl InMemoryLayer {
         let inner = self.inner.read().unwrap();
         assert!(!inner.writeable);
 
+        let predecessor = inner.predecessor.clone();
+
         if let Some(drop_lsn) = inner.drop_lsn {
             let delta_layer = DeltaLayer::create(
                 self.conf,
@@ -706,7 +690,8 @@ impl InMemoryLayer {
                 self.start_lsn,
                 drop_lsn,
                 true,
-                inner.page_versions.ordered_page_version_iter(None),
+                predecessor,
+                inner.page_versions.iter(),
                 inner.segsizes.clone(),
             )?;
             trace!(
@@ -715,20 +700,26 @@ impl InMemoryLayer {
                 self.start_lsn,
                 drop_lsn
             );
-            return Ok(LayersOnDisk {
-                delta_layers: vec![delta_layer],
-                image_layers: Vec::new(),
-            });
+            return Ok(vec![Arc::new(delta_layer)]);
         }
 
         let end_lsn = self.end_lsn.unwrap();
 
-        let mut before_page_versions = inner.page_versions.ordered_page_version_iter(Some(end_lsn));
+        let mut before_segsizes = BTreeMap::new();
+        for (lsn, size) in inner.segsizes.iter() {
+            if *lsn <= end_lsn {
+                before_segsizes.insert(*lsn, *size);
+            }
+        }
+        let mut before_page_versions = inner.page_versions.iter().filter(|tup| {
+            let ((_blknum, lsn), _pv) = tup;
+
+            *lsn < end_lsn
+        });
+
+        let mut frozen_layers: Vec<Arc<dyn Layer>> = Vec::new();
 
-        let mut delta_layers = Vec::new();
         if self.start_lsn != end_lsn {
-            let (before_segsizes, _after_segsizes) = inner.segsizes.split_at(&Lsn(end_lsn.0 + 1));
-
             // Write the page versions before the cutoff to disk.
             let delta_layer = DeltaLayer::create(
                 self.conf,
@@ -738,10 +729,11 @@ impl InMemoryLayer {
                 self.start_lsn,
                 end_lsn,
                 false,
+                predecessor,
                 before_page_versions,
                 before_segsizes,
             )?;
-            delta_layers.push(delta_layer);
+            frozen_layers.push(Arc::new(delta_layer));
             trace!(
                 "freeze: created delta layer {} {}-{}",
                 self.seg,
@@ -756,11 +748,14 @@ impl InMemoryLayer {
 
         // Write a new base image layer at the cutoff point
         let image_layer = ImageLayer::create_from_src(self.conf, timeline, self, end_lsn)?;
+        frozen_layers.push(Arc::new(image_layer));
         trace!("freeze: created image layer {} at {}", self.seg, end_lsn);
 
-        Ok(LayersOnDisk {
-            delta_layers,
-            image_layers: vec![image_layer],
-        })
+        Ok(frozen_layers)
+    }
+
+    pub fn update_predecessor(&self, predecessor: Arc<dyn Layer>) -> Option<Arc<dyn Layer>> {
+        let mut inner = self.inner.write().unwrap();
+        inner.predecessor.replace(predecessor)
     }
 }

pageserver/src/layered_repository/interval_tree.rs

@@ -283,7 +283,6 @@ mod tests {
             write!(f, "{}", self.val)
         }
     }
-    #[rustfmt::skip]
     fn assert_search(
         tree: &IntervalTree<MockItem>,
         key: u32,
@@ -292,20 +291,24 @@ mod tests {
         if let Some(v) = tree.search(key) {
             let vstr = v.to_string();
 
-            assert!(!expected.is_empty(), "search with {} returned {}, expected None", key, v);
-            assert!(
-                expected.contains(&vstr.as_str()),
-                "search with {} returned {}, expected one of: {:?}",
-                key, v, expected,
-            );
+            if expected.is_empty() {
+                panic!("search with {} returned {}, expected None", key, v);
+            }
 
+            if !expected.contains(&vstr.as_str()) {
+                panic!(
+                    "search with {} returned {}, expected one of: {:?}",
+                    key, v, expected
+                );
+            }
             Some(v)
         } else {
-            assert!(
-                expected.is_empty(),
-                "search with {} returned None, expected one of {:?}",
-                key, expected
-            );
+            if !expected.is_empty() {
+                panic!(
+                    "search with {} returned None, expected one of {:?}",
+                    key, expected
+                );
+            }
             None
         }
     }

pageserver/src/layered_repository/page_versions.rs

@@ -1,182 +0,0 @@
-use std::{collections::HashMap, ops::RangeBounds, slice};
-
-use zenith_utils::{accum::Accum, lsn::Lsn, vec_map::VecMap};
-
-use super::storage_layer::PageVersion;
-
-const EMPTY_SLICE: &[(Lsn, PageVersion)] = &[];
-
-#[derive(Debug, Default)]
-pub struct PageVersions(HashMap<u32, VecMap<Lsn, PageVersion>>);
-
-impl PageVersions {
-    pub fn is_empty(&self) -> bool {
-        self.0.is_empty()
-    }
-
-    pub fn append_or_update_last(
-        &mut self,
-        blknum: u32,
-        lsn: Lsn,
-        page_version: PageVersion,
-    ) -> Option<PageVersion> {
-        let map = self.0.entry(blknum).or_insert_with(VecMap::default);
-        map.append_or_update_last(lsn, page_version).unwrap()
-    }
-
-    /// Get all [`PageVersion`]s in a block
-    pub fn get_block_slice(&self, blknum: u32) -> &[(Lsn, PageVersion)] {
-        self.0
-            .get(&blknum)
-            .map(VecMap::as_slice)
-            .unwrap_or(EMPTY_SLICE)
-    }
-
-    /// Get a range of [`PageVersions`] in a block
-    pub fn get_block_lsn_range<R: RangeBounds<Lsn>>(
-        &self,
-        blknum: u32,
-        range: R,
-    ) -> &[(Lsn, PageVersion)] {
-        self.0
-            .get(&blknum)
-            .map(|vec_map| vec_map.slice_range(range))
-            .unwrap_or(EMPTY_SLICE)
-    }
-
-    /// Split the page version map into two.
-    ///
-    /// Left contains everything up to and not including [`cutoff_lsn`].
-    /// Right contains [`cutoff_lsn`] and everything after.
-    pub fn split_at(&self, cutoff_lsn: Lsn, after_oldest_lsn: &mut Accum<Lsn>) -> (Self, Self) {
-        let mut before_blocks = HashMap::new();
-        let mut after_blocks = HashMap::new();
-
-        for (blknum, vec_map) in self.0.iter() {
-            let (before_versions, after_versions) = vec_map.split_at(&cutoff_lsn);
-
-            if !before_versions.is_empty() {
-                let old = before_blocks.insert(*blknum, before_versions);
-                assert!(old.is_none());
-            }
-
-            if !after_versions.is_empty() {
-                let (first_lsn, _first_pv) = &after_versions.as_slice()[0];
-                after_oldest_lsn.accum(std::cmp::min, *first_lsn);
-
-                let old = after_blocks.insert(*blknum, after_versions);
-                assert!(old.is_none());
-            }
-        }
-
-        (Self(before_blocks), Self(after_blocks))
-    }
-
-    /// Iterate through [`PageVersion`]s in (block, lsn) order.
-    /// If a [`cutoff_lsn`] is set, only show versions with `lsn < cutoff_lsn`
-    pub fn ordered_page_version_iter(&self, cutoff_lsn: Option<Lsn>) -> OrderedPageVersionIter<'_> {
-        let mut ordered_blocks: Vec<u32> = self.0.keys().cloned().collect();
-        ordered_blocks.sort_unstable();
-
-        let slice = ordered_blocks
-            .first()
-            .map(|&blknum| self.get_block_slice(blknum))
-            .unwrap_or(EMPTY_SLICE);
-
-        OrderedPageVersionIter {
-            page_versions: self,
-            ordered_blocks,
-            cur_block_idx: 0,
-            cutoff_lsn,
-            cur_slice_iter: slice.iter(),
-        }
-    }
-}
-
-pub struct OrderedPageVersionIter<'a> {
-    page_versions: &'a PageVersions,
-
-    ordered_blocks: Vec<u32>,
-    cur_block_idx: usize,
-
-    cutoff_lsn: Option<Lsn>,
-
-    cur_slice_iter: slice::Iter<'a, (Lsn, PageVersion)>,
-}
-
-impl OrderedPageVersionIter<'_> {
-    fn is_lsn_before_cutoff(&self, lsn: &Lsn) -> bool {
-        if let Some(cutoff_lsn) = self.cutoff_lsn.as_ref() {
-            lsn < cutoff_lsn
-        } else {
-            true
-        }
-    }
-}
-
-impl<'a> Iterator for OrderedPageVersionIter<'a> {
-    type Item = (u32, Lsn, &'a PageVersion);
-
-    fn next(&mut self) -> Option<Self::Item> {
-        loop {
-            if let Some((lsn, page_version)) = self.cur_slice_iter.next() {
-                if self.is_lsn_before_cutoff(lsn) {
-                    let blknum = self.ordered_blocks[self.cur_block_idx];
-                    return Some((blknum, *lsn, page_version));
-                }
-            }
-
-            let next_block_idx = self.cur_block_idx + 1;
-            let blknum: u32 = *self.ordered_blocks.get(next_block_idx)?;
-            self.cur_block_idx = next_block_idx;
-            self.cur_slice_iter = self.page_versions.get_block_slice(blknum).iter();
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    const EMPTY_PAGE_VERSION: PageVersion = PageVersion {
-        page_image: None,
-        record: None,
-    };
-
-    #[test]
-    fn test_ordered_iter() {
-        let mut page_versions = PageVersions::default();
-        const BLOCKS: u32 = 1000;
-        const LSNS: u64 = 50;
-
-        for blknum in 0..BLOCKS {
-            for lsn in 0..LSNS {
-                let old = page_versions.append_or_update_last(blknum, Lsn(lsn), EMPTY_PAGE_VERSION);
-                assert!(old.is_none());
-            }
-        }
-
-        let mut iter = page_versions.ordered_page_version_iter(None);
-        for blknum in 0..BLOCKS {
-            for lsn in 0..LSNS {
-                let (actual_blknum, actual_lsn, _pv) = iter.next().unwrap();
-                assert_eq!(actual_blknum, blknum);
-                assert_eq!(Lsn(lsn), actual_lsn);
-            }
-        }
-        assert!(iter.next().is_none());
-        assert!(iter.next().is_none()); // should be robust against excessive next() calls
-
-        const CUTOFF_LSN: Lsn = Lsn(30);
-        let mut iter = page_versions.ordered_page_version_iter(Some(CUTOFF_LSN));
-        for blknum in 0..BLOCKS {
-            for lsn in 0..CUTOFF_LSN.0 {
-                let (actual_blknum, actual_lsn, _pv) = iter.next().unwrap();
-                assert_eq!(actual_blknum, blknum);
-                assert_eq!(Lsn(lsn), actual_lsn);
-            }
-        }
-        assert!(iter.next().is_none());
-        assert!(iter.next().is_none()); // should be robust against excessive next() calls
-    }
-}

pageserver/src/layered_repository/storage_layer.rs

@@ -10,6 +10,7 @@ use bytes::Bytes;
 use serde::{Deserialize, Serialize};
 use std::fmt;
 use std::path::PathBuf;
+use std::sync::Arc;
 
 use zenith_utils::lsn::Lsn;
 
@@ -86,9 +87,9 @@ pub struct PageReconstructData {
 pub enum PageReconstructResult {
     /// Got all the data needed to reconstruct the requested page
     Complete,
-    /// This layer didn't contain all the required data, the caller should look up
-    /// the predecessor layer at the returned LSN and collect more data from there.
-    Continue(Lsn),
+    /// This layer didn't contain all the required data, the caller should collect
+    /// more data from the returned predecessor layer at the returned LSN.
+    Continue(Lsn, Arc<dyn Layer>),
     /// This layer didn't contain data needed to reconstruct the page version at
     /// the returned LSN. This is usually considered an error, but might be OK
     /// in some circumstances.
@@ -123,6 +124,10 @@ pub trait Layer: Send + Sync {
     /// Is the segment represented by this layer dropped by PostgreSQL?
     fn is_dropped(&self) -> bool;
 
+    /// Gets the physical location of the layer on disk.
+    /// Some layers, such as in-memory, might not have the location.
+    fn path(&self) -> Option<PathBuf>;
+
     /// Filename used to store this layer on disk. (Even in-memory layers
     /// implement this, to print a handy unique identifier for the layer for
     /// log messages, even though they're never not on disk.)
@@ -140,8 +145,8 @@ pub trait Layer: Send + Sync {
     ///
     /// See PageReconstructResult for possible return values. The collected data
     /// is appended to reconstruct_data; the caller should pass an empty struct
-    /// on first call. If this returns PageReconstructResult::Continue, look up
-    /// the predecessor layer and call again with the same 'reconstruct_data'
+    /// on first call. If this returns PageReconstructResult::Continue, call
+    /// again on the returned predecessor layer with the same 'reconstruct_data'
     /// to collect more data.
     fn get_page_reconstruct_data(
         &self,

pageserver/src/lib.rs

@@ -13,7 +13,7 @@ pub mod http;
 pub mod layered_repository;
 pub mod page_service;
 pub mod relish;
-pub mod relish_storage;
+mod relish_storage;
 pub mod repository;
 pub mod restore_local_repo;
 pub mod tenant_mgr;
@@ -40,7 +40,6 @@ pub mod defaults {
     pub const DEFAULT_GC_PERIOD: Duration = Duration::from_secs(100);
 
     pub const DEFAULT_SUPERUSER: &str = "zenith_admin";
-    pub const DEFAULT_RELISH_STORAGE_MAX_CONCURRENT_SYNC_LIMITS: usize = 100;
 }
 
 lazy_static! {
@@ -126,6 +125,10 @@ impl PageServerConf {
         self.timeline_path(timelineid, tenantid).join("ancestor")
     }
 
+    fn wal_dir_path(&self, timelineid: &ZTimelineId, tenantid: &ZTenantId) -> PathBuf {
+        self.timeline_path(timelineid, tenantid).join("wal")
+    }
+
     //
     // Postgres distribution paths
     //
@@ -165,37 +168,18 @@ impl PageServerConf {
 
 /// External relish storage configuration, enough for creating a client for that storage.
 #[derive(Debug, Clone)]
-pub struct RelishStorageConfig {
-    /// Limits the number of concurrent sync operations between pageserver and relish storage.
-    pub max_concurrent_sync: usize,
-    /// The storage connection configuration.
-    pub storage: RelishStorageKind,
-}
-
-/// A kind of a relish storage to connect to, with its connection configuration.
-#[derive(Debug, Clone)]
-pub enum RelishStorageKind {
-    /// Storage based on local file system.
-    /// Specify a root folder to place all stored relish data into.
+pub enum RelishStorageConfig {
+    /// Root folder to place all stored relish data into.
     LocalFs(PathBuf),
-    /// AWS S3 based storage, storing all relishes into the root
-    /// of the S3 bucket from the config.
     AwsS3(S3Config),
 }
 
 /// AWS S3 bucket coordinates and access credentials to manage the bucket contents (read and write).
 #[derive(Clone)]
 pub struct S3Config {
-    /// Name of the bucket to connect to.
     pub bucket_name: String,
-    /// The region where the bucket is located at.
     pub bucket_region: String,
-    /// "Login" to use when connecting to bucket.
-    /// Can be empty for cases like AWS k8s IAM
-    /// where we can allow certain pods to connect
-    /// to the bucket directly without any credentials.
     pub access_key_id: Option<String>,
-    /// "Password" to use when connecting to bucket.
     pub secret_access_key: Option<String>,
 }
 

pageserver/src/page_service.rs

@@ -194,7 +194,7 @@ pub fn thread_main(
         let local_auth = auth.clone();
         thread::spawn(move || {
             if let Err(err) = page_service_conn_main(conf, local_auth, socket, auth_type) {
-                error!("page server thread exiting with error: {:#}", err);
+                error!("error: {}", err);
             }
         });
     }

pageserver/src/relish_storage.rs

@@ -8,41 +8,14 @@
 
 mod local_fs;
 mod rust_s3;
-/// A queue-based storage with the background machinery behind it to synchronize
-/// local page server layer files with external storage.
-mod synced_storage;
+/// A queue and the background machinery behind it to upload
+/// local page server layer files to external storage.
+pub mod storage_uploader;
 
-use std::{path::Path, thread};
+use std::path::Path;
 
 use anyhow::Context;
 
-pub use self::synced_storage::schedule_timeline_upload;
-use self::{local_fs::LocalFs, rust_s3::RustS3};
-use crate::{PageServerConf, RelishStorageKind};
-
-pub fn run_storage_sync_thread(
-    config: &'static PageServerConf,
-) -> anyhow::Result<Option<thread::JoinHandle<anyhow::Result<()>>>> {
-    match &config.relish_storage_config {
-        Some(relish_storage_config) => {
-            let max_concurrent_sync = relish_storage_config.max_concurrent_sync;
-            match &relish_storage_config.storage {
-                RelishStorageKind::LocalFs(root) => synced_storage::run_storage_sync_thread(
-                    config,
-                    LocalFs::new(root.clone())?,
-                    max_concurrent_sync,
-                ),
-                RelishStorageKind::AwsS3(s3_config) => synced_storage::run_storage_sync_thread(
-                    config,
-                    RustS3::new(s3_config)?,
-                    max_concurrent_sync,
-                ),
-            }
-        }
-        None => Ok(None),
-    }
-}
-
 /// Storage (potentially remote) API to manage its state.
 #[async_trait::async_trait]
 pub trait RelishStorage: Send + Sync {

pageserver/src/relish_storage/rust_s3.rs

@@ -5,10 +5,9 @@ use std::path::Path;
 use anyhow::Context;
 use s3::{bucket::Bucket, creds::Credentials, region::Region};
 
-use crate::{
-    relish_storage::{strip_workspace_prefix, RelishStorage},
-    S3Config,
-};
+use crate::{relish_storage::strip_workspace_prefix, S3Config};
+
+use super::RelishStorage;
 
 const S3_FILE_SEPARATOR: char = '/';
 

pageserver/src/relish_storage/storage_uploader.rs

@@ -0,0 +1,116 @@
+use std::{
+    collections::VecDeque,
+    path::{Path, PathBuf},
+    sync::{Arc, Mutex},
+    thread,
+};
+
+use zenith_utils::zid::ZTimelineId;
+
+use crate::{relish_storage::RelishStorage, RelishStorageConfig};
+
+use super::{local_fs::LocalFs, rust_s3::RustS3};
+
+pub struct QueueBasedRelishUploader {
+    upload_queue: Arc<Mutex<VecDeque<(ZTimelineId, PathBuf)>>>,
+}
+
+impl QueueBasedRelishUploader {
+    pub fn new(
+        config: &RelishStorageConfig,
+        page_server_workdir: &'static Path,
+    ) -> anyhow::Result<Self> {
+        let upload_queue = Arc::new(Mutex::new(VecDeque::new()));
+        let _handle = match config {
+            RelishStorageConfig::LocalFs(root) => {
+                let relish_storage = LocalFs::new(root.clone())?;
+                create_upload_thread(
+                    Arc::clone(&upload_queue),
+                    relish_storage,
+                    page_server_workdir,
+                )?
+            }
+            RelishStorageConfig::AwsS3(s3_config) => {
+                let relish_storage = RustS3::new(s3_config)?;
+                create_upload_thread(
+                    Arc::clone(&upload_queue),
+                    relish_storage,
+                    page_server_workdir,
+                )?
+            }
+        };
+
+        Ok(Self { upload_queue })
+    }
+
+    pub fn schedule_upload(&self, timeline_id: ZTimelineId, relish_path: PathBuf) {
+        self.upload_queue
+            .lock()
+            .unwrap()
+            .push_back((timeline_id, relish_path))
+    }
+}
+
+fn create_upload_thread<P, S: 'static + RelishStorage<RelishStoragePath = P>>(
+    upload_queue: Arc<Mutex<VecDeque<(ZTimelineId, PathBuf)>>>,
+    relish_storage: S,
+    page_server_workdir: &'static Path,
+) -> std::io::Result<thread::JoinHandle<()>> {
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()?;
+    thread::Builder::new()
+        .name("Queue based relish uploader".to_string())
+        .spawn(move || loop {
+            runtime.block_on(async {
+                upload_loop_step(&upload_queue, &relish_storage, page_server_workdir).await;
+            })
+        })
+}
+
+async fn upload_loop_step<P, S: 'static + RelishStorage<RelishStoragePath = P>>(
+    upload_queue: &Mutex<VecDeque<(ZTimelineId, PathBuf)>>,
+    relish_storage: &S,
+    page_server_workdir: &Path,
+) {
+    let mut queue_accessor = upload_queue.lock().unwrap();
+    log::debug!("current upload queue length: {}", queue_accessor.len());
+    let next_upload = queue_accessor.pop_front();
+    drop(queue_accessor);
+
+    let (relish_timeline_id, relish_local_path) = match next_upload {
+        Some(data) => data,
+        None => {
+            // Don't spin and allow others to use the queue.
+            // In future, could be improved to be more clever about delays depending on relish upload stats
+            thread::sleep(std::time::Duration::from_secs(1));
+            return;
+        }
+    };
+
+    if let Err(e) = upload_relish(relish_storage, page_server_workdir, &relish_local_path).await {
+        log::error!(
+            "Failed to upload relish '{}' for timeline {}, reason: {}",
+            relish_local_path.display(),
+            relish_timeline_id,
+            e
+        );
+        upload_queue
+            .lock()
+            .unwrap()
+            .push_back((relish_timeline_id, relish_local_path))
+    } else {
+        log::debug!("Relish successfully uploaded");
+    }
+}
+
+async fn upload_relish<P, S: RelishStorage<RelishStoragePath = P>>(
+    relish_storage: &S,
+    page_server_workdir: &Path,
+    relish_local_path: &Path,
+) -> anyhow::Result<()> {
+    let destination = S::derive_destination(page_server_workdir, relish_local_path)?;
+    relish_storage
+        .upload_relish(relish_local_path, &destination)
+        .await
+}

pageserver/src/relish_storage/synced_storage.rs

@@ -1,52 +0,0 @@
-use std::time::Duration;
-use std::{collections::BinaryHeap, sync::Mutex, thread};
-
-use crate::{relish_storage::RelishStorage, PageServerConf};
-
-lazy_static::lazy_static! {
-    static ref UPLOAD_QUEUE: Mutex<BinaryHeap<SyncTask>> = Mutex::new(BinaryHeap::new());
-}
-
-pub fn schedule_timeline_upload(_local_timeline: ()) {
-    // UPLOAD_QUEUE
-    //     .lock()
-    //     .unwrap()
-    //     .push(SyncTask::Upload(local_timeline))
-}
-
-#[derive(Debug, PartialEq, Eq, PartialOrd, Ord)]
-enum SyncTask {}
-
-pub fn run_storage_sync_thread<
-    P: std::fmt::Debug,
-    S: 'static + RelishStorage<RelishStoragePath = P>,
->(
-    config: &'static PageServerConf,
-    relish_storage: S,
-    max_concurrent_sync: usize,
-) -> anyhow::Result<Option<thread::JoinHandle<anyhow::Result<()>>>> {
-    let runtime = tokio::runtime::Builder::new_current_thread()
-        .enable_all()
-        .build()?;
-
-    let handle = thread::Builder::new()
-        .name("Queue based relish storage sync".to_string())
-        .spawn(move || loop {
-            let mut queue_accessor = UPLOAD_QUEUE.lock().unwrap();
-            log::debug!("Upload queue length: {}", queue_accessor.len());
-            let next_task = queue_accessor.pop();
-            drop(queue_accessor);
-            match next_task {
-                Some(task) => runtime.block_on(async {
-                    // suppress warnings
-                    let _ = (config, task, &relish_storage, max_concurrent_sync);
-                    todo!("omitted for brevity")
-                }),
-                None => {
-                    thread::sleep(Duration::from_secs(1));
-                    continue;
-                }
-            }
-        })?;
-    Ok(Some(handle))
-}

pageserver/src/repository.rs

@@ -209,21 +209,15 @@ impl WALRecord {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::layered_repository::{LayeredRepository, METADATA_FILE_NAME};
+    use crate::layered_repository::LayeredRepository;
     use crate::walredo::{WalRedoError, WalRedoManager};
     use crate::PageServerConf;
-    use hex_literal::hex;
     use postgres_ffi::pg_constants;
     use postgres_ffi::xlog_utils::SIZEOF_CHECKPOINT;
     use std::fs;
-    use std::path::PathBuf;
+    use std::str::FromStr;
     use zenith_utils::zid::ZTenantId;
 
-    const TIMELINE_ID: ZTimelineId =
-        ZTimelineId::from_array(hex!("11223344556677881122334455667788"));
-    const NEW_TIMELINE_ID: ZTimelineId =
-        ZTimelineId::from_array(hex!("AA223344556677881122334455667788"));
-
     /// Arbitrary relation tag, for testing.
     const TESTREL_A: RelishTag = RelishTag::Relation(RelTag {
         spcnode: 0,
@@ -259,53 +253,39 @@ mod tests {
     static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; 8192]);
     static ZERO_CHECKPOINT: Bytes = Bytes::from_static(&[0u8; SIZEOF_CHECKPOINT]);
 
-    struct RepoHarness {
-        conf: &'static PageServerConf,
-        tenant_id: ZTenantId,
-    }
+    fn get_test_repo(test_name: &str) -> Result<Box<dyn Repository>> {
+        let repo_dir = PageServerConf::test_repo_dir(test_name);
+        let _ = fs::remove_dir_all(&repo_dir);
+        fs::create_dir_all(&repo_dir)?;
+        fs::create_dir_all(&repo_dir.join("timelines"))?;
 
-    impl RepoHarness {
-        fn create(test_name: &'static str) -> Result<Self> {
-            let repo_dir = PageServerConf::test_repo_dir(test_name);
-            let _ = fs::remove_dir_all(&repo_dir);
-            fs::create_dir_all(&repo_dir)?;
-            fs::create_dir_all(&repo_dir.join("timelines"))?;
+        let conf = PageServerConf::dummy_conf(repo_dir);
+        // Make a static copy of the config. This can never be free'd, but that's
+        // OK in a test.
+        let conf: &'static PageServerConf = Box::leak(Box::new(conf));
+        let tenantid = ZTenantId::generate();
+        fs::create_dir_all(conf.tenant_path(&tenantid)).unwrap();
 
-            let conf = PageServerConf::dummy_conf(repo_dir);
-            // Make a static copy of the config. This can never be free'd, but that's
-            // OK in a test.
-            let conf: &'static PageServerConf = Box::leak(Box::new(conf));
+        let walredo_mgr = TestRedoManager {};
 
-            let tenant_id = ZTenantId::generate();
-            fs::create_dir_all(conf.tenant_path(&tenant_id))?;
+        let repo = Box::new(LayeredRepository::new(
+            conf,
+            Arc::new(walredo_mgr),
+            tenantid,
+        ));
 
-            Ok(Self { conf, tenant_id })
-        }
-
-        fn load(&self) -> Box<dyn Repository> {
-            let walredo_mgr = Arc::new(TestRedoManager);
-
-            Box::new(LayeredRepository::new(
-                self.conf,
-                walredo_mgr,
-                self.tenant_id,
-                false,
-            ))
-        }
-
-        fn timeline_path(&self, timeline_id: &ZTimelineId) -> PathBuf {
-            self.conf.timeline_path(timeline_id, &self.tenant_id)
-        }
+        Ok(repo)
     }
 
     #[test]
     fn test_relsize() -> Result<()> {
-        let repo = RepoHarness::create("test_relsize")?.load();
+        let repo = get_test_repo("test_relsize")?;
         // get_timeline() with non-existent timeline id should fail
         //repo.get_timeline("11223344556677881122334455667788");
 
         // Create timeline to work on
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;
 
         tline.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
         tline.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
@@ -419,10 +399,11 @@ mod tests {
     // and then created it again within the same layer.
     #[test]
     fn test_drop_extend() -> Result<()> {
-        let repo = RepoHarness::create("test_drop_extend")?.load();
+        let repo = get_test_repo("test_drop_extend")?;
 
         // Create timeline to work on
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;
 
         tline.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
         tline.advance_last_record_lsn(Lsn(0x20));
@@ -455,10 +436,11 @@ mod tests {
     // and then extended it again within the same layer.
     #[test]
     fn test_truncate_extend() -> Result<()> {
-        let repo = RepoHarness::create("test_truncate_extend")?.load();
+        let repo = get_test_repo("test_truncate_extend")?;
 
         // Create timeline to work on
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;
 
         //from storage_layer.rs
         const RELISH_SEG_SIZE: u32 = 10 * 1024 * 1024 / 8192;
@@ -555,8 +537,9 @@ mod tests {
     /// split into multiple 1 GB segments in Postgres.
     #[test]
     fn test_large_rel() -> Result<()> {
-        let repo = RepoHarness::create("test_large_rel")?.load();
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
+        let repo = get_test_repo("test_large_rel")?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;
 
         let mut lsn = 0x10;
         for blknum in 0..pg_constants::RELSEG_SIZE + 1 {
@@ -617,8 +600,9 @@ mod tests {
     ///
     #[test]
     fn test_list_rels_drop() -> Result<()> {
-        let repo = RepoHarness::create("test_list_rels_drop")?.load();
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
+        let repo = get_test_repo("test_list_rels_drop")?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;
         const TESTDB: u32 = 111;
 
         // Import initial dummy checkpoint record, otherwise the get_timeline() call
@@ -636,8 +620,9 @@ mod tests {
         assert!(tline.list_rels(0, TESTDB, Lsn(0x30))?.contains(&TESTREL_A));
 
         // Create a branch, check that the relation is visible there
-        repo.branch_timeline(TIMELINE_ID, NEW_TIMELINE_ID, Lsn(0x30))?;
-        let newtline = repo.get_timeline(NEW_TIMELINE_ID)?;
+        let newtimelineid = ZTimelineId::from_str("AA223344556677881122334455667788").unwrap();
+        repo.branch_timeline(timelineid, newtimelineid, Lsn(0x30))?;
+        let newtline = repo.get_timeline(newtimelineid)?;
 
         assert!(newtline
             .list_rels(0, TESTDB, Lsn(0x30))?
@@ -657,7 +642,7 @@ mod tests {
 
         // Run checkpoint and garbage collection and check that it's still not visible
         newtline.checkpoint()?;
-        repo.gc_iteration(Some(NEW_TIMELINE_ID), 0, true)?;
+        repo.gc_iteration(Some(newtimelineid), 0, true)?;
 
         assert!(!newtline
             .list_rels(0, TESTDB, Lsn(0x40))?
@@ -671,8 +656,9 @@ mod tests {
     ///
     #[test]
     fn test_branch() -> Result<()> {
-        let repo = RepoHarness::create("test_branch")?.load();
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
+        let repo = get_test_repo("test_branch")?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;
 
         // Import initial dummy checkpoint record, otherwise the get_timeline() call
         // after branching fails below
@@ -690,8 +676,9 @@ mod tests {
         assert_current_logical_size(&tline, Lsn(0x40));
 
         // Branch the history, modify relation differently on the new timeline
-        repo.branch_timeline(TIMELINE_ID, NEW_TIMELINE_ID, Lsn(0x30))?;
-        let newtline = repo.get_timeline(NEW_TIMELINE_ID)?;
+        let newtimelineid = ZTimelineId::from_str("AA223344556677881122334455667788").unwrap();
+        repo.branch_timeline(timelineid, newtimelineid, Lsn(0x30))?;
+        let newtline = repo.get_timeline(newtimelineid)?;
 
         newtline.put_page_image(TESTREL_A, 0, Lsn(0x40), TEST_IMG("bar blk 0 at 4"))?;
         newtline.advance_last_record_lsn(Lsn(0x40));
@@ -719,89 +706,8 @@ mod tests {
         Ok(())
     }
 
-    #[test]
-    fn corrupt_metadata() -> Result<()> {
-        const TEST_NAME: &str = "corrupt_metadata";
-        let harness = RepoHarness::create(TEST_NAME)?;
-        let repo = harness.load();
-
-        repo.create_empty_timeline(TIMELINE_ID)?;
-        drop(repo);
-
-        let metadata_path = harness.timeline_path(&TIMELINE_ID).join(METADATA_FILE_NAME);
-
-        assert!(metadata_path.is_file());
-
-        let mut metadata_bytes = std::fs::read(&metadata_path)?;
-        assert_eq!(metadata_bytes.len(), 512);
-        metadata_bytes[512 - 4 - 2] ^= 1;
-        std::fs::write(metadata_path, metadata_bytes)?;
-
-        let new_repo = harness.load();
-        let err = new_repo.get_timeline(TIMELINE_ID).err().unwrap();
-        assert!(err.to_string().contains("checksum"));
-
-        Ok(())
-    }
-
-    #[test]
-    fn future_layerfiles() -> Result<()> {
-        const TEST_NAME: &str = "future_layerfiles";
-        let harness = RepoHarness::create(TEST_NAME)?;
-        let repo = harness.load();
-
-        repo.create_empty_timeline(TIMELINE_ID)?;
-        drop(repo);
-
-        let timeline_path = harness.timeline_path(&TIMELINE_ID);
-
-        let make_empty_file = |filename: &str| -> std::io::Result<()> {
-            let path = timeline_path.join(filename);
-
-            assert!(!path.exists());
-            std::fs::write(&path, &[])?;
-
-            Ok(())
-        };
-
-        let image_filename = format!("pg_control_0_{:016X}", 8000);
-        let delta_filename = format!("pg_control_0_{:016X}_{:016X}", 8000, 8008);
-
-        make_empty_file(&image_filename)?;
-        make_empty_file(&delta_filename)?;
-
-        let new_repo = harness.load();
-        new_repo.get_timeline(TIMELINE_ID).unwrap();
-        drop(new_repo);
-
-        let check_old = |filename: &str, num: u32| {
-            let path = timeline_path.join(filename);
-            assert!(!path.exists());
-
-            let backup_path = timeline_path.join(format!("{}.{}.old", filename, num));
-            assert!(backup_path.exists());
-        };
-
-        check_old(&image_filename, 0);
-        check_old(&delta_filename, 0);
-
-        make_empty_file(&image_filename)?;
-        make_empty_file(&delta_filename)?;
-
-        let new_repo = harness.load();
-        new_repo.get_timeline(TIMELINE_ID).unwrap();
-        drop(new_repo);
-
-        check_old(&image_filename, 0);
-        check_old(&delta_filename, 0);
-        check_old(&image_filename, 1);
-        check_old(&delta_filename, 1);
-
-        Ok(())
-    }
-
     // Mock WAL redo manager that doesn't do much
-    struct TestRedoManager;
+    struct TestRedoManager {}
 
     impl WalRedoManager for TestRedoManager {
         fn request_redo(

pageserver/src/restore_local_repo.rs

@@ -11,7 +11,7 @@ use std::fs::File;
 use std::io::Read;
 use std::path::Path;
 
-use anyhow::{bail, Result};
+use anyhow::Result;
 use bytes::{Buf, Bytes};
 
 use crate::relish::*;
@@ -173,7 +173,8 @@ fn import_relfile(
                     break;
                 }
                 _ => {
-                    bail!("error reading file {}: {:#}", path.display(), e);
+                    error!("error reading file: {:?} ({})", path, e);
+                    break;
                 }
             },
         };
@@ -267,7 +268,8 @@ fn import_slru_file(timeline: &dyn Timeline, lsn: Lsn, slru: SlruKind, path: &Pa
                     break;
                 }
                 _ => {
-                    bail!("error reading file {}: {:#}", path.display(), e);
+                    error!("error reading file: {:?} ({})", path, e);
+                    break;
                 }
             },
         };

pageserver/src/tenant_mgr.rs

@@ -9,84 +9,46 @@ use crate::PageServerConf;
 use anyhow::{anyhow, bail, Context, Result};
 use lazy_static::lazy_static;
 use log::info;
-use std::collections::hash_map::Entry;
 use std::collections::HashMap;
 use std::fs;
 use std::str::FromStr;
-use std::sync::{Arc, Mutex, MutexGuard};
+use std::sync::{Arc, Mutex};
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 lazy_static! {
-    static ref REPOSITORY: Mutex<HashMap<ZTenantId, Arc<dyn Repository>>> =
+    pub static ref REPOSITORY: Mutex<HashMap<ZTenantId, Arc<dyn Repository>>> =
         Mutex::new(HashMap::new());
 }
 
-fn access_repository() -> MutexGuard<'static, HashMap<ZTenantId, Arc<dyn Repository>>> {
-    REPOSITORY.lock().unwrap()
-}
-
 pub fn init(conf: &'static PageServerConf) {
-    let mut m = access_repository();
+    let mut m = REPOSITORY.lock().unwrap();
+
     for dir_entry in fs::read_dir(conf.tenants_path()).unwrap() {
         let tenantid =
             ZTenantId::from_str(dir_entry.unwrap().file_name().to_str().unwrap()).unwrap();
-        let repo = init_repo(conf, tenantid);
+
+        // Set up a WAL redo manager, for applying WAL records.
+        let walredo_mgr = PostgresRedoManager::new(conf, tenantid);
+
+        // Set up an object repository, for actual data storage.
+        let repo = Arc::new(LayeredRepository::new(
+            conf,
+            Arc::new(walredo_mgr),
+            tenantid,
+        ));
+        LayeredRepository::launch_checkpointer_thread(conf, repo.clone());
+        LayeredRepository::launch_gc_thread(conf, repo.clone());
+
         info!("initialized storage for tenant: {}", &tenantid);
         m.insert(tenantid, repo);
     }
 }
 
-fn init_repo(conf: &'static PageServerConf, tenant_id: ZTenantId) -> Arc<LayeredRepository> {
-    // Set up a WAL redo manager, for applying WAL records.
-    let walredo_mgr = PostgresRedoManager::new(conf, tenant_id);
-
-    // Set up an object repository, for actual data storage.
-    let repo = Arc::new(LayeredRepository::new(
-        conf,
-        Arc::new(walredo_mgr),
-        tenant_id,
-        true,
-    ));
-    LayeredRepository::launch_checkpointer_thread(conf, repo.clone());
-    LayeredRepository::launch_gc_thread(conf, repo.clone());
-    repo
-}
-
-// TODO kb Currently unused function, will later be used when the relish storage downloads a new layer.
-// Relevant PR: https://github.com/zenithdb/zenith/pull/686
-pub fn register_relish_download(
-    conf: &'static PageServerConf,
-    tenant_id: ZTenantId,
-    timeline_id: ZTimelineId,
-) {
-    log::info!(
-        "Registering new download, tenant id {}, timeline id: {}",
-        tenant_id,
-        timeline_id
-    );
-    match access_repository().entry(tenant_id) {
-        Entry::Occupied(o) => init_timeline(o.get().as_ref(), timeline_id),
-        Entry::Vacant(v) => {
-            log::info!("New repo initialized");
-            let new_repo = init_repo(conf, tenant_id);
-            init_timeline(new_repo.as_ref(), timeline_id);
-            v.insert(new_repo);
-        }
-    }
-}
-
-fn init_timeline(repo: &dyn Repository, timeline_id: ZTimelineId) {
-    match repo.get_timeline(timeline_id) {
-        Ok(_timeline) => log::info!("Successfully initialized timeline {}", timeline_id),
-        Err(e) => log::error!("Failed to init timeline {}, reason: {:#}", timeline_id, e),
-    }
-}
-
 pub fn create_repository_for_tenant(
     conf: &'static PageServerConf,
     tenantid: ZTenantId,
 ) -> Result<()> {
-    let mut m = access_repository();
+    let mut m = REPOSITORY.lock().unwrap();
 
     // First check that the tenant doesn't exist already
     if m.get(&tenantid).is_some() {
@@ -100,10 +62,15 @@ pub fn create_repository_for_tenant(
     Ok(())
 }
 
+pub fn insert_repository_for_tenant(tenantid: ZTenantId, repo: Arc<dyn Repository>) {
+    let o = &mut REPOSITORY.lock().unwrap();
+    o.insert(tenantid, repo);
+}
+
 pub fn get_repository_for_tenant(tenantid: ZTenantId) -> Result<Arc<dyn Repository>> {
-    access_repository()
-        .get(&tenantid)
-        .map(Arc::clone)
+    let o = &REPOSITORY.lock().unwrap();
+    o.get(&tenantid)
+        .map(|repo| Arc::clone(repo))
         .ok_or_else(|| anyhow!("repository not found for tenant name {}", tenantid))
 }
 

pageserver/src/walreceiver.rs

@@ -10,22 +10,23 @@ use crate::restore_local_repo;
 use crate::tenant_mgr;
 use crate::waldecoder::*;
 use crate::PageServerConf;
-use anyhow::{bail, Error, Result};
+use anyhow::{Error, Result};
 use lazy_static::lazy_static;
 use log::*;
 use postgres::fallible_iterator::FallibleIterator;
 use postgres::replication::ReplicationIter;
 use postgres::{Client, NoTls, SimpleQueryMessage, SimpleQueryRow};
+use postgres_ffi::xlog_utils::*;
 use postgres_ffi::*;
 use postgres_protocol::message::backend::ReplicationMessage;
 use postgres_types::PgLsn;
-use std::cell::Cell;
+use std::cmp::{max, min};
 use std::collections::HashMap;
+use std::fs;
 use std::str::FromStr;
 use std::sync::Mutex;
 use std::thread;
 use std::thread::sleep;
-use std::thread_local;
 use std::time::{Duration, SystemTime};
 use zenith_utils::lsn::Lsn;
 use zenith_utils::zid::ZTenantId;
@@ -43,13 +44,6 @@ lazy_static! {
         Mutex::new(HashMap::new());
 }
 
-thread_local! {
-    // Boolean that is true only for WAL receiver threads
-    //
-    // This is used in `wait_lsn` to guard against usage that might lead to a deadlock.
-    pub(crate) static IS_WAL_RECEIVER: Cell<bool> = Cell::new(false);
-}
-
 // Launch a new WAL receiver, or tell one that's running about change in connection string
 pub fn launch_wal_receiver(
     conf: &'static PageServerConf,
@@ -70,10 +64,12 @@ pub fn launch_wal_receiver(
             receivers.insert(timelineid, receiver);
 
             // Also launch a new thread to handle this connection
+            //
+            // NOTE: This thread name is checked in the assertion in wait_lsn. If you change
+            // this, make sure you update the assertion too.
             let _walreceiver_thread = thread::Builder::new()
                 .name("WAL receiver thread".into())
                 .spawn(move || {
-                    IS_WAL_RECEIVER.with(|c| c.set(true));
                     thread_main(conf, timelineid, tenantid);
                 })
                 .unwrap();
@@ -122,7 +118,7 @@ fn thread_main(conf: &'static PageServerConf, timelineid: ZTimelineId, tenantid:
 }
 
 fn walreceiver_main(
-    _conf: &PageServerConf,
+    conf: &PageServerConf,
     timelineid: ZTimelineId,
     wal_producer_connstr: &str,
     tenantid: ZTenantId,
@@ -162,7 +158,7 @@ fn walreceiver_main(
     let mut startpoint = last_rec_lsn;
 
     if startpoint == Lsn(0) {
-        bail!("No previous WAL position");
+        error!("No previous WAL position");
     }
 
     // There might be some padding after the last full record, skip it.
@@ -192,6 +188,7 @@ fn walreceiver_main(
                 let data = xlog_data.data();
                 let startlsn = Lsn::from(xlog_data.wal_start());
                 let endlsn = startlsn + data.len() as u64;
+                let prev_last_rec_lsn = last_rec_lsn;
 
                 trace!("received XLogData between {} and {}", startlsn, endlsn);
 
@@ -232,6 +229,34 @@ fn walreceiver_main(
                     last_rec_lsn = lsn;
                 }
 
+                // Somewhat arbitrarily, if we have at least 10 complete wal segments (16 MB each),
+                // "checkpoint" the repository to flush all the changes from WAL we've processed
+                // so far to disk. After this, we don't need the original WAL anymore, and it
+                // can be removed. This is probably too aggressive for production, but it's useful
+                // to expose bugs now.
+                //
+                // TODO: We don't actually dare to remove the WAL. It's useful for debugging,
+                // and we might it for logical decoding other things in the future. Although
+                // we should also be able to fetch it back from the WAL safekeepers or S3 if
+                // needed.
+                if prev_last_rec_lsn.segment_number(pg_constants::WAL_SEGMENT_SIZE)
+                    != last_rec_lsn.segment_number(pg_constants::WAL_SEGMENT_SIZE)
+                {
+                    info!("switched segment {} to {}", prev_last_rec_lsn, last_rec_lsn);
+                    let (oldest_segno, newest_segno) = find_wal_file_range(
+                        conf,
+                        &timelineid,
+                        pg_constants::WAL_SEGMENT_SIZE,
+                        last_rec_lsn,
+                        &tenantid,
+                    )?;
+
+                    if newest_segno - oldest_segno >= 10 {
+                        // TODO: This is where we could remove WAL older than last_rec_lsn.
+                        //remove_wal_files(timelineid, pg_constants::WAL_SEGMENT_SIZE, last_rec_lsn)?;
+                    }
+                }
+
                 if !caught_up && endlsn >= end_of_wal {
                     info!("caught up at LSN {}", endlsn);
                     caught_up = true;
@@ -253,7 +278,7 @@ fn walreceiver_main(
                 );
 
                 if reply_requested {
-                    Some(last_rec_lsn)
+                    Some(timeline.get_last_record_lsn())
                 } else {
                     None
                 }
@@ -277,6 +302,47 @@ fn walreceiver_main(
     Ok(())
 }
 
+fn find_wal_file_range(
+    conf: &PageServerConf,
+    timeline: &ZTimelineId,
+    wal_seg_size: usize,
+    written_upto: Lsn,
+    tenant: &ZTenantId,
+) -> Result<(u64, u64)> {
+    let written_upto_segno = written_upto.segment_number(wal_seg_size);
+
+    let mut oldest_segno = written_upto_segno;
+    let mut newest_segno = written_upto_segno;
+    // Scan the wal directory, and count how many WAL filed we could remove
+    let wal_dir = conf.wal_dir_path(timeline, tenant);
+    for entry in fs::read_dir(wal_dir)? {
+        let entry = entry?;
+        let path = entry.path();
+
+        if path.is_dir() {
+            continue;
+        }
+
+        let filename = path.file_name().unwrap().to_str().unwrap();
+
+        if IsXLogFileName(filename) {
+            let (segno, _tli) = XLogFromFileName(filename, wal_seg_size);
+
+            if segno > written_upto_segno {
+                // that's strange.
+                warn!("there is a WAL file from future at {}", path.display());
+                continue;
+            }
+
+            oldest_segno = min(oldest_segno, segno);
+            newest_segno = max(newest_segno, segno);
+        }
+    }
+    // FIXME: would be good to assert that there are no gaps in the WAL files
+
+    Ok((oldest_segno, newest_segno))
+}
+
 /// Data returned from the postgres `IDENTIFY_SYSTEM` command
 ///
 /// See the [postgres docs] for more details.

pageserver/src/walredo.rs

@@ -298,11 +298,9 @@ impl PostgresRedoManager {
                     // Transaction manager stuff
                     let rec_segno = match rel {
                         RelishTag::Slru { slru, segno } => {
-                            assert!(
-                                slru == SlruKind::Clog,
-                                "Not valid XACT relish tag {:?}",
-                                rel
-                            );
+                            if slru != SlruKind::Clog {
+                                panic!("Not valid XACT relish tag {:?}", rel);
+                            }
                             segno
                         }
                         _ => panic!("Not valid XACT relish tag {:?}", rel),
@@ -422,7 +420,7 @@ impl PostgresRedoManager {
         );
 
         if let Err(e) = apply_result {
-            error!("could not apply WAL records: {:#}", e);
+            error!("could not apply WAL records: {}", e);
             result = Err(WalRedoError::IoError(e));
         } else {
             let img = apply_result.unwrap();
@@ -460,7 +458,7 @@ impl PostgresRedoProcess {
         if datadir.exists() {
             info!("directory {:?} exists, removing", &datadir);
             if let Err(e) = fs::remove_dir_all(&datadir) {
-                error!("could not remove old wal-redo-datadir: {:#}", e);
+                error!("could not remove old wal-redo-datadir: {:?}", e);
             }
         }
         info!("running initdb in {:?}", datadir.display());

proxy/src/cplane_api.rs

@@ -1,6 +1,6 @@
-use anyhow::{bail, Context, Result};
+use anyhow::{bail, Result};
 use serde::{Deserialize, Serialize};
-use std::net::{SocketAddr, ToSocketAddrs};
+use std::net::{IpAddr, SocketAddr};
 
 pub struct CPlaneApi {
     auth_endpoint: &'static str,
@@ -8,7 +8,7 @@ pub struct CPlaneApi {
 
 #[derive(Serialize, Deserialize, Debug)]
 pub struct DatabaseInfo {
-    pub host: String,
+    pub host: IpAddr, // TODO: allow host name here too
     pub port: u16,
     pub dbname: String,
     pub user: String,
@@ -16,13 +16,8 @@ pub struct DatabaseInfo {
 }
 
 impl DatabaseInfo {
-    pub fn socket_addr(&self) -> Result<SocketAddr> {
-        let host_port = format!("{}:{}", self.host, self.port);
-        host_port
-            .to_socket_addrs()
-            .with_context(|| format!("cannot resolve {} to SocketAddr", host_port))?
-            .next()
-            .ok_or_else(|| anyhow::Error::msg("cannot resolve at least one SocketAddr"))
+    pub fn socket_addr(&self) -> SocketAddr {
+        SocketAddr::new(self.host, self.port)
     }
 
     pub fn conn_string(&self) -> String {
@@ -41,14 +36,12 @@ impl CPlaneApi {
     pub fn authenticate_proxy_request(
         &self,
         user: &str,
-        database: &str,
         md5_response: &[u8],
         salt: &[u8; 4],
     ) -> Result<DatabaseInfo> {
         let mut url = reqwest::Url::parse(self.auth_endpoint)?;
         url.query_pairs_mut()
             .append_pair("login", user)
-            .append_pair("database", database)
             .append_pair("md5response", std::str::from_utf8(md5_response)?)
             .append_pair("salt", &hex::encode(salt));
 

proxy/src/proxy.rs

@@ -176,7 +176,6 @@ impl ProxyConnection {
 
             match self.cplane.authenticate_proxy_request(
                 self.user.as_str(),
-                self.database.as_str(),
                 md5_response,
                 &self.md5_salt,
             ) {
@@ -248,7 +247,7 @@ databases without opening the browser.
 
 /// Create a TCP connection to a postgres database, authenticate with it, and receive the ReadyForQuery message
 async fn connect_to_db(db_info: DatabaseInfo) -> anyhow::Result<tokio::net::TcpStream> {
-    let mut socket = tokio::net::TcpStream::connect(db_info.socket_addr()?).await?;
+    let mut socket = tokio::net::TcpStream::connect(db_info.socket_addr()).await?;
     let config = db_info.conn_string().parse::<tokio_postgres::Config>()?;
     let _ = config.connect_raw(&mut socket, NoTls).await?;
     Ok(socket)

test_runner/batch_others/test_branch_behind.py

@@ -1,6 +1,9 @@
 import subprocess
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
@@ -13,7 +16,7 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     zenith_cli.run(["branch", "test_branch_behind", "empty"])
 
     pgmain = postgres.create_start('test_branch_behind')
-    print("postgres is running on 'test_branch_behind' branch")
+    log.info("postgres is running on 'test_branch_behind' branch")
 
     main_pg_conn = pgmain.connect()
     main_cur = main_pg_conn.cursor()
@@ -27,17 +30,17 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     ''')
     main_cur.execute('SELECT pg_current_wal_insert_lsn()')
     lsn_a = main_cur.fetchone()[0]
-    print('LSN after 100 rows: ' + lsn_a)
+    log.info(f'LSN after 100 rows: {lsn_a}')
 
     # Insert some more rows. (This generates enough WAL to fill a few segments.)
     main_cur.execute('''
         INSERT INTO foo
             SELECT 'long string to consume some space' || g
-            FROM generate_series(1, 200000) g
+            FROM generate_series(1, 100000) g
     ''')
     main_cur.execute('SELECT pg_current_wal_insert_lsn()')
     lsn_b = main_cur.fetchone()[0]
-    print('LSN after 200100 rows: ' + lsn_b)
+    log.info(f'LSN after 100100 rows: {lsn_b}')
 
     # Branch at the point where only 100 rows were inserted
     zenith_cli.run(["branch", "test_branch_behind_hundred", "test_branch_behind@" + lsn_a])
@@ -46,15 +49,15 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     main_cur.execute('''
         INSERT INTO foo
             SELECT 'long string to consume some space' || g
-            FROM generate_series(1, 200000) g
+            FROM generate_series(1, 100000) g
     ''')
     main_cur.execute('SELECT pg_current_wal_insert_lsn()')
 
     main_cur.execute('SELECT pg_current_wal_insert_lsn()')
     lsn_c = main_cur.fetchone()[0]
-    print('LSN after 400100 rows: ' + lsn_c)
+    log.info(f'LSN after 200100 rows: {lsn_c}')
 
-    # Branch at the point where only 200100 rows were inserted
+    # Branch at the point where only 200 rows were inserted
     zenith_cli.run(["branch", "test_branch_behind_more", "test_branch_behind@" + lsn_b])
 
     pg_hundred = postgres.create_start("test_branch_behind_hundred")
@@ -70,11 +73,11 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     more_pg_conn = pg_more.connect()
     more_cur = more_pg_conn.cursor()
     more_cur.execute('SELECT count(*) FROM foo')
-    assert more_cur.fetchone() == (200100, )
+    assert more_cur.fetchone() == (100100, )
 
     # All the rows are visible on the main branch
     main_cur.execute('SELECT count(*) FROM foo')
-    assert main_cur.fetchone() == (400100, )
+    assert main_cur.fetchone() == (200100, )
 
     # Check bad lsn's for branching
 
@@ -89,4 +92,4 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     try:
         zenith_cli.run(["branch", "test_branch_preinitdb", "test_branch_behind@0/42"])
     except subprocess.CalledProcessError:
-        print("Branch creation with pre-initdb LSN failed (as expected)")
+        log.info("Branch creation with pre-initdb LSN failed (as expected)")

test_runner/batch_others/test_clog_truncate.py

@@ -5,6 +5,10 @@ from contextlib import closing
 
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -24,7 +28,7 @@ def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     ]
 
     pg = postgres.create_start('test_clog_truncate', config_lines=config)
-    print('postgres is running on test_clog_truncate branch')
+    log.info('postgres is running on test_clog_truncate branch')
 
     # Install extension containing function needed for test
     pg.safe_psql('CREATE EXTENSION zenith_test_utils')
@@ -33,22 +37,22 @@ def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
             cur.execute('select test_consume_xids(1000*1000*10);')
-            print('xids consumed')
+            log.info('xids consumed')
 
             # call a checkpoint to trigger TruncateSubtrans
             cur.execute('CHECKPOINT;')
 
             # ensure WAL flush
             cur.execute('select txid_current()')
-            print(cur.fetchone())
+            log.info(cur.fetchone())
 
     # wait for autovacuum to truncate the pg_xact
     # XXX Is it worth to add a timeout here?
     pg_xact_0000_path = os.path.join(pg.pg_xact_dir_path(), '0000')
-    print("pg_xact_0000_path = " + pg_xact_0000_path)
+    log.info(f"pg_xact_0000_path = {pg_xact_0000_path}")
 
     while os.path.isfile(pg_xact_0000_path):
-        print("file exists. wait for truncation. " "pg_xact_0000_path = " + pg_xact_0000_path)
+        log.info(f"file exists. wait for truncation. " "pg_xact_0000_path = {pg_xact_0000_path}")
         time.sleep(5)
 
     # checkpoint to advance latest lsn
@@ -59,14 +63,14 @@ def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
             lsn_after_truncation = cur.fetchone()[0]
 
     # create new branch after clog truncation and start a compute node on it
-    print('create branch at lsn_after_truncation ' + lsn_after_truncation)
+    log.info(f'create branch at lsn_after_truncation {lsn_after_truncation}')
     zenith_cli.run(
         ["branch", "test_clog_truncate_new", "test_clog_truncate@" + lsn_after_truncation])
 
     pg2 = postgres.create_start('test_clog_truncate_new')
-    print('postgres is running on test_clog_truncate_new branch')
+    log.info('postgres is running on test_clog_truncate_new branch')
 
     # check that new node doesn't contain truncated segment
     pg_xact_0000_path_new = os.path.join(pg2.pg_xact_dir_path(), '0000')
-    print("pg_xact_0000_path_new = " + pg_xact_0000_path_new)
+    log.info("pg_xact_0000_path_new = " + pg_xact_0000_path_new)
     assert os.path.isfile(pg_xact_0000_path_new) is False

test_runner/batch_others/test_config.py

@@ -2,6 +2,10 @@ from contextlib import closing
 
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -14,7 +18,7 @@ def test_config(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFact
 
     # change config
     pg = postgres.create_start('test_config', config_lines=['log_min_messages=debug1'])
-    print('postgres is running on test_config branch')
+    log.info('postgres is running on test_config branch')
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:

test_runner/batch_others/test_createdropdb.py

@@ -4,6 +4,10 @@ import pathlib
 from contextlib import closing
 from fixtures.zenith_fixtures import ZenithPageserver, PostgresFactory, ZenithCli, check_restored_datadir_content
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -19,7 +23,7 @@ def test_createdb(
     zenith_cli.run(["branch", "test_createdb", "empty"])
 
     pg = postgres.create_start('test_createdb')
-    print("postgres is running on 'test_createdb' branch")
+    log.info("postgres is running on 'test_createdb' branch")
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
@@ -53,7 +57,7 @@ def test_dropdb(
     zenith_cli.run(["branch", "test_dropdb", "empty"])
 
     pg = postgres.create_start('test_dropdb')
-    print("postgres is running on 'test_dropdb' branch")
+    log.info("postgres is running on 'test_dropdb' branch")
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
@@ -88,13 +92,13 @@ def test_dropdb(
 
     # Test that database subdir exists on the branch before drop
     dbpath = pathlib.Path(pg_before.pgdata_dir) / 'base' / str(dboid)
-    print(dbpath)
+    log.info(dbpath)
 
     assert os.path.isdir(dbpath) == True
 
     # Test that database subdir doesn't exist on the branch after drop
     dbpath = pathlib.Path(pg_after.pgdata_dir) / 'base' / str(dboid)
-    print(dbpath)
+    log.info(dbpath)
 
     assert os.path.isdir(dbpath) == False
 

test_runner/batch_others/test_createuser.py

@@ -2,6 +2,10 @@ from contextlib import closing
 
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -12,7 +16,7 @@ def test_createuser(zenith_cli, pageserver: ZenithPageserver, postgres: Postgres
     zenith_cli.run(["branch", "test_createuser", "empty"])
 
     pg = postgres.create_start('test_createuser')
-    print("postgres is running on 'test_createuser' branch")
+    log.info("postgres is running on 'test_createuser' branch")
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:

test_runner/batch_others/test_multixact.py

@@ -1,5 +1,9 @@
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver, check_restored_datadir_content
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -15,7 +19,7 @@ def test_multixact(pageserver: ZenithPageserver, postgres: PostgresFactory,
     zenith_cli.run(["branch", "test_multixact", "empty"])
     pg = postgres.create_start('test_multixact')
 
-    print("postgres is running on 'test_multixact' branch")
+    log.info("postgres is running on 'test_multixact' branch")
     pg_conn = pg.connect()
     cur = pg_conn.cursor()
 
@@ -55,7 +59,7 @@ def test_multixact(pageserver: ZenithPageserver, postgres: PostgresFactory,
     zenith_cli.run(["branch", "test_multixact_new", "test_multixact@" + lsn])
     pg_new = postgres.create_start('test_multixact_new')
 
-    print("postgres is running on 'test_multixact_new' branch")
+    log.info("postgres is running on 'test_multixact_new' branch")
     pg_new_conn = pg_new.connect()
     cur_new = pg_new_conn.cursor()
 

test_runner/batch_others/test_old_request_lsn.py

@@ -2,6 +2,10 @@ from contextlib import closing
 
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 #
@@ -18,7 +22,7 @@ def test_old_request_lsn(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
     # Create a branch for us
     zenith_cli.run(["branch", "test_old_request_lsn", "empty"])
     pg = postgres.create_start('test_old_request_lsn')
-    print('postgres is running on test_old_request_lsn branch')
+    log.info('postgres is running on test_old_request_lsn branch')
 
     pg_conn = pg.connect()
     cur = pg_conn.cursor()
@@ -46,7 +50,7 @@ def test_old_request_lsn(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
         from pg_settings where name = 'shared_buffers'
     ''')
     row = cur.fetchone()
-    print(f'shared_buffers is {row[0]}, table size {row[1]}');
+    log.info(f'shared_buffers is {row[0]}, table size {row[1]}');
     assert int(row[0]) < int(row[1])
 
     cur.execute('VACUUM foo');

test_runner/batch_others/test_pageserver_restart.py

@@ -6,6 +6,10 @@ from contextlib import closing
 from multiprocessing import Process, Value
 from fixtures.zenith_fixtures import WalAcceptorFactory, ZenithPageserver, PostgresFactory
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 # Check that dead minority doesn't prevent the commits: execute insert n_inserts
@@ -40,7 +44,7 @@ def test_pageserver_restart(zenith_cli, pageserver: ZenithPageserver, postgres:
         from pg_settings where name = 'shared_buffers'
     ''')
     row = cur.fetchone()
-    print("shared_buffers is {}, table size {}", row[0], row[1]);
+    log.info(f"shared_buffers is {row[0]}, table size {row[1]}");
     assert int(row[0]) < int(row[1])
 
     # Stop and restart pageserver. This is a more or less graceful shutdown, although

test_runner/batch_others/test_pgbench.py

@@ -1,5 +1,9 @@
 from fixtures.zenith_fixtures import PostgresFactory
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -8,7 +12,7 @@ def test_pgbench(postgres: PostgresFactory, pg_bin, zenith_cli):
     zenith_cli.run(["branch", "test_pgbench", "empty"])
 
     pg = postgres.create_start('test_pgbench')
-    print("postgres is running on 'test_pgbench' branch")
+    log.info("postgres is running on 'test_pgbench' branch")
 
     connstr = pg.connstr()
 

test_runner/batch_others/test_restart_compute.py

@@ -3,13 +3,20 @@ import pytest
 from contextlib import closing
 from fixtures.zenith_fixtures import ZenithPageserver, PostgresFactory
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
 #
 # Test restarting and recreating a postgres instance
 #
-@pytest.mark.parametrize('with_wal_acceptors', [False, True])
+# XXX: with_wal_acceptors=True fails now, would be fixed with
+# `postgres --sync-walkeepers` patches.
+#
+@pytest.mark.parametrize('with_wal_acceptors', [False])
 def test_restart_compute(
         zenith_cli,
         pageserver: ZenithPageserver,
@@ -27,7 +34,7 @@ def test_restart_compute(
 
     pg = postgres.create_start('test_restart_compute',
                                wal_acceptors=wal_acceptor_connstrs)
-    print("postgres is running on 'test_restart_compute' branch")
+    log.info("postgres is running on 'test_restart_compute' branch")
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
@@ -36,7 +43,7 @@ def test_restart_compute(
             cur.execute('SELECT sum(key) FROM t')
             r = cur.fetchone()
             assert r == (5000050000, )
-            print("res = ", r)
+            log.info(f"res = {r}")
 
     # Remove data directory and restart
     pg.stop_and_destroy().create_start('test_restart_compute',
@@ -49,7 +56,7 @@ def test_restart_compute(
             cur.execute('SELECT sum(key) FROM t')
             r = cur.fetchone()
             assert r == (5000050000, )
-            print("res = ", r)
+            log.info(f"res = {r}")
 
             # Insert another row
             cur.execute("INSERT INTO t VALUES (100001, 'payload2')")
@@ -57,7 +64,7 @@ def test_restart_compute(
 
             r = cur.fetchone()
             assert r == (100001, )
-            print("res = ", r)
+            log.info(f"res = {r}")
 
     # Again remove data directory and restart
     pg.stop_and_destroy().create_start('test_restart_compute',
@@ -72,7 +79,7 @@ def test_restart_compute(
 
             r = cur.fetchone()
             assert r == (100001, )
-            print("res = ", r)
+            log.info(f"res = {r}")
 
     # And again remove data directory and restart
     pg.stop_and_destroy().create_start('test_restart_compute',
@@ -85,4 +92,4 @@ def test_restart_compute(
 
             r = cur.fetchone()
             assert r == (100001, )
-            print("res = ", r)
+            log.info(f"res = {r}")

test_runner/batch_others/test_snapfiles_gc.py

@@ -2,12 +2,16 @@ from contextlib import closing
 import psycopg2.extras
 import time;
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 def print_gc_result(row):
-    print("GC duration {elapsed} ms".format_map(row));
-    print("  REL    total: {layer_relfiles_total}, needed_by_cutoff {layer_relfiles_needed_by_cutoff}, needed_by_branches: {layer_relfiles_needed_by_branches}, not_updated: {layer_relfiles_not_updated}, needed_as_tombstone {layer_relfiles_needed_as_tombstone}, removed: {layer_relfiles_removed}, dropped: {layer_relfiles_dropped}".format_map(row))
-    print("  NONREL total: {layer_nonrelfiles_total}, needed_by_cutoff {layer_nonrelfiles_needed_by_cutoff}, needed_by_branches: {layer_nonrelfiles_needed_by_branches}, not_updated: {layer_nonrelfiles_not_updated}, needed_as_tombstone {layer_nonrelfiles_needed_as_tombstone}, removed: {layer_nonrelfiles_removed}, dropped: {layer_nonrelfiles_dropped}".format_map(row))
+    log.info("GC duration {elapsed} ms".format_map(row));
+    log.info("  REL    total: {layer_relfiles_total}, needed_by_cutoff {layer_relfiles_needed_by_cutoff}, needed_by_branches: {layer_relfiles_needed_by_branches}, not_updated: {layer_relfiles_not_updated}, needed_as_tombstone {layer_relfiles_needed_as_tombstone}, removed: {layer_relfiles_removed}, dropped: {layer_relfiles_dropped}".format_map(row))
+    log.info("  NONREL total: {layer_nonrelfiles_total}, needed_by_cutoff {layer_nonrelfiles_needed_by_cutoff}, needed_by_branches: {layer_nonrelfiles_needed_by_branches}, not_updated: {layer_nonrelfiles_not_updated}, needed_as_tombstone {layer_nonrelfiles_needed_as_tombstone}, removed: {layer_nonrelfiles_removed}, dropped: {layer_nonrelfiles_dropped}".format_map(row))
 
 
 #
@@ -35,7 +39,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
 
                     cur.execute("select relfilenode from pg_class where oid = 'foo'::regclass");
                     row = cur.fetchone();
-                    print("relfilenode is {}", row[0]);
+                    log.info(f"relfilenode is {row[0]}");
 
                     # Run GC, to clear out any garbage left behind in the catalogs by
                     # the CREATE TABLE command. We want to have a clean slate with no garbage
@@ -50,7 +54,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     # update to confuse our numbers either.
                     cur.execute("DELETE FROM foo")
 
-                    print("Running GC before test")
+                    log.info("Running GC before test")
                     pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
                     print_gc_result(row);
@@ -61,7 +65,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     # Insert a row and run GC. Checkpoint should freeze the layer
                     # so that there is only the most recent image layer left for the rel,
                     # removing the old image and delta layer.
-                    print("Inserting one row and running GC")
+                    log.info("Inserting one row and running GC")
                     cur.execute("INSERT INTO foo VALUES (1)")
                     pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
@@ -73,7 +77,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     # Insert two more rows and run GC.
                     # This should create new image and delta layer file with the new contents, and
                     # then remove the old one image and the just-created delta layer.
-                    print("Inserting two more rows and running GC")
+                    log.info("Inserting two more rows and running GC")
                     cur.execute("INSERT INTO foo VALUES (2)")
                     cur.execute("INSERT INTO foo VALUES (3)")
 
@@ -85,7 +89,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     assert row['layer_relfiles_dropped'] == 0
 
                     # Do it again. Should again create two new layer files and remove old ones.
-                    print("Inserting two more rows and running GC")
+                    log.info("Inserting two more rows and running GC")
                     cur.execute("INSERT INTO foo VALUES (2)")
                     cur.execute("INSERT INTO foo VALUES (3)")
 
@@ -97,7 +101,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     assert row['layer_relfiles_dropped'] == 0
 
                     # Run GC again, with no changes in the database. Should not remove anything.
-                    print("Run GC again, with nothing to do")
+                    log.info("Run GC again, with nothing to do")
                     pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
                     print_gc_result(row);
@@ -108,7 +112,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     #
                     # Test DROP TABLE checks that relation data and metadata was deleted by GC from object storage
                     #
-                    print("Drop table and run GC again");
+                    log.info("Drop table and run GC again");
                     cur.execute("DROP TABLE foo")
 
                     pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")

test_runner/batch_others/test_timeline_size.py

@@ -3,6 +3,9 @@ from uuid import UUID
 import psycopg2.extras
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
 
 def test_timeline_size(
     zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin
@@ -15,7 +18,7 @@ def test_timeline_size(
     assert res["current_logical_size"] == res["current_logical_size_non_incremental"]
 
     pgmain = postgres.create_start("test_timeline_size")
-    print("postgres is running on 'test_timeline_size' branch")
+    log.info("postgres is running on 'test_timeline_size' branch")
 
     with closing(pgmain.connect()) as conn:
         with conn.cursor() as cur:

test_runner/batch_others/test_twophase.py

@@ -2,6 +2,9 @@ import os
 
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver, PgBin
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
@@ -13,7 +16,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
     zenith_cli.run(["branch", "test_twophase", "empty"])
 
     pg = postgres.create_start('test_twophase', config_lines=['max_prepared_transactions=5'])
-    print("postgres is running on 'test_twophase' branch")
+    log.info("postgres is running on 'test_twophase' branch")
 
     conn = pg.connect()
     cur = conn.cursor()
@@ -45,7 +48,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
     cur.execute('CHECKPOINT')
 
     twophase_files = os.listdir(pg.pg_twophase_dir_path())
-    print(twophase_files)
+    log.info(twophase_files)
     assert len(twophase_files) == 4
 
     cur.execute("COMMIT PREPARED 'insert_three'")
@@ -53,7 +56,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
     cur.execute('CHECKPOINT')
 
     twophase_files = os.listdir(pg.pg_twophase_dir_path())
-    print(twophase_files)
+    log.info(twophase_files)
     assert len(twophase_files) == 2
 
     # Create a branch with the transaction in prepared state
@@ -67,7 +70,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
 
     # Check that we restored only needed twophase files
     twophase_files2 = os.listdir(pg2.pg_twophase_dir_path())
-    print(twophase_files2)
+    log.info(twophase_files2)
     assert twophase_files2.sort() == twophase_files.sort()
 
     conn2 = pg2.connect()

test_runner/batch_others/test_vm_bits.py

@@ -1,5 +1,9 @@
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 #
@@ -11,7 +15,7 @@ def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, p
     zenith_cli.run(["branch", "test_vm_bit_clear", "empty"])
     pg = postgres.create_start('test_vm_bit_clear')
 
-    print("postgres is running on 'test_vm_bit_clear' branch")
+    log.info("postgres is running on 'test_vm_bit_clear' branch")
     pg_conn = pg.connect()
     cur = pg_conn.cursor()
 
@@ -63,7 +67,7 @@ def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, p
     # server at the right point-in-time avoids that full-page image.
     pg_new = postgres.create_start('test_vm_bit_clear_new')
 
-    print("postgres is running on 'test_vm_bit_clear_new' branch")
+    log.info("postgres is running on 'test_vm_bit_clear_new' branch")
     pg_new_conn = pg_new.connect()
     cur_new = pg_new_conn.cursor()
 

test_runner/batch_others/test_wal_acceptor.py

@@ -10,6 +10,10 @@ from multiprocessing import Process, Value
 from fixtures.zenith_fixtures import WalAcceptorFactory, ZenithPageserver, PostgresFactory, PgBin
 from fixtures.utils import lsn_to_hex, mkdir_if_needed
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -284,10 +288,10 @@ def test_sync_safekeepers(repo_dir: str, pg_bin: PgBin, wa_factory: WalAcceptorF
         )
         lsn_hex = lsn_to_hex(res["inserted_wal"]["end_lsn"])
         lsn_after_append.append(lsn_hex)
-        print(f"safekeeper[{i}] lsn after append: {lsn_hex}")
+        log.info(f"safekeeper[{i}] lsn after append: {lsn_hex}")
 
     # run sync safekeepers
     lsn_after_sync = pg.sync_safekeepers()
-    print(f"lsn after sync = {lsn_after_sync}")
+    log.info(f"lsn after sync = {lsn_after_sync}")
 
     assert all(lsn_after_sync == lsn for lsn in lsn_after_append)

test_runner/batch_others/test_wal_acceptor_async.py

@@ -4,7 +4,10 @@ import random
 
 from fixtures.zenith_fixtures import WalAcceptor, WalAcceptorFactory, ZenithPageserver, PostgresFactory, Postgres
 from typing import List
-from fixtures.utils import debug_print
+
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root.wal_acceptor_async')
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
@@ -63,18 +66,18 @@ class WorkerStats(object):
         self.counters[worker_id] += 1
 
     def check_progress(self):
-        debug_print("Workers progress: {}".format(self.counters))
+        log.debug("Workers progress: {}".format(self.counters))
 
         # every worker should finish at least one tx
         assert all(cnt > 0 for cnt in self.counters)
 
         progress = sum(self.counters)
-        print('All workers made {} transactions'.format(progress))
+        log.info('All workers made {} transactions'.format(progress))
 
 
 async def run_random_worker(stats: WorkerStats, pg: Postgres, worker_id, n_accounts, max_transfer):
     pg_conn = await pg.connect_async()
-    debug_print('Started worker {}'.format(worker_id))
+    log.debug('Started worker {}'.format(worker_id))
 
     while stats.running:
         from_uid = random.randint(0, n_accounts - 1)
@@ -84,9 +87,9 @@ async def run_random_worker(stats: WorkerStats, pg: Postgres, worker_id, n_accou
         await bank_transfer(pg_conn, from_uid, to_uid, amount)
         stats.inc_progress(worker_id)
 
-        debug_print('Executed transfer({}) {} => {}'.format(amount, from_uid, to_uid))
+        log.debug('Executed transfer({}) {} => {}'.format(amount, from_uid, to_uid))
 
-    debug_print('Finished worker {}'.format(worker_id))
+    log.debug('Finished worker {}'.format(worker_id))
 
     await pg_conn.close()
 
@@ -134,7 +137,7 @@ async def run_restarts_under_load(pg: Postgres, acceptors: List[WalAcceptor], n_
 
         victim.start()
 
-    print('Iterations are finished, exiting coroutines...')
+    log.info('Iterations are finished, exiting coroutines...')
     stats.running = False
     # await all workers
     await asyncio.gather(*workers)

test_runner/batch_pg_regress/test_zenith_regress.py

@@ -3,6 +3,10 @@ import os
 from fixtures.utils import mkdir_if_needed
 from fixtures.zenith_fixtures import PageserverPort, PostgresFactory, check_restored_datadir_content
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
@@ -38,7 +42,7 @@ def test_zenith_regress(postgres: PostgresFactory, pg_bin, zenith_cli, test_outp
         '--inputdir={}'.format(src_path),
     ]
 
-    print(pg_regress_command)
+    log.info(pg_regress_command)
     env = {
         'PGPORT': str(pg.port),
         'PGUSER': pg.username,

test_runner/fixtures/benchmark_fixture.py

@@ -1,5 +1,3 @@
-from pprint import pprint
-
 import os
 import re
 import timeit
@@ -136,34 +134,9 @@ class ZenithBenchmarker:
         # The metric should be an integer, as it's a number of bytes. But in general
         # all prometheus metrics are floats. So to be pedantic, read it as a float
         # and round to integer.
-        matches = re.search(r'^pageserver_disk_io_bytes{io_operation="write"} (\S+)$', all_metrics,
-                            re.MULTILINE)
+        matches = re.search(r'pageserver_disk_io_bytes{io_operation="write"} (\S+)', all_metrics)
         return int(round(float(matches.group(1))))
 
-    def get_peak_mem(self, pageserver) -> int:
-        """
-        Fetch the "maxrss" metric from the pageserver
-        """
-        # Fetch all the exposed prometheus metrics from page server
-        all_metrics = pageserver.http_client().get_metrics()
-        # See comment in get_io_writes()
-        matches = re.search(r'^pageserver_maxrss_kb (\S+)$', all_metrics,
-                            re.MULTILINE)
-        return int(round(float(matches.group(1))))
-
-    def get_timeline_size(self, repo_dir: str, tenantid: str, timelineid: str):
-        """
-        Calculate the on-disk size of a timeline
-        """
-        path = "{}/tenants/{}/timelines/{}".format(repo_dir, tenantid, timelineid)
-
-        totalbytes = 0
-        for root, dirs, files in os.walk(path):
-            for name in files:
-                totalbytes += os.path.getsize(os.path.join(root, name))
-
-        return totalbytes
-
     @contextmanager
     def record_pageserver_writes(self, pageserver, metric_name):
         """

test_runner/fixtures/log_helper.py

@@ -0,0 +1,37 @@
+import logging
+import logging.config
+import time
+
+# timestamp in UTC+-00:00 aka GMT
+class UTCFormatter(logging.Formatter):
+    converter = time.gmtime
+
+LOGGING = {
+    "version": 1,
+    "formatters": {
+    "defaultFormatter": {
+        "()": UTCFormatter,
+        "format": "%(asctime)s.%(msecs)-3d %(levelname)s [%(filename)s:%(lineno)d] %(message)s",
+        "datefmt": "%Y-%m-%d %H:%M:%S"
+    }
+    },
+    "handlers": {
+        "console": {
+            "class": "logging.StreamHandler",
+            "formatter": "defaultFormatter",
+            "level": "DEBUG",
+            "stream": "ext://sys.stderr"
+        }
+    },
+    "loggers": {
+        "root": {
+            "level": "DEBUG",
+            "handlers": ["console"]
+        },
+        "root.wal_acceptor_async": { 
+            "level": "INFO" # lot of logs on DEBUG level
+        }
+    }
+}
+
+logging.config.dictConfig(LOGGING)

test_runner/fixtures/utils.py

@@ -3,6 +3,9 @@ import subprocess
 
 from typing import Any, List
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
 
 def get_self_dir() -> str:
     """ Get the path to the directory where this script lives. """
@@ -39,7 +42,7 @@ def subprocess_capture(capture_dir: str, cmd: List[str], **kwargs: Any) -> str:
 
     with open(stdout_filename, 'w') as stdout_f:
         with open(stderr_filename, 'w') as stderr_f:
-            print('(capturing output to "{}.stdout")'.format(base))
+            log.info('(capturing output to "{}.stdout")'.format(base))
             subprocess.run(cmd, **kwargs, stdout=stdout_f, stderr=stderr_f)
 
     return basepath
@@ -58,14 +61,6 @@ def global_counter() -> int:
     _global_counter += 1
     return _global_counter
 
-def debug_print(*args, **kwargs) -> None:
-    """ Print to the console if TEST_DEBUG_PRINT is set in env.
-    
-    All parameters are passed to print().
-    """
-    if os.environ.get('TEST_DEBUG_PRINT') is not None:
-        print(*args, **kwargs)
-
 def lsn_to_hex(num: int) -> str:
     """ Convert lsn from int to standard hex notation. """
     return "{:X}/{:X}".format(num >> 32, num & 0xffffffff)

test_runner/fixtures/zenith_fixtures.py

@@ -27,6 +27,11 @@ from typing_extensions import Literal
 import requests
 
 from .utils import (get_self_dir, mkdir_if_needed, subprocess_capture)
+
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 """
 This file contains pytest fixtures. A fixture is a test resource that can be
 summoned by placing its name in the test's arguments.
@@ -71,9 +76,7 @@ def pytest_configure(config):
         # This is bad; we don't want any of those processes polluting the
         # result of the test.
         # NOTE this shows as an internal pytest error, there might be a better way
-        raise Exception(
-            'Found interfering processes running. Stop all Zenith pageservers, nodes, WALs, as well as stand-alone Postgres.'
-        )
+        raise Exception('found interfering processes running')
 
 
 def determine_scope(fixture_name: str, config: Any) -> str:
@@ -188,13 +191,13 @@ class ZenithCli:
 
         >>> result = zenith_cli.run(...)
         >>> assert result.stderr == ""
-        >>> print(result.stdout)
+        >>> log.info(result.stdout)
         """
 
         assert type(arguments) == list
 
         args = [self.bin_zenith] + arguments
-        print('Running command "{}"'.format(' '.join(args)))
+        log.info('Running command "{}"'.format(' '.join(args)))
 
         # Interceipt CalledProcessError and print more info
         try:
@@ -211,7 +214,7 @@ class ZenithCli:
               stdout: {exc.stdout}
               stderr: {exc.stderr}
             """
-            print(msg)
+            log.info(msg)
 
             raise Exception(msg) from exc
 
@@ -419,7 +422,7 @@ class ZenithPageserver(PgProtocol):
 def pageserver_port(port_distributor: PortDistributor) -> PageserverPort:
     pg = port_distributor.get_port()
     http = port_distributor.get_port()
-    print(f"pageserver_port: pg={pg} http={http}")
+    log.info(f"pageserver_port: pg={pg} http={http}")
     return PageserverPort(pg=pg, http=http)
 
 
@@ -443,7 +446,7 @@ def pageserver(zenith_cli: ZenithCli, repo_dir: str, pageserver_port: Pageserver
     yield ps
 
     # After the yield comes any cleanup code we need.
-    print('Starting pageserver cleanup')
+    log.info('Starting pageserver cleanup')
     ps.stop()
 
 class PgBin:
@@ -481,7 +484,7 @@ class PgBin:
         """
 
         self._fixpath(command)
-        print('Running command "{}"'.format(' '.join(command)))
+        log.info('Running command "{}"'.format(' '.join(command)))
         env = self._build_env(env)
         subprocess.run(command, env=env, cwd=cwd, check=True)
 
@@ -498,7 +501,7 @@ class PgBin:
         """
 
         self._fixpath(command)
-        print('Running command "{}"'.format(' '.join(command)))
+        log.info('Running command "{}"'.format(' '.join(command)))
         env = self._build_env(env)
         return subprocess_capture(self.log_dir, command, env=env, cwd=cwd, check=True, **kwargs)
 
@@ -566,12 +569,12 @@ class Postgres(PgProtocol):
 
         assert self.branch is not None
 
-        print(f"Starting postgres on branch {self.branch}")
+        log.info(f"Starting postgres on branch {self.branch}")
 
         run_result = self.zenith_cli.run(['pg', 'start', self.branch, f'--tenantid={self.tenant_id}', f'--port={self.port}'])
         self.running = True
 
-        print(f"stdout: {run_result.stdout}")
+        log.info(f"stdout: {run_result.stdout}")
 
         return self
 
@@ -786,7 +789,7 @@ def postgres(zenith_cli: ZenithCli, initial_tenant: str, repo_dir: str, pg_bin:
     yield pgfactory
 
     # After the yield comes any cleanup code we need.
-    print('Starting postgres cleanup')
+    log.info('Starting postgres cleanup')
     pgfactory.stop_all()
 
 def read_pid(path: Path):
@@ -794,17 +797,12 @@ def read_pid(path: Path):
     return int(path.read_text())
 
 
-@dataclass
-class WalAcceptorPort:
-    pg: int
-    http: int
-
 @dataclass
 class WalAcceptor:
     """ An object representing a running wal acceptor daemon. """
     wa_bin_path: Path
     data_dir: Path
-    port: WalAcceptorPort
+    port: int
     num: int # identifier for logging
     pageserver_port: int
     auth_token: Optional[str] = None
@@ -816,14 +814,13 @@ class WalAcceptor:
 
         cmd = [str(self.wa_bin_path)]
         cmd.extend(["-D", str(self.data_dir)])
-        cmd.extend(["--listen-pg", f"localhost:{self.port.pg}"])
-        cmd.extend(["--listen-http", f"localhost:{self.port.http}"])
+        cmd.extend(["-l", f"localhost:{self.port}"])
         cmd.append("--daemonize")
         cmd.append("--no-sync")
         # Tell page server it can receive WAL from this WAL safekeeper
         cmd.extend(["--pageserver", f"localhost:{self.pageserver_port}"])
         cmd.extend(["--recall", "1 second"])
-        print('Running command "{}"'.format(' '.join(cmd)))
+        log.info('Running command "{}"'.format(' '.join(cmd)))
         env = {'PAGESERVER_AUTH_TOKEN': self.auth_token} if self.auth_token else None
         subprocess.run(cmd, check=True, env=env)
 
@@ -852,10 +849,10 @@ class WalAcceptor:
         return pid
 
     def stop(self) -> 'WalAcceptor':
-        print('Stopping wal acceptor {}'.format(self.num))
+        log.info('Stopping wal acceptor {}'.format(self.num))
         pid = self.get_pid()
         if pid is None:
-            print("Wal acceptor {} is not running".format(self.num))
+            log.info("Wal acceptor {} is not running".format(self.num))
             return self
 
         try:
@@ -874,17 +871,17 @@ class WalAcceptor:
 
         # "replication=0" hacks psycopg not to send additional queries
         # on startup, see https://github.com/psycopg/psycopg2/pull/482
-        connstr = f"host=localhost port={self.port.pg} replication=0 options='-c ztimelineid={timeline_id} ztenantid={tenant_id}'"
+        connstr = f"host=localhost port={self.port} replication=0 options='-c ztimelineid={timeline_id} ztenantid={tenant_id}'"
 
         with closing(psycopg2.connect(connstr)) as conn:
             # server doesn't support transactions
             conn.autocommit = True
             with conn.cursor() as cur:
                 request_json = json.dumps(request)
-                print(f"JSON_CTRL request on port {self.port.pg}: {request_json}")
+                log.info(f"JSON_CTRL request on port {self.port}: {request_json}")
                 cur.execute("JSON_CTRL " + request_json)
                 all = cur.fetchall()
-                print(f"JSON_CTRL response: {all[0][0]}")
+                log.info(f"JSON_CTRL response: {all[0][0]}")
                 return json.loads(all[0][0])
 
 class WalAcceptorFactory:
@@ -904,10 +901,7 @@ class WalAcceptorFactory:
         wa = WalAcceptor(
             wa_bin_path=self.wa_bin_path,
             data_dir=self.data_dir / "wal_acceptor_{}".format(wa_num),
-            port=WalAcceptorPort(
-                pg=self.port_distributor.get_port(),
-                http=self.port_distributor.get_port(),
-            ),
+            port=self.port_distributor.get_port(),
             num=wa_num,
             pageserver_port=self.pageserver_port,
             auth_token=auth_token,
@@ -931,7 +925,7 @@ class WalAcceptorFactory:
 
     def get_connstrs(self) -> str:
         """ Get list of wal acceptor endpoints suitable for wal_acceptors GUC  """
-        return ','.join(["localhost:{}".format(wa.port.pg) for wa in self.instances])
+        return ','.join(["localhost:{}".format(wa.port) for wa in self.instances])
 
 
 @zenfixture
@@ -945,7 +939,7 @@ def wa_factory(zenith_binpath: str, repo_dir: str, pageserver_port: PageserverPo
     )
     yield wafactory
     # After the yield comes any cleanup code we need.
-    print('Starting wal acceptors cleanup')
+    log.info('Starting wal acceptors cleanup')
     wafactory.stop_all()
 
 
@@ -954,7 +948,7 @@ def base_dir() -> str:
     """ find the base directory (currently this is the git root) """
 
     base_dir = os.path.normpath(os.path.join(get_self_dir(), '../..'))
-    print('\nbase_dir is', base_dir)
+    log.info(f'base_dir is {base_dir}')
     return base_dir
 
 
@@ -983,7 +977,7 @@ def test_output_dir(request: Any, top_output_dir: str) -> str:
         test_name = 'shared'
 
     test_output_dir = os.path.join(top_output_dir, test_name)
-    print('test_output_dir is', test_output_dir)
+    log.info(f'test_output_dir is {test_output_dir}')
     shutil.rmtree(test_output_dir, ignore_errors=True)
     mkdir_if_needed(test_output_dir)
     return test_output_dir
@@ -1025,7 +1019,7 @@ def pg_distrib_dir(base_dir: str) -> str:
         pg_dir = env_postgres_bin
     else:
         pg_dir = os.path.normpath(os.path.join(base_dir, DEFAULT_POSTGRES_DIR))
-    print('postgres dir is', pg_dir)
+    log.info(f'postgres dir is {pg_dir}')
     if not os.path.exists(os.path.join(pg_dir, 'bin/postgres')):
         raise Exception('postgres not found at "{}"'.format(pg_dir))
     return pg_dir
@@ -1064,7 +1058,7 @@ def list_files_to_compare(pgdata_dir: str):
                 pgdata_files.append(rel_file)
 
     pgdata_files.sort()
-    print(pgdata_files)
+    log.info(pgdata_files)
     return pgdata_files
 
 # pg is the existing and running compute node, that we want to compare with a basebackup
@@ -1110,9 +1104,9 @@ def check_restored_datadir_content(zenith_cli: ZenithCli, test_output_dir: str,
                                                 restored_dir_path,
                                                 pgdata_files,
                                                 shallow=False)
-    print('filecmp result mismatch and error lists:')
-    print(mismatch)
-    print(error)
+    log.info('filecmp result mismatch and error lists:')
+    log.info(mismatch)
+    log.info(error)
 
     for f in mismatch:
 

test_runner/performance/test_bulk_insert.py

@@ -2,8 +2,25 @@ import os
 from contextlib import closing
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 
+def get_timeline_size(repo_dir: str, tenantid: str, timelineid: str):
+    path = "{}/tenants/{}/timelines/{}".format(repo_dir, tenantid, timelineid)
+
+    totalbytes = 0
+    for root, dirs, files in os.walk(path):
+        for name in files:
+            totalbytes += os.path.getsize(os.path.join(root, name))
+
+        if 'wal' in dirs:
+            dirs.remove('wal')  # don't visit 'wal' subdirectory
+
+    return totalbytes
+
 #
 # Run bulk INSERT test.
 #
@@ -12,14 +29,13 @@ pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 # 1. Time to INSERT 5 million rows
 # 2. Disk writes
 # 3. Disk space used
-# 4. Peak memory usage
 #
 def test_bulk_insert(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
     # Create a branch for us
     zenith_cli.run(["branch", "test_bulk_insert", "empty"])
 
     pg = postgres.create_start('test_bulk_insert')
-    print("postgres is running on 'test_bulk_insert' branch")
+    log.info("postgres is running on 'test_bulk_insert' branch")
 
     # Open a connection directly to the page server that we'll use to force
     # flushing the layers to disk
@@ -43,9 +59,6 @@ def test_bulk_insert(postgres: PostgresFactory, pageserver: ZenithPageserver, pg
                     # time and I/O
                     pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
 
-            # Record peak memory usage
-            zenbenchmark.record("peak_mem", zenbenchmark.get_peak_mem(pageserver) / 1024, 'MB')
-
             # Report disk space used by the repository
-            timeline_size = zenbenchmark.get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
+            timeline_size = get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
             zenbenchmark.record('size', timeline_size / (1024*1024), 'MB')

test_runner/performance/test_bulk_tenant_create.py

@@ -1,58 +0,0 @@
-import timeit
-import pytest
-
-from fixtures.zenith_fixtures import (
-    TenantFactory,
-    ZenithCli,
-    PostgresFactory,
-)
-
-pytest_plugins = ("fixtures.benchmark_fixture")
-
-# Run bulk tenant creation test.
-#
-# Collects metrics:
-#
-# 1. Time to create {1,10,50} tenants
-# 2. Average creation time per tenant
-
-
-@pytest.mark.parametrize('tenants_count', [1, 5, 10])
-@pytest.mark.parametrize('use_wal_acceptors', ['with_wa', 'without_wa'])
-def test_bulk_tenant_create(
-    zenith_cli: ZenithCli,
-    tenant_factory: TenantFactory,
-    postgres: PostgresFactory,
-    wa_factory,
-    use_wal_acceptors: str,
-    tenants_count: int,
-    zenbenchmark,
-):
-    """Measure tenant creation time (with and without wal acceptors)"""
-
-    time_slices = []
-
-    for i in range(tenants_count):
-        start = timeit.default_timer()
-
-        tenant = tenant_factory.create()
-        zenith_cli.run([
-            "branch", f"test_bulk_tenant_create_{tenants_count}_{i}_{use_wal_acceptors}", "main",
-            f"--tenantid={tenant}"
-        ])
-
-        if use_wal_acceptors == 'with_wa':
-            wa_factory.start_n_new(3)
-
-        pg_tenant = postgres.create_start(
-            f"test_bulk_tenant_create_{tenants_count}_{i}_{use_wal_acceptors}",
-            tenant,
-            wal_acceptors=wa_factory.get_connstrs() if use_wal_acceptors == 'with_wa' else None,
-        )
-
-        end = timeit.default_timer()
-        time_slices.append(end - start)
-
-        pg_tenant.stop()
-
-    zenbenchmark.record('tenant_creation_time', sum(time_slices) / len(time_slices), 's')

test_runner/performance/test_perf_pgbench.py

@@ -2,8 +2,25 @@ import os
 from contextlib import closing
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 
+def get_timeline_size(repo_dir: str, tenantid: str, timelineid: str):
+    path = "{}/tenants/{}/timelines/{}".format(repo_dir, tenantid, timelineid)
+
+    totalbytes = 0
+    for root, dirs, files in os.walk(path):
+        for name in files:
+            totalbytes += os.path.getsize(os.path.join(root, name))
+
+        if 'wal' in dirs:
+            dirs.remove('wal')  # don't visit 'wal' subdirectory
+
+    return totalbytes
+
 #
 # Run a very short pgbench test.
 #
@@ -18,7 +35,7 @@ def test_pgbench(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin
     zenith_cli.run(["branch", "test_pgbench_perf", "empty"])
 
     pg = postgres.create_start('test_pgbench_perf')
-    print("postgres is running on 'test_pgbench_perf' branch")
+    log.info("postgres is running on 'test_pgbench_perf' branch")
 
     # Open a connection directly to the page server that we'll use to force
     # flushing the layers to disk
@@ -51,5 +68,5 @@ def test_pgbench(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin
     pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
 
     # Report disk space used by the repository
-    timeline_size = zenbenchmark.get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
+    timeline_size = get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
     zenbenchmark.record('size', timeline_size / (1024*1024), 'MB')

test_runner/performance/test_write_amplification.py

@@ -1,74 +0,0 @@
-# Demonstrate Write Amplification with naive oldest-first layer checkpointing
-# algorithm.
-#
-# In each iteration of the test, we create a new table that's slightly under 10
-# MB in size (10 MB is the current "segment size" used by the page server). Then
-# we make a tiny update to all the tables already created. This creates a WAL
-# pattern where you have a lot of updates on one segment (the newly created
-# one), alternating with a small updates on all relations. This is the worst
-# case scenario for the naive checkpointing policy where we write out the layers
-# in LSN order, writing the oldest layer first. That creates a new 10 MB image
-# layer to be created for each of those small updates.  This is the Write
-# Amplification problem at its finest.
-import os
-from contextlib import closing
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-
-pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
-
-def test_write_amplification(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
-    # Create a branch for us
-    zenith_cli.run(["branch", "test_write_amplification", "empty"])
-
-    pg = postgres.create_start('test_write_amplification')
-    print("postgres is running on 'test_write_amplification' branch")
-
-    # Open a connection directly to the page server that we'll use to force
-    # flushing the layers to disk
-    psconn = pageserver.connect();
-    pscur = psconn.cursor()
-
-    with closing(pg.connect()) as conn:
-        with conn.cursor() as cur:
-            # Get the timeline ID of our branch. We need it for the 'do_gc' command
-            cur.execute("SHOW zenith.zenith_timeline")
-            timeline = cur.fetchone()[0]
-
-            with zenbenchmark.record_pageserver_writes(pageserver, 'pageserver_writes'):
-                with zenbenchmark.record_duration('run'):
-
-                    # NOTE: Because each iteration updates every table already created,
-                    # the runtime and write amplification is O(n^2), where n is the
-                    # number of iterations.
-                    for i in range(25):
-                        cur.execute(f'''
-                        CREATE TABLE tbl{i} AS
-                            SELECT g as i, 'long string to consume some space' || g as t
-                            FROM generate_series(1, 100000) g
-                        ''')
-                        cur.execute(f"create index on tbl{i} (i);")
-                        for j in range(1, i):
-                            cur.execute(f"delete from tbl{j} where i = {i}")
-
-                        # Force checkpointing. As of this writing, we don't have
-                        # a back-pressure mechanism, and the page server cannot
-                        # keep up digesting and checkpointing the WAL at the
-                        # rate that it is generated. If we don't force a
-                        # checkpoint, the WAL will just accumulate in memory
-                        # until you hit OOM error. So in effect, we use much
-                        # more memory to hold the incoming WAL, and write them
-                        # out in larger batches than we'd really want. Using
-                        # more memory hides the write amplification problem this
-                        # test tries to demonstrate.
-                        #
-                        # The write amplification problem is real, and using
-                        # more memory isn't the right solution. We could
-                        # demonstrate the effect also by generating the WAL
-                        # slower, adding some delays in this loop.  But forcing
-                        # the the checkpointing and GC makes the test go faster,
-                        # with the same total I/O effect.
-                        pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
-
-            # Report disk space used by the repository
-            timeline_size = zenbenchmark.get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
-            zenbenchmark.record('size', timeline_size / (1024*1024), 'MB')

test_runner/test_broken.py

@@ -1,6 +1,10 @@
 import pytest
 import os
 
+import logging
+import fixtures.log_helper  # configures loggers
+log = logging.getLogger('root')
+
 pytest_plugins = ("fixtures.zenith_fixtures")
 """
 Use this test to see what happens when tests fail.
@@ -22,7 +26,7 @@ def test_broken(zenith_cli, pageserver, postgres, pg_bin):
     zenith_cli.run(["branch", "test_broken", "empty"])
 
     postgres.create_start("test_broken")
-    print('postgres is running')
+    log.info('postgres is running')
 
-    print('THIS NEXT COMMAND WILL FAIL:')
+    log.info('THIS NEXT COMMAND WILL FAIL:')
     pg_bin.run('pgbench -i_am_a_broken_test'.split())

walkeeper/Cargo.toml

@@ -28,11 +28,9 @@ humantime = "2.1.0"
 walkdir = "2"
 serde = { version = "1.0", features = ["derive"] }
 hex = "0.4.3"
-const_format = "0.2.21"
 
 # FIXME: 'pageserver' is needed for ZTimelineId. Refactor
 pageserver = { path = "../pageserver" }
 postgres_ffi = { path = "../postgres_ffi" }
 workspace_hack = { path = "../workspace_hack" }
-zenith_metrics = { path = "../zenith_metrics" }
 zenith_utils = { path = "../zenith_utils" }

walkeeper/README

@@ -76,43 +76,6 @@ safekeepers.
 See README_PROTO.md for a more detailed desription of the consensus
 protocol. spec/ contains TLA+ specification of it.
 
-# Q&A
-
-Q: Why have a separate service instead of connecting Page Server directly to a
-   primary PostgreSQL node?
-A: Page Server is a single server which can be lost. As our primary
-   fault-tolerant storage is S3, we do not want to wait for it before
-   committing a transaction. The WAL service acts as a temporary fault-tolerant
-   storage for recent data before it gets to the Page Server and then finally
-   to S3. Whenever WALs and pages are committed to S3, WAL's storage can be
-   trimmed.
-
-Q: What if the compute node evicts a page, needs it back, but the page is yet
-   to reach the Page Server?
-A: If the compute node has evicted a page, all changes from that page are
-   already committed, i.e. they are saved on majority of WAL safekeepers. These
-   WAL records will eventually reach the Page Server. The Page Server notes
-   that the compute note requests pages with a very recent LSN and will not
-   respond to the compute node until it a corresponding WAL is received from WAL
-   safekeepers.
-
-Q: How long may Page Server wait for?
-A: Not too long, hopefully. If a page is evicted, it probably was not used for
-   a while, so the WAL service have had enough time to push changes to the Page
-   Server. There may be issues if there is no backpressure and compute node with
-   WAL service run ahead of Page Server, though.
-   There is no backpressure right now, so you may even see some spurious
-   timeouts in tests.
-
-Q: How do WAL safekeepers communicate with each other?
-A: They may only send each other messages via the compute node, they never
-   communicate directly with each other.
-
-Q: Why have a consensus algorithm if there is only a single compute node?
-A: Actually there may be moments with multiple PostgreSQL nodes running at the
-   same time. E.g. we are bringing one up and one down. We would like to avoid
-   simultaneous writes from different nodes, so there should be a consensus on
-   who is the primary node.
 
 # Terminology
 

walkeeper/src/bin/wal_acceptor.rs

@@ -3,23 +3,18 @@
 //
 use anyhow::Result;
 use clap::{App, Arg};
-use const_format::formatcp;
 use daemonize::Daemonize;
 use log::*;
 use std::env;
-use std::net::TcpListener;
 use std::path::{Path, PathBuf};
 use std::thread;
-use zenith_utils::http::endpoint;
 use zenith_utils::logging;
 
-use walkeeper::defaults::{DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_PG_LISTEN_ADDR};
 use walkeeper::s3_offload;
 use walkeeper::wal_service;
 use walkeeper::WalAcceptorConf;
 
 fn main() -> Result<()> {
-    zenith_metrics::set_common_metrics_prefix("safekeeper");
     let arg_matches = App::new("Zenith wal_acceptor")
         .about("Store WAL stream to local file system and push it to WAL receivers")
         .arg(
@@ -30,18 +25,11 @@ fn main() -> Result<()> {
                 .help("Path to the WAL acceptor data directory"),
         )
         .arg(
-            Arg::with_name("listen-pg")
+            Arg::with_name("listen")
                 .short("l")
-                .long("listen-pg")
-                .alias("listen") // for compatibility
+                .long("listen")
                 .takes_value(true)
-                .help(formatcp!("listen for incoming WAL data connections on ip:port (default: {DEFAULT_PG_LISTEN_ADDR})")),
-        )
-        .arg(
-            Arg::with_name("listen-http")
-                .long("listen-http")
-                .takes_value(true)
-                .help(formatcp!("http endpoint address for metrics on ip:port (default: {DEFAULT_HTTP_LISTEN_ADDR})")),
+                .help("listen for incoming connections on ip:port (default: 127.0.0.1:5454)"),
         )
         .arg(
             Arg::with_name("pageserver")
@@ -82,8 +70,7 @@ fn main() -> Result<()> {
         daemonize: false,
         no_sync: false,
         pageserver_addr: None,
-        listen_pg_addr: DEFAULT_PG_LISTEN_ADDR.to_string(),
-        listen_http_addr: DEFAULT_HTTP_LISTEN_ADDR.to_string(),
+        listen_addr: "localhost:5454".to_string(),
         ttl: None,
         recall_period: None,
         pageserver_auth_token: env::var("PAGESERVER_AUTH_TOKEN").ok(),
@@ -104,12 +91,8 @@ fn main() -> Result<()> {
         conf.daemonize = true;
     }
 
-    if let Some(addr) = arg_matches.value_of("listen-pg") {
-        conf.listen_pg_addr = addr.to_owned();
-    }
-
-    if let Some(addr) = arg_matches.value_of("listen-http") {
-        conf.listen_http_addr = addr.to_owned();
+    if let Some(addr) = arg_matches.value_of("listen") {
+        conf.listen_addr = addr.to_owned();
     }
 
     if let Some(addr) = arg_matches.value_of("pageserver") {
@@ -131,11 +114,6 @@ fn start_wal_acceptor(conf: WalAcceptorConf) -> Result<()> {
     let log_filename = conf.data_dir.join("wal_acceptor.log");
     let (_scope_guard, log_file) = logging::init(log_filename, conf.daemonize)?;
 
-    let http_listener = TcpListener::bind(conf.listen_http_addr.clone()).map_err(|e| {
-        error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
-        e
-    })?;
-
     if conf.daemonize {
         info!("daemonizing...");
 
@@ -158,16 +136,6 @@ fn start_wal_acceptor(conf: WalAcceptorConf) -> Result<()> {
 
     let mut threads = Vec::new();
 
-    let http_endpoint_thread = thread::Builder::new()
-        .name("http_endpoint_thread".into())
-        .spawn(move || {
-            // No authentication at all: read-only metrics only, early stage.
-            let router = endpoint::make_router();
-            endpoint::serve_thread_main(router, http_listener).unwrap();
-        })
-        .unwrap();
-    threads.push(http_endpoint_thread);
-
     if conf.ttl.is_some() {
         let s3_conf = conf.clone();
         let s3_offload_thread = thread::Builder::new()

walkeeper/src/lib.rs

@@ -11,23 +11,12 @@ pub mod send_wal;
 pub mod timeline;
 pub mod wal_service;
 
-pub mod defaults {
-    use const_format::formatcp;
-
-    pub const DEFAULT_PG_LISTEN_PORT: u16 = 5454;
-    pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");
-
-    pub const DEFAULT_HTTP_LISTEN_PORT: u16 = 7676;
-    pub const DEFAULT_HTTP_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_HTTP_LISTEN_PORT}");
-}
-
 #[derive(Debug, Clone)]
 pub struct WalAcceptorConf {
     pub data_dir: PathBuf,
     pub daemonize: bool,
     pub no_sync: bool,
-    pub listen_pg_addr: String,
-    pub listen_http_addr: String,
+    pub listen_addr: String,
     pub pageserver_addr: Option<String>,
     // TODO (create issue) this is temporary, until protocol between PG<->SK<->PS rework
     pub pageserver_auth_token: Option<String>,

walkeeper/src/receive_wal.rs

@@ -42,7 +42,7 @@ fn request_callback(conf: WalAcceptorConf, timelineid: ZTimelineId, tenantid: ZT
     );
 
     // use Config parsing because SockAddr parsing doesnt allow to use host names instead of ip addresses
-    let me_connstr = format!("postgresql://no_user@{}/no_db", conf.listen_pg_addr);
+    let me_connstr = format!("postgresql://no_user@{}/no_db", conf.listen_addr);
     let me_conf: Config = me_connstr.parse().unwrap();
     let (host, port) = connection_host_port(&me_conf);
     let callme = format!(

walkeeper/src/safekeeper.rs

@@ -15,11 +15,8 @@ use std::cmp::min;
 use std::io;
 use std::io::Read;
 
-use lazy_static::lazy_static;
-
 use crate::replication::HotStandbyFeedback;
 use postgres_ffi::xlog_utils::MAX_SEND_SIZE;
-use zenith_metrics::{register_gauge_vec, Gauge, GaugeVec};
 use zenith_utils::bin_ser::LeSer;
 use zenith_utils::lsn::Lsn;
 use zenith_utils::pq_proto::SystemId;
@@ -278,40 +275,8 @@ impl AcceptorProposerMessage {
 pub trait Storage {
     /// Persist safekeeper state on disk, optionally syncing it.
     fn persist(&mut self, s: &SafeKeeperState, sync: bool) -> Result<()>;
-    /// Write piece of wal in buf to disk and sync it.
-    fn write_wal(&mut self, server: &ServerInfo, startpos: Lsn, buf: &[u8]) -> Result<()>;
-}
-
-lazy_static! {
-    // The prometheus crate does not support u64 yet, i64 only (see `IntGauge`).
-    // i64 is faster than f64, so update to u64 when available.
-    static ref FLUSH_LSN_GAUGE: GaugeVec = register_gauge_vec!(
-        "safekeeper_flush_lsn",
-        "Current flush_lsn, grouped by timeline",
-        &["ztli"]
-    )
-    .expect("Failed to register safekeeper_flush_lsn int gauge vec");
-    static ref COMMIT_LSN_GAUGE: GaugeVec = register_gauge_vec!(
-        "safekeeper_commit_lsn",
-        "Current commit_lsn (not necessarily persisted to disk), grouped by timeline",
-        &["ztli"]
-    )
-    .expect("Failed to register safekeeper_commit_lsn int gauge vec");
-}
-
-struct SafeKeeperMetrics {
-    flush_lsn: Gauge,
-    commit_lsn: Gauge,
-}
-
-impl SafeKeeperMetrics {
-    fn new(ztli: ZTimelineId) -> SafeKeeperMetrics {
-        let ztli_str = format!("{}", ztli);
-        SafeKeeperMetrics {
-            flush_lsn: FLUSH_LSN_GAUGE.with_label_values(&[&ztli_str]),
-            commit_lsn: COMMIT_LSN_GAUGE.with_label_values(&[&ztli_str]),
-        }
-    }
+    /// Write piece of wal in buf to disk.
+    fn write_wal(&mut self, s: &SafeKeeperState, startpos: Lsn, buf: &[u8]) -> Result<()>;
 }
 
 /// SafeKeeper which consumes events (messages from compute) and provides
@@ -321,8 +286,6 @@ pub struct SafeKeeper<ST: Storage> {
     /// Established by reading wal.
     pub flush_lsn: Lsn,
     pub tli: u32,
-    // Cached metrics so we don't have to recompute labels on each update.
-    metrics: Option<SafeKeeperMetrics>,
     /// not-yet-flushed pairs of same named fields in s.*
     pub commit_lsn: Lsn,
     pub truncate_lsn: Lsn,
@@ -341,7 +304,6 @@ where
         SafeKeeper {
             flush_lsn,
             tli,
-            metrics: None,
             commit_lsn: state.commit_lsn,
             truncate_lsn: state.truncate_lsn,
             storage,
@@ -393,8 +355,6 @@ where
         self.s.server.wal_seg_size = msg.wal_seg_size;
         self.storage.persist(&self.s, true)?;
 
-        self.metrics = Some(SafeKeeperMetrics::new(self.s.server.ztli));
-
         info!(
             "processed greeting from proposer {:?}, sending term {:?}",
             msg.proposer_id, self.s.acceptor_state.term
@@ -463,7 +423,7 @@ where
         let mut last_rec_lsn = Lsn(0);
         if !msg.wal_data.is_empty() {
             self.storage
-                .write_wal(&self.s.server, msg.h.begin_lsn, &msg.wal_data)?;
+                .write_wal(&self.s, msg.h.begin_lsn, &msg.wal_data)?;
 
             // figure out last record's end lsn for reporting (if we got the
             // whole record)
@@ -518,11 +478,6 @@ where
         }
         if last_rec_lsn > self.flush_lsn {
             self.flush_lsn = last_rec_lsn;
-            self.metrics
-                .as_ref()
-                .unwrap()
-                .flush_lsn
-                .set(u64::from(self.flush_lsn) as f64);
         }
 
         // Advance commit_lsn taking into account what we have locally. xxx this
@@ -540,11 +495,6 @@ where
             sync_control_file |=
                 commit_lsn >= msg.h.epoch_start_lsn && self.s.commit_lsn < msg.h.epoch_start_lsn;
             self.commit_lsn = commit_lsn;
-            self.metrics
-                .as_ref()
-                .unwrap()
-                .commit_lsn
-                .set(u64::from(self.commit_lsn) as f64);
         }
 
         self.truncate_lsn = msg.h.truncate_lsn;
@@ -596,7 +546,7 @@ mod tests {
             Ok(())
         }
 
-        fn write_wal(&mut self, _server: &ServerInfo, _startpos: Lsn, _buf: &[u8]) -> Result<()> {
+        fn write_wal(&mut self, _s: &SafeKeeperState, _startpos: Lsn, _buf: &[u8]) -> Result<()> {
             Ok(())
         }
     }

walkeeper/src/timeline.rs

@@ -18,8 +18,8 @@ use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 use crate::replication::{HotStandbyFeedback, END_REPLICATION_MARKER};
 use crate::safekeeper::{
-    AcceptorProposerMessage, ProposerAcceptorMessage, SafeKeeper, SafeKeeperState, ServerInfo,
-    Storage, SK_FORMAT_VERSION, SK_MAGIC,
+    AcceptorProposerMessage, ProposerAcceptorMessage, SafeKeeper, SafeKeeperState, Storage,
+    SK_FORMAT_VERSION, SK_MAGIC,
 };
 use crate::WalAcceptorConf;
 use postgres_ffi::xlog_utils::{XLogFileName, XLOG_BLCKSZ};
@@ -32,7 +32,7 @@ struct SharedState {
     sk: SafeKeeper<FileStorage>,
     /// For receiving-sending wal cooperation
     /// quorum commit LSN we've notified walsenders about
-    notified_commit_lsn: Lsn,
+    commit_lsn: Lsn,
     /// combined hot standby feedback from all replicas
     hs_feedback: HotStandbyFeedback,
 }
@@ -72,7 +72,7 @@ impl SharedState {
         };
 
         Ok(Self {
-            notified_commit_lsn: Lsn(0),
+            commit_lsn: Lsn(0),
             sk: SafeKeeper::new(Lsn(flush_lsn), tli, storage, state),
             hs_feedback: HotStandbyFeedback {
                 ts: 0,
@@ -186,7 +186,7 @@ impl Timeline {
     pub fn wait_for_lsn(&self, lsn: Lsn) -> Lsn {
         let mut shared_state = self.mutex.lock().unwrap();
         loop {
-            let commit_lsn = shared_state.notified_commit_lsn;
+            let commit_lsn = shared_state.commit_lsn;
             // This must be `>`, not `>=`.
             if commit_lsn > lsn {
                 return commit_lsn;
@@ -198,8 +198,8 @@ impl Timeline {
     // Notify caught-up WAL senders about new WAL data received
     pub fn notify_wal_senders(&self, commit_lsn: Lsn) {
         let mut shared_state = self.mutex.lock().unwrap();
-        if shared_state.notified_commit_lsn < commit_lsn {
-            shared_state.notified_commit_lsn = commit_lsn;
+        if shared_state.commit_lsn < commit_lsn {
+            shared_state.commit_lsn = commit_lsn;
             self.cond.notify_all();
         }
     }
@@ -337,14 +337,14 @@ impl Storage for FileStorage {
         Ok(())
     }
 
-    fn write_wal(&mut self, server: &ServerInfo, startpos: Lsn, buf: &[u8]) -> Result<()> {
+    fn write_wal(&mut self, s: &SafeKeeperState, startpos: Lsn, buf: &[u8]) -> Result<()> {
         let mut bytes_left: usize = buf.len();
         let mut bytes_written: usize = 0;
         let mut partial;
         let mut start_pos = startpos;
         const ZERO_BLOCK: &[u8] = &[0u8; XLOG_BLCKSZ];
-        let wal_seg_size = server.wal_seg_size as usize;
-        let ztli = server.ztli;
+        let wal_seg_size = s.server.wal_seg_size as usize;
+        let ztli = s.server.ztli;
 
         /* Extract WAL location for this block */
         let mut xlogoff = start_pos.segment_offset(wal_seg_size) as usize;
@@ -365,7 +365,7 @@ impl Storage for FileStorage {
             /* Open file */
             let segno = start_pos.segment_number(wal_seg_size);
             // note: we basically don't support changing pg timeline
-            let wal_file_name = XLogFileName(server.tli, segno, wal_seg_size);
+            let wal_file_name = XLogFileName(s.server.tli, segno, wal_seg_size);
             let wal_file_path = self
                 .conf
                 .data_dir

walkeeper/src/wal_service.rs

@@ -13,9 +13,9 @@ use zenith_utils::postgres_backend::{AuthType, PostgresBackend};
 
 /// Accept incoming TCP connections and spawn them into a background thread.
 pub fn thread_main(conf: WalAcceptorConf) -> Result<()> {
-    info!("Starting wal acceptor on {}", conf.listen_pg_addr);
-    let listener = TcpListener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
-        error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
+    info!("Starting wal acceptor on {}", conf.listen_addr);
+    let listener = TcpListener::bind(conf.listen_addr.clone()).map_err(|e| {
+        error!("failed to bind to address {}: {}", conf.listen_addr, e);
         e
     })?;
 

zenith_metrics/Cargo.toml

@@ -7,4 +7,3 @@ edition = "2018"
 prometheus = {version = "0.12", default_features=false} # removes protobuf dependency
 libc = "0.2"
 lazy_static = "1.4"
-once_cell = "1.8.0"

zenith_metrics/src/lib.rs

@@ -3,10 +3,7 @@
 //! Otherwise, we might not see all metrics registered via
 //! a default registry.
 use lazy_static::lazy_static;
-use once_cell::race::OnceBox;
 pub use prometheus::{exponential_buckets, linear_buckets};
-pub use prometheus::{register_gauge, Gauge};
-pub use prometheus::{register_gauge_vec, GaugeVec};
 pub use prometheus::{register_histogram, Histogram};
 pub use prometheus::{register_histogram_vec, HistogramVec};
 pub use prometheus::{register_int_counter, IntCounter};
@@ -23,55 +20,17 @@ pub use wrappers::{CountedReader, CountedWriter};
 /// Metrics gathering is a relatively simple and standalone operation, so
 /// it might be fine to do it this way to keep things simple.
 pub fn gather() -> Vec<prometheus::proto::MetricFamily> {
-    update_rusage_metrics();
+    update_io_metrics();
     prometheus::gather()
 }
 
-static COMMON_METRICS_PREFIX: OnceBox<&str> = OnceBox::new();
-
-/// Sets a prefix which will be used for all common metrics, typically a service
-/// name like 'pageserver'. Should be executed exactly once in the beginning of
-/// any executable which uses common metrics.
-pub fn set_common_metrics_prefix(prefix: &'static str) {
-    // Not unwrap() because metrics may be initialized after multiple threads have been started.
-    COMMON_METRICS_PREFIX
-        .set(prefix.into())
-        .unwrap_or_else(|_| {
-            eprintln!(
-                "set_common_metrics_prefix() was called second time with '{}', exiting",
-                prefix
-            );
-            std::process::exit(1);
-        });
-}
-
-/// Prepends a prefix to a common metric name so they are distinguished between
-/// different services, see <https://github.com/zenithdb/zenith/pull/681>
-/// A call to set_common_metrics_prefix() is necessary prior to calling this.
-pub fn new_common_metric_name(unprefixed_metric_name: &str) -> String {
-    // Not unwrap() because metrics may be initialized after multiple threads have been started.
-    format!(
-        "{}_{}",
-        COMMON_METRICS_PREFIX.get().unwrap_or_else(|| {
-            eprintln!("set_common_metrics_prefix() was not called, but metrics are used, exiting");
-            std::process::exit(1);
-        }),
-        unprefixed_metric_name
-    )
-}
-
 lazy_static! {
     static ref DISK_IO_BYTES: IntGaugeVec = register_int_gauge_vec!(
-        new_common_metric_name("disk_io_bytes"),
+        "pageserver_disk_io_bytes",
         "Bytes written and read from disk, grouped by the operation (read|write)",
         &["io_operation"]
     )
     .expect("Failed to register disk i/o bytes int gauge vec");
-    static ref MAXRSS_KB: IntGauge = register_int_gauge!(
-        new_common_metric_name("maxrss_kb"),
-        "Memory usage (Maximum Resident Set Size)"
-    )
-    .expect("Failed to register maxrss_kb int gauge");
 }
 
 // Records I/O stats in a "cross-platform" way.
@@ -83,7 +42,7 @@ lazy_static! {
 // We know the size of the block, so we can determine the I/O bytes out of it.
 // The value might be not 100% exact, but should be fine for Prometheus metrics in this case.
 #[allow(clippy::unnecessary_cast)]
-fn update_rusage_metrics() {
+fn update_io_metrics() {
     let rusage_stats = get_rusage_stats();
 
     const BYTES_IN_BLOCK: i64 = 512;
@@ -93,7 +52,6 @@ fn update_rusage_metrics() {
     DISK_IO_BYTES
         .with_label_values(&["write"])
         .set(rusage_stats.ru_oublock * BYTES_IN_BLOCK);
-    MAXRSS_KB.set(rusage_stats.ru_maxrss);
 }
 
 fn get_rusage_stats() -> libc::rusage {

zenith_utils/Cargo.toml

@@ -18,7 +18,6 @@ serde = { version = "1.0", features = ["derive"] }
 serde_json = "1"
 thiserror = "1.0"
 tokio = "1.11"
-crossbeam-utils = "0.8.5"
 
 slog-async = "2.6.0"
 slog-stdlog = "4.1.0"
@@ -39,4 +38,3 @@ rustls-split = "0.2.1"
 hex-literal = "0.3"
 bytes = "1.0"
 webpki = "0.21"
-tempfile = "3.2"

zenith_utils/src/batch_fsync.rs

@@ -1,45 +0,0 @@
-use crossbeam_utils::thread;
-
-use std::{
-    fs::File,
-    io,
-    path::PathBuf,
-    sync::atomic::{AtomicUsize, Ordering},
-};
-
-pub fn batch_fsync(paths: &[PathBuf]) -> std::io::Result<()> {
-    let next = AtomicUsize::new(0);
-
-    let num_threads = std::cmp::min(paths.len() / 2, 256);
-
-    if num_threads <= 1 {
-        for path in paths {
-            let file = File::open(&path)?;
-            file.sync_all()?;
-        }
-
-        return Ok(());
-    }
-
-    thread::scope(|s| -> io::Result<()> {
-        let mut handles = Vec::new();
-        for _ in 0..num_threads {
-            handles.push(s.spawn(|_| loop {
-                let idx = next.fetch_add(1, Ordering::Relaxed);
-                if idx >= paths.len() {
-                    return io::Result::Ok(());
-                }
-
-                let file = File::open(&paths[idx])?;
-                file.sync_all()?;
-            }));
-        }
-
-        for handle in handles {
-            handle.join().unwrap()?;
-        }
-
-        Ok(())
-    })
-    .unwrap()
-}

zenith_utils/src/crashsafe_dir.rs

@@ -1,125 +0,0 @@
-use std::{
-    fs::{self, File},
-    io,
-    path::Path,
-};
-
-/// Similar to [`std::fs::create_dir`], except we fsync the
-/// created directory and its parent.
-pub fn create_dir(path: impl AsRef<Path>) -> io::Result<()> {
-    let path = path.as_ref();
-
-    fs::create_dir(path)?;
-    File::open(path)?.sync_all()?;
-
-    if let Some(parent) = path.parent() {
-        File::open(parent)?.sync_all()
-    } else {
-        Err(io::Error::new(
-            io::ErrorKind::InvalidInput,
-            "can't find parent",
-        ))
-    }
-}
-
-/// Similar to [`std::fs::create_dir_all`], except we fsync all
-/// newly created directories and the pre-existing parent.
-pub fn create_dir_all(path: impl AsRef<Path>) -> io::Result<()> {
-    let mut path = path.as_ref();
-
-    let mut dirs_to_create = Vec::new();
-
-    // Figure out which directories we need to create.
-    loop {
-        match path.metadata() {
-            Ok(metadata) if metadata.is_dir() => break,
-            Ok(_) => {
-                return Err(io::Error::new(
-                    io::ErrorKind::AlreadyExists,
-                    format!("non-directory found in path: {:?}", path),
-                ));
-            }
-            Err(ref e) if e.kind() == io::ErrorKind::NotFound => {}
-            Err(e) => return Err(e),
-        }
-
-        dirs_to_create.push(path);
-
-        match path.parent() {
-            Some(parent) => path = parent,
-            None => {
-                return Err(io::Error::new(
-                    io::ErrorKind::InvalidInput,
-                    "can't find parent",
-                ))
-            }
-        }
-    }
-
-    // Create directories from parent to child.
-    for &path in dirs_to_create.iter().rev() {
-        fs::create_dir(path)?;
-    }
-
-    // Fsync the created directories from child to parent.
-    for &path in dirs_to_create.iter() {
-        File::open(path)?.sync_all()?;
-    }
-
-    // If we created any new directories, fsync the parent.
-    if !dirs_to_create.is_empty() {
-        File::open(path)?.sync_all()?;
-    }
-
-    Ok(())
-}
-
-#[cfg(test)]
-mod tests {
-    use tempfile::tempdir;
-
-    use super::*;
-
-    #[test]
-    fn test_create_dir_fsyncd() {
-        let dir = tempdir().unwrap();
-
-        let existing_dir_path = dir.path();
-        let err = create_dir(existing_dir_path).unwrap_err();
-        assert_eq!(err.kind(), io::ErrorKind::AlreadyExists);
-
-        let child_dir = existing_dir_path.join("child");
-        create_dir(child_dir).unwrap();
-
-        let nested_child_dir = existing_dir_path.join("child1").join("child2");
-        let err = create_dir(nested_child_dir).unwrap_err();
-        assert_eq!(err.kind(), io::ErrorKind::NotFound);
-    }
-
-    #[test]
-    fn test_create_dir_all_fsyncd() {
-        let dir = tempdir().unwrap();
-
-        let existing_dir_path = dir.path();
-        create_dir_all(existing_dir_path).unwrap();
-
-        let child_dir = existing_dir_path.join("child");
-        assert!(!child_dir.exists());
-        create_dir_all(&child_dir).unwrap();
-        assert!(child_dir.exists());
-
-        let nested_child_dir = existing_dir_path.join("child1").join("child2");
-        assert!(!nested_child_dir.exists());
-        create_dir_all(&nested_child_dir).unwrap();
-        assert!(nested_child_dir.exists());
-
-        let file_path = existing_dir_path.join("file");
-        std::fs::write(&file_path, b"").unwrap();
-
-        let err = create_dir_all(&file_path).unwrap_err();
-        assert_eq!(err.kind(), io::ErrorKind::AlreadyExists);
-
-        let invalid_dir_path = file_path.join("folder");
-        create_dir_all(&invalid_dir_path).unwrap_err();
-    }
-}

zenith_utils/src/http/endpoint.rs

@@ -9,14 +9,14 @@ use routerify::ext::RequestExt;
 use routerify::RequestInfo;
 use routerify::{Middleware, Router, RouterBuilder, RouterService};
 use std::net::TcpListener;
-use zenith_metrics::{new_common_metric_name, register_int_counter, IntCounter};
+use zenith_metrics::{register_int_counter, IntCounter};
 use zenith_metrics::{Encoder, TextEncoder};
 
 use super::error::ApiError;
 
 lazy_static! {
     static ref SERVE_METRICS_COUNT: IntCounter = register_int_counter!(
-        new_common_metric_name("serve_metrics_count"),
+        "pageserver_serve_metrics_count",
         "Number of metric requests made"
     )
     .expect("failed to define a metric");

zenith_utils/src/lib.rs

@@ -8,9 +8,6 @@ pub mod lsn;
 /// SeqWait allows waiting for a future sequence number to arrive
 pub mod seqwait;
 
-/// append only ordered map implemented with a Vec
-pub mod vec_map;
-
 // Async version of SeqWait. Currently unused.
 // pub mod seqwait_async;
 
@@ -21,9 +18,6 @@ pub mod pq_proto;
 // dealing with connstring parsing and handy access to it's parts
 pub mod connstring;
 
-// helper functions for creating and fsyncing directories/trees
-pub mod crashsafe_dir;
-
 // common authentication routines
 pub mod auth;
 
@@ -40,6 +34,3 @@ pub mod logging;
 
 // Misc
 pub mod accum;
-
-/// Utility for quickly fsyncing many files at once
-pub mod batch_fsync;

zenith_utils/src/lsn.rs

@@ -192,7 +192,9 @@ impl AtomicLsn {
     /// This operation will panic on overflow.
     pub fn fetch_add(&self, val: u64) -> Lsn {
         let prev = self.inner.fetch_add(val, Ordering::AcqRel);
-        assert!(prev.checked_add(val).is_some(), "AtomicLsn overflow");
+        if prev.checked_add(val).is_none() {
+            panic!("AtomicLsn overflow");
+        }
         Lsn(prev)
     }
 

zenith_utils/src/vec_map.rs

@@ -1,293 +0,0 @@
-use std::{cmp::Ordering, ops::RangeBounds};
-
-use serde::{Deserialize, Serialize};
-
-/// Ordered map datastructure implemented in a Vec.
-/// Append only - can only add keys that are larger than the
-/// current max key.
-#[derive(Clone, Debug, Serialize, Deserialize)]
-pub struct VecMap<K, V>(Vec<(K, V)>);
-
-impl<K, V> Default for VecMap<K, V> {
-    fn default() -> Self {
-        VecMap(Default::default())
-    }
-}
-
-#[derive(Debug)]
-pub struct InvalidKey;
-
-impl<K: Ord, V> VecMap<K, V> {
-    pub fn is_empty(&self) -> bool {
-        self.0.is_empty()
-    }
-
-    pub fn as_slice(&self) -> &[(K, V)] {
-        self.0.as_slice()
-    }
-
-    /// This function may panic if given a range where the lower bound is
-    /// greater than the upper bound.
-    pub fn slice_range<R: RangeBounds<K>>(&self, range: R) -> &[(K, V)] {
-        use std::ops::Bound::*;
-
-        let binary_search = |k: &K| self.0.binary_search_by_key(&k, extract_key);
-
-        let start_idx = match range.start_bound() {
-            Unbounded => 0,
-            Included(k) => binary_search(k).unwrap_or_else(std::convert::identity),
-            Excluded(k) => match binary_search(k) {
-                Ok(idx) => idx + 1,
-                Err(idx) => idx,
-            },
-        };
-
-        let end_idx = match range.end_bound() {
-            Unbounded => self.0.len(),
-            Included(k) => match binary_search(k) {
-                Ok(idx) => idx + 1,
-                Err(idx) => idx,
-            },
-            Excluded(k) => binary_search(k).unwrap_or_else(std::convert::identity),
-        };
-
-        &self.0[start_idx..end_idx]
-    }
-
-    /// Add a key value pair to the map.
-    /// If `key` is less than or equal to the current maximum key
-    /// the pair will not be added and InvalidKey error will be returned.
-    pub fn append(&mut self, key: K, value: V) -> Result<(), InvalidKey> {
-        if let Some((last_key, _last_value)) = self.0.last() {
-            if &key <= last_key {
-                return Err(InvalidKey);
-            }
-        }
-
-        self.0.push((key, value));
-        Ok(())
-    }
-
-    /// Update the maximum key value pair or add a new key value pair to the map.
-    /// If `key` is less than the current maximum key no updates or additions
-    /// will occur and InvalidKey error will be returned.
-    pub fn append_or_update_last(&mut self, key: K, mut value: V) -> Result<Option<V>, InvalidKey> {
-        if let Some((last_key, last_value)) = self.0.last_mut() {
-            match key.cmp(last_key) {
-                Ordering::Less => return Err(InvalidKey),
-                Ordering::Equal => {
-                    std::mem::swap(last_value, &mut value);
-                    return Ok(Some(value));
-                }
-                Ordering::Greater => {}
-            }
-        }
-
-        self.0.push((key, value));
-        Ok(None)
-    }
-
-    /// Split the map into two.
-    ///
-    /// The left map contains everything before `cutoff` (exclusive).
-    /// Right map contains `cutoff` and everything after (inclusive).
-    pub fn split_at(&self, cutoff: &K) -> (Self, Self)
-    where
-        K: Clone,
-        V: Clone,
-    {
-        let split_idx = self
-            .0
-            .binary_search_by_key(&cutoff, extract_key)
-            .unwrap_or_else(std::convert::identity);
-
-        (
-            VecMap(self.0[..split_idx].to_vec()),
-            VecMap(self.0[split_idx..].to_vec()),
-        )
-    }
-
-    /// Move items from `other` to the end of `self`, leaving `other` empty.
-    /// If any keys in `other` is less than or equal to any key in `self`,
-    /// `InvalidKey` error will be returned and no mutation will occur.
-    pub fn extend(&mut self, other: &mut Self) -> Result<(), InvalidKey> {
-        let self_last_opt = self.0.last().map(extract_key);
-        let other_first_opt = other.0.last().map(extract_key);
-
-        if let (Some(self_last), Some(other_first)) = (self_last_opt, other_first_opt) {
-            if self_last >= other_first {
-                return Err(InvalidKey);
-            }
-        }
-
-        self.0.append(&mut other.0);
-
-        Ok(())
-    }
-}
-
-fn extract_key<K, V>(entry: &(K, V)) -> &K {
-    &entry.0
-}
-
-#[cfg(test)]
-mod tests {
-    use std::{collections::BTreeMap, ops::Bound};
-
-    use super::VecMap;
-
-    #[test]
-    fn unbounded_range() {
-        let mut vec = VecMap::default();
-        vec.append(0, ()).unwrap();
-
-        assert_eq!(vec.slice_range(0..0), &[]);
-    }
-
-    #[test]
-    #[should_panic]
-    fn invalid_ordering_range() {
-        let mut vec = VecMap::default();
-        vec.append(0, ()).unwrap();
-
-        #[allow(clippy::reversed_empty_ranges)]
-        vec.slice_range(1..0);
-    }
-
-    #[test]
-    fn range_tests() {
-        let mut vec = VecMap::default();
-        vec.append(0, ()).unwrap();
-        vec.append(2, ()).unwrap();
-        vec.append(4, ()).unwrap();
-
-        assert_eq!(vec.slice_range(0..0), &[]);
-        assert_eq!(vec.slice_range(0..1), &[(0, ())]);
-        assert_eq!(vec.slice_range(0..2), &[(0, ())]);
-        assert_eq!(vec.slice_range(0..3), &[(0, ()), (2, ())]);
-
-        assert_eq!(vec.slice_range(..0), &[]);
-        assert_eq!(vec.slice_range(..1), &[(0, ())]);
-
-        assert_eq!(vec.slice_range(..3), &[(0, ()), (2, ())]);
-        assert_eq!(vec.slice_range(..3), &[(0, ()), (2, ())]);
-
-        assert_eq!(vec.slice_range(0..=0), &[(0, ())]);
-        assert_eq!(vec.slice_range(0..=1), &[(0, ())]);
-        assert_eq!(vec.slice_range(0..=2), &[(0, ()), (2, ())]);
-        assert_eq!(vec.slice_range(0..=3), &[(0, ()), (2, ())]);
-
-        assert_eq!(vec.slice_range(..=0), &[(0, ())]);
-        assert_eq!(vec.slice_range(..=1), &[(0, ())]);
-        assert_eq!(vec.slice_range(..=2), &[(0, ()), (2, ())]);
-        assert_eq!(vec.slice_range(..=3), &[(0, ()), (2, ())]);
-    }
-
-    struct BoundIter {
-        min: i32,
-        max: i32,
-
-        next: Option<Bound<i32>>,
-    }
-
-    impl BoundIter {
-        fn new(min: i32, max: i32) -> Self {
-            Self {
-                min,
-                max,
-
-                next: Some(Bound::Unbounded),
-            }
-        }
-    }
-
-    impl Iterator for BoundIter {
-        type Item = Bound<i32>;
-
-        fn next(&mut self) -> Option<Self::Item> {
-            let cur = self.next?;
-
-            self.next = match &cur {
-                Bound::Unbounded => Some(Bound::Included(self.min)),
-                Bound::Included(x) => {
-                    if *x >= self.max {
-                        Some(Bound::Excluded(self.min))
-                    } else {
-                        Some(Bound::Included(x + 1))
-                    }
-                }
-                Bound::Excluded(x) => {
-                    if *x >= self.max {
-                        None
-                    } else {
-                        Some(Bound::Excluded(x + 1))
-                    }
-                }
-            };
-
-            Some(cur)
-        }
-    }
-
-    #[test]
-    fn range_exhaustive() {
-        let map: BTreeMap<i32, ()> = (1..=7).step_by(2).map(|x| (x, ())).collect();
-        let mut vec = VecMap::default();
-        for &key in map.keys() {
-            vec.append(key, ()).unwrap();
-        }
-
-        const RANGE_MIN: i32 = 0;
-        const RANGE_MAX: i32 = 8;
-        for lower_bound in BoundIter::new(RANGE_MIN, RANGE_MAX) {
-            let ub_min = match lower_bound {
-                Bound::Unbounded => RANGE_MIN,
-                Bound::Included(x) => x,
-                Bound::Excluded(x) => x + 1,
-            };
-            for upper_bound in BoundIter::new(ub_min, RANGE_MAX) {
-                let map_range: Vec<(i32, ())> = map
-                    .range((lower_bound, upper_bound))
-                    .map(|(&x, _)| (x, ()))
-                    .collect();
-                let vec_slice = vec.slice_range((lower_bound, upper_bound));
-
-                assert_eq!(map_range, vec_slice);
-            }
-        }
-    }
-
-    #[test]
-    fn extend() {
-        let mut left = VecMap::default();
-        left.append(0, ()).unwrap();
-        assert_eq!(left.as_slice(), &[(0, ())]);
-
-        let mut empty = VecMap::default();
-        left.extend(&mut empty).unwrap();
-        assert_eq!(left.as_slice(), &[(0, ())]);
-        assert_eq!(empty.as_slice(), &[]);
-
-        let mut right = VecMap::default();
-        right.append(1, ()).unwrap();
-
-        left.extend(&mut right).unwrap();
-
-        assert_eq!(left.as_slice(), &[(0, ()), (1, ())]);
-        assert_eq!(right.as_slice(), &[]);
-
-        let mut zero_map = VecMap::default();
-        zero_map.append(0, ()).unwrap();
-
-        left.extend(&mut zero_map).unwrap_err();
-        assert_eq!(left.as_slice(), &[(0, ()), (1, ())]);
-        assert_eq!(zero_map.as_slice(), &[(0, ())]);
-
-        let mut one_map = VecMap::default();
-        one_map.append(1, ()).unwrap();
-
-        left.extend(&mut one_map).unwrap_err();
-        assert_eq!(left.as_slice(), &[(0, ()), (1, ())]);
-        assert_eq!(one_map.as_slice(), &[(1, ())]);
-    }
-}

zenith_utils/src/zid.rs

@@ -78,10 +78,6 @@ macro_rules! zid_newtype {
             pub fn generate() -> Self {
                 $t(ZId::generate())
             }
-
-            pub const fn from_array(b: [u8; 16]) -> Self {
-                $t(ZId(b))
-            }
         }
 
         impl FromStr for $t {