pageserver - add naive layer IDs

Also fixes 'zenith safekeeper restart -m immediate'. The stop-mode was
previously ignored.

.circleci/config.yml

@@ -1,13 +1,13 @@
 version: 2.1
 
-orbs:
-  python: circleci/python@1.4.0
-
 executors:
   zenith-build-executor:
     resource_class: xlarge
     docker:
-      - image: cimg/rust:1.52.1
+      - image: cimg/rust:1.55.0
+  zenith-python-executor:
+    docker:
+      - image: cimg/python:3.7.10  # Oldest available 3.7 with Ubuntu 20.04 (for GLIBC and Rust) at CirlceCI
 
 jobs:
   check-codestyle:
@@ -110,7 +110,7 @@ jobs:
             # Require an exact match. While an out of date cache might speed up the build,
             # there's no way to clean out old packages, so the cache grows every time something
             # changes.
-            - v03-rust-cache-deps-<< parameters.build_type >>-{{ checksum "Cargo.lock" }}
+            - v04-rust-cache-deps-<< parameters.build_type >>-{{ checksum "Cargo.lock" }}
 
         # Build the rust code, including test binaries
       - run:
@@ -128,7 +128,7 @@ jobs:
 
       - save_cache:
           name: Save rust cache
-          key: v03-rust-cache-deps-<< parameters.build_type >>-{{ checksum "Cargo.lock" }}
+          key: v04-rust-cache-deps-<< parameters.build_type >>-{{ checksum "Cargo.lock" }}
           paths:
             - ~/.cargo/registry
             - ~/.cargo/git
@@ -182,9 +182,27 @@ jobs:
           paths:
             - "*"
 
+  check-python:
+    executor: zenith-python-executor
+    steps:
+      - checkout
+      - run:
+          name: Install deps
+          working_directory: test_runner
+          command: pipenv --python 3.7 install --dev
+      - run:
+          name: Run yapf to ensure code format
+          when: always
+          working_directory: test_runner
+          command: pipenv run yapf --recursive --diff .
+      - run:
+          name: Run mypy to check types
+          when: always
+          working_directory: test_runner
+          command: pipenv run mypy .
+
   run-pytest:
-    #description: "Run pytest"
-    executor: python/default
+    executor: zenith-python-executor
     parameters:
       # pytest args to specify the tests to run.
       #
@@ -219,11 +237,9 @@ jobs:
           steps:
             - run: git submodule update --init --depth 1
       - run:
-          name: Install pipenv & deps
+          name: Install deps
           working_directory: test_runner
-          command: |
-            pip install pipenv
-            pipenv install
+          command: pipenv --python 3.7 install
       - run:
           name: Run pytest
           working_directory: test_runner
@@ -241,8 +257,6 @@ jobs:
             if << parameters.run_in_parallel >>; then
               EXTRA_PARAMS="-n4 $EXTRA_PARAMS"
             fi;
-            ./netstat-script.sh &
-            NS_PID=$!
             # Run the tests.
             #
             # The junit.xml file allows CircleCI to display more fine-grained test information
@@ -254,8 +268,6 @@ jobs:
             # -s is not used to prevent pytest from capturing output, because tests are running
             # in parallel and logs are mixed between different tests
             pipenv run pytest --junitxml=$TEST_OUTPUT/junit.xml --tb=short --verbose -rA $TEST_SELECTION $EXTRA_PARAMS
-            kill $NS_PID
-            awk '/===/ {if (count) print count; print; count=0; next} {count++} END {print count}' $TEST_OUTPUT/netstat.stdout > $TEST_OUTPUT/netstat_stats.stdout
       - 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
@@ -337,6 +349,7 @@ workflows:
   build_and_test:
     jobs:
       - check-codestyle
+      - check-python
       - build-postgres:
           name: build-postgres-<< matrix.build_type >>
           matrix:

Dockerfile

@@ -38,8 +38,6 @@ RUN apt-get update && apt-get -yq install libreadline-dev libseccomp-dev openssl
 
 COPY --from=build /zenith/target/release/pageserver /usr/local/bin
 COPY --from=build /zenith/target/release/safekeeper /usr/local/bin
-# TODO: temporary alias for compatibility, see https://github.com/zenithdb/zenith/pull/740
-RUN ln -s /usr/local/bin/safekeeper /usr/local/bin/wal_acceptor
 COPY --from=build /zenith/target/release/proxy /usr/local/bin
 COPY --from=pg-build /zenith/tmp_install postgres_install
 COPY docker-entrypoint.sh /docker-entrypoint.sh

Dockerfile.alpine

@@ -82,8 +82,6 @@ RUN apk add --update openssl build-base libseccomp-dev
 RUN apk --no-cache --update --repository https://dl-cdn.alpinelinux.org/alpine/edge/testing add rocksdb
 COPY --from=build /zenith/target/release/pageserver /usr/local/bin
 COPY --from=build /zenith/target/release/safekeeper /usr/local/bin
-# TODO: temporary alias for compatibility, see https://github.com/zenithdb/zenith/pull/740
-RUN ln -s /usr/local/bin/safekeeper /usr/local/bin/wal_acceptor
 COPY --from=build /zenith/target/release/proxy /usr/local/bin
 COPY --from=pg-build /zenith/tmp_install /usr/local
 COPY docker-entrypoint.sh /docker-entrypoint.sh

README.md

@@ -28,12 +28,12 @@ apt install build-essential libtool libreadline-dev zlib1g-dev flex bison libsec
 libssl-dev clang pkg-config libpq-dev
 ```
 
-[Rust] 1.52 or later is also required.
+[Rust] 1.55 or later is also required.
 
 To run the `psql` client, install the `postgresql-client` package or modify `PATH` and `LD_LIBRARY_PATH` to include `tmp_install/bin` and `tmp_install/lib`, respectively.
 
 To run the integration tests (not required to use the code), install
-Python (3.6 or higher), and install python3 packages with `pipenv` using `pipenv install` in the project directory.
+Python (3.7 or higher), and install python3 packages with `pipenv` using `pipenv install` in the project directory.
 
 2. Build zenith and patched postgres
 ```sh
@@ -47,17 +47,26 @@ make -j5
 # Create repository in .zenith with proper paths to binaries and data
 # Later that would be responsibility of a package install script
 > ./target/debug/zenith init
+initializing tenantid c03ba6b7ad4c5e9cf556f059ade44229
+created initial timeline 5b014a9e41b4b63ce1a1febc04503636 timeline.lsn 0/169C3C8
+created main branch
 pageserver init succeeded
 
-# start pageserver
+# start pageserver and safekeeper
 > ./target/debug/zenith start
-Starting pageserver at '127.0.0.1:64000' in .zenith
+Starting pageserver at 'localhost:64000' in '.zenith'
 Pageserver started
+initializing for single for 7676
+Starting safekeeper at 'localhost:5454' in '.zenith/safekeepers/single'
+Safekeeper started
 
-# start postgres on top on the pageserver
+# start postgres compute node
 > ./target/debug/zenith pg start main
-Starting postgres node at 'host=127.0.0.1 port=55432 user=stas'
+Starting new postgres main on main...
+Extracting base backup to create postgres instance: path=.zenith/pgdatadirs/tenants/c03ba6b7ad4c5e9cf556f059ade44229/main port=55432
+Starting postgres node at 'host=127.0.0.1 port=55432 user=zenith_admin dbname=postgres'
 waiting for server to start.... done
+server started
 
 # check list of running postgres instances
 > ./target/debug/zenith pg list
@@ -108,10 +117,9 @@ 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:
+6. If you want to run tests afterwards (see below), you have to stop all the running the pageserver, safekeeper and postgres instances
+   you have just started. You can stop them all with one command:
 ```sh
-> ./target/debug/zenith pg stop migration_check
-> ./target/debug/zenith pg stop main
 > ./target/debug/zenith stop
 ```
 

control_plane/Cargo.toml

@@ -18,7 +18,7 @@ regex = "1"
 anyhow = "1.0"
 thiserror = "1"
 bytes = "1.0.1"
-nix = "0.20"
+nix = "0.23"
 url = "2.2.2"
 hex = { version = "0.4.3", features = ["serde"] }
 reqwest = { version = "0.11", features = ["blocking", "json"] }

control_plane/safekeepers.conf

@@ -0,0 +1,20 @@
+# Page server and three safekeepers.
+[pageserver]
+pg_port = 64000
+http_port = 9898
+auth_type = 'Trust'
+
+[[safekeepers]]
+name = 'sk1'
+pg_port = 5454
+http_port = 7676
+
+[[safekeepers]]
+name = 'sk2'
+pg_port = 5455
+http_port = 7677
+
+[[safekeepers]]
+name = 'sk3'
+pg_port = 5456
+http_port = 7678

control_plane/simple.conf

@@ -0,0 +1,11 @@
+# Minimal zenith environment with one safekeeper. This is equivalent to the built-in
+# defaults that you get with no --config
+[pageserver]
+pg_port = 64000
+http_port = 9898
+auth_type = 'Trust'
+
+[[safekeepers]]
+name = 'single'
+pg_port = 5454
+http_port = 7676

control_plane/src/compute.rs

@@ -39,8 +39,6 @@ impl ComputeControlPlane {
     // |  |- <tenant_id>
     // |  |   |- <branch name>
     pub fn load(env: LocalEnv) -> Result<ComputeControlPlane> {
-        // TODO: since pageserver do not have config file yet we believe here that
-        // it is running on default port. Change that when pageserver will have config.
         let pageserver = Arc::new(PageServerNode::from_env(&env));
 
         let mut nodes = BTreeMap::default();
@@ -75,40 +73,59 @@ impl ComputeControlPlane {
             .unwrap_or(self.base_port)
     }
 
-    pub fn local(local_env: &LocalEnv, pageserver: &Arc<PageServerNode>) -> ComputeControlPlane {
-        ComputeControlPlane {
-            base_port: 65431,
-            pageserver: Arc::clone(pageserver),
-            nodes: BTreeMap::new(),
-            env: local_env.clone(),
+    // FIXME: see also parse_point_in_time in branches.rs.
+    fn parse_point_in_time(
+        &self,
+        tenantid: ZTenantId,
+        s: &str,
+    ) -> Result<(ZTimelineId, Option<Lsn>)> {
+        let mut strings = s.split('@');
+        let name = strings.next().unwrap();
+
+        let lsn: Option<Lsn>;
+        if let Some(lsnstr) = strings.next() {
+            lsn = Some(
+                Lsn::from_str(lsnstr)
+                    .with_context(|| "invalid LSN in point-in-time specification")?,
+            );
+        } else {
+            lsn = None
         }
+
+        // Resolve the timeline ID, given the human-readable branch name
+        let timeline_id = self
+            .pageserver
+            .branch_get_by_name(&tenantid, name)?
+            .timeline_id;
+
+        Ok((timeline_id, lsn))
     }
 
     pub fn new_node(
         &mut self,
         tenantid: ZTenantId,
-        branch_name: &str,
+        name: &str,
+        timeline_spec: &str,
         port: Option<u16>,
     ) -> Result<Arc<PostgresNode>> {
-        let timeline_id = self
-            .pageserver
-            .branch_get_by_name(&tenantid, branch_name)?
-            .timeline_id;
+        // Resolve the human-readable timeline spec into timeline ID and LSN
+        let (timelineid, lsn) = self.parse_point_in_time(tenantid, timeline_spec)?;
 
         let port = port.unwrap_or_else(|| self.get_port());
         let node = Arc::new(PostgresNode {
-            name: branch_name.to_owned(),
+            name: name.to_owned(),
             address: SocketAddr::new("127.0.0.1".parse().unwrap(), port),
             env: self.env.clone(),
             pageserver: Arc::clone(&self.pageserver),
             is_test: false,
-            timelineid: timeline_id,
+            timelineid,
+            lsn,
             tenantid,
             uses_wal_proposer: false,
         });
 
         node.create_pgdata()?;
-        node.setup_pg_conf(self.env.auth_type)?;
+        node.setup_pg_conf(self.env.pageserver.auth_type)?;
 
         self.nodes
             .insert((tenantid, node.name.clone()), Arc::clone(&node));
@@ -127,6 +144,7 @@ pub struct PostgresNode {
     pageserver: Arc<PageServerNode>,
     is_test: bool,
     pub timelineid: ZTimelineId,
+    pub lsn: Option<Lsn>, // if it's a read-only node. None for primary
     pub tenantid: ZTenantId,
     uses_wal_proposer: bool,
 }
@@ -161,9 +179,12 @@ impl PostgresNode {
         let port: u16 = conf.parse_field("port", &context)?;
         let timelineid: ZTimelineId = conf.parse_field("zenith.zenith_timeline", &context)?;
         let tenantid: ZTenantId = conf.parse_field("zenith.zenith_tenant", &context)?;
-
         let uses_wal_proposer = conf.get("wal_acceptors").is_some();
 
+        // parse recovery_target_lsn, if any
+        let recovery_target_lsn: Option<Lsn> =
+            conf.parse_field_optional("recovery_target_lsn", &context)?;
+
         // ok now
         Ok(PostgresNode {
             address: SocketAddr::new("127.0.0.1".parse().unwrap(), port),
@@ -172,12 +193,13 @@ impl PostgresNode {
             pageserver: Arc::clone(pageserver),
             is_test: false,
             timelineid,
+            lsn: recovery_target_lsn,
             tenantid,
             uses_wal_proposer,
         })
     }
 
-    fn sync_walkeepers(&self) -> Result<Lsn> {
+    fn sync_safekeepers(&self) -> Result<Lsn> {
         let pg_path = self.env.pg_bin_dir().join("postgres");
         let sync_handle = Command::new(pg_path)
             .arg("--sync-safekeepers")
@@ -202,7 +224,7 @@ impl PostgresNode {
         }
 
         let lsn = Lsn::from_str(std::str::from_utf8(&sync_output.stdout)?.trim())?;
-        println!("Walkeepers synced on {}", lsn);
+        println!("Safekeepers synced on {}", lsn);
         Ok(lsn)
     }
 
@@ -233,7 +255,7 @@ impl PostgresNode {
         // Read the archive directly from the `CopyOutReader`
         tar::Archive::new(copyreader)
             .unpack(&self.pgdata())
-            .with_context(|| "extracting page backup failed")?;
+            .with_context(|| "extracting base backup failed")?;
 
         Ok(())
     }
@@ -301,11 +323,30 @@ impl PostgresNode {
         conf.append("zenith.page_server_connstring", &pageserver_connstr);
         conf.append("zenith.zenith_tenant", &self.tenantid.to_string());
         conf.append("zenith.zenith_timeline", &self.timelineid.to_string());
+        if let Some(lsn) = self.lsn {
+            conf.append("recovery_target_lsn", &lsn.to_string());
+        }
         conf.append_line("");
 
-        // Configure the node to stream WAL directly to the pageserver
-        conf.append("synchronous_standby_names", "pageserver"); // TODO: add a new function arg?
-        conf.append("zenith.callmemaybe_connstring", &self.connstr());
+        if !self.env.safekeepers.is_empty() {
+            // Configure the node to connect to the safekeepers
+            conf.append("synchronous_standby_names", "walproposer");
+
+            let wal_acceptors = self
+                .env
+                .safekeepers
+                .iter()
+                .map(|sk| format!("localhost:{}", sk.pg_port))
+                .collect::<Vec<String>>()
+                .join(",");
+            conf.append("wal_acceptors", &wal_acceptors);
+        } else {
+            // Configure the node to stream WAL directly to the pageserver
+            // This isn't really a supported configuration, but can be useful for
+            // testing.
+            conf.append("synchronous_standby_names", "pageserver");
+            conf.append("zenith.callmemaybe_connstring", &self.connstr());
+        }
 
         let mut file = File::create(self.pgdata().join("postgresql.conf"))?;
         file.write_all(conf.to_string().as_bytes())?;
@@ -314,12 +355,14 @@ impl PostgresNode {
     }
 
     fn load_basebackup(&self) -> Result<()> {
-        let lsn = if self.uses_wal_proposer {
+        let backup_lsn = if let Some(lsn) = self.lsn {
+            Some(lsn)
+        } else if self.uses_wal_proposer {
             // LSN 0 means that it is bootstrap and we need to download just
             // latest data from the pageserver. That is a bit clumsy but whole bootstrap
             // procedure evolves quite actively right now, so let's think about it again
             // when things would be more stable (TODO).
-            let lsn = self.sync_walkeepers()?;
+            let lsn = self.sync_safekeepers()?;
             if lsn == Lsn(0) {
                 None
             } else {
@@ -329,7 +372,7 @@ impl PostgresNode {
             None
         };
 
-        self.do_basebackup(lsn)?;
+        self.do_basebackup(backup_lsn)?;
 
         Ok(())
     }
@@ -406,6 +449,10 @@ impl PostgresNode {
         // 3. Load basebackup
         self.load_basebackup()?;
 
+        if self.lsn.is_some() {
+            File::create(self.pgdata().join("standby.signal"))?;
+        }
+
         // 4. Finally start the compute node postgres
         println!("Starting postgres node at '{}'", self.connstr());
         self.pg_ctl(&["start"], auth_token)

control_plane/src/lib.rs

@@ -13,6 +13,7 @@ use std::path::Path;
 pub mod compute;
 pub mod local_env;
 pub mod postgresql_conf;
+pub mod safekeeper;
 pub mod storage;
 
 /// Read a PID file

control_plane/src/local_env.rs

@@ -7,46 +7,102 @@
 use anyhow::{Context, Result};
 use serde::{Deserialize, Serialize};
 use std::env;
+use std::fmt::Write;
 use std::fs;
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 use std::process::{Command, Stdio};
-use zenith_utils::auth::{encode_from_key_path, Claims, Scope};
+use zenith_utils::auth::{encode_from_key_file, Claims, Scope};
 use zenith_utils::postgres_backend::AuthType;
 use zenith_utils::zid::ZTenantId;
 
 //
-// This data structures represent deserialized zenith CLI config
+// This data structures represents zenith CLI config
+//
+// It is deserialized from the .zenith/config file, or the config file passed
+// to 'zenith init --config=<path>' option. See control_plane/simple.conf for
+// an example.
 //
 #[derive(Serialize, Deserialize, Clone, Debug)]
 pub struct LocalEnv {
-    // Pageserver connection settings
-    pub pageserver_pg_port: u16,
-    pub pageserver_http_port: u16,
-
-    // Base directory for both pageserver and compute nodes
+    // Base directory for all the nodes (the pageserver, safekeepers and
+    // compute nodes).
+    //
+    // This is not stored in the config file. Rather, this is the path where the
+    // config file itself is. It is read from the ZENITH_REPO_DIR env variable or
+    // '.zenith' if not given.
+    #[serde(skip)]
     pub base_data_dir: PathBuf,
 
     // Path to postgres distribution. It's expected that "bin", "include",
     // "lib", "share" from postgres distribution are there. If at some point
     // in time we will be able to run against vanilla postgres we may split that
     // to four separate paths and match OS-specific installation layout.
+    #[serde(default)]
     pub pg_distrib_dir: PathBuf,
 
     // Path to pageserver binary.
+    #[serde(default)]
     pub zenith_distrib_dir: PathBuf,
 
-    // keeping tenant id in config to reduce copy paste when running zenith locally with single tenant
-    #[serde(with = "hex")]
-    pub tenantid: ZTenantId,
+    // Default tenant ID to use with the 'zenith' command line utility, when
+    // --tenantid is not explicitly specified.
+    #[serde(with = "opt_tenantid_serde")]
+    #[serde(default)]
+    pub default_tenantid: Option<ZTenantId>,
 
-    // jwt auth token used for communication with pageserver
-    pub auth_token: String,
+    // used to issue tokens during e.g pg start
+    #[serde(default)]
+    pub private_key_path: PathBuf,
+
+    pub pageserver: PageServerConf,
+
+    #[serde(default)]
+    pub safekeepers: Vec<SafekeeperConf>,
+}
+
+#[derive(Serialize, Deserialize, Clone, Debug)]
+#[serde(default)]
+pub struct PageServerConf {
+    // Pageserver connection settings
+    pub pg_port: u16,
+    pub http_port: u16,
 
     // used to determine which auth type is used
     pub auth_type: AuthType,
 
-    // used to issue tokens during e.g pg start
-    pub private_key_path: PathBuf,
+    // jwt auth token used for communication with pageserver
+    pub auth_token: String,
+}
+
+impl Default for PageServerConf {
+    fn default() -> Self {
+        Self {
+            pg_port: 0,
+            http_port: 0,
+            auth_type: AuthType::Trust,
+            auth_token: "".to_string(),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Clone, Debug)]
+#[serde(default)]
+pub struct SafekeeperConf {
+    pub name: String,
+    pub pg_port: u16,
+    pub http_port: u16,
+    pub sync: bool,
+}
+
+impl Default for SafekeeperConf {
+    fn default() -> Self {
+        Self {
+            name: "".to_string(),
+            pg_port: 0,
+            http_port: 0,
+            sync: true,
+        }
+    }
 }
 
 impl LocalEnv {
@@ -62,6 +118,10 @@ impl LocalEnv {
         Ok(self.zenith_distrib_dir.join("pageserver"))
     }
 
+    pub fn safekeeper_bin(&self) -> Result<PathBuf> {
+        Ok(self.zenith_distrib_dir.join("safekeeper"))
+    }
+
     pub fn pg_data_dirs_path(&self) -> PathBuf {
         self.base_data_dir.join("pgdatadirs").join("tenants")
     }
@@ -76,6 +136,187 @@ impl LocalEnv {
     pub fn pageserver_data_dir(&self) -> PathBuf {
         self.base_data_dir.clone()
     }
+
+    pub fn safekeeper_data_dir(&self, node_name: &str) -> PathBuf {
+        self.base_data_dir.join("safekeepers").join(node_name)
+    }
+
+    /// Create a LocalEnv from a config file.
+    ///
+    /// Unlike 'load_config', this function fills in any defaults that are missing
+    /// from the config file.
+    pub fn create_config(toml: &str) -> Result<LocalEnv> {
+        let mut env: LocalEnv = toml::from_str(toml)?;
+
+        // Find postgres binaries.
+        // Follow POSTGRES_DISTRIB_DIR if set, otherwise look in "tmp_install".
+        if env.pg_distrib_dir == Path::new("") {
+            if let Some(postgres_bin) = env::var_os("POSTGRES_DISTRIB_DIR") {
+                env.pg_distrib_dir = postgres_bin.into();
+            } else {
+                let cwd = env::current_dir()?;
+                env.pg_distrib_dir = cwd.join("tmp_install")
+            }
+        }
+        if !env.pg_distrib_dir.join("bin/postgres").exists() {
+            anyhow::bail!(
+                "Can't find postgres binary at {}",
+                env.pg_distrib_dir.display()
+            );
+        }
+
+        // Find zenith binaries.
+        if env.zenith_distrib_dir == Path::new("") {
+            env.zenith_distrib_dir = env::current_exe()?.parent().unwrap().to_owned();
+        }
+        if !env.zenith_distrib_dir.join("pageserver").exists() {
+            anyhow::bail!("Can't find pageserver binary.");
+        }
+        if !env.zenith_distrib_dir.join("safekeeper").exists() {
+            anyhow::bail!("Can't find safekeeper binary.");
+        }
+
+        // If no initial tenant ID was given, generate it.
+        if env.default_tenantid.is_none() {
+            env.default_tenantid = Some(ZTenantId::generate());
+        }
+
+        env.base_data_dir = base_path();
+
+        Ok(env)
+    }
+
+    /// Locate and load config
+    pub fn load_config() -> Result<LocalEnv> {
+        let repopath = base_path();
+
+        if !repopath.exists() {
+            anyhow::bail!(
+                "Zenith config is not found in {}. You need to run 'zenith init' first",
+                repopath.to_str().unwrap()
+            );
+        }
+
+        // TODO: check that it looks like a zenith repository
+
+        // load and parse file
+        let config = fs::read_to_string(repopath.join("config"))?;
+        let mut env: LocalEnv = toml::from_str(config.as_str())?;
+
+        env.base_data_dir = repopath;
+
+        Ok(env)
+    }
+
+    // this function is used only for testing purposes in CLI e g generate tokens during init
+    pub fn generate_auth_token(&self, claims: &Claims) -> Result<String> {
+        let private_key_path = if self.private_key_path.is_absolute() {
+            self.private_key_path.to_path_buf()
+        } else {
+            self.base_data_dir.join(&self.private_key_path)
+        };
+
+        let key_data = fs::read(private_key_path)?;
+        encode_from_key_file(claims, &key_data)
+    }
+
+    //
+    // Initialize a new Zenith repository
+    //
+    pub fn init(&mut self) -> Result<()> {
+        // check if config already exists
+        let base_path = &self.base_data_dir;
+        if base_path == Path::new("") {
+            anyhow::bail!("repository base path is missing");
+        }
+        if base_path.exists() {
+            anyhow::bail!(
+                "directory '{}' already exists. Perhaps already initialized?",
+                base_path.to_str().unwrap()
+            );
+        }
+
+        fs::create_dir(&base_path)?;
+
+        // generate keys for jwt
+        // openssl genrsa -out private_key.pem 2048
+        let private_key_path;
+        if self.private_key_path == PathBuf::new() {
+            private_key_path = base_path.join("auth_private_key.pem");
+            let keygen_output = Command::new("openssl")
+                .arg("genrsa")
+                .args(&["-out", private_key_path.to_str().unwrap()])
+                .arg("2048")
+                .stdout(Stdio::null())
+                .output()
+                .with_context(|| "failed to generate auth private key")?;
+            if !keygen_output.status.success() {
+                anyhow::bail!(
+                    "openssl failed: '{}'",
+                    String::from_utf8_lossy(&keygen_output.stderr)
+                );
+            }
+            self.private_key_path = Path::new("auth_private_key.pem").to_path_buf();
+
+            let public_key_path = base_path.join("auth_public_key.pem");
+            // openssl rsa -in private_key.pem -pubout -outform PEM -out public_key.pem
+            let keygen_output = Command::new("openssl")
+                .arg("rsa")
+                .args(&["-in", private_key_path.to_str().unwrap()])
+                .arg("-pubout")
+                .args(&["-outform", "PEM"])
+                .args(&["-out", public_key_path.to_str().unwrap()])
+                .stdout(Stdio::null())
+                .output()
+                .with_context(|| "failed to generate auth private key")?;
+            if !keygen_output.status.success() {
+                anyhow::bail!(
+                    "openssl failed: '{}'",
+                    String::from_utf8_lossy(&keygen_output.stderr)
+                );
+            }
+        }
+
+        self.pageserver.auth_token =
+            self.generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
+
+        fs::create_dir_all(self.pg_data_dirs_path())?;
+
+        for safekeeper in self.safekeepers.iter() {
+            fs::create_dir_all(self.safekeeper_data_dir(&safekeeper.name))?;
+        }
+
+        let mut conf_content = String::new();
+
+        // Currently, the user first passes a config file with 'zenith init --config=<path>'
+        // We read that in, in `create_config`, and fill any missing defaults. Then it's saved
+        // to .zenith/config. TODO: We lose any formatting and comments along the way, which is
+        // a bit sad.
+        write!(
+            &mut conf_content,
+            r#"# This file describes a locale deployment of the page server
+# and safekeeeper node. It is read by the 'zenith' command-line
+# utility.
+"#
+        )?;
+
+        // Convert the LocalEnv to a toml file.
+        //
+        // This could be as simple as this:
+        //
+        // conf_content += &toml::to_string_pretty(env)?;
+        //
+        // But it results in a "values must be emitted before tables". I'm not sure
+        // why, AFAICS the table, i.e. 'safekeepers: Vec<SafekeeperConf>' is last.
+        // Maybe rust reorders the fields to squeeze avoid padding or something?
+        // In any case, converting to toml::Value first, and serializing that, works.
+        // See https://github.com/alexcrichton/toml-rs/issues/142
+        conf_content += &toml::to_string_pretty(&toml::Value::try_from(&self)?)?;
+
+        fs::write(base_path.join("config"), conf_content)?;
+
+        Ok(())
+    }
 }
 
 fn base_path() -> PathBuf {
@@ -85,118 +326,29 @@ fn base_path() -> PathBuf {
     }
 }
 
-//
-// Initialize a new Zenith repository
-//
-pub fn init(
-    pageserver_pg_port: u16,
-    pageserver_http_port: u16,
-    tenantid: ZTenantId,
-    auth_type: AuthType,
-) -> Result<()> {
-    // check if config already exists
-    let base_path = base_path();
-    if base_path.exists() {
-        anyhow::bail!(
-            "{} already exists. Perhaps already initialized?",
-            base_path.to_str().unwrap()
-        );
+/// Serde routines for Option<ZTenantId>. The serialized form is a hex string.
+mod opt_tenantid_serde {
+    use serde::{Deserialize, Deserializer, Serialize, Serializer};
+    use std::str::FromStr;
+    use zenith_utils::zid::ZTenantId;
+
+    pub fn serialize<S>(tenantid: &Option<ZTenantId>, ser: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        tenantid.map(|t| t.to_string()).serialize(ser)
     }
-    fs::create_dir(&base_path)?;
 
-    // ok, now check that expected binaries are present
-
-    // Find postgres binaries. Follow POSTGRES_DISTRIB_DIR if set, otherwise look in "tmp_install".
-    let pg_distrib_dir: PathBuf = {
-        if let Some(postgres_bin) = env::var_os("POSTGRES_DISTRIB_DIR") {
-            postgres_bin.into()
-        } else {
-            let cwd = env::current_dir()?;
-            cwd.join("tmp_install")
+    pub fn deserialize<'de, D>(des: D) -> Result<Option<ZTenantId>, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let s: Option<String> = Option::deserialize(des)?;
+        if let Some(s) = s {
+            return Ok(Some(
+                ZTenantId::from_str(&s).map_err(serde::de::Error::custom)?,
+            ));
         }
-    };
-    if !pg_distrib_dir.join("bin/postgres").exists() {
-        anyhow::bail!("Can't find postgres binary at {:?}", pg_distrib_dir);
+        Ok(None)
     }
-
-    // generate keys for jwt
-    // openssl genrsa -out private_key.pem 2048
-    let private_key_path = base_path.join("auth_private_key.pem");
-    let keygen_output = Command::new("openssl")
-        .arg("genrsa")
-        .args(&["-out", private_key_path.to_str().unwrap()])
-        .arg("2048")
-        .stdout(Stdio::null())
-        .output()
-        .with_context(|| "failed to generate auth private key")?;
-    if !keygen_output.status.success() {
-        anyhow::bail!(
-            "openssl failed: '{}'",
-            String::from_utf8_lossy(&keygen_output.stderr)
-        );
-    }
-
-    let public_key_path = base_path.join("auth_public_key.pem");
-    // openssl rsa -in private_key.pem -pubout -outform PEM -out public_key.pem
-    let keygen_output = Command::new("openssl")
-        .arg("rsa")
-        .args(&["-in", private_key_path.to_str().unwrap()])
-        .arg("-pubout")
-        .args(&["-outform", "PEM"])
-        .args(&["-out", public_key_path.to_str().unwrap()])
-        .stdout(Stdio::null())
-        .output()
-        .with_context(|| "failed to generate auth private key")?;
-    if !keygen_output.status.success() {
-        anyhow::bail!(
-            "openssl failed: '{}'",
-            String::from_utf8_lossy(&keygen_output.stderr)
-        );
-    }
-
-    let auth_token =
-        encode_from_key_path(&Claims::new(None, Scope::PageServerApi), &private_key_path)?;
-
-    // Find zenith binaries.
-    let zenith_distrib_dir = env::current_exe()?.parent().unwrap().to_owned();
-    if !zenith_distrib_dir.join("pageserver").exists() {
-        anyhow::bail!("Can't find pageserver binary.",);
-    }
-
-    let conf = LocalEnv {
-        pageserver_pg_port,
-        pageserver_http_port,
-        pg_distrib_dir,
-        zenith_distrib_dir,
-        base_data_dir: base_path,
-        tenantid,
-        auth_token,
-        auth_type,
-        private_key_path,
-    };
-
-    fs::create_dir_all(conf.pg_data_dirs_path())?;
-
-    let toml = toml::to_string_pretty(&conf)?;
-    fs::write(conf.base_data_dir.join("config"), toml)?;
-
-    Ok(())
-}
-
-// Locate and load config
-pub fn load_config() -> Result<LocalEnv> {
-    let repopath = base_path();
-
-    if !repopath.exists() {
-        anyhow::bail!(
-            "Zenith config is not found in {}. You need to run 'zenith init' first",
-            repopath.to_str().unwrap()
-        );
-    }
-
-    // TODO: check that it looks like a zenith repository
-
-    // load and parse file
-    let config = fs::read_to_string(repopath.join("config"))?;
-    toml::from_str(config.as_str()).map_err(|e| e.into())
 }

control_plane/src/postgresql_conf.rs

@@ -83,6 +83,22 @@ impl PostgresConf {
             .with_context(|| format!("could not parse '{}' option {}", field_name, context))
     }
 
+    pub fn parse_field_optional<T>(&self, field_name: &str, context: &str) -> Result<Option<T>>
+    where
+        T: FromStr,
+        <T as FromStr>::Err: std::error::Error + Send + Sync + 'static,
+    {
+        if let Some(val) = self.get(field_name) {
+            let result = val
+                .parse::<T>()
+                .with_context(|| format!("could not parse '{}' option {}", field_name, context))?;
+
+            Ok(Some(result))
+        } else {
+            Ok(None)
+        }
+    }
+
     ///
     /// Note: if you call this multiple times for the same option, the config
     /// file will a line for each call. It would be nice to have a function

control_plane/src/safekeeper.rs

@@ -0,0 +1,277 @@
+use std::io::Write;
+use std::net::TcpStream;
+use std::path::PathBuf;
+use std::process::Command;
+use std::sync::Arc;
+use std::time::Duration;
+use std::{io, result, thread};
+
+use anyhow::bail;
+use nix::errno::Errno;
+use nix::sys::signal::{kill, Signal};
+use nix::unistd::Pid;
+use postgres::Config;
+use reqwest::blocking::{Client, RequestBuilder, Response};
+use reqwest::{IntoUrl, Method};
+use thiserror::Error;
+use zenith_utils::http::error::HttpErrorBody;
+use zenith_utils::postgres_backend::AuthType;
+
+use crate::local_env::{LocalEnv, SafekeeperConf};
+use crate::read_pidfile;
+use crate::storage::PageServerNode;
+use zenith_utils::connstring::connection_address;
+use zenith_utils::connstring::connection_host_port;
+
+#[derive(Error, Debug)]
+pub enum SafekeeperHttpError {
+    #[error("Reqwest error: {0}")]
+    Transport(#[from] reqwest::Error),
+
+    #[error("Error: {0}")]
+    Response(String),
+}
+
+type Result<T> = result::Result<T, SafekeeperHttpError>;
+
+pub trait ResponseErrorMessageExt: Sized {
+    fn error_from_body(self) -> Result<Self>;
+}
+
+impl ResponseErrorMessageExt for Response {
+    fn error_from_body(self) -> Result<Self> {
+        let status = self.status();
+        if !(status.is_client_error() || status.is_server_error()) {
+            return Ok(self);
+        }
+
+        // reqwest do not export it's error construction utility functions, so lets craft the message ourselves
+        let url = self.url().to_owned();
+        Err(SafekeeperHttpError::Response(
+            match self.json::<HttpErrorBody>() {
+                Ok(err_body) => format!("Error: {}", err_body.msg),
+                Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
+            },
+        ))
+    }
+}
+
+//
+// Control routines for safekeeper.
+//
+// Used in CLI and tests.
+//
+#[derive(Debug)]
+pub struct SafekeeperNode {
+    pub name: String,
+
+    pub conf: SafekeeperConf,
+
+    pub pg_connection_config: Config,
+    pub env: LocalEnv,
+    pub http_client: Client,
+    pub http_base_url: String,
+
+    pub pageserver: Arc<PageServerNode>,
+}
+
+impl SafekeeperNode {
+    pub fn from_env(env: &LocalEnv, conf: &SafekeeperConf) -> SafekeeperNode {
+        let pageserver = Arc::new(PageServerNode::from_env(env));
+
+        println!("initializing for {} for {}", conf.name, conf.http_port);
+
+        SafekeeperNode {
+            name: conf.name.clone(),
+            conf: conf.clone(),
+            pg_connection_config: Self::safekeeper_connection_config(conf.pg_port),
+            env: env.clone(),
+            http_client: Client::new(),
+            http_base_url: format!("http://localhost:{}/v1", conf.http_port),
+            pageserver,
+        }
+    }
+
+    /// Construct libpq connection string for connecting to this safekeeper.
+    fn safekeeper_connection_config(port: u16) -> Config {
+        // TODO safekeeper authentication not implemented yet
+        format!("postgresql://no_user@localhost:{}/no_db", port)
+            .parse()
+            .unwrap()
+    }
+
+    pub fn datadir_path(&self) -> PathBuf {
+        self.env.safekeeper_data_dir(&self.name)
+    }
+
+    pub fn pid_file(&self) -> PathBuf {
+        self.datadir_path().join("safekeeper.pid")
+    }
+
+    pub fn start(&self) -> anyhow::Result<()> {
+        print!(
+            "Starting safekeeper at '{}' in '{}'",
+            connection_address(&self.pg_connection_config),
+            self.datadir_path().display()
+        );
+        io::stdout().flush().unwrap();
+
+        // Configure connection to page server
+        //
+        // FIXME: We extract the host and port from the connection string instead of using
+        // the connection string directly, because the 'safekeeper' binary expects
+        // host:port format. That's a bit silly when we already have a full libpq connection
+        // string at hand.
+        let pageserver_conn = {
+            let (host, port) = connection_host_port(&self.pageserver.pg_connection_config);
+            format!("{}:{}", host, port)
+        };
+
+        let listen_pg = format!("localhost:{}", self.conf.pg_port);
+        let listen_http = format!("localhost:{}", self.conf.http_port);
+
+        let mut cmd: &mut Command = &mut Command::new(self.env.safekeeper_bin()?);
+        cmd = cmd
+            .args(&["-D", self.datadir_path().to_str().unwrap()])
+            .args(&["--listen-pg", &listen_pg])
+            .args(&["--listen-http", &listen_http])
+            .args(&["--pageserver", &pageserver_conn])
+            .args(&["--recall", "1 second"])
+            .arg("--daemonize")
+            .env_clear()
+            .env("RUST_BACKTRACE", "1");
+        if !self.conf.sync {
+            cmd = cmd.arg("--no-sync");
+        }
+
+        if self.env.pageserver.auth_type == AuthType::ZenithJWT {
+            cmd.env("PAGESERVER_AUTH_TOKEN", &self.env.pageserver.auth_token);
+        }
+
+        if !cmd.status()?.success() {
+            bail!(
+                "Safekeeper failed to start. See '{}' for details.",
+                self.datadir_path().join("safekeeper.log").display()
+            );
+        }
+
+        // It takes a while for the safekeeper to start up. Wait until it is
+        // open for business.
+        const RETRIES: i8 = 15;
+        for retries in 1..RETRIES {
+            match self.check_status() {
+                Ok(_) => {
+                    println!("\nSafekeeper started");
+                    return Ok(());
+                }
+                Err(err) => {
+                    match err {
+                        SafekeeperHttpError::Transport(err) => {
+                            if err.is_connect() && retries < 5 {
+                                print!(".");
+                                io::stdout().flush().unwrap();
+                            } else {
+                                if retries == 5 {
+                                    println!() // put a line break after dots for second message
+                                }
+                                println!(
+                                    "Safekeeper not responding yet, err {} retrying ({})...",
+                                    err, retries
+                                );
+                            }
+                        }
+                        SafekeeperHttpError::Response(msg) => {
+                            bail!("safekeeper failed to start: {} ", msg)
+                        }
+                    }
+                    thread::sleep(Duration::from_secs(1));
+                }
+            }
+        }
+        bail!("safekeeper failed to start in {} seconds", RETRIES);
+    }
+
+    ///
+    /// Stop the server.
+    ///
+    /// If 'immediate' is true, we use SIGQUIT, killing the process immediately.
+    /// Otherwise we use SIGTERM, triggering a clean shutdown
+    ///
+    /// If the server is not running, returns success
+    ///
+    pub fn stop(&self, immediate: bool) -> anyhow::Result<()> {
+        let pid_file = self.pid_file();
+        if !pid_file.exists() {
+            println!("Safekeeper {} is already stopped", self.name);
+            return Ok(());
+        }
+        let pid = read_pidfile(&pid_file)?;
+        let pid = Pid::from_raw(pid);
+
+        let sig = if immediate {
+            println!("Stop safekeeper immediately");
+            Signal::SIGQUIT
+        } else {
+            println!("Stop safekeeper gracefully");
+            Signal::SIGTERM
+        };
+        match kill(pid, sig) {
+            Ok(_) => (),
+            Err(Errno::ESRCH) => {
+                println!(
+                    "Safekeeper with pid {} does not exist, but a PID file was found",
+                    pid
+                );
+                return Ok(());
+            }
+            Err(err) => bail!(
+                "Failed to send signal to safekeeper with pid {}: {}",
+                pid,
+                err.desc()
+            ),
+        }
+
+        let address = connection_address(&self.pg_connection_config);
+
+        // TODO Remove this "timeout" and handle it on caller side instead.
+        // Shutting down may take a long time,
+        // if safekeeper flushes a lot of data
+        for _ in 0..100 {
+            if let Err(_e) = TcpStream::connect(&address) {
+                println!("Safekeeper stopped receiving connections");
+
+                //Now check status
+                match self.check_status() {
+                    Ok(_) => {
+                        println!("Safekeeper status is OK. Wait a bit.");
+                        thread::sleep(Duration::from_secs(1));
+                    }
+                    Err(err) => {
+                        println!("Safekeeper status is: {}", err);
+                        return Ok(());
+                    }
+                }
+            } else {
+                println!("Safekeeper still receives connections");
+                thread::sleep(Duration::from_secs(1));
+            }
+        }
+
+        bail!("Failed to stop safekeeper with pid {}", pid);
+    }
+
+    fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> RequestBuilder {
+        // TODO: authentication
+        //if self.env.auth_type == AuthType::ZenithJWT {
+        //    builder = builder.bearer_auth(&self.env.safekeeper_auth_token)
+        //}
+        self.http_client.request(method, url)
+    }
+
+    pub fn check_status(&self) -> Result<()> {
+        self.http_request(Method::GET, format!("{}/{}", self.http_base_url, "status"))
+            .send()?
+            .error_from_body()?;
+        Ok(())
+    }
+}

control_plane/src/storage.rs

@@ -6,6 +6,7 @@ use std::time::Duration;
 use std::{io, result, thread};
 
 use anyhow::{anyhow, bail};
+use nix::errno::Errno;
 use nix::sys::signal::{kill, Signal};
 use nix::unistd::Pid;
 use pageserver::http::models::{BranchCreateRequest, TenantCreateRequest};
@@ -20,6 +21,7 @@ use zenith_utils::zid::ZTenantId;
 use crate::local_env::LocalEnv;
 use crate::read_pidfile;
 use pageserver::branches::BranchInfo;
+use pageserver::tenant_mgr::TenantInfo;
 use zenith_utils::connstring::connection_address;
 
 #[derive(Error, Debug)]
@@ -62,7 +64,6 @@ impl ResponseErrorMessageExt for Response {
 //
 #[derive(Debug)]
 pub struct PageServerNode {
-    pub kill_on_exit: bool,
     pub pg_connection_config: Config,
     pub env: LocalEnv,
     pub http_client: Client,
@@ -71,34 +72,34 @@ pub struct PageServerNode {
 
 impl PageServerNode {
     pub fn from_env(env: &LocalEnv) -> PageServerNode {
-        let password = if env.auth_type == AuthType::ZenithJWT {
-            &env.auth_token
+        let password = if env.pageserver.auth_type == AuthType::ZenithJWT {
+            &env.pageserver.auth_token
         } else {
             ""
         };
 
         PageServerNode {
-            kill_on_exit: false,
             pg_connection_config: Self::pageserver_connection_config(
                 password,
-                env.pageserver_pg_port,
+                env.pageserver.pg_port,
             ),
             env: env.clone(),
             http_client: Client::new(),
-            http_base_url: format!("http://127.0.0.1:{}/v1", env.pageserver_http_port),
+            http_base_url: format!("http://localhost:{}/v1", env.pageserver.http_port),
         }
     }
 
+    /// Construct libpq connection string for connecting to the pageserver.
     fn pageserver_connection_config(password: &str, port: u16) -> Config {
-        format!("postgresql://no_user:{}@127.0.0.1:{}/no_db", password, port)
+        format!("postgresql://no_user:{}@localhost:{}/no_db", password, port)
             .parse()
             .unwrap()
     }
 
-    pub fn init(&self, create_tenant: Option<&str>, enable_auth: bool) -> anyhow::Result<()> {
+    pub fn init(&self, create_tenant: Option<&str>) -> anyhow::Result<()> {
         let mut cmd = Command::new(self.env.pageserver_bin()?);
-        let listen_pg = format!("127.0.0.1:{}", self.env.pageserver_pg_port);
-        let listen_http = format!("127.0.0.1:{}", self.env.pageserver_http_port);
+        let listen_pg = format!("localhost:{}", self.env.pageserver.pg_port);
+        let listen_http = format!("localhost:{}", self.env.pageserver.http_port);
         let mut args = vec![
             "--init",
             "-D",
@@ -111,10 +112,11 @@ impl PageServerNode {
             &listen_http,
         ];
 
-        if enable_auth {
+        let auth_type_str = &self.env.pageserver.auth_type.to_string();
+        if self.env.pageserver.auth_type != AuthType::Trust {
             args.extend(&["--auth-validation-public-key-path", "auth_public_key.pem"]);
-            args.extend(&["--auth-type", "ZenithJWT"]);
         }
+        args.extend(&["--auth-type", auth_type_str]);
 
         if let Some(tenantid) = create_tenant {
             args.extend(&["--create-tenant", tenantid])
@@ -152,7 +154,7 @@ impl PageServerNode {
 
         let mut cmd = Command::new(self.env.pageserver_bin()?);
         cmd.args(&["-D", self.repo_path().to_str().unwrap()])
-            .arg("-d")
+            .arg("--daemonize")
             .env_clear()
             .env("RUST_BACKTRACE", "1");
 
@@ -199,19 +201,43 @@ impl PageServerNode {
         bail!("pageserver failed to start in {} seconds", RETRIES);
     }
 
+    ///
+    /// Stop the server.
+    ///
+    /// If 'immediate' is true, we use SIGQUIT, killing the process immediately.
+    /// Otherwise we use SIGTERM, triggering a clean shutdown
+    ///
+    /// If the server is not running, returns success
+    ///
     pub fn stop(&self, immediate: bool) -> anyhow::Result<()> {
-        let pid = read_pidfile(&self.pid_file())?;
-        let pid = Pid::from_raw(pid);
-        if immediate {
+        let pid_file = self.pid_file();
+        if !pid_file.exists() {
+            println!("Pageserver is already stopped");
+            return Ok(());
+        }
+        let pid = Pid::from_raw(read_pidfile(&pid_file)?);
+
+        let sig = if immediate {
             println!("Stop pageserver immediately");
-            if kill(pid, Signal::SIGQUIT).is_err() {
-                bail!("Failed to kill pageserver with pid {}", pid);
-            }
+            Signal::SIGQUIT
         } else {
             println!("Stop pageserver gracefully");
-            if kill(pid, Signal::SIGTERM).is_err() {
-                bail!("Failed to stop pageserver with pid {}", pid);
+            Signal::SIGTERM
+        };
+        match kill(pid, sig) {
+            Ok(_) => (),
+            Err(Errno::ESRCH) => {
+                println!(
+                    "Pageserver with pid {} does not exist, but a PID file was found",
+                    pid
+                );
+                return Ok(());
             }
+            Err(err) => bail!(
+                "Failed to send signal to pageserver with pid {}: {}",
+                pid,
+                err.desc()
+            ),
         }
 
         let address = connection_address(&self.pg_connection_config);
@@ -256,8 +282,8 @@ impl PageServerNode {
 
     fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> RequestBuilder {
         let mut builder = self.http_client.request(method, url);
-        if self.env.auth_type == AuthType::ZenithJWT {
-            builder = builder.bearer_auth(&self.env.auth_token)
+        if self.env.pageserver.auth_type == AuthType::ZenithJWT {
+            builder = builder.bearer_auth(&self.env.pageserver.auth_token)
         }
         builder
     }
@@ -269,7 +295,7 @@ impl PageServerNode {
         Ok(())
     }
 
-    pub fn tenant_list(&self) -> Result<Vec<String>> {
+    pub fn tenant_list(&self) -> Result<Vec<TenantInfo>> {
         Ok(self
             .http_request(Method::GET, format!("{}/{}", self.http_base_url, "tenant"))
             .send()?
@@ -332,12 +358,3 @@ impl PageServerNode {
             .json()?)
     }
 }
-
-impl Drop for PageServerNode {
-    fn drop(&mut self) {
-        // TODO Looks like this flag is never set
-        if self.kill_on_exit {
-            let _ = self.stop(true);
-        }
-    }
-}

pageserver/Cargo.toml

@@ -17,7 +17,7 @@ lazy_static = "1.4.0"
 log = "0.4.14"
 clap = "2.33.0"
 daemonize = "0.4.1"
-tokio = { version = "1.11", features = ["process", "macros", "fs", "rt"] }
+tokio = { version = "1.11", features = ["process", "macros", "fs", "rt", "io-util"] }
 postgres-types = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
 postgres-protocol = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
 postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
@@ -37,6 +37,7 @@ async-trait = "0.1"
 const_format = "0.2.21"
 tracing = "0.1.27"
 signal-hook = {version = "0.3.10", features = ["extended-siginfo"] }
+url = "2"
 
 postgres_ffi = { path = "../postgres_ffi" }
 zenith_metrics = { path = "../zenith_metrics" }
@@ -45,3 +46,4 @@ workspace_hack = { path = "../workspace_hack" }
 
 [dev-dependencies]
 hex-literal = "0.3"
+tempfile = "3.2"

pageserver/README.md

@@ -41,7 +41,7 @@ Legend:
 +--+
 
 ....
-.  .   Component that we will need, but doesn't exist at the moment. A TODO.
+.  .   Component at its early development phase.
 ....
 
 --->   Data flow
@@ -116,13 +116,49 @@ Remove old on-disk layer files that are no longer needed according to the
 PITR retention policy
 
 
-TODO: Backup service
---------------------
+### Backup service
 
-The backup service is responsible for periodically pushing the chunks to S3.
+The backup service, responsible for storing pageserver recovery data externally.
 
-TODO: How/when do restore from S3? Whenever we get a GetPage@LSN request for
-a chunk we don't currently have? Or when an external Control Plane tells us?
+Currently, pageserver stores its files in a filesystem directory it's pointed to.
+That working directory could be rather ephemeral for such cases as "a pageserver pod running in k8s with no persistent volumes attached".
+Therefore, the server interacts with external, more reliable storage to back up and restore its state.
+
+The code for storage support is extensible and can support arbitrary ones as long as they implement a certain Rust trait.
+There are the following implementations present:
+* local filesystem — to use in tests mainly
+* AWS S3           - to use in production
+
+Implementation details are covered in the [storage readme](./src/relish_storage/README.md) and corresponding Rust file docs.
+
+The backup service is disabled by default and can be enabled to interact with a single remote storage.
+
+CLI examples:
+* Local FS: `${PAGESERVER_BIN} --relish-storage-local-path="/some/local/path/"`
+* AWS S3  : `${PAGESERVER_BIN} --relish-storage-s3-bucket="some-sample-bucket" --relish-storage-region="eu-north-1" --relish-storage-access-key="SOMEKEYAAAAASADSAH*#" --relish-storage-secret-access-key="SOMEsEcReTsd292v"`
+
+For Amazon AWS S3, a key id and secret access key could be located in `~/.aws/credentials` if awscli was ever configured to work with the desired bucket, on the AWS Settings page for a certain user. Also note, that the bucket names does not contain any protocols when used on AWS.
+For local S3 installations, refer to the their documentation for name format and credentials.
+
+Similar to other pageserver settings, toml config file can be used to configure either of the storages as backup backup targets.
+Required sections are:
+
+```toml
+[relish_storage]
+local_path = '/Users/someonetoignore/Downloads/tmp_dir/'
+```
+
+or
+
+```toml
+[relish_storage]
+bucket_name = 'some-sample-bucket'
+bucket_region = 'eu-north-1'
+access_key_id = 'SOMEKEYAAAAASADSAH*#'
+secret_access_key = 'SOMEsEcReTsd292v'
+```
+
+Also, `AWS_SECRET_ACCESS_KEY` and `AWS_ACCESS_KEY_ID` variables can be used to specify the credentials instead of any of the ways above.
 
 TODO: Sharding
 --------------------

pageserver/src/basebackup.rs

@@ -13,6 +13,7 @@
 use anyhow::Result;
 use bytes::{BufMut, BytesMut};
 use log::*;
+use std::fmt::Write as FmtWrite;
 use std::io;
 use std::io::Write;
 use std::sync::Arc;
@@ -83,7 +84,7 @@ impl<'a> Basebackup<'a> {
 
         info!(
             "taking basebackup lsn={}, prev_lsn={}",
-            backup_prev, backup_lsn
+            backup_lsn, backup_prev
         );
 
         Ok(Basebackup {
@@ -248,13 +249,7 @@ impl<'a> Basebackup<'a> {
         let mut pg_control = ControlFileData::decode(&pg_control_bytes)?;
         let mut checkpoint = CheckPoint::decode(&checkpoint_bytes)?;
 
-        // Generate new pg_control and WAL needed for bootstrap
-        let checkpoint_segno = self.lsn.segment_number(pg_constants::WAL_SEGMENT_SIZE);
-        let checkpoint_lsn = XLogSegNoOffsetToRecPtr(
-            checkpoint_segno,
-            XLOG_SIZE_OF_XLOG_LONG_PHD as u32,
-            pg_constants::WAL_SEGMENT_SIZE,
-        );
+        // Generate new pg_control needed for bootstrap
         checkpoint.redo = normalize_lsn(self.lsn, pg_constants::WAL_SEGMENT_SIZE).0;
 
         //reset some fields we don't want to preserve
@@ -263,19 +258,24 @@ impl<'a> Basebackup<'a> {
         checkpoint.oldestActiveXid = 0;
 
         //save new values in pg_control
-        pg_control.checkPoint = checkpoint_lsn;
+        pg_control.checkPoint = 0;
         pg_control.checkPointCopy = checkpoint;
         pg_control.state = pg_constants::DB_SHUTDOWNED;
 
         // add zenith.signal file
-        let xl_prev = if self.prev_record_lsn == Lsn(0) {
-            0xBAD0 // magic value to indicate that we don't know prev_lsn
+        let mut zenith_signal = String::new();
+        if self.prev_record_lsn == Lsn(0) {
+            if self.lsn == self.timeline.get_ancestor_lsn() {
+                write!(zenith_signal, "PREV LSN: none")?;
+            } else {
+                write!(zenith_signal, "PREV LSN: invalid")?;
+            }
         } else {
-            self.prev_record_lsn.0
-        };
+            write!(zenith_signal, "PREV LSN: {}", self.prev_record_lsn)?;
+        }
         self.ar.append(
-            &new_tar_header("zenith.signal", 8)?,
-            &xl_prev.to_le_bytes()[..],
+            &new_tar_header("zenith.signal", zenith_signal.len() as u64)?,
+            zenith_signal.as_bytes(),
         )?;
 
         //send pg_control
@@ -284,14 +284,15 @@ impl<'a> Basebackup<'a> {
         self.ar.append(&header, &pg_control_bytes[..])?;
 
         //send wal segment
+        let segno = self.lsn.segment_number(pg_constants::WAL_SEGMENT_SIZE);
         let wal_file_name = XLogFileName(
             1, // FIXME: always use Postgres timeline 1
-            checkpoint_segno,
+            segno,
             pg_constants::WAL_SEGMENT_SIZE,
         );
         let wal_file_path = format!("pg_wal/{}", wal_file_name);
         let header = new_tar_header(&wal_file_path, pg_constants::WAL_SEGMENT_SIZE as u64)?;
-        let wal_seg = generate_wal_segment(&pg_control);
+        let wal_seg = generate_wal_segment(segno, pg_control.system_identifier);
         assert!(wal_seg.len() == pg_constants::WAL_SEGMENT_SIZE);
         self.ar.append(&header, &wal_seg[..])?;
         Ok(())

pageserver/src/bin/pageserver.rs

@@ -2,17 +2,15 @@
 // Main entry point for the Page Server executable
 //
 
-use pageserver::defaults::*;
 use serde::{Deserialize, Serialize};
 use std::{
     env,
-    net::TcpListener,
     path::{Path, PathBuf},
     str::FromStr,
     thread,
 };
 use tracing::*;
-use zenith_utils::{auth::JwtAuth, logging, postgres_backend::AuthType};
+use zenith_utils::{auth::JwtAuth, logging, postgres_backend::AuthType, tcp_listener};
 
 use anyhow::{bail, ensure, Context, Result};
 use signal_hook::consts::signal::*;
@@ -28,13 +26,8 @@ use clap::{App, Arg, ArgMatches};
 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,
+    branches, defaults::*, http, page_service, relish_storage, tenant_mgr, PageServerConf,
+    RelishStorageConfig, RelishStorageKind, S3Config, LOG_FILE_NAME,
 };
 use zenith_utils::http::endpoint;
 use zenith_utils::postgres_backend;
@@ -42,7 +35,7 @@ use zenith_utils::postgres_backend;
 use const_format::formatcp;
 
 /// String arguments that can be declared via CLI or config file
-#[derive(Serialize, Deserialize)]
+#[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
 struct CfgFileParams {
     listen_pg_addr: Option<String>,
     listen_http_addr: Option<String>,
@@ -53,12 +46,21 @@ struct CfgFileParams {
     pg_distrib_dir: Option<String>,
     auth_validation_public_key_path: Option<String>,
     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>,
+    /////////////////////////////////
+    //// Don't put `Option<String>` and other "simple" values below.
+    ////
+    /// `Option<RelishStorage>` is a <a href='https://toml.io/en/v1.0.0#table'>table</a> in TOML.
+    /// Values in TOML cannot be defined after tables (other tables can),
+    /// and [`toml`] crate serializes all fields in the order of their appearance.
+    ////////////////////////////////
+    relish_storage: Option<RelishStorage>,
 }
 
-#[derive(Serialize, Deserialize, Clone)]
+#[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
+// Without this attribute, enums with values won't be serialized by the `toml` library (but can be deserialized nonetheless!).
+// See https://github.com/alexcrichton/toml-rs/blob/6c162e6562c3e432bf04c82a3d1d789d80761a86/examples/enum_external.rs for the examples
+#[serde(untagged)]
 enum RelishStorage {
     Local {
         local_path: String,
@@ -477,13 +479,13 @@ fn start_pageserver(conf: &'static PageServerConf) -> Result<()> {
         "Starting pageserver http handler on {}",
         conf.listen_http_addr
     );
-    let http_listener = TcpListener::bind(conf.listen_http_addr.clone())?;
+    let http_listener = tcp_listener::bind(conf.listen_http_addr.clone())?;
 
     info!(
         "Starting pageserver pg protocol handler on {}",
         conf.listen_pg_addr
     );
-    let pageserver_listener = TcpListener::bind(conf.listen_pg_addr.clone())?;
+    let pageserver_listener = tcp_listener::bind(conf.listen_pg_addr.clone())?;
 
     if conf.daemonize {
         info!("daemonizing...");
@@ -552,7 +554,7 @@ fn start_pageserver(conf: &'static PageServerConf) -> Result<()> {
                 info!("Got SIGQUIT. Terminate pageserver in immediate shutdown mode");
                 exit(111);
             }
-            SIGTERM => {
+            SIGINT | SIGTERM => {
                 info!("Got SIGINT/SIGTERM. Terminate gracefully in fast shutdown mode");
                 // Terminate postgres backends
                 postgres_backend::set_pgbackend_shutdown_requested();
@@ -577,11 +579,142 @@ fn start_pageserver(conf: &'static PageServerConf) -> Result<()> {
                 info!("Pageserver shut down successfully completed");
                 exit(0);
             }
-            _ => {
-                debug!("Unknown signal.");
+            unknown_signal => {
+                debug!("Unknown signal {}", unknown_signal);
             }
         }
     }
 
     Ok(())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn page_server_conf_toml_serde() {
+        let params = CfgFileParams {
+            listen_pg_addr: Some("listen_pg_addr_VALUE".to_string()),
+            listen_http_addr: Some("listen_http_addr_VALUE".to_string()),
+            checkpoint_distance: Some("checkpoint_distance_VALUE".to_string()),
+            checkpoint_period: Some("checkpoint_period_VALUE".to_string()),
+            gc_horizon: Some("gc_horizon_VALUE".to_string()),
+            gc_period: Some("gc_period_VALUE".to_string()),
+            pg_distrib_dir: Some("pg_distrib_dir_VALUE".to_string()),
+            auth_validation_public_key_path: Some(
+                "auth_validation_public_key_path_VALUE".to_string(),
+            ),
+            auth_type: Some("auth_type_VALUE".to_string()),
+            relish_storage: Some(RelishStorage::Local {
+                local_path: "relish_storage_local_VALUE".to_string(),
+            }),
+            relish_storage_max_concurrent_sync: Some(
+                "relish_storage_max_concurrent_sync_VALUE".to_string(),
+            ),
+        };
+
+        let toml_string = toml::to_string(&params).expect("Failed to serialize correct config");
+        let toml_pretty_string =
+            toml::to_string_pretty(&params).expect("Failed to serialize correct config");
+        assert_eq!(
+            r#"listen_pg_addr = 'listen_pg_addr_VALUE'
+listen_http_addr = 'listen_http_addr_VALUE'
+checkpoint_distance = 'checkpoint_distance_VALUE'
+checkpoint_period = 'checkpoint_period_VALUE'
+gc_horizon = 'gc_horizon_VALUE'
+gc_period = 'gc_period_VALUE'
+pg_distrib_dir = 'pg_distrib_dir_VALUE'
+auth_validation_public_key_path = 'auth_validation_public_key_path_VALUE'
+auth_type = 'auth_type_VALUE'
+relish_storage_max_concurrent_sync = 'relish_storage_max_concurrent_sync_VALUE'
+
+[relish_storage]
+local_path = 'relish_storage_local_VALUE'
+"#,
+            toml_pretty_string
+        );
+
+        let params_from_serialized: CfgFileParams = toml::from_str(&toml_string)
+            .expect("Failed to deserialize the serialization result of the config");
+        let params_from_serialized_pretty: CfgFileParams = toml::from_str(&toml_pretty_string)
+            .expect("Failed to deserialize the prettified serialization result of the config");
+        assert!(
+            params_from_serialized == params,
+            "Expected the same config in the end of config -> serialize -> deserialize chain"
+        );
+        assert!(
+            params_from_serialized_pretty == params,
+            "Expected the same config in the end of config -> serialize pretty -> deserialize chain"
+        );
+    }
+
+    #[test]
+    fn credentials_omitted_during_serialization() {
+        let params = CfgFileParams {
+            listen_pg_addr: Some("listen_pg_addr_VALUE".to_string()),
+            listen_http_addr: Some("listen_http_addr_VALUE".to_string()),
+            checkpoint_distance: Some("checkpoint_distance_VALUE".to_string()),
+            checkpoint_period: Some("checkpoint_period_VALUE".to_string()),
+            gc_horizon: Some("gc_horizon_VALUE".to_string()),
+            gc_period: Some("gc_period_VALUE".to_string()),
+            pg_distrib_dir: Some("pg_distrib_dir_VALUE".to_string()),
+            auth_validation_public_key_path: Some(
+                "auth_validation_public_key_path_VALUE".to_string(),
+            ),
+            auth_type: Some("auth_type_VALUE".to_string()),
+            relish_storage: Some(RelishStorage::AwsS3 {
+                bucket_name: "bucket_name_VALUE".to_string(),
+                bucket_region: "bucket_region_VALUE".to_string(),
+                access_key_id: Some("access_key_id_VALUE".to_string()),
+                secret_access_key: Some("secret_access_key_VALUE".to_string()),
+            }),
+            relish_storage_max_concurrent_sync: Some(
+                "relish_storage_max_concurrent_sync_VALUE".to_string(),
+            ),
+        };
+
+        let toml_string = toml::to_string(&params).expect("Failed to serialize correct config");
+        let toml_pretty_string =
+            toml::to_string_pretty(&params).expect("Failed to serialize correct config");
+        assert_eq!(
+            r#"listen_pg_addr = 'listen_pg_addr_VALUE'
+listen_http_addr = 'listen_http_addr_VALUE'
+checkpoint_distance = 'checkpoint_distance_VALUE'
+checkpoint_period = 'checkpoint_period_VALUE'
+gc_horizon = 'gc_horizon_VALUE'
+gc_period = 'gc_period_VALUE'
+pg_distrib_dir = 'pg_distrib_dir_VALUE'
+auth_validation_public_key_path = 'auth_validation_public_key_path_VALUE'
+auth_type = 'auth_type_VALUE'
+relish_storage_max_concurrent_sync = 'relish_storage_max_concurrent_sync_VALUE'
+
+[relish_storage]
+bucket_name = 'bucket_name_VALUE'
+bucket_region = 'bucket_region_VALUE'
+"#,
+            toml_pretty_string
+        );
+
+        let params_from_serialized: CfgFileParams = toml::from_str(&toml_string)
+            .expect("Failed to deserialize the serialization result of the config");
+        let params_from_serialized_pretty: CfgFileParams = toml::from_str(&toml_pretty_string)
+            .expect("Failed to deserialize the prettified serialization result of the config");
+
+        let mut expected_params = params;
+        expected_params.relish_storage = Some(RelishStorage::AwsS3 {
+            bucket_name: "bucket_name_VALUE".to_string(),
+            bucket_region: "bucket_region_VALUE".to_string(),
+            access_key_id: None,
+            secret_access_key: None,
+        });
+        assert!(
+            params_from_serialized == expected_params,
+            "Expected the config without credentials in the end of a 'config -> serialize -> deserialize' chain"
+        );
+        assert!(
+            params_from_serialized_pretty == expected_params,
+            "Expected the config without credentials in the end of a 'config -> serialize pretty -> deserialize' chain"
+        );
+    }
+}

pageserver/src/branches.rs

@@ -4,7 +4,7 @@
 // TODO: move all paths construction to conf impl
 //
 
-use anyhow::{bail, ensure, Context, Result};
+use anyhow::{bail, Context, Result};
 use postgres_ffi::ControlFileData;
 use serde::{Deserialize, Serialize};
 use std::{
@@ -23,6 +23,7 @@ use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 use crate::tenant_mgr;
 use crate::walredo::WalRedoManager;
+use crate::CheckpointConfig;
 use crate::{repository::Repository, PageServerConf};
 use crate::{restore_local_repo, LOG_FILE_NAME};
 
@@ -35,7 +36,7 @@ pub struct BranchInfo {
     pub ancestor_id: Option<String>,
     pub ancestor_lsn: Option<String>,
     pub current_logical_size: usize,
-    pub current_logical_size_non_incremental: usize,
+    pub current_logical_size_non_incremental: Option<usize>,
 }
 
 impl BranchInfo {
@@ -44,6 +45,7 @@ impl BranchInfo {
         conf: &PageServerConf,
         tenantid: &ZTenantId,
         repo: &Arc<dyn Repository>,
+        include_non_incremental_logical_size: bool,
     ) -> Result<Self> {
         let name = path
             .as_ref()
@@ -78,6 +80,14 @@ impl BranchInfo {
             );
         }
 
+        // non incremental size calculation can be heavy, so let it be optional
+        // needed for tests to check size calculation
+        let current_logical_size_non_incremental = include_non_incremental_logical_size
+            .then(|| {
+                timeline.get_current_logical_size_non_incremental(timeline.get_last_record_lsn())
+            })
+            .transpose()?;
+
         Ok(BranchInfo {
             name,
             timeline_id,
@@ -85,8 +95,7 @@ impl BranchInfo {
             ancestor_id,
             ancestor_lsn,
             current_logical_size: timeline.get_current_logical_size(),
-            current_logical_size_non_incremental: timeline
-                .get_current_logical_size_non_incremental(timeline.get_last_record_lsn())?,
+            current_logical_size_non_incremental,
         })
     }
 }
@@ -230,7 +239,7 @@ fn bootstrap_timeline(
         timeline.writer().as_ref(),
         lsn,
     )?;
-    timeline.checkpoint()?;
+    timeline.checkpoint(CheckpointConfig::Forced)?;
 
     println!(
         "created initial timeline {} timeline.lsn {}",
@@ -248,19 +257,11 @@ fn bootstrap_timeline(
     Ok(())
 }
 
-pub(crate) fn get_tenants(conf: &PageServerConf) -> Result<Vec<String>> {
-    let tenants_dir = conf.tenants_path();
-
-    std::fs::read_dir(&tenants_dir)?
-        .map(|dir_entry_res| {
-            let dir_entry = dir_entry_res?;
-            ensure!(dir_entry.file_type()?.is_dir());
-            Ok(dir_entry.file_name().to_str().unwrap().to_owned())
-        })
-        .collect()
-}
-
-pub(crate) fn get_branches(conf: &PageServerConf, tenantid: &ZTenantId) -> Result<Vec<BranchInfo>> {
+pub(crate) fn get_branches(
+    conf: &PageServerConf,
+    tenantid: &ZTenantId,
+    include_non_incremental_logical_size: bool,
+) -> Result<Vec<BranchInfo>> {
     let repo = tenant_mgr::get_repository_for_tenant(*tenantid)?;
 
     // Each branch has a corresponding record (text file) in the refs/branches
@@ -270,7 +271,13 @@ pub(crate) fn get_branches(conf: &PageServerConf, tenantid: &ZTenantId) -> Resul
     std::fs::read_dir(&branches_dir)?
         .map(|dir_entry_res| {
             let dir_entry = dir_entry_res?;
-            BranchInfo::from_path(dir_entry.path(), conf, tenantid, &repo)
+            BranchInfo::from_path(
+                dir_entry.path(),
+                conf,
+                tenantid,
+                &repo,
+                include_non_incremental_logical_size,
+            )
         })
         .collect()
 }
@@ -332,7 +339,7 @@ pub(crate) fn create_branch(
         ancestor_id: None,
         ancestor_lsn: None,
         current_logical_size: 0,
-        current_logical_size_non_incremental: 0,
+        current_logical_size_non_incremental: Some(0),
     })
 }
 

pageserver/src/http/openapi_spec.yml

@@ -25,6 +25,11 @@ paths:
         schema:
           type: string
           format: hex
+      - name: include-non-incremental-logical-size
+        in: query
+        schema:
+          type: string
+          description: Controls calculation of current_logical_size_non_incremental
     get:
       description: Get branches for tenant
       responses:
@@ -73,6 +78,11 @@ paths:
         required: true
         schema:
           type: string
+      - name: include-non-incremental-logical-size
+        in: query
+        schema:
+          type: string
+          description: Controls calculation of current_logical_size_non_incremental
     get:
       description: Get branches for tenant
       responses:
@@ -164,13 +174,13 @@ paths:
       description: Get tenants list
       responses:
         "200":
-          description: OK
+          description: TenantInfo
           content:
             application/json:
               schema:
                 type: array
                 items:
-                  type: string
+                  $ref: "#/components/schemas/TenantInfo"
         "401":
           description: Unauthorized Error
           content:
@@ -243,6 +253,16 @@ components:
       scheme: bearer
       bearerFormat: JWT
   schemas:
+    TenantInfo:
+      type: object
+      required:
+        - id
+        - state
+      properties:
+        id:
+          type: string
+        state:
+          type: string
     BranchInfo:
       type: object
       required:
@@ -250,7 +270,6 @@ components:
         - timeline_id
         - latest_valid_lsn
         - current_logical_size
-        - current_logical_size_non_incremental
       properties:
         name:
           type: string

pageserver/src/http/routes.rs

@@ -86,31 +86,59 @@ async fn branch_create_handler(mut request: Request<Body>) -> Result<Response<Bo
     Ok(json_response(StatusCode::CREATED, response_data)?)
 }
 
+// Gate non incremental logical size calculation behind a flag
+// after pgbench -i -s100 calculation took 28ms so if multiplied by the number of timelines
+// and tenants it can take noticeable amount of time. Also the value currently used only in tests
+fn get_include_non_incremental_logical_size(request: &Request<Body>) -> bool {
+    request
+        .uri()
+        .query()
+        .map(|v| {
+            url::form_urlencoded::parse(v.as_bytes())
+                .into_owned()
+                .any(|(param, _)| param == "include-non-incremental-logical-size")
+        })
+        .unwrap_or(false)
+}
+
 async fn branch_list_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenantid: ZTenantId = parse_request_param(&request, "tenant_id")?;
 
+    let include_non_incremental_logical_size = get_include_non_incremental_logical_size(&request);
+
     check_permission(&request, Some(tenantid))?;
 
     let response_data = tokio::task::spawn_blocking(move || {
         let _enter = info_span!("branch_list", tenant = %tenantid).entered();
-        crate::branches::get_branches(get_config(&request), &tenantid)
+        crate::branches::get_branches(
+            get_config(&request),
+            &tenantid,
+            include_non_incremental_logical_size,
+        )
     })
     .await
     .map_err(ApiError::from_err)??;
     Ok(json_response(StatusCode::OK, response_data)?)
 }
 
-// TODO add to swagger
 async fn branch_detail_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenantid: ZTenantId = parse_request_param(&request, "tenant_id")?;
     let branch_name: String = get_request_param(&request, "branch_name")?.to_string();
     let conf = get_state(&request).conf;
     let path = conf.branch_path(&branch_name, &tenantid);
 
+    let include_non_incremental_logical_size = get_include_non_incremental_logical_size(&request);
+
     let response_data = tokio::task::spawn_blocking(move || {
         let _enter = info_span!("branch_detail", tenant = %tenantid, branch=%branch_name).entered();
         let repo = tenant_mgr::get_repository_for_tenant(tenantid)?;
-        BranchInfo::from_path(path, conf, &tenantid, &repo)
+        BranchInfo::from_path(
+            path,
+            conf,
+            &tenantid,
+            &repo,
+            include_non_incremental_logical_size,
+        )
     })
     .await
     .map_err(ApiError::from_err)??;
@@ -124,7 +152,7 @@ async fn tenant_list_handler(request: Request<Body>) -> Result<Response<Body>, A
 
     let response_data = tokio::task::spawn_blocking(move || {
         let _enter = info_span!("tenant_list").entered();
-        crate::branches::get_tenants(get_config(&request))
+        crate::tenant_mgr::list_tenants()
     })
     .await
     .map_err(ApiError::from_err)??;

pageserver/src/layered_repository.rs

@@ -16,13 +16,11 @@ use bookfile::Book;
 use bytes::Bytes;
 use lazy_static::lazy_static;
 use postgres_ffi::pg_constants::BLCKSZ;
-use serde::{Deserialize, Serialize};
 use tracing::*;
 
 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::io::Write;
@@ -30,9 +28,9 @@ use std::ops::{Bound::Included, Deref};
 use std::path::{Path, PathBuf};
 use std::sync::atomic::{AtomicUsize, Ordering};
 use std::sync::{Arc, Mutex, MutexGuard};
-use std::thread::JoinHandle;
 use std::time::{Duration, Instant};
 
+use self::metadata::{metadata_path, TimelineMetadata};
 use crate::relish::*;
 use crate::relish_storage::schedule_timeline_upload;
 use crate::repository::{GcResult, Repository, Timeline, TimelineWriter, WALRecord};
@@ -40,6 +38,7 @@ use crate::tenant_mgr;
 use crate::walreceiver;
 use crate::walreceiver::IS_WAL_RECEIVER;
 use crate::walredo::WalRedoManager;
+use crate::CheckpointConfig;
 use crate::PageServerConf;
 use crate::{ZTenantId, ZTimelineId};
 
@@ -47,7 +46,6 @@ use zenith_metrics::{
     register_histogram, register_int_gauge_vec, Histogram, IntGauge, IntGaugeVec,
 };
 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;
@@ -59,6 +57,7 @@ mod image_layer;
 mod inmemory_layer;
 mod interval_tree;
 mod layer_map;
+pub mod metadata;
 mod page_versions;
 mod storage_layer;
 
@@ -111,8 +110,9 @@ 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";
+/// Parts of the `.zenith/tenants/<tenantid>/timelines/<timelineid>` directory prefix.
+pub const TENANTS_SEGMENT_NAME: &str = "tenants";
+pub const TIMELINES_SEGMENT_NAME: &str = "timelines";
 
 ///
 /// Repository consists of multiple timelines. Keep them in a hash table.
@@ -142,12 +142,7 @@ impl Repository for LayeredRepository {
         // Create the timeline directory, and write initial metadata to file.
         crashsafe_dir::create_dir_all(self.conf.timeline_path(&timelineid, &self.tenantid))?;
 
-        let metadata = TimelineMetadata {
-            disk_consistent_lsn: Lsn(0),
-            prev_record_lsn: None,
-            ancestor_timeline: None,
-            ancestor_lsn: Lsn(0),
-        };
+        let metadata = TimelineMetadata::new(Lsn(0), None, None, Lsn(0));
         Self::save_metadata(self.conf, timelineid, self.tenantid, &metadata, true)?;
 
         let timeline = LayeredTimeline::new(
@@ -186,12 +181,7 @@ impl Repository for LayeredRepository {
         // Create the metadata file, noting the ancestor of the new timeline.
         // There is initially no data in it, but all the read-calls know to look
         // into the ancestor.
-        let metadata = TimelineMetadata {
-            disk_consistent_lsn: start_lsn,
-            prev_record_lsn: dst_prev,
-            ancestor_timeline: Some(src),
-            ancestor_lsn: start_lsn,
-        };
+        let metadata = TimelineMetadata::new(start_lsn, dst_prev, Some(src), start_lsn);
         crashsafe_dir::create_dir_all(self.conf.timeline_path(&dst, &self.tenantid))?;
         Self::save_metadata(self.conf, dst, self.tenantid, &metadata, true)?;
 
@@ -216,6 +206,22 @@ impl Repository for LayeredRepository {
             })
     }
 
+    fn checkpoint_iteration(&self, cconf: CheckpointConfig) -> Result<()> {
+        {
+            let timelines = self.timelines.lock().unwrap();
+
+            for (timelineid, timeline) in timelines.iter() {
+                let _entered =
+                    info_span!("checkpoint", timeline = %timelineid, tenant = %self.tenantid)
+                        .entered();
+
+                timeline.checkpoint(cconf)?;
+            }
+        }
+
+        Ok(())
+    }
+
     // Wait for all threads to complete and persist repository data before pageserver shutdown.
     fn shutdown(&self) -> Result<()> {
         trace!("LayeredRepository shutdown for tenant {}", self.tenantid);
@@ -225,7 +231,7 @@ impl Repository for LayeredRepository {
             walreceiver::stop_wal_receiver(*timelineid);
             // Wait for syncing data to disk
             trace!("repo shutdown. checkpoint timeline {}", timelineid);
-            timeline.checkpoint()?;
+            timeline.checkpoint(CheckpointConfig::Forced)?;
 
             //TODO Wait for walredo process to shutdown too
         }
@@ -247,14 +253,14 @@ 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;
+                let disk_consistent_lsn = metadata.disk_consistent_lsn();
 
                 // Recurse to look up the ancestor timeline.
                 //
                 // TODO: If you have a very deep timeline history, this could become
                 // expensive. Perhaps delay this until we need to look up a page in
                 // ancestor.
-                let ancestor = if let Some(ancestor_timelineid) = metadata.ancestor_timeline {
+                let ancestor = if let Some(ancestor_timelineid) = metadata.ancestor_timeline() {
                     Some(self.get_timeline_locked(ancestor_timelineid, timelines)?)
                 } else {
                     None
@@ -266,7 +272,7 @@ impl LayeredRepository {
 
                 let mut timeline = LayeredTimeline::new(
                     self.conf,
-                    metadata,
+                    metadata.clone(),
                     ancestor,
                     timelineid,
                     self.tenantid,
@@ -276,15 +282,9 @@ impl LayeredRepository {
                 )?;
 
                 // List the layers on disk, and load them into the layer map
-                let _loaded_layers = timeline.load_layer_map(disk_consistent_lsn)?;
+                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,
-                    // );
+                    schedule_timeline_upload(self.tenantid, timelineid, loaded_layers, metadata);
                 }
 
                 // needs to be after load_layer_map
@@ -312,90 +312,6 @@ impl LayeredRepository {
         }
     }
 
-    ///
-    /// Launch the checkpointer thread in given repository.
-    ///
-    pub fn launch_checkpointer_thread(
-        conf: &'static PageServerConf,
-        rc: Arc<LayeredRepository>,
-    ) -> JoinHandle<()> {
-        std::thread::Builder::new()
-            .name("Checkpointer thread".into())
-            .spawn(move || {
-                // FIXME: relaunch it? Panic is not good.
-                rc.checkpoint_loop(conf).expect("Checkpointer thread died");
-            })
-            .unwrap()
-    }
-
-    ///
-    /// Checkpointer thread's main loop
-    ///
-    fn checkpoint_loop(&self, conf: &'static PageServerConf) -> Result<()> {
-        while !tenant_mgr::shutdown_requested() {
-            std::thread::sleep(conf.checkpoint_period);
-            info!("checkpointer thread for tenant {} waking up", self.tenantid);
-
-            // checkpoint timelines that have accumulated more than CHECKPOINT_DISTANCE
-            // bytes of WAL since last checkpoint.
-            {
-                let timelines = self.timelines.lock().unwrap();
-                for (timelineid, timeline) in timelines.iter() {
-                    let _entered =
-                        info_span!("checkpoint", timeline = %timelineid, tenant = %self.tenantid)
-                            .entered();
-
-                    STORAGE_TIME
-                        .with_label_values(&["checkpoint_timed"])
-                        .observe_closure_duration(|| {
-                            timeline.checkpoint_internal(conf.checkpoint_distance, false)
-                        })?
-                }
-                // release lock on 'timelines'
-            }
-        }
-        trace!("Checkpointer thread shut down");
-        Ok(())
-    }
-
-    ///
-    /// Launch the GC thread in given repository.
-    ///
-    pub fn launch_gc_thread(
-        conf: &'static PageServerConf,
-        rc: Arc<LayeredRepository>,
-    ) -> JoinHandle<()> {
-        std::thread::Builder::new()
-            .name("GC thread".into())
-            .spawn(move || {
-                // FIXME: relaunch it? Panic is not good.
-                rc.gc_loop(conf).expect("GC thread died");
-            })
-            .unwrap()
-    }
-
-    ///
-    /// GC thread's main loop
-    ///
-    fn gc_loop(&self, conf: &'static PageServerConf) -> Result<()> {
-        while !tenant_mgr::shutdown_requested() {
-            // Garbage collect old files that are not needed for PITR anymore
-            if conf.gc_horizon > 0 {
-                self.gc_iteration(None, conf.gc_horizon, false).unwrap();
-            }
-
-            // TODO Write it in more adequate way using
-            // condvar.wait_timeout() or something
-            let mut sleep_time = conf.gc_period.as_secs();
-            while sleep_time > 0 && !tenant_mgr::shutdown_requested() {
-                sleep_time -= 1;
-                std::thread::sleep(Duration::from_secs(1));
-            }
-            info!("gc thread for tenant {} waking up", self.tenantid);
-        }
-        Ok(())
-    }
-
     /// Save timeline metadata to file
     fn save_metadata(
         conf: &'static PageServerConf,
@@ -412,13 +328,7 @@ impl LayeredRepository {
             .create_new(first_save)
             .open(&path)?;
 
-        let mut metadata_bytes = 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));
+        let metadata_bytes = data.to_bytes().context("Failed to get metadata bytes")?;
 
         if file.write(&metadata_bytes)? != metadata_bytes.len() {
             bail!("Could not write all the metadata bytes in a single call");
@@ -445,20 +355,7 @@ impl LayeredRepository {
     ) -> 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 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)?;
-        assert!(data.disk_consistent_lsn.is_aligned());
-
-        Ok(data)
+        TimelineMetadata::from_bytes(&metadata_bytes)
     }
 
     //
@@ -568,7 +465,7 @@ impl LayeredRepository {
                 // so that they too can be garbage collected. That's
                 // used in tests, so we want as deterministic results as possible.
                 if checkpoint_before_gc {
-                    timeline.checkpoint()?;
+                    timeline.checkpoint(CheckpointConfig::Forced)?;
                     info!("timeline {} checkpoint_before_gc done", timelineid);
                 }
 
@@ -583,29 +480,6 @@ impl LayeredRepository {
     }
 }
 
-/// Metadata stored on disk for each timeline
-///
-/// The fields correspond to the values we hold in memory, in LayeredTimeline.
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct TimelineMetadata {
-    disk_consistent_lsn: Lsn,
-
-    // This is only set if we know it. We track it in memory when the page
-    // server is running, but we only track the value corresponding to
-    // 'last_record_lsn', not 'disk_consistent_lsn' which can lag behind by a
-    // lot. We only store it in the metadata file when we flush *all* the
-    // in-memory data so that 'last_record_lsn' is the same as
-    // 'disk_consistent_lsn'.  That's OK, because after page server restart, as
-    // soon as we reprocess at least one record, we will have a valid
-    // 'prev_record_lsn' value in memory again. This is only really needed when
-    // doing a clean shutdown, so that there is no more WAL beyond
-    // 'disk_consistent_lsn'
-    prev_record_lsn: Option<Lsn>,
-
-    ancestor_timeline: Option<ZTimelineId>,
-    ancestor_lsn: Lsn,
-}
-
 pub struct LayeredTimeline {
     conf: &'static PageServerConf,
 
@@ -678,6 +552,10 @@ pub struct LayeredTimeline {
 
 /// Public interface functions
 impl Timeline for LayeredTimeline {
+    fn get_ancestor_lsn(&self) -> Lsn {
+        self.ancestor_lsn
+    }
+
     /// 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
@@ -691,8 +569,8 @@ impl Timeline for LayeredTimeline {
             .wait_for_timeout(lsn, TIMEOUT)
             .with_context(|| {
                 format!(
-                    "Timed out while waiting for WAL record at LSN {} to arrive",
-                    lsn
+                    "Timed out while waiting for WAL record at LSN {} to arrive, disk consistent LSN={}",
+                    lsn, self.get_disk_consistent_lsn()
                 )
             })?;
 
@@ -848,11 +726,15 @@ impl Timeline for LayeredTimeline {
     /// Public entry point for checkpoint(). All the logic is in the private
     /// checkpoint_internal function, this public facade just wraps it for
     /// metrics collection.
-    fn checkpoint(&self) -> Result<()> {
-        STORAGE_TIME
-            .with_label_values(&["checkpoint_force"])
-            //pass checkpoint_distance=0 to force checkpoint
-            .observe_closure_duration(|| self.checkpoint_internal(0, true))
+    fn checkpoint(&self, cconf: CheckpointConfig) -> Result<()> {
+        match cconf {
+            CheckpointConfig::Forced => STORAGE_TIME
+                .with_label_values(&["forced checkpoint"])
+                .observe_closure_duration(|| self.checkpoint_internal(0)),
+            CheckpointConfig::Distance(distance) => STORAGE_TIME
+                .with_label_values(&["checkpoint"])
+                .observe_closure_duration(|| self.checkpoint_internal(distance)),
+        }
     }
 
     fn get_last_record_lsn(&self) -> Lsn {
@@ -906,6 +788,10 @@ impl Timeline for LayeredTimeline {
         Ok(total_blocks * BLCKSZ as usize)
     }
 
+    fn get_disk_consistent_lsn(&self) -> Lsn {
+        self.disk_consistent_lsn.load()
+    }
+
     fn writer<'a>(&'a self) -> Box<dyn TimelineWriter + 'a> {
         Box::new(LayeredTimelineWriter {
             tl: self,
@@ -942,13 +828,13 @@ impl LayeredTimeline {
 
             // initialize in-memory 'last_record_lsn' from 'disk_consistent_lsn'.
             last_record_lsn: SeqWait::new(RecordLsn {
-                last: metadata.disk_consistent_lsn,
-                prev: metadata.prev_record_lsn.unwrap_or(Lsn(0)),
+                last: metadata.disk_consistent_lsn(),
+                prev: metadata.prev_record_lsn().unwrap_or(Lsn(0)),
             }),
-            disk_consistent_lsn: AtomicLsn::new(metadata.disk_consistent_lsn.0),
+            disk_consistent_lsn: AtomicLsn::new(metadata.disk_consistent_lsn().0),
 
             ancestor_timeline: ancestor,
-            ancestor_lsn: metadata.ancestor_lsn,
+            ancestor_lsn: metadata.ancestor_lsn(),
             current_logical_size: AtomicUsize::new(current_logical_size),
             current_logical_size_gauge,
             upload_relishes,
@@ -1227,7 +1113,7 @@ impl LayeredTimeline {
     /// Flush to disk all data that was written with the put_* functions
     ///
     /// NOTE: This has nothing to do with checkpoint in PostgreSQL.
-    fn checkpoint_internal(&self, checkpoint_distance: u64, forced: bool) -> Result<()> {
+    fn checkpoint_internal(&self, checkpoint_distance: u64) -> Result<()> {
         let mut write_guard = self.write_lock.lock().unwrap();
         let mut layers = self.layers.lock().unwrap();
 
@@ -1258,10 +1144,6 @@ impl LayeredTimeline {
         while let Some((oldest_layer, oldest_generation)) = layers.peek_oldest_open() {
             let oldest_pending_lsn = oldest_layer.get_oldest_pending_lsn();
 
-            if tenant_mgr::shutdown_requested() && !forced {
-                return Ok(());
-            }
-
             // Does this layer need freezing?
             //
             // Write out all in-memory layers that contain WAL older than CHECKPOINT_DISTANCE.
@@ -1340,50 +1222,48 @@ impl LayeredTimeline {
             timeline_dir.sync_all()?;
         }
 
-        // 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.
+        // If we were able to advance 'disk_consistent_lsn', save it the metadata file.
+        // After crash, we will restart WAL streaming and processing from that point.
+        let old_disk_consistent_lsn = self.disk_consistent_lsn.load();
+        if disk_consistent_lsn != old_disk_consistent_lsn {
+            assert!(disk_consistent_lsn > old_disk_consistent_lsn);
 
-        // We can only save a valid 'prev_record_lsn' value on disk if we
-        // flushed *all* in-memory changes to disk. We only track
-        // 'prev_record_lsn' in memory for the latest processed record, so we
-        // don't remember what the correct value that corresponds to some old
-        // LSN is. But if we flush everything, then the value corresponding
-        // current 'last_record_lsn' is correct and we can store it on disk.
-        let ondisk_prev_record_lsn = if disk_consistent_lsn == last_record_lsn {
-            Some(prev_record_lsn)
-        } else {
-            None
-        };
+            // We can only save a valid 'prev_record_lsn' value on disk if we
+            // flushed *all* in-memory changes to disk. We only track
+            // 'prev_record_lsn' in memory for the latest processed record, so we
+            // don't remember what the correct value that corresponds to some old
+            // LSN is. But if we flush everything, then the value corresponding
+            // current 'last_record_lsn' is correct and we can store it on disk.
+            let ondisk_prev_record_lsn = if disk_consistent_lsn == last_record_lsn {
+                Some(prev_record_lsn)
+            } else {
+                None
+            };
 
-        let ancestor_timelineid = self.ancestor_timeline.as_ref().map(|x| x.timelineid);
+            let ancestor_timelineid = self.ancestor_timeline.as_ref().map(|x| x.timelineid);
 
-        let metadata = TimelineMetadata {
-            disk_consistent_lsn,
-            prev_record_lsn: ondisk_prev_record_lsn,
-            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 = TimelineMetadata::new(
+                disk_consistent_lsn,
+                ondisk_prev_record_lsn,
+                ancestor_timelineid,
+                self.ancestor_lsn,
+            );
+
+            LayeredRepository::save_metadata(
+                self.conf,
+                self.timelineid,
+                self.tenantid,
+                &metadata,
+                false,
+            )?;
+            if self.upload_relishes {
+                schedule_timeline_upload(self.tenantid, self.timelineid, layer_uploads, metadata);
+            }
+
+            // Also update the in-memory copy
+            self.disk_consistent_lsn.store(disk_consistent_lsn);
         }
 
-        // Also update the in-memory copy
-        self.disk_consistent_lsn.store(disk_consistent_lsn);
-
         Ok(())
     }
 
@@ -1935,15 +1815,6 @@ 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<()> {

pageserver/src/layered_repository/delta_layer.rs

@@ -442,12 +442,7 @@ impl DeltaLayer {
     }
 
     fn open_book(&self) -> Result<(PathBuf, Book<File>)> {
-        let path = Self::path_for(
-            &self.path_or_conf,
-            self.timelineid,
-            self.tenantid,
-            &self.layer_name(),
-        );
+        let path = self.path();
 
         let file = File::open(&path)?;
         let book = Book::new(file)?;

pageserver/src/layered_repository/filename.rs

@@ -13,7 +13,7 @@ use anyhow::Result;
 use log::*;
 use zenith_utils::lsn::Lsn;
 
-use super::METADATA_FILE_NAME;
+use super::metadata::METADATA_FILE_NAME;
 
 // Note: LayeredTimeline::load_layer_map() relies on this sort order
 #[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone)]

pageserver/src/layered_repository/layer_map.rs

@@ -17,6 +17,7 @@ use anyhow::Result;
 use lazy_static::lazy_static;
 use std::cmp::Ordering;
 use std::collections::{BinaryHeap, HashMap};
+use std::sync::atomic::{self, AtomicU64};
 use std::sync::Arc;
 use zenith_metrics::{register_int_gauge, IntGauge};
 use zenith_utils::lsn::Lsn;
@@ -30,6 +31,17 @@ lazy_static! {
             .expect("failed to define a metric");
 }
 
+static NEXT_LAYER_ID: AtomicU64 = AtomicU64::new(0);
+
+#[derive(PartialEq, Eq, Hash, Clone, Copy)]
+pub struct LayerId(u64);
+
+impl LayerId {
+    fn next() -> LayerId {
+        Self(NEXT_LAYER_ID.fetch_add(1, atomic::Ordering::Relaxed))
+    }
+}
+
 ///
 /// LayerMap tracks what layers exist on a timeline.
 ///
@@ -43,6 +55,8 @@ pub struct LayerMap {
     /// contains the oldest WAL record.
     open_layers: BinaryHeap<OpenLayerEntry>,
 
+    open_layers_by_id: HashMap<LayerId, Arc<InMemoryLayer>>,
+
     /// Generation number, used to distinguish newly inserted entries in the
     /// binary heap from older entries during checkpoint.
     current_generation: u64,
@@ -71,10 +85,15 @@ impl LayerMap {
         segentry.open.as_ref().map(Arc::clone)
     }
 
+    #[allow(dead_code)]
+    pub fn get_open_by_id(&self, layer_id: &LayerId) -> Option<Arc<InMemoryLayer>> {
+        self.open_layers_by_id.get(layer_id).cloned()
+    }
+
     ///
     /// Insert an open in-memory layer
     ///
-    pub fn insert_open(&mut self, layer: Arc<InMemoryLayer>) {
+    pub fn insert_open(&mut self, layer: Arc<InMemoryLayer>) -> LayerId {
         let segentry = self.segs.entry(layer.get_seg_tag()).or_default();
 
         segentry.update_open(Arc::clone(&layer));
@@ -86,15 +105,23 @@ impl LayerMap {
         // in the middle of a WAL record.
         assert!(oldest_pending_lsn.is_aligned());
 
+        let id = LayerId::next();
+
         // Also add it to the binary heap
         let open_layer_entry = OpenLayerEntry {
             oldest_pending_lsn: layer.get_oldest_pending_lsn(),
-            layer,
+            layer: Arc::clone(&layer),
             generation: self.current_generation,
+            id,
         };
         self.open_layers.push(open_layer_entry);
 
+        let old_layer = self.open_layers_by_id.insert(id, layer);
+        assert!(old_layer.is_none());
+
         NUM_INMEMORY_LAYERS.inc();
+
+        id
     }
 
     /// Remove the oldest in-memory layer
@@ -114,6 +141,8 @@ impl LayerMap {
             assert!(oldest_entry.layer.is_dropped());
         }
 
+        self.open_layers_by_id.remove(&oldest_entry.id).unwrap();
+
         NUM_INMEMORY_LAYERS.dec();
     }
 
@@ -319,6 +348,7 @@ struct OpenLayerEntry {
     pub oldest_pending_lsn: Lsn, // copy of layer.get_oldest_pending_lsn()
     pub generation: u64,
     pub layer: Arc<InMemoryLayer>,
+    id: LayerId,
 }
 impl Ord for OpenLayerEntry {
     fn cmp(&self, other: &Self) -> Ordering {

pageserver/src/layered_repository/metadata.rs

@@ -0,0 +1,202 @@
+//! Every image of a certain timeline from [`crate::layered_repository::LayeredRepository`]
+//! has a metadata that needs to be stored persistently.
+//!
+//! Later, the file gets is used in [`crate::relish_storage::storage_sync`] as a part of
+//! external storage import and export operations.
+//!
+//! The module contains all structs and related helper methods related to timeline metadata.
+
+use std::{convert::TryInto, path::PathBuf};
+
+use anyhow::ensure;
+use zenith_utils::{
+    bin_ser::BeSer,
+    lsn::Lsn,
+    zid::{ZTenantId, ZTimelineId},
+};
+
+use crate::{
+    layered_repository::{METADATA_CHECKSUM_SIZE, METADATA_MAX_DATA_SIZE, METADATA_MAX_SAFE_SIZE},
+    PageServerConf,
+};
+
+/// The name of the metadata file pageserver creates per timeline.
+pub const METADATA_FILE_NAME: &str = "metadata";
+
+/// Metadata stored on disk for each timeline
+///
+/// The fields correspond to the values we hold in memory, in LayeredTimeline.
+#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord)]
+pub struct TimelineMetadata {
+    disk_consistent_lsn: Lsn,
+    // This is only set if we know it. We track it in memory when the page
+    // server is running, but we only track the value corresponding to
+    // 'last_record_lsn', not 'disk_consistent_lsn' which can lag behind by a
+    // lot. We only store it in the metadata file when we flush *all* the
+    // in-memory data so that 'last_record_lsn' is the same as
+    // 'disk_consistent_lsn'.  That's OK, because after page server restart, as
+    // soon as we reprocess at least one record, we will have a valid
+    // 'prev_record_lsn' value in memory again. This is only really needed when
+    // doing a clean shutdown, so that there is no more WAL beyond
+    // 'disk_consistent_lsn'
+    prev_record_lsn: Option<Lsn>,
+    ancestor_timeline: Option<ZTimelineId>,
+    ancestor_lsn: Lsn,
+}
+
+/// Points to a place in pageserver's local directory,
+/// where certain timeline's metadata file should be located.
+pub fn metadata_path(
+    conf: &'static PageServerConf,
+    timelineid: ZTimelineId,
+    tenantid: ZTenantId,
+) -> PathBuf {
+    conf.timeline_path(&timelineid, &tenantid)
+        .join(METADATA_FILE_NAME)
+}
+
+impl TimelineMetadata {
+    pub fn new(
+        disk_consistent_lsn: Lsn,
+        prev_record_lsn: Option<Lsn>,
+        ancestor_timeline: Option<ZTimelineId>,
+        ancestor_lsn: Lsn,
+    ) -> Self {
+        Self {
+            disk_consistent_lsn,
+            prev_record_lsn,
+            ancestor_timeline,
+            ancestor_lsn,
+        }
+    }
+
+    pub fn from_bytes(metadata_bytes: &[u8]) -> anyhow::Result<Self> {
+        ensure!(
+            metadata_bytes.len() == METADATA_MAX_SAFE_SIZE,
+            "metadata bytes size is wrong"
+        );
+
+        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,
+            "metadata checksum mismatch"
+        );
+
+        let data = TimelineMetadata::from(serialize::DeTimelineMetadata::des_prefix(data)?);
+        assert!(data.disk_consistent_lsn.is_aligned());
+
+        Ok(data)
+    }
+
+    pub fn to_bytes(&self) -> anyhow::Result<Vec<u8>> {
+        let serializeable_metadata = serialize::SeTimelineMetadata::from(self);
+        let mut metadata_bytes = serialize::SeTimelineMetadata::ser(&serializeable_metadata)?;
+        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));
+        Ok(metadata_bytes)
+    }
+
+    /// [`Lsn`] that corresponds to the corresponding timeline directory
+    /// contents, stored locally in the pageserver workdir.
+    pub fn disk_consistent_lsn(&self) -> Lsn {
+        self.disk_consistent_lsn
+    }
+
+    pub fn prev_record_lsn(&self) -> Option<Lsn> {
+        self.prev_record_lsn
+    }
+
+    pub fn ancestor_timeline(&self) -> Option<ZTimelineId> {
+        self.ancestor_timeline
+    }
+
+    pub fn ancestor_lsn(&self) -> Lsn {
+        self.ancestor_lsn
+    }
+}
+
+/// This module is for direct conversion of metadata to bytes and back.
+/// For a certain metadata, besides the conversion a few verification steps has to
+/// be done, so all serde derives are hidden from the user, to avoid accidental
+/// verification-less metadata creation.
+mod serialize {
+    use serde::{Deserialize, Serialize};
+    use zenith_utils::{lsn::Lsn, zid::ZTimelineId};
+
+    use super::TimelineMetadata;
+
+    #[derive(Serialize)]
+    pub(super) struct SeTimelineMetadata<'a> {
+        disk_consistent_lsn: &'a Lsn,
+        prev_record_lsn: &'a Option<Lsn>,
+        ancestor_timeline: &'a Option<ZTimelineId>,
+        ancestor_lsn: &'a Lsn,
+    }
+
+    impl<'a> From<&'a TimelineMetadata> for SeTimelineMetadata<'a> {
+        fn from(other: &'a TimelineMetadata) -> Self {
+            Self {
+                disk_consistent_lsn: &other.disk_consistent_lsn,
+                prev_record_lsn: &other.prev_record_lsn,
+                ancestor_timeline: &other.ancestor_timeline,
+                ancestor_lsn: &other.ancestor_lsn,
+            }
+        }
+    }
+
+    #[derive(Deserialize)]
+    pub(super) struct DeTimelineMetadata {
+        disk_consistent_lsn: Lsn,
+        prev_record_lsn: Option<Lsn>,
+        ancestor_timeline: Option<ZTimelineId>,
+        ancestor_lsn: Lsn,
+    }
+
+    impl From<DeTimelineMetadata> for TimelineMetadata {
+        fn from(other: DeTimelineMetadata) -> Self {
+            Self {
+                disk_consistent_lsn: other.disk_consistent_lsn,
+                prev_record_lsn: other.prev_record_lsn,
+                ancestor_timeline: other.ancestor_timeline,
+                ancestor_lsn: other.ancestor_lsn,
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::repository::repo_harness::TIMELINE_ID;
+
+    use super::*;
+
+    #[test]
+    fn metadata_serializes_correctly() {
+        let original_metadata = TimelineMetadata {
+            disk_consistent_lsn: Lsn(0x200),
+            prev_record_lsn: Some(Lsn(0x100)),
+            ancestor_timeline: Some(TIMELINE_ID),
+            ancestor_lsn: Lsn(0),
+        };
+
+        let metadata_bytes = original_metadata
+            .to_bytes()
+            .expect("Should serialize correct metadata to bytes");
+
+        let deserialized_metadata = TimelineMetadata::from_bytes(&metadata_bytes)
+            .expect("Should deserialize its own bytes");
+
+        assert_eq!(
+            deserialized_metadata, original_metadata,
+            "Metadata that was serialized to bytes and deserialized back should not change"
+        );
+    }
+}

pageserver/src/lib.rs

@@ -1,3 +1,4 @@
+use layered_repository::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME};
 use zenith_utils::postgres_backend::AuthType;
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
@@ -17,6 +18,7 @@ pub mod relish_storage;
 pub mod repository;
 pub mod restore_local_repo;
 pub mod tenant_mgr;
+pub mod tenant_threads;
 pub mod waldecoder;
 pub mod walreceiver;
 pub mod walredo;
@@ -91,7 +93,7 @@ impl PageServerConf {
     //
 
     fn tenants_path(&self) -> PathBuf {
-        self.workdir.join("tenants")
+        self.workdir.join(TENANTS_SEGMENT_NAME)
     }
 
     fn tenant_path(&self, tenantid: &ZTenantId) -> PathBuf {
@@ -115,7 +117,7 @@ impl PageServerConf {
     }
 
     fn timelines_path(&self, tenantid: &ZTenantId) -> PathBuf {
-        self.tenant_path(tenantid).join("timelines")
+        self.tenant_path(tenantid).join(TIMELINES_SEGMENT_NAME)
     }
 
     fn timeline_path(&self, timelineid: &ZTimelineId, tenantid: &ZTenantId) -> PathBuf {
@@ -163,6 +165,15 @@ impl PageServerConf {
     }
 }
 
+/// Config for the Repository checkpointer
+#[derive(Debug, Clone, Copy)]
+pub enum CheckpointConfig {
+    // Flush in-memory data that is older than this
+    Distance(u64),
+    // Flush all in-memory data
+    Forced,
+}
+
 /// External relish storage configuration, enough for creating a client for that storage.
 #[derive(Debug, Clone)]
 pub struct RelishStorageConfig {

pageserver/src/page_service.rs

@@ -630,14 +630,16 @@ impl postgres_backend::Handler for PageServerHandler {
 
             let tenantid = ZTenantId::from_str(caps.get(1).unwrap().as_str())?;
 
-            let branches = crate::branches::get_branches(self.conf, &tenantid)?;
+            // since these handlers for tenant/branch commands are deprecated (in favor of http based ones)
+            // just use false in place of include non incremental logical size
+            let branches = crate::branches::get_branches(self.conf, &tenantid, false)?;
             let branches_buf = serde_json::to_vec(&branches)?;
 
             pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
                 .write_message_noflush(&BeMessage::DataRow(&[Some(&branches_buf)]))?
                 .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
         } else if query_string.starts_with("tenant_list") {
-            let tenants = crate::branches::get_tenants(self.conf)?;
+            let tenants = crate::tenant_mgr::list_tenants()?;
             let tenants_buf = serde_json::to_vec(&tenants)?;
 
             pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?

pageserver/src/relish_storage.rs

@@ -1,60 +1,138 @@
-//! Abstractions for the page server to store its relish layer data in the external storage.
+//! A set of generic storage abstractions for the page server to use when backing up and restoring its state from the external storage.
+//! This particular module serves as a public API border between pageserver and the internal storage machinery.
+//! No other modules from this tree are supposed to be used directly by the external code.
 //!
-//! Main purpose of this module subtree is to provide a set of abstractions to manage the storage state
-//! in a way, optimal for page server.
+//! There are a few components the storage machinery consists of:
+//! * [`RelishStorage`] trait a CRUD-like generic abstraction to use for adapting external storages with a few implementations:
+//!     * [`local_fs`] allows to use local file system as an external storage
+//!     * [`rust_s3`] uses AWS S3 bucket entirely as an external storage
 //!
-//! The abstractions hide multiple custom external storage API implementations,
-//! such as AWS S3, local filesystem, etc., located in the submodules.
+//! * synchronization logic at [`storage_sync`] module that keeps pageserver state (both runtime one and the workdir files) and storage state in sync.
+//!
+//! * public API via to interact with the external world: [`run_storage_sync_thread`] and [`schedule_timeline_upload`]
+//!
+//! Here's a schematic overview of all interactions relish storage and the rest of the pageserver perform:
+//!
+//! +------------------------+                                    +--------->-------+
+//! |                        |  - - - (init async loop) - - - ->  |                 |
+//! |                        |                                    |                 |
+//! |                        |  ------------------------------->  |      async      |
+//! |       pageserver       |   (schedule frozen layer upload)   | upload/download |
+//! |                        |                                    |      loop       |
+//! |                        |  <-------------------------------  |                 |
+//! |                        |    (register downloaded layers)    |                 |
+//! +------------------------+                                    +---------<-------+
+//!                                                                         |
+//!                                                                         |
+//!                                          CRUD layer file operations     |
+//!                                     (upload/download/delete/list, etc.) |
+//!                                                                         V
+//!                                                            +------------------------+
+//!                                                            |                        |
+//!                                                            | [`RelishStorage`] impl |
+//!                                                            |                        |
+//!                                                            | pageserver assumes it  |
+//!                                                            | owns exclusive write   |
+//!                                                            | access to this storage |
+//!                                                            +------------------------+
+//!
+//! First, during startup, the pageserver inits the storage sync thread with the async loop, or leaves the loop unitialised, if configured so.
+//! Some time later, during pageserver checkpoints, in-memory data is flushed onto disk along with its metadata.
+//! If the storage sync loop was successfully started before, pageserver schedules the new image uploads after every checkpoint.
+//! See [`crate::layered_repository`] for the upload calls and the adjacent logic.
+//!
+//! The storage logic considers `image` as a set of local files, fully representing a certain timeline at given moment (identified with `disk_consistent_lsn`).
+//! Timeline can change its state, by adding more files on disk and advancing its `disk_consistent_lsn`: this happens after pageserver checkpointing and is followed
+//! by the storage upload, if enabled.
+//! When a certain image gets uploaded, the sync loop remembers the fact, preventing further reuploads of the same image state.
+//! No files are deleted from either local or remote storage, only the missing ones locally/remotely get downloaded/uploaded, local metadata file will be overwritten
+//! when the newer timeline is downloaded.
+//!
+//! Meanwhile, the loop inits the storage connection and checks the remote files stored.
+//! This is done once at startup only, relying on the fact that pageserver uses the storage alone (ergo, nobody else uploads the files to the storage but this server).
+//! Based on the remote image data, the storage sync logic queues image downloads, while accepting any potential upload tasks from pageserver and managing the tasks by their priority.
+//! On the image download, a [`crate::tenant_mgr::register_relish_download`] function is called to register the new image in pageserver, initializing all related threads and internal state.
+//!
+//! When the pageserver terminates, the upload loop finishes a current image sync task (if any) and exits.
+//!
+//! NOTES:
+//! * pageserver assumes it has exclusive write access to the relish storage. If supported, the way multiple pageservers can be separated in the same storage
+//! (i.e. using different directories in the local filesystem external storage), but totally up to the storage implementation and not covered with the trait API.
+//!
+//! * the uploads do not happen right after pageserver startup, they are registered when
+//!     1. pageserver does the checkpoint, which happens further in the future after the server start
+//!     2. pageserver loads the timeline from disk for the first time
+//!
+//! * the uploads do not happen right after the upload registration: the sync loop might be occupied with other tasks, or tasks with bigger priority could be waiting already
+//!
+//! * all synchronization tasks (including the public API to register uploads and downloads and the sync queue management) happens on an image scale: a big set of relish files,
+//! enough to represent (and recover, if needed) a certain timeline state. On the contrary, all internal storage CRUD calls are made per reilsh file from those images.
+//! This way, the synchronization is able to download the image partially, if some state was synced before, but exposes correctly synced images only.
 
 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;
+mod storage_sync;
 
-use std::{path::Path, thread};
+use std::{
+    path::{Path, PathBuf},
+    thread,
+};
 
-use anyhow::Context;
+use anyhow::{anyhow, ensure, Context};
+use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
-pub use self::synced_storage::schedule_timeline_upload;
-use self::{local_fs::LocalFs, rust_s3::RustS3};
-use crate::{PageServerConf, RelishStorageKind};
+pub use self::storage_sync::schedule_timeline_upload;
+use self::{local_fs::LocalFs, rust_s3::S3};
+use crate::{
+    layered_repository::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME},
+    PageServerConf, RelishStorageKind,
+};
 
+/// Based on the config, initiates the remote storage connection and starts a separate thread
+/// that ensures that pageserver and the remote storage are in sync with each other.
+/// If no external configuraion connection given, no thread or storage initialization is done.
 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(
+            let handle = match &relish_storage_config.storage {
+                RelishStorageKind::LocalFs(root) => storage_sync::spawn_storage_sync_thread(
                     config,
-                    LocalFs::new(root.clone())?,
+                    LocalFs::new(root.clone(), &config.workdir)?,
                     max_concurrent_sync,
                 ),
-                RelishStorageKind::AwsS3(s3_config) => synced_storage::run_storage_sync_thread(
+                RelishStorageKind::AwsS3(s3_config) => storage_sync::spawn_storage_sync_thread(
                     config,
-                    RustS3::new(s3_config)?,
+                    S3::new(s3_config, &config.workdir)?,
                     max_concurrent_sync,
                 ),
-            }
+            };
+            handle.map(Some)
         }
         None => Ok(None),
     }
 }
 
 /// Storage (potentially remote) API to manage its state.
+/// This storage tries to be unaware of any layered repository context,
+/// providing basic CRUD operations with storage files.
 #[async_trait::async_trait]
-pub trait RelishStorage: Send + Sync {
+trait RelishStorage: Send + Sync {
+    /// A way to uniquely reference relish in the remote storage.
     type RelishStoragePath;
 
-    fn derive_destination(
-        page_server_workdir: &Path,
-        relish_local_path: &Path,
-    ) -> anyhow::Result<Self::RelishStoragePath>;
+    /// Attempts to derive the storage path out of the local path, if the latter is correct.
+    fn storage_path(&self, local_path: &Path) -> anyhow::Result<Self::RelishStoragePath>;
 
+    /// Gets the layered storage information about the given entry.
+    fn info(&self, storage_path: &Self::RelishStoragePath) -> anyhow::Result<RemoteRelishInfo>;
+
+    /// Lists all items the storage has right now.
     async fn list_relishes(&self) -> anyhow::Result<Vec<Self::RelishStoragePath>>;
 
+    /// Streams the remote storage entry contents into the buffered writer given, returns the filled writer.
     async fn download_relish<W: 'static + std::io::Write + Send>(
         &self,
         from: &Self::RelishStoragePath,
@@ -65,6 +143,7 @@ pub trait RelishStorage: Send + Sync {
 
     async fn delete_relish(&self, path: &Self::RelishStoragePath) -> anyhow::Result<()>;
 
+    /// Streams the local file contents into remote into the remote storage entry.
     async fn upload_relish<R: tokio::io::AsyncRead + std::marker::Unpin + Send>(
         &self,
         from: &mut tokio::io::BufReader<R>,
@@ -72,16 +151,173 @@ pub trait RelishStorage: Send + Sync {
     ) -> anyhow::Result<()>;
 }
 
-fn strip_workspace_prefix<'a>(
-    page_server_workdir: &'a Path,
-    relish_local_path: &'a Path,
-) -> anyhow::Result<&'a Path> {
-    relish_local_path
-        .strip_prefix(page_server_workdir)
-        .with_context(|| {
+/// Information about a certain remote storage entry.
+#[derive(Debug, PartialEq, Eq)]
+struct RemoteRelishInfo {
+    tenant_id: ZTenantId,
+    timeline_id: ZTimelineId,
+    /// Path in the pageserver workdir where the file should go to.
+    download_destination: PathBuf,
+    is_metadata: bool,
+}
+
+fn strip_path_prefix<'a>(prefix: &'a Path, path: &'a Path) -> anyhow::Result<&'a Path> {
+    if prefix == path {
+        anyhow::bail!(
+            "Prefix and the path are equal, cannot strip: '{}'",
+            prefix.display()
+        )
+    } else {
+        path.strip_prefix(prefix).with_context(|| {
             format!(
-                "Unexpected: relish local path '{}' is not relevant to server workdir",
-                relish_local_path.display(),
+                "Path '{}' is not prefixed with '{}'",
+                path.display(),
+                prefix.display(),
             )
         })
+    }
+}
+
+fn parse_ids_from_path<'a, R: std::fmt::Display>(
+    path_segments: impl Iterator<Item = &'a str>,
+    path_log_representation: &R,
+) -> anyhow::Result<(ZTenantId, ZTimelineId)> {
+    let mut segments = path_segments.skip_while(|&segment| segment != TENANTS_SEGMENT_NAME);
+    let tenants_segment = segments.next().ok_or_else(|| {
+        anyhow!(
+            "Found no '{}' segment in the storage path '{}'",
+            TENANTS_SEGMENT_NAME,
+            path_log_representation
+        )
+    })?;
+    ensure!(
+        tenants_segment == TENANTS_SEGMENT_NAME,
+        "Failed to extract '{}' segment from storage path '{}'",
+        TENANTS_SEGMENT_NAME,
+        path_log_representation
+    );
+    let tenant_id = segments
+        .next()
+        .ok_or_else(|| {
+            anyhow!(
+                "Found no tenant id in the storage path '{}'",
+                path_log_representation
+            )
+        })?
+        .parse::<ZTenantId>()
+        .with_context(|| {
+            format!(
+                "Failed to parse tenant id from storage path '{}'",
+                path_log_representation
+            )
+        })?;
+
+    let timelines_segment = segments.next().ok_or_else(|| {
+        anyhow!(
+            "Found no '{}' segment in the storage path '{}'",
+            TIMELINES_SEGMENT_NAME,
+            path_log_representation
+        )
+    })?;
+    ensure!(
+        timelines_segment == TIMELINES_SEGMENT_NAME,
+        "Failed to extract '{}' segment from storage path '{}'",
+        TIMELINES_SEGMENT_NAME,
+        path_log_representation
+    );
+    let timeline_id = segments
+        .next()
+        .ok_or_else(|| {
+            anyhow!(
+                "Found no timeline id in the storage path '{}'",
+                path_log_representation
+            )
+        })?
+        .parse::<ZTimelineId>()
+        .with_context(|| {
+            format!(
+                "Failed to parse timeline id from storage path '{}'",
+                path_log_representation
+            )
+        })?;
+
+    Ok((tenant_id, timeline_id))
+}
+
+/// A set of common test utils to share in unit tests inside the module tree.
+#[cfg(test)]
+mod test_utils {
+    use std::path::{Path, PathBuf};
+
+    use anyhow::ensure;
+
+    use crate::{
+        layered_repository::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME},
+        repository::repo_harness::{RepoHarness, TIMELINE_ID},
+    };
+
+    /// Gives a timeline path with pageserver workdir stripped off.
+    pub fn relative_timeline_path(harness: &RepoHarness) -> anyhow::Result<PathBuf> {
+        let timeline_path = harness.timeline_path(&TIMELINE_ID);
+        Ok(timeline_path
+            .strip_prefix(&harness.conf.workdir)?
+            .to_path_buf())
+    }
+
+    /// Creates a path with custom tenant id in one of its segments.
+    /// Useful for emulating paths with wrong ids.
+    pub fn custom_tenant_id_path(
+        path_with_tenant_id: &Path,
+        new_tenant_id: &str,
+    ) -> anyhow::Result<PathBuf> {
+        let mut new_path = PathBuf::new();
+        let mut is_tenant_id = false;
+        let mut tenant_id_replaced = false;
+        for segment in path_with_tenant_id {
+            match segment.to_str() {
+                Some(TENANTS_SEGMENT_NAME) => is_tenant_id = true,
+                Some(_tenant_id_str) if is_tenant_id => {
+                    is_tenant_id = false;
+                    new_path.push(new_tenant_id);
+                    tenant_id_replaced = true;
+                    continue;
+                }
+                _ => {}
+            }
+            new_path.push(segment)
+        }
+
+        ensure!(tenant_id_replaced, "Found no tenant id segment to replace");
+        Ok(new_path)
+    }
+
+    /// Creates a path with custom timeline id in one of its segments.
+    /// Useful for emulating paths with wrong ids.
+    pub fn custom_timeline_id_path(
+        path_with_timeline_id: &Path,
+        new_timeline_id: &str,
+    ) -> anyhow::Result<PathBuf> {
+        let mut new_path = PathBuf::new();
+        let mut is_timeline_id = false;
+        let mut timeline_id_replaced = false;
+        for segment in path_with_timeline_id {
+            match segment.to_str() {
+                Some(TIMELINES_SEGMENT_NAME) => is_timeline_id = true,
+                Some(_timeline_id_str) if is_timeline_id => {
+                    is_timeline_id = false;
+                    new_path.push(new_timeline_id);
+                    timeline_id_replaced = true;
+                    continue;
+                }
+                _ => {}
+            }
+            new_path.push(segment)
+        }
+
+        ensure!(
+            timeline_id_replaced,
+            "Found no timeline id segment to replace"
+        );
+        Ok(new_path)
+    }
 }

pageserver/src/relish_storage/README.md

@@ -0,0 +1,82 @@
+# Non-implementation details
+
+This document describes the current state of the backup system in pageserver, existing limitations and concerns, why some things are done the way they are the future development plans.
+Detailed description on how the synchronization works and how it fits into the rest of the pageserver can be found in the [storage module](./../relish_storage.rs) and its submodules.
+Ideally, this document should disappear after current implementation concerns are mitigated, with the remaining useful knowledge bits moved into rustdocs.
+
+## Approach
+
+Backup functionality is a new component, appeared way after the core DB functionality was implemented.
+Pageserver layer functionality is also quite volatile at the moment, there's a risk its local file management changes over time.
+
+To avoid adding more chaos into that, backup functionality is currently designed as a relatively standalone component, with the majority of its logic placed in a standalone async loop.
+This way, the backups are managed in background, not affecting directly other pageserver parts: this way the backup and restoration process may lag behind, but eventually keep up with the reality. To track that, a set of prometheus metrics is exposed from pageserver.
+
+## What's done
+
+Current implementation
+* provides remote storage wrappers for AWS S3 and local FS
+* uploads layers, frozen by pageserver checkpoint thread
+* downloads and registers layers, found on the remote storage, but missing locally
+
+No good optimisations or performance testing is done, the feature is disabled by default and gets polished over time.
+It's planned to deal with all questions that are currently on and prepare the feature to be enabled by default in cloud environments.
+
+### Peculiarities
+
+As mentioned, the backup component is rather new and under development currently, so not all things are done properly from the start.
+Here's the list of known compromises with comments:
+
+* Remote storage model is the same as the `tenants/` directory contents of the pageserver's local workdir storage.
+This is relatively simple to implement, but may be costly to use in AWS S3: an initial data image contains ~782 relish file and a metadata file, ~31 MB combined.
+AWS charges per API call and for traffic either, layers are expected to be updated frequently, so this model most probably is ineffective.
+Additionally, pageservers might need to migrate images between tenants, which does not improve the situation.
+
+Storage sync API operates images when backing up or restoring a backup, so we're fluent to repack the layer contents the way we want to, which most probably will be done later.
+
+* no proper file comparison
+
+Currently, every layer contains `Lsn` in their name, to map the data it holds against a certain DB state.
+Then the images with same ids and different `Lsn`'s are compared, files are considered equal if their local file paths are equal (for remote files, "local file path" is their download destination).
+No file contents assertion is done currently, but should be.
+AWS S3 returns file checksums during the `list` operation, so that can be used to ensure the backup consistency, but that needs further research and, since current pageserver impl also needs to deal with layer file checksums.
+
+For now, due to this, we consider local workdir files as source of truth, not removing them ever and adjusting remote files instead, if image files mismatch.
+
+* no proper retry management
+
+Now, the storage sync attempts to redo the upload/download operation for the image files that failed.
+No proper task eviction or backpressure is implemented currently: the tasks will stay in the queue forever, reattempting the downloads.
+
+This will be fixed when more details on the file consistency model will be agreed on.
+
+* sad rust-s3 api
+
+rust-s3 is not very pleasant to use:
+1. it returns `anyhow::Result` and it's hard to distinguish "missing file" cases from "no connection" one, for instance
+2. at least one function it its API that we need (`get_object_stream`) has `async` keyword and blocks (!), see details [here](https://github.com/zenithdb/zenith/pull/752#discussion_r728373091)
+3. it's a prerelease library with unclear maintenance status
+4. noisy on debug level
+
+But it's already used in the project, so for now it's reused to avoid bloating the dependency tree.
+Based on previous evaluation, even `rusoto-s3` could be a better choice over this library, but needs further benchmarking.
+
+
+* gc and branches are ignored
+
+So far, we don't consider non-main images and don't adjust the remote storage based on GC thread loop results.
+Only checkpointer loop affects the remote storage.
+
+* more layers should be downloaded on demand
+
+Since we download and load remote layers into pageserver, there's a possibility a need for those layers' ancestors arise.
+Most probably, every downloaded image's ancestor is not present in locally too, but currently there's no logic for downloading such ancestors and their metadata,
+so the pageserver is unable to respond property on requests to such ancestors.
+
+To implement the downloading, more `tenant_mgr` refactoring is needed to properly handle web requests for layers and handle the state changes.
+[Here](https://github.com/zenithdb/zenith/pull/689#issuecomment-931216193) are the details about initial state management updates needed.
+
+* no IT tests
+
+Automated S3 testing is lacking currently, due to no convenient way to enable backups during the tests.
+After it's fixed, benchmark runs should also be carried out to find bottlenecks.

pageserver/src/relish_storage/local_fs.rs

@@ -1,13 +1,11 @@
 //! Local filesystem relish storage.
+//! Multiple pageservers can use the same "storage" of this kind by using different storage roots.
 //!
-//! Page server already stores layer data on the server, when freezing it.
-//! This storage serves a way to
-//!
-//! * test things locally simply
-//! * allow to compabre both binary sets
-//! * help validating the relish storage API
+//! This storage used in pageserver tests, but can also be used in cases when a certain persistent
+//! volume is mounted to the local FS.
 
 use std::{
+    ffi::OsStr,
     future::Future,
     io::Write,
     path::{Path, PathBuf},
@@ -16,25 +14,32 @@ use std::{
 
 use anyhow::{bail, Context};
 use tokio::{fs, io};
+use tracing::*;
 
-use super::{strip_workspace_prefix, RelishStorage};
+use crate::layered_repository::metadata::METADATA_FILE_NAME;
+
+use super::{parse_ids_from_path, strip_path_prefix, RelishStorage, RemoteRelishInfo};
 
 pub struct LocalFs {
+    pageserver_workdir: &'static Path,
     root: PathBuf,
 }
 
 impl LocalFs {
-    /// Atetmpts to create local FS relish storage, also creates the directory provided, if not exists.
-    pub fn new(root: PathBuf) -> anyhow::Result<Self> {
+    /// Attempts to create local FS relish storage, along with the storage root directory.
+    pub fn new(root: PathBuf, pageserver_workdir: &'static Path) -> anyhow::Result<Self> {
         if !root.exists() {
             std::fs::create_dir_all(&root).with_context(|| {
                 format!(
-                    "Failed to create all directories in the given root path {}",
+                    "Failed to create all directories in the given root path '{}'",
                     root.display(),
                 )
             })?;
         }
-        Ok(Self { root })
+        Ok(Self {
+            pageserver_workdir,
+            root,
+        })
     }
 
     fn resolve_in_storage(&self, path: &Path) -> anyhow::Result<PathBuf> {
@@ -55,11 +60,29 @@ impl LocalFs {
 impl RelishStorage for LocalFs {
     type RelishStoragePath = PathBuf;
 
-    fn derive_destination(
-        page_server_workdir: &Path,
-        relish_local_path: &Path,
-    ) -> anyhow::Result<Self::RelishStoragePath> {
-        Ok(strip_workspace_prefix(page_server_workdir, relish_local_path)?.to_path_buf())
+    fn storage_path(&self, local_path: &Path) -> anyhow::Result<Self::RelishStoragePath> {
+        Ok(self.root.join(
+            strip_path_prefix(self.pageserver_workdir, local_path)
+                .context("local path does not belong to this storage")?,
+        ))
+    }
+
+    fn info(&self, storage_path: &Self::RelishStoragePath) -> anyhow::Result<RemoteRelishInfo> {
+        let is_metadata =
+            storage_path.file_name().and_then(OsStr::to_str) == Some(METADATA_FILE_NAME);
+        let relative_path = strip_path_prefix(&self.root, storage_path)
+            .context("local path does not belong to this storage")?;
+        let download_destination = self.pageserver_workdir.join(relative_path);
+        let (tenant_id, timeline_id) = parse_ids_from_path(
+            relative_path.iter().filter_map(|segment| segment.to_str()),
+            &relative_path.display(),
+        )?;
+        Ok(RemoteRelishInfo {
+            tenant_id,
+            timeline_id,
+            download_destination,
+            is_metadata,
+        })
     }
 
     async fn list_relishes(&self) -> anyhow::Result<Vec<Self::RelishStoragePath>> {
@@ -72,6 +95,7 @@ impl RelishStorage for LocalFs {
         mut to: std::io::BufWriter<W>,
     ) -> anyhow::Result<std::io::BufWriter<W>> {
         let file_path = self.resolve_in_storage(from)?;
+
         if file_path.exists() && file_path.is_file() {
             let updated_buffer = tokio::task::spawn_blocking(move || {
                 let mut source = std::io::BufReader::new(
@@ -104,7 +128,7 @@ impl RelishStorage for LocalFs {
     async fn delete_relish(&self, path: &Self::RelishStoragePath) -> anyhow::Result<()> {
         let file_path = self.resolve_in_storage(path)?;
         if file_path.exists() && file_path.is_file() {
-            Ok(tokio::fs::remove_file(file_path).await?)
+            Ok(fs::remove_file(file_path).await?)
         } else {
             bail!(
                 "File '{}' either does not exist or is not a file",
@@ -152,12 +176,12 @@ where
         if directory_path.exists() {
             if directory_path.is_dir() {
                 let mut paths = Vec::new();
-                let mut dir_contents = tokio::fs::read_dir(directory_path).await?;
+                let mut dir_contents = fs::read_dir(directory_path).await?;
                 while let Some(dir_entry) = dir_contents.next_entry().await? {
                     let file_type = dir_entry.file_type().await?;
                     let entry_path = dir_entry.path();
                     if file_type.is_symlink() {
-                        log::debug!("{:?} us a symlink, skipping", entry_path)
+                        debug!("{:?} us a symlink, skipping", entry_path)
                     } else if file_type.is_dir() {
                         paths.extend(get_all_files(entry_path).await?.into_iter())
                     } else {
@@ -183,7 +207,370 @@ async fn create_target_directory(target_file_path: &Path) -> anyhow::Result<()>
         ),
     };
     if !target_dir.exists() {
-        tokio::fs::create_dir_all(target_dir).await?;
+        fs::create_dir_all(target_dir).await?;
     }
     Ok(())
 }
+
+#[cfg(test)]
+mod pure_tests {
+    use crate::{
+        layered_repository::metadata::METADATA_FILE_NAME,
+        relish_storage::test_utils::{
+            custom_tenant_id_path, custom_timeline_id_path, relative_timeline_path,
+        },
+        repository::repo_harness::{RepoHarness, TIMELINE_ID},
+    };
+
+    use super::*;
+
+    #[test]
+    fn storage_path_positive() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("storage_path_positive")?;
+        let storage_root = PathBuf::from("somewhere").join("else");
+        let storage = LocalFs {
+            pageserver_workdir: &repo_harness.conf.workdir,
+            root: storage_root.clone(),
+        };
+
+        let local_path = repo_harness.timeline_path(&TIMELINE_ID).join("relish_name");
+        let expected_path = storage_root.join(local_path.strip_prefix(&repo_harness.conf.workdir)?);
+
+        assert_eq!(
+            expected_path,
+            storage.storage_path(&local_path).expect("Matching path should map to storage path normally"),
+            "Relish paths from pageserver workdir should be stored in local fs storage with the same path they have relative to the workdir"
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn storage_path_negatives() -> anyhow::Result<()> {
+        #[track_caller]
+        fn storage_path_error(storage: &LocalFs, mismatching_path: &Path) -> String {
+            match storage.storage_path(mismatching_path) {
+                Ok(wrong_path) => panic!(
+                    "Expected path '{}' to error, but got storage path: {:?}",
+                    mismatching_path.display(),
+                    wrong_path,
+                ),
+                Err(e) => format!("{:?}", e),
+            }
+        }
+
+        let repo_harness = RepoHarness::create("storage_path_negatives")?;
+        let storage_root = PathBuf::from("somewhere").join("else");
+        let storage = LocalFs {
+            pageserver_workdir: &repo_harness.conf.workdir,
+            root: storage_root,
+        };
+
+        let error_string = storage_path_error(&storage, &repo_harness.conf.workdir);
+        assert!(error_string.contains("does not belong to this storage"));
+        assert!(error_string.contains(repo_harness.conf.workdir.to_str().unwrap()));
+
+        let mismatching_path_str = "/something/else";
+        let error_message = storage_path_error(&storage, Path::new(mismatching_path_str));
+        assert!(
+            error_message.contains(mismatching_path_str),
+            "Error should mention wrong path"
+        );
+        assert!(
+            error_message.contains(repo_harness.conf.workdir.to_str().unwrap()),
+            "Error should mention server workdir"
+        );
+        assert!(error_message.contains("does not belong to this storage"));
+
+        Ok(())
+    }
+
+    #[test]
+    fn info_positive() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("info_positive")?;
+        let storage_root = PathBuf::from("somewhere").join("else");
+        let storage = LocalFs {
+            pageserver_workdir: &repo_harness.conf.workdir,
+            root: storage_root.clone(),
+        };
+
+        let name = "not a metadata";
+        let local_path = repo_harness.timeline_path(&TIMELINE_ID).join(name);
+        assert_eq!(
+            RemoteRelishInfo {
+                tenant_id: repo_harness.tenant_id,
+                timeline_id: TIMELINE_ID,
+                download_destination: local_path.clone(),
+                is_metadata: false,
+            },
+            storage
+                .info(&storage_root.join(local_path.strip_prefix(&repo_harness.conf.workdir)?))
+                .expect("For a valid input, valid S3 info should be parsed"),
+            "Should be able to parse metadata out of the correctly named remote delta relish"
+        );
+
+        let local_metadata_path = repo_harness
+            .timeline_path(&TIMELINE_ID)
+            .join(METADATA_FILE_NAME);
+        let remote_metadata_path = storage.storage_path(&local_metadata_path)?;
+        assert_eq!(
+            RemoteRelishInfo {
+                tenant_id: repo_harness.tenant_id,
+                timeline_id: TIMELINE_ID,
+                download_destination: local_metadata_path,
+                is_metadata: true,
+            },
+            storage
+                .info(&remote_metadata_path)
+                .expect("For a valid input, valid S3 info should be parsed"),
+            "Should be able to parse metadata out of the correctly named remote metadata file"
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn info_negatives() -> anyhow::Result<()> {
+        #[track_caller]
+        #[allow(clippy::ptr_arg)] // have to use &PathBuf due to `storage.info` parameter requirements
+        fn storage_info_error(storage: &LocalFs, storage_path: &PathBuf) -> String {
+            match storage.info(storage_path) {
+                Ok(wrong_info) => panic!(
+                    "Expected storage path input {:?} to cause an error, but got relish info: {:?}",
+                    storage_path, wrong_info,
+                ),
+                Err(e) => format!("{:?}", e),
+            }
+        }
+
+        let repo_harness = RepoHarness::create("info_negatives")?;
+        let storage_root = PathBuf::from("somewhere").join("else");
+        let storage = LocalFs {
+            pageserver_workdir: &repo_harness.conf.workdir,
+            root: storage_root.clone(),
+        };
+
+        let totally_wrong_path = "wrong_wrong_wrong";
+        let error_message = storage_info_error(&storage, &PathBuf::from(totally_wrong_path));
+        assert!(error_message.contains(totally_wrong_path));
+
+        let relative_timeline_path = relative_timeline_path(&repo_harness)?;
+
+        let relative_relish_path =
+            custom_tenant_id_path(&relative_timeline_path, "wrong_tenant_id")?
+                .join("wrong_tenant_id_name");
+        let wrong_tenant_id_path = storage_root.join(&relative_relish_path);
+        let error_message = storage_info_error(&storage, &wrong_tenant_id_path);
+        assert!(
+            error_message.contains(relative_relish_path.to_str().unwrap()),
+            "Error message '{}' does not contain the expected substring",
+            error_message
+        );
+
+        let relative_relish_path =
+            custom_timeline_id_path(&relative_timeline_path, "wrong_timeline_id")?
+                .join("wrong_timeline_id_name");
+        let wrong_timeline_id_path = storage_root.join(&relative_relish_path);
+        let error_message = storage_info_error(&storage, &wrong_timeline_id_path);
+        assert!(
+            error_message.contains(relative_relish_path.to_str().unwrap()),
+            "Error message '{}' does not contain the expected substring",
+            error_message
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn download_destination_matches_original_path() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("download_destination_matches_original_path")?;
+        let original_path = repo_harness.timeline_path(&TIMELINE_ID).join("some name");
+
+        let storage_root = PathBuf::from("somewhere").join("else");
+        let dummy_storage = LocalFs {
+            pageserver_workdir: &repo_harness.conf.workdir,
+            root: storage_root,
+        };
+
+        let storage_path = dummy_storage.storage_path(&original_path)?;
+        let download_destination = dummy_storage.info(&storage_path)?.download_destination;
+
+        assert_eq!(
+            original_path, download_destination,
+            "'original path -> storage path -> matching fs path' transformation should produce the same path as the input one for the correct path"
+        );
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod fs_tests {
+    use crate::{
+        relish_storage::test_utils::relative_timeline_path, repository::repo_harness::RepoHarness,
+    };
+
+    use super::*;
+
+    use tempfile::tempdir;
+
+    #[tokio::test]
+    async fn upload_relish() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("upload_relish")?;
+        let storage = create_storage()?;
+
+        let mut source = create_file_for_upload(
+            &storage.pageserver_workdir.join("whatever"),
+            "whatever_contents",
+        )
+        .await?;
+        let target_path = PathBuf::from("/").join("somewhere").join("else");
+        match storage.upload_relish(&mut source, &target_path).await {
+            Ok(()) => panic!("Should not allow storing files with wrong target path"),
+            Err(e) => {
+                let message = format!("{:?}", e);
+                assert!(message.contains(&target_path.display().to_string()));
+                assert!(message.contains("does not belong to the current storage"));
+            }
+        }
+        assert!(storage.list_relishes().await?.is_empty());
+
+        let target_path_1 = upload_dummy_file(&repo_harness, &storage, "upload_1").await?;
+        assert_eq!(
+            storage.list_relishes().await?,
+            vec![target_path_1.clone()],
+            "Should list a single file after first upload"
+        );
+
+        let target_path_2 = upload_dummy_file(&repo_harness, &storage, "upload_2").await?;
+        assert_eq!(
+            list_relishes_sorted(&storage).await?,
+            vec![target_path_1.clone(), target_path_2.clone()],
+            "Should list a two different files after second upload"
+        );
+
+        // match storage.upload_relish(&mut source, &target_path_1).await {
+        //     Ok(()) => panic!("Should not allow reuploading storage files"),
+        //     Err(e) => {
+        //         let message = format!("{:?}", e);
+        //         assert!(message.contains(&target_path_1.display().to_string()));
+        //         assert!(message.contains("File exists"));
+        //     }
+        // }
+        assert_eq!(
+            list_relishes_sorted(&storage).await?,
+            vec![target_path_1, target_path_2],
+            "Should list a two different files after all upload attempts"
+        );
+
+        Ok(())
+    }
+
+    fn create_storage() -> anyhow::Result<LocalFs> {
+        let pageserver_workdir = Box::leak(Box::new(tempdir()?.path().to_owned()));
+        let storage = LocalFs::new(tempdir()?.path().to_owned(), pageserver_workdir)?;
+        Ok(storage)
+    }
+
+    #[tokio::test]
+    async fn download_relish() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("download_relish")?;
+        let storage = create_storage()?;
+        let upload_name = "upload_1";
+        let upload_target = upload_dummy_file(&repo_harness, &storage, upload_name).await?;
+
+        let contents_bytes = storage
+            .download_relish(&upload_target, std::io::BufWriter::new(Vec::new()))
+            .await?
+            .into_inner()?;
+        let contents = String::from_utf8(contents_bytes)?;
+        assert_eq!(
+            dummy_contents(upload_name),
+            contents,
+            "We should upload and download the same contents"
+        );
+
+        let non_existing_path = PathBuf::from("somewhere").join("else");
+        match storage
+            .download_relish(&non_existing_path, std::io::BufWriter::new(Vec::new()))
+            .await
+        {
+            Ok(_) => panic!("Should not allow downloading non-existing storage files"),
+            Err(e) => {
+                let error_string = e.to_string();
+                assert!(error_string.contains("does not exist"));
+                assert!(error_string.contains(&non_existing_path.display().to_string()));
+            }
+        }
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn delete_relish() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("delete_relish")?;
+        let storage = create_storage()?;
+        let upload_name = "upload_1";
+        let upload_target = upload_dummy_file(&repo_harness, &storage, upload_name).await?;
+
+        storage.delete_relish(&upload_target).await?;
+        assert!(storage.list_relishes().await?.is_empty());
+
+        match storage.delete_relish(&upload_target).await {
+            Ok(()) => panic!("Should not allow deleting non-existing storage files"),
+            Err(e) => {
+                let error_string = e.to_string();
+                assert!(error_string.contains("does not exist"));
+                assert!(error_string.contains(&upload_target.display().to_string()));
+            }
+        }
+        Ok(())
+    }
+
+    async fn upload_dummy_file(
+        harness: &RepoHarness,
+        storage: &LocalFs,
+        name: &str,
+    ) -> anyhow::Result<PathBuf> {
+        let storage_path = storage
+            .root
+            .join(relative_timeline_path(harness)?)
+            .join(name);
+        storage
+            .upload_relish(
+                &mut create_file_for_upload(
+                    &storage.pageserver_workdir.join(name),
+                    &dummy_contents(name),
+                )
+                .await?,
+                &storage_path,
+            )
+            .await?;
+        Ok(storage_path)
+    }
+
+    async fn create_file_for_upload(
+        path: &Path,
+        contents: &str,
+    ) -> anyhow::Result<io::BufReader<fs::File>> {
+        std::fs::create_dir_all(path.parent().unwrap())?;
+        let mut file_for_writing = std::fs::OpenOptions::new()
+            .write(true)
+            .create_new(true)
+            .open(path)?;
+        write!(file_for_writing, "{}", contents)?;
+        drop(file_for_writing);
+        Ok(io::BufReader::new(
+            fs::OpenOptions::new().read(true).open(&path).await?,
+        ))
+    }
+
+    fn dummy_contents(name: &str) -> String {
+        format!("contents for {}", name)
+    }
+
+    async fn list_relishes_sorted(storage: &LocalFs) -> anyhow::Result<Vec<PathBuf>> {
+        let mut relishes = storage.list_relishes().await?;
+        relishes.sort();
+        Ok(relishes)
+    }
+}

pageserver/src/relish_storage/rust_s3.rs

@@ -1,35 +1,45 @@
-//! A wrapper around AWS S3 client library `rust_s3` to be used a relish storage.
+//! AWS S3 relish storage wrapper around `rust_s3` library.
+//! Currently does not allow multiple pageservers to use the same bucket concurrently: relishes are
+//! placed in the root of the bucket.
 
-use std::io::Write;
-use std::path::Path;
+use std::{
+    io::Write,
+    path::{Path, PathBuf},
+};
 
 use anyhow::Context;
 use s3::{bucket::Bucket, creds::Credentials, region::Region};
 
 use crate::{
-    relish_storage::{strip_workspace_prefix, RelishStorage},
+    layered_repository::metadata::METADATA_FILE_NAME,
+    relish_storage::{parse_ids_from_path, strip_path_prefix, RelishStorage, RemoteRelishInfo},
     S3Config,
 };
 
 const S3_FILE_SEPARATOR: char = '/';
 
-#[derive(Debug)]
+#[derive(Debug, Eq, PartialEq)]
 pub struct S3ObjectKey(String);
 
 impl S3ObjectKey {
     fn key(&self) -> &str {
         &self.0
     }
+
+    fn download_destination(&self, pageserver_workdir: &Path) -> PathBuf {
+        pageserver_workdir.join(self.0.split(S3_FILE_SEPARATOR).collect::<PathBuf>())
+    }
 }
 
 /// AWS S3 relish storage.
-pub struct RustS3 {
+pub struct S3 {
+    pageserver_workdir: &'static Path,
     bucket: Bucket,
 }
 
-impl RustS3 {
+impl S3 {
     /// Creates the relish storage, errors if incorrect AWS S3 configuration provided.
-    pub fn new(aws_config: &S3Config) -> anyhow::Result<Self> {
+    pub fn new(aws_config: &S3Config, pageserver_workdir: &'static Path) -> anyhow::Result<Self> {
         let region = aws_config
             .bucket_region
             .parse::<Region>()
@@ -49,19 +59,17 @@ impl RustS3 {
                 credentials,
             )
             .context("Failed to create the s3 bucket")?,
+            pageserver_workdir,
         })
     }
 }
 
 #[async_trait::async_trait]
-impl RelishStorage for RustS3 {
+impl RelishStorage for S3 {
     type RelishStoragePath = S3ObjectKey;
 
-    fn derive_destination(
-        page_server_workdir: &Path,
-        relish_local_path: &Path,
-    ) -> anyhow::Result<Self::RelishStoragePath> {
-        let relative_path = strip_workspace_prefix(page_server_workdir, relish_local_path)?;
+    fn storage_path(&self, local_path: &Path) -> anyhow::Result<Self::RelishStoragePath> {
+        let relative_path = strip_path_prefix(self.pageserver_workdir, local_path)?;
         let mut key = String::new();
         for segment in relative_path {
             key.push(S3_FILE_SEPARATOR);
@@ -70,6 +78,21 @@ impl RelishStorage for RustS3 {
         Ok(S3ObjectKey(key))
     }
 
+    fn info(&self, storage_path: &Self::RelishStoragePath) -> anyhow::Result<RemoteRelishInfo> {
+        let storage_path_key = &storage_path.0;
+        let is_metadata =
+            storage_path_key.ends_with(&format!("{}{}", S3_FILE_SEPARATOR, METADATA_FILE_NAME));
+        let download_destination = storage_path.download_destination(self.pageserver_workdir);
+        let (tenant_id, timeline_id) =
+            parse_ids_from_path(storage_path_key.split(S3_FILE_SEPARATOR), storage_path_key)?;
+        Ok(RemoteRelishInfo {
+            tenant_id,
+            timeline_id,
+            download_destination,
+            is_metadata,
+        })
+    }
+
     async fn list_relishes(&self) -> anyhow::Result<Vec<Self::RelishStoragePath>> {
         let list_response = self
             .bucket
@@ -101,11 +124,11 @@ impl RelishStorage for RustS3 {
             ))
         } else {
             tokio::task::spawn_blocking(move || {
-                to.flush().context("Failed to fluch the downoad buffer")?;
+                to.flush().context("Failed to flush the download buffer")?;
                 Ok::<_, anyhow::Error>(to)
             })
             .await
-            .context("Failed to joim the download buffer flush task")?
+            .context("Failed to join the download buffer flush task")?
         }
     }
 
@@ -115,9 +138,9 @@ impl RelishStorage for RustS3 {
             .delete_object(path.key())
             .await
             .with_context(|| format!("Failed to delete s3 object with key {}", path.key()))?;
-        if code != 200 {
+        if code != 204 {
             Err(anyhow::format_err!(
-                "Received non-200 exit code during deleting object with key '{}', code: {}",
+                "Received non-204 exit code during deleting object with key '{}', code: {}",
                 path.key(),
                 code
             ))
@@ -147,3 +170,226 @@ impl RelishStorage for RustS3 {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use crate::{
+        relish_storage::test_utils::{
+            custom_tenant_id_path, custom_timeline_id_path, relative_timeline_path,
+        },
+        repository::repo_harness::{RepoHarness, TIMELINE_ID},
+    };
+
+    use super::*;
+
+    #[test]
+    fn download_destination() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("download_destination")?;
+
+        let local_path = repo_harness.timeline_path(&TIMELINE_ID).join("test_name");
+        let relative_path = local_path.strip_prefix(&repo_harness.conf.workdir)?;
+
+        let key = S3ObjectKey(format!(
+            "{}{}",
+            S3_FILE_SEPARATOR,
+            relative_path
+                .iter()
+                .map(|segment| segment.to_str().unwrap())
+                .collect::<Vec<_>>()
+                .join(&S3_FILE_SEPARATOR.to_string()),
+        ));
+
+        assert_eq!(
+            local_path,
+            key.download_destination(&repo_harness.conf.workdir),
+            "Download destination should consist of s3 path joined with the pageserver workdir prefix"
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn storage_path_positive() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("storage_path_positive")?;
+
+        let segment_1 = "matching";
+        let segment_2 = "relish";
+        let local_path = &repo_harness.conf.workdir.join(segment_1).join(segment_2);
+        let expected_key = S3ObjectKey(format!(
+            "{SEPARATOR}{}{SEPARATOR}{}",
+            segment_1,
+            segment_2,
+            SEPARATOR = S3_FILE_SEPARATOR,
+        ));
+
+        let actual_key = dummy_storage(&repo_harness.conf.workdir)
+            .storage_path(local_path)
+            .expect("Matching path should map to S3 path normally");
+        assert_eq!(
+            expected_key,
+            actual_key,
+            "S3 key from the matching path should contain all segments after the workspace prefix, separated with S3 separator"
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn storage_path_negatives() -> anyhow::Result<()> {
+        #[track_caller]
+        fn storage_path_error(storage: &S3, mismatching_path: &Path) -> String {
+            match storage.storage_path(mismatching_path) {
+                Ok(wrong_key) => panic!(
+                    "Expected path '{}' to error, but got S3 key: {:?}",
+                    mismatching_path.display(),
+                    wrong_key,
+                ),
+                Err(e) => e.to_string(),
+            }
+        }
+
+        let repo_harness = RepoHarness::create("storage_path_negatives")?;
+        let storage = dummy_storage(&repo_harness.conf.workdir);
+
+        let error_message = storage_path_error(&storage, &repo_harness.conf.workdir);
+        assert!(
+            error_message.contains("Prefix and the path are equal"),
+            "Message '{}' does not contain the required string",
+            error_message
+        );
+
+        let mismatching_path = PathBuf::from("somewhere").join("else");
+        let error_message = storage_path_error(&storage, &mismatching_path);
+        assert!(
+            error_message.contains(mismatching_path.to_str().unwrap()),
+            "Error should mention wrong path"
+        );
+        assert!(
+            error_message.contains(repo_harness.conf.workdir.to_str().unwrap()),
+            "Error should mention server workdir"
+        );
+        assert!(
+            error_message.contains("is not prefixed with"),
+            "Message '{}' does not contain a required string",
+            error_message
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn info_positive() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("info_positive")?;
+        let storage = dummy_storage(&repo_harness.conf.workdir);
+        let relative_timeline_path = relative_timeline_path(&repo_harness)?;
+
+        let s3_key = create_s3_key(&relative_timeline_path.join("not a metadata"));
+        assert_eq!(
+            RemoteRelishInfo {
+                tenant_id: repo_harness.tenant_id,
+                timeline_id: TIMELINE_ID,
+                download_destination: s3_key.download_destination(&repo_harness.conf.workdir),
+                is_metadata: false,
+            },
+            storage
+                .info(&s3_key)
+                .expect("For a valid input, valid S3 info should be parsed"),
+            "Should be able to parse metadata out of the correctly named remote delta relish"
+        );
+
+        let s3_key = create_s3_key(&relative_timeline_path.join(METADATA_FILE_NAME));
+        assert_eq!(
+            RemoteRelishInfo {
+                tenant_id: repo_harness.tenant_id,
+                timeline_id: TIMELINE_ID,
+                download_destination: s3_key.download_destination(&repo_harness.conf.workdir),
+                is_metadata: true,
+            },
+            storage
+                .info(&s3_key)
+                .expect("For a valid input, valid S3 info should be parsed"),
+            "Should be able to parse metadata out of the correctly named remote metadata file"
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn info_negatives() -> anyhow::Result<()> {
+        #[track_caller]
+        fn storage_info_error(storage: &S3, s3_key: &S3ObjectKey) -> String {
+            match storage.info(s3_key) {
+                Ok(wrong_info) => panic!(
+                    "Expected key {:?} to error, but got relish info: {:?}",
+                    s3_key, wrong_info,
+                ),
+                Err(e) => e.to_string(),
+            }
+        }
+
+        let repo_harness = RepoHarness::create("info_negatives")?;
+        let storage = dummy_storage(&repo_harness.conf.workdir);
+        let relative_timeline_path = relative_timeline_path(&repo_harness)?;
+
+        let totally_wrong_path = "wrong_wrong_wrong";
+        let error_message =
+            storage_info_error(&storage, &S3ObjectKey(totally_wrong_path.to_string()));
+        assert!(error_message.contains(totally_wrong_path));
+
+        let wrong_tenant_id = create_s3_key(
+            &custom_tenant_id_path(&relative_timeline_path, "wrong_tenant_id")?.join("name"),
+        );
+        let error_message = storage_info_error(&storage, &wrong_tenant_id);
+        assert!(error_message.contains(&wrong_tenant_id.0));
+
+        let wrong_timeline_id = create_s3_key(
+            &custom_timeline_id_path(&relative_timeline_path, "wrong_timeline_id")?.join("name"),
+        );
+        let error_message = storage_info_error(&storage, &wrong_timeline_id);
+        assert!(error_message.contains(&wrong_timeline_id.0));
+
+        Ok(())
+    }
+
+    #[test]
+    fn download_destination_matches_original_path() -> anyhow::Result<()> {
+        let repo_harness = RepoHarness::create("download_destination_matches_original_path")?;
+        let original_path = repo_harness.timeline_path(&TIMELINE_ID).join("some name");
+
+        let dummy_storage = dummy_storage(&repo_harness.conf.workdir);
+
+        let key = dummy_storage.storage_path(&original_path)?;
+        let download_destination = dummy_storage.info(&key)?.download_destination;
+
+        assert_eq!(
+            original_path, download_destination,
+            "'original path -> storage key -> matching fs path' transformation should produce the same path as the input one for the correct path"
+        );
+
+        Ok(())
+    }
+
+    fn dummy_storage(pageserver_workdir: &'static Path) -> S3 {
+        S3 {
+            pageserver_workdir,
+            bucket: Bucket::new(
+                "dummy-bucket",
+                "us-east-1".parse().unwrap(),
+                Credentials::anonymous().unwrap(),
+            )
+            .unwrap(),
+        }
+    }
+
+    fn create_s3_key(relative_relish_path: &Path) -> S3ObjectKey {
+        S3ObjectKey(
+            relative_relish_path
+                .iter()
+                .fold(String::new(), |mut path_string, segment| {
+                    path_string.push(S3_FILE_SEPARATOR);
+                    path_string.push_str(segment.to_str().unwrap());
+                    path_string
+                }),
+        )
+    }
+}

pageserver/src/relish_storage/synced_storage.rs

@@ -1,57 +0,0 @@
-use std::time::Duration;
-use std::{collections::BinaryHeap, sync::Mutex, thread};
-
-use crate::tenant_mgr;
-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 || {
-            while !tenant_mgr::shutdown_requested() {
-                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;
-                    }
-                }
-            }
-            log::debug!("Queue based relish storage sync thread shut down");
-            Ok(())
-        })?;
-    Ok(Some(handle))
-}

pageserver/src/repository.rs

@@ -1,4 +1,5 @@
 use crate::relish::*;
+use crate::CheckpointConfig;
 use anyhow::Result;
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use serde::{Deserialize, Serialize};
@@ -24,9 +25,9 @@ pub trait Repository: Send + Sync {
     /// Branch a timeline
     fn branch_timeline(&self, src: ZTimelineId, dst: ZTimelineId, start_lsn: Lsn) -> Result<()>;
 
-    /// perform one garbage collection iteration.
-    /// garbage collection is periodically performed by gc thread,
-    /// but it can be explicitly requested through page server api.
+    /// perform one garbage collection iteration, removing old data files from disk.
+    /// this funtion is periodically called by gc thread.
+    /// also it can be explicitly requested through page server api 'do_gc' command.
     ///
     /// 'timelineid' specifies the timeline to GC, or None for all.
     /// `horizon` specifies delta from last lsn to preserve all object versions (pitr interval).
@@ -39,6 +40,10 @@ pub trait Repository: Send + Sync {
         horizon: u64,
         checkpoint_before_gc: bool,
     ) -> Result<GcResult>;
+
+    /// perform one checkpoint iteration, flushing in-memory data on disk.
+    /// this function is periodically called by checkponter thread.
+    fn checkpoint_iteration(&self, cconf: CheckpointConfig) -> Result<()>;
 }
 
 ///
@@ -119,6 +124,9 @@ pub trait Timeline: Send + Sync {
     /// Get a list of all existing non-relational objects
     fn list_nonrels(&self, lsn: Lsn) -> Result<HashSet<RelishTag>>;
 
+    /// Get the LSN where this branch was created
+    fn get_ancestor_lsn(&self) -> Lsn;
+
     //------------------------------------------------------------------------------
     // Public PUT functions, to update the repository with new page versions.
     //
@@ -131,6 +139,7 @@ pub trait Timeline: Send + Sync {
     fn get_last_record_lsn(&self) -> Lsn;
     fn get_prev_record_lsn(&self) -> Lsn;
     fn get_start_lsn(&self) -> Lsn;
+    fn get_disk_consistent_lsn(&self) -> Lsn;
 
     /// Mutate the timeline with a [`TimelineWriter`].
     fn writer<'a>(&'a self) -> Box<dyn TimelineWriter + 'a>;
@@ -140,7 +149,7 @@ pub trait Timeline: Send + Sync {
     ///
     /// NOTE: This has nothing to do with checkpoint in PostgreSQL. We don't
     /// know anything about them here in the repository.
-    fn checkpoint(&self) -> Result<()>;
+    fn checkpoint(&self, cconf: CheckpointConfig) -> Result<()>;
 
     /// Retrieve current logical size of the timeline
     ///
@@ -210,27 +219,115 @@ impl WALRecord {
     }
 }
 
+#[cfg(test)]
+pub mod repo_harness {
+    use std::{fs, path::PathBuf};
+
+    use crate::{
+        layered_repository::{LayeredRepository, TIMELINES_SEGMENT_NAME},
+        walredo::{WalRedoError, WalRedoManager},
+        PageServerConf,
+    };
+
+    use super::*;
+    use hex_literal::hex;
+    use zenith_utils::zid::ZTenantId;
+
+    pub const TIMELINE_ID: ZTimelineId =
+        ZTimelineId::from_array(hex!("11223344556677881122334455667788"));
+    pub const NEW_TIMELINE_ID: ZTimelineId =
+        ZTimelineId::from_array(hex!("AA223344556677881122334455667788"));
+
+    /// Convenience function to create a page image with given string as the only content
+    #[allow(non_snake_case)]
+    pub fn TEST_IMG(s: &str) -> Bytes {
+        let mut buf = BytesMut::new();
+        buf.extend_from_slice(s.as_bytes());
+        buf.resize(8192, 0);
+
+        buf.freeze()
+    }
+
+    pub struct RepoHarness {
+        pub conf: &'static PageServerConf,
+        pub tenant_id: ZTenantId,
+    }
+
+    impl RepoHarness {
+        pub 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_SEGMENT_NAME))?;
+
+            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 tenant_id = ZTenantId::generate();
+            fs::create_dir_all(conf.tenant_path(&tenant_id))?;
+
+            Ok(Self { conf, tenant_id })
+        }
+
+        pub fn load(&self) -> Box<dyn Repository> {
+            let walredo_mgr = Arc::new(TestRedoManager);
+
+            Box::new(LayeredRepository::new(
+                self.conf,
+                walredo_mgr,
+                self.tenant_id,
+                false,
+            ))
+        }
+
+        pub fn timeline_path(&self, timeline_id: &ZTimelineId) -> PathBuf {
+            self.conf.timeline_path(timeline_id, &self.tenant_id)
+        }
+    }
+
+    // Mock WAL redo manager that doesn't do much
+    struct TestRedoManager;
+
+    impl WalRedoManager for TestRedoManager {
+        fn request_redo(
+            &self,
+            rel: RelishTag,
+            blknum: u32,
+            lsn: Lsn,
+            base_img: Option<Bytes>,
+            records: Vec<(Lsn, WALRecord)>,
+        ) -> Result<Bytes, WalRedoError> {
+            let s = format!(
+                "redo for {} blk {} to get to {}, with {} and {} records",
+                rel,
+                blknum,
+                lsn,
+                if base_img.is_some() {
+                    "base image"
+                } else {
+                    "no base image"
+                },
+                records.len()
+            );
+            println!("{}", s);
+            Ok(TEST_IMG(&s))
+        }
+    }
+}
+
 ///
 /// Tests that should work the same with any Repository/Timeline implementation.
 ///
 #[allow(clippy::bool_assert_comparison)]
 #[cfg(test)]
 mod tests {
-    use super::*;
-    use crate::layered_repository::{LayeredRepository, METADATA_FILE_NAME};
-    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 zenith_utils::zid::ZTenantId;
+    use crate::layered_repository::metadata::METADATA_FILE_NAME;
 
-    const TIMELINE_ID: ZTimelineId =
-        ZTimelineId::from_array(hex!("11223344556677881122334455667788"));
-    const NEW_TIMELINE_ID: ZTimelineId =
-        ZTimelineId::from_array(hex!("AA223344556677881122334455667788"));
+    use super::repo_harness::*;
+    use super::*;
+    use postgres_ffi::{pg_constants, xlog_utils::SIZEOF_CHECKPOINT};
 
     /// Arbitrary relation tag, for testing.
     const TESTREL_A: RelishTag = RelishTag::Relation(RelTag {
@@ -246,16 +343,6 @@ mod tests {
         forknum: 0,
     });
 
-    /// Convenience function to create a page image with given string as the only content
-    #[allow(non_snake_case)]
-    fn TEST_IMG(s: &str) -> Bytes {
-        let mut buf = BytesMut::new();
-        buf.extend_from_slice(s.as_bytes());
-        buf.resize(8192, 0);
-
-        buf.freeze()
-    }
-
     fn assert_current_logical_size(timeline: &Arc<dyn Timeline>, lsn: Lsn) {
         let incremental = timeline.get_current_logical_size();
         let non_incremental = timeline
@@ -267,45 +354,6 @@ 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,
-    }
-
-    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 tenant_id = ZTenantId::generate();
-            fs::create_dir_all(conf.tenant_path(&tenant_id))?;
-
-            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)
-        }
-    }
-
     #[test]
     fn test_relsize() -> Result<()> {
         let repo = RepoHarness::create("test_relsize")?.load();
@@ -672,7 +720,7 @@ mod tests {
             .contains(&TESTREL_A));
 
         // Run checkpoint and garbage collection and check that it's still not visible
-        newtline.checkpoint()?;
+        newtline.checkpoint(CheckpointConfig::Forced)?;
         repo.gc_iteration(Some(NEW_TIMELINE_ID), 0, true)?;
 
         assert!(!newtline
@@ -817,33 +865,4 @@ mod tests {
 
         Ok(())
     }
-
-    // Mock WAL redo manager that doesn't do much
-    struct TestRedoManager;
-
-    impl WalRedoManager for TestRedoManager {
-        fn request_redo(
-            &self,
-            rel: RelishTag,
-            blknum: u32,
-            lsn: Lsn,
-            base_img: Option<Bytes>,
-            records: Vec<(Lsn, WALRecord)>,
-        ) -> Result<Bytes, WalRedoError> {
-            let s = format!(
-                "redo for {} blk {} to get to {}, with {} and {} records",
-                rel,
-                blknum,
-                lsn,
-                if base_img.is_some() {
-                    "base image"
-                } else {
-                    "no base image"
-                },
-                records.len()
-            );
-            println!("{}", s);
-            Ok(TEST_IMG(&s))
-        }
-    }
 }

pageserver/src/restore_local_repo.rs

@@ -7,10 +7,10 @@ use postgres_ffi::nonrelfile_utils::slru_may_delete_clogsegment;
 use std::cmp::min;
 use std::fs;
 use std::fs::File;
-use std::io::Read;
-use std::path::Path;
+use std::io::{Read, Seek, SeekFrom};
+use std::path::{Path, PathBuf};
 
-use anyhow::{bail, Result};
+use anyhow::{anyhow, bail, Result};
 use bytes::{Buf, Bytes};
 use tracing::*;
 
@@ -37,6 +37,8 @@ pub fn import_timeline_from_postgres_datadir(
     writer: &dyn TimelineWriter,
     lsn: Lsn,
 ) -> Result<()> {
+    let mut pg_control: Option<ControlFileData> = None;
+
     // Scan 'global'
     for direntry in fs::read_dir(path.join("global"))? {
         let direntry = direntry?;
@@ -44,7 +46,7 @@ pub fn import_timeline_from_postgres_datadir(
             None => continue,
 
             Some("pg_control") => {
-                import_control_file(writer, lsn, &direntry.path())?;
+                pg_control = Some(import_control_file(writer, lsn, &direntry.path())?);
             }
             Some("pg_filenode.map") => import_nonrel_file(
                 writer,
@@ -127,6 +129,18 @@ pub fn import_timeline_from_postgres_datadir(
 
     writer.advance_last_record_lsn(lsn);
 
+    // Import WAL. This is needed even when starting from a shutdown checkpoint, because
+    // this reads the checkpoint record itself, advancing the tip of the timeline to
+    // *after* the checkpoint record. And crucially, it initializes the 'prev_lsn'
+    let pg_control = pg_control.ok_or_else(|| anyhow!("pg_control file not found"))?;
+    import_wal(
+        &path.join("pg_wal"),
+        writer,
+        Lsn(pg_control.checkPointCopy.redo),
+        lsn,
+        &mut pg_control.checkPointCopy.clone(),
+    )?;
+
     Ok(())
 }
 
@@ -212,7 +226,11 @@ fn import_nonrel_file(
 ///
 /// The control file is imported as is, but we also extract the checkpoint record
 /// from it and store it separated.
-fn import_control_file(timeline: &dyn TimelineWriter, lsn: Lsn, path: &Path) -> Result<()> {
+fn import_control_file(
+    timeline: &dyn TimelineWriter,
+    lsn: Lsn,
+    path: &Path,
+) -> Result<ControlFileData> {
     let mut file = File::open(path)?;
     let mut buffer = Vec::new();
     // read the whole file
@@ -233,7 +251,7 @@ fn import_control_file(timeline: &dyn TimelineWriter, lsn: Lsn, path: &Path) ->
     let checkpoint_bytes = pg_control.checkPointCopy.encode();
     timeline.put_page_image(RelishTag::Checkpoint, 0, lsn, checkpoint_bytes)?;
 
-    Ok(())
+    Ok(pg_control)
 }
 
 ///
@@ -285,6 +303,104 @@ fn import_slru_file(
     Ok(())
 }
 
+/// Scan PostgreSQL WAL files in given directory and load all records between
+/// 'startpoint' and 'endpoint' into the repository.
+fn import_wal(
+    walpath: &Path,
+    timeline: &dyn TimelineWriter,
+    startpoint: Lsn,
+    endpoint: Lsn,
+    checkpoint: &mut CheckPoint,
+) -> Result<()> {
+    let mut waldecoder = WalStreamDecoder::new(startpoint);
+
+    let mut segno = startpoint.segment_number(pg_constants::WAL_SEGMENT_SIZE);
+    let mut offset = startpoint.segment_offset(pg_constants::WAL_SEGMENT_SIZE);
+    let mut last_lsn = startpoint;
+
+    while last_lsn <= endpoint {
+        // FIXME: assume postgresql tli 1 for now
+        let filename = XLogFileName(1, segno, pg_constants::WAL_SEGMENT_SIZE);
+        let mut buf = Vec::new();
+
+        // Read local file
+        let mut path = walpath.join(&filename);
+
+        // It could be as .partial
+        if !PathBuf::from(&path).exists() {
+            path = walpath.join(filename + ".partial");
+        }
+
+        // Slurp the WAL file
+        let mut file = File::open(&path)?;
+
+        if offset > 0 {
+            file.seek(SeekFrom::Start(offset as u64))?;
+        }
+
+        let nread = file.read_to_end(&mut buf)?;
+        if nread != pg_constants::WAL_SEGMENT_SIZE - offset as usize {
+            // Maybe allow this for .partial files?
+            error!("read only {} bytes from WAL file", nread);
+        }
+
+        waldecoder.feed_bytes(&buf);
+
+        let mut nrecords = 0;
+        while last_lsn <= endpoint {
+            if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
+                let mut checkpoint_modified = false;
+
+                let decoded = decode_wal_record(recdata.clone());
+                save_decoded_record(
+                    checkpoint,
+                    &mut checkpoint_modified,
+                    timeline,
+                    &decoded,
+                    recdata,
+                    lsn,
+                )?;
+                last_lsn = lsn;
+
+                if checkpoint_modified {
+                    let checkpoint_bytes = checkpoint.encode();
+                    timeline.put_page_image(
+                        RelishTag::Checkpoint,
+                        0,
+                        last_lsn,
+                        checkpoint_bytes,
+                    )?;
+                }
+
+                // Now that this record has been fully handled, including updating the
+                // checkpoint data, let the repository know that it is up-to-date to this LSN
+                timeline.advance_last_record_lsn(last_lsn);
+                nrecords += 1;
+
+                trace!("imported record at {} (end {})", lsn, endpoint);
+            }
+        }
+
+        debug!("imported {} records up to {}", nrecords, last_lsn);
+
+        segno += 1;
+        offset = 0;
+    }
+
+    if last_lsn != startpoint {
+        debug!(
+            "reached end of WAL at {}, updating checkpoint info",
+            last_lsn
+        );
+
+        timeline.advance_last_record_lsn(last_lsn);
+    } else {
+        info!("no WAL to import at {}", last_lsn);
+    }
+
+    Ok(())
+}
+
 ///
 /// Helper function to parse a WAL record and call the Timeline's PUT functions for all the
 /// relations/pages that the record affects.

pageserver/src/tenant_mgr.rs

@@ -4,54 +4,89 @@
 use crate::branches;
 use crate::layered_repository::LayeredRepository;
 use crate::repository::{Repository, Timeline};
+use crate::tenant_threads;
 use crate::walredo::PostgresRedoManager;
 use crate::PageServerConf;
 use anyhow::{anyhow, bail, Context, Result};
 use lazy_static::lazy_static;
 use log::{debug, info};
-use std::collections::hash_map::Entry;
+use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
+use std::fmt;
 use std::fs;
 use std::str::FromStr;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, Mutex, MutexGuard};
-use std::thread::JoinHandle;
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 lazy_static! {
-    static ref REPOSITORY: Mutex<HashMap<ZTenantId, Arc<dyn Repository>>> =
-        Mutex::new(HashMap::new());
+    static ref TENANTS: Mutex<HashMap<ZTenantId, Tenant>> = Mutex::new(HashMap::new());
 }
 
-fn access_repository() -> MutexGuard<'static, HashMap<ZTenantId, Arc<dyn Repository>>> {
-    REPOSITORY.lock().unwrap()
-}
-struct TenantHandleEntry {
-    checkpointer_handle: Option<JoinHandle<()>>,
-    gc_handle: Option<JoinHandle<()>>,
+struct Tenant {
+    state: TenantState,
+    repo: Option<Arc<dyn Repository>>,
 }
 
-// Logically these handles belong to Repository,
-// but it's just simpler to store them separately
-lazy_static! {
-    static ref TENANT_HANDLES: Mutex<HashMap<ZTenantId, TenantHandleEntry>> =
-        Mutex::new(HashMap::new());
+#[derive(Debug, Serialize, Deserialize, Clone, Copy, PartialEq, Eq)]
+pub enum TenantState {
+    // This tenant only exists in cloud storage. It cannot be accessed.
+    CloudOnly,
+    // This tenant exists in cloud storage, and we are currently downloading it to local disk.
+    // It cannot be accessed yet, not until it's been fully downloaded to local disk.
+    Downloading,
+    // All data for this tenant is complete on local disk, but we haven't loaded the Repository,
+    // Timeline and Layer structs into memory yet, so it cannot be accessed yet.
+    //Ready,
+    // This tenant exists on local disk, and the layer map has been loaded into memory.
+    // The local disk might have some newer files that don't exist in cloud storage yet.
+    Active,
+    // Tenant is active, but there is no walreceiver connection.
+    Idle,
+    // This tenant exists on local disk, and the layer map has been loaded into memory.
+    // The local disk might have some newer files that don't exist in cloud storage yet.
+    // The tenant cannot be accessed anymore for any reason, but graceful shutdown.
+    Stopping,
+}
+
+impl fmt::Display for TenantState {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            TenantState::CloudOnly => f.write_str("CloudOnly"),
+            TenantState::Downloading => f.write_str("Downloading"),
+            TenantState::Active => f.write_str("Active"),
+            TenantState::Idle => f.write_str("Idle"),
+            TenantState::Stopping => f.write_str("Stopping"),
+        }
+    }
+}
+
+fn access_tenants() -> MutexGuard<'static, HashMap<ZTenantId, Tenant>> {
+    TENANTS.lock().unwrap()
 }
 
 static SHUTDOWN_REQUESTED: AtomicBool = AtomicBool::new(false);
 
 pub fn init(conf: &'static PageServerConf) {
-    let mut m = access_repository();
     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);
+
+        {
+            let mut m = access_tenants();
+            let tenant = Tenant {
+                state: TenantState::CloudOnly,
+                repo: None,
+            };
+            m.insert(tenantid, tenant);
+        }
+
+        init_repo(conf, tenantid);
         info!("initialized storage for tenant: {}", &tenantid);
-        m.insert(tenantid, repo);
     }
 }
 
-fn init_repo(conf: &'static PageServerConf, tenant_id: ZTenantId) -> Arc<LayeredRepository> {
+fn init_repo(conf: &'static PageServerConf, tenant_id: ZTenantId) {
     // Set up a WAL redo manager, for applying WAL records.
     let walredo_mgr = PostgresRedoManager::new(conf, tenant_id);
 
@@ -63,22 +98,15 @@ fn init_repo(conf: &'static PageServerConf, tenant_id: ZTenantId) -> Arc<Layered
         true,
     ));
 
-    let checkpointer_handle = LayeredRepository::launch_checkpointer_thread(conf, repo.clone());
-    let gc_handle = LayeredRepository::launch_gc_thread(conf, repo.clone());
+    let mut m = access_tenants();
+    let tenant = m.get_mut(&tenant_id).unwrap();
+    tenant.repo = Some(repo);
+    tenant.state = TenantState::Active;
 
-    let mut handles = TENANT_HANDLES.lock().unwrap();
-    let h = TenantHandleEntry {
-        checkpointer_handle: Some(checkpointer_handle),
-        gc_handle: Some(gc_handle),
-    };
-
-    handles.insert(tenant_id, h);
-
-    repo
+    // TODO Start these threads only if tenant actively receives some WAL
+    tenant_threads::start_tenant_threads(conf, tenant_id);
 }
 
-// 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,
@@ -89,15 +117,29 @@ pub fn register_relish_download(
         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);
+
+    {
+        let mut m = access_tenants();
+        let tenant = m.entry(tenant_id).or_insert_with(|| Tenant {
+            state: TenantState::Downloading,
+            repo: None,
+        });
+        tenant.state = TenantState::Downloading;
+        match &tenant.repo {
+            Some(repo) => {
+                init_timeline(repo.as_ref(), timeline_id);
+                tenant.state = TenantState::Active;
+                return;
+            }
+            None => log::warn!("Initialize new repo"),
         }
+        tenant.state = TenantState::Active;
     }
+
+    // init repo updates Tenant state
+    init_repo(conf, tenant_id);
+    let new_repo = get_repository_for_tenant(tenant_id).unwrap();
+    init_timeline(new_repo.as_ref(), timeline_id);
 }
 
 fn init_timeline(repo: &dyn Repository, timeline_id: ZTimelineId) {
@@ -112,27 +154,23 @@ pub fn shutdown_requested() -> bool {
     SHUTDOWN_REQUESTED.load(Ordering::Relaxed)
 }
 
-pub fn stop_tenant_threads(tenantid: ZTenantId) {
-    let mut handles = TENANT_HANDLES.lock().unwrap();
-    if let Some(h) = handles.get_mut(&tenantid) {
-        h.checkpointer_handle.take().map(JoinHandle::join);
-        debug!("checkpointer for tenant {} has stopped", tenantid);
-        h.gc_handle.take().map(JoinHandle::join);
-        debug!("gc for tenant {} has stopped", tenantid);
-    }
-}
-
 pub fn shutdown_all_tenants() -> Result<()> {
     SHUTDOWN_REQUESTED.swap(true, Ordering::Relaxed);
 
-    let tenants = list_tenants()?;
-    for tenantid in tenants {
-        stop_tenant_threads(tenantid);
+    let tenantids = list_tenantids()?;
+
+    for tenantid in &tenantids {
+        set_tenant_state(*tenantid, TenantState::Stopping)?;
+    }
+
+    for tenantid in tenantids {
+        // Wait for checkpointer and GC to finish their job
+        tenant_threads::wait_for_tenant_threads_to_stop(tenantid);
+
         let repo = get_repository_for_tenant(tenantid)?;
         debug!("shutdown tenant {}", tenantid);
         repo.shutdown()?;
     }
-
     Ok(())
 }
 
@@ -140,25 +178,67 @@ pub fn create_repository_for_tenant(
     conf: &'static PageServerConf,
     tenantid: ZTenantId,
 ) -> Result<()> {
-    let mut m = access_repository();
-
-    // First check that the tenant doesn't exist already
-    if m.get(&tenantid).is_some() {
-        bail!("tenant {} already exists", tenantid);
+    {
+        let mut m = access_tenants();
+        // First check that the tenant doesn't exist already
+        if m.get(&tenantid).is_some() {
+            bail!("tenant {} already exists", tenantid);
+        }
+        let tenant = Tenant {
+            state: TenantState::CloudOnly,
+            repo: None,
+        };
+        m.insert(tenantid, tenant);
     }
+
     let wal_redo_manager = Arc::new(PostgresRedoManager::new(conf, tenantid));
     let repo = branches::create_repo(conf, tenantid, wal_redo_manager)?;
 
-    m.insert(tenantid, repo);
+    let mut m = access_tenants();
+    let tenant = m.get_mut(&tenantid).unwrap();
+    tenant.repo = Some(repo);
+    tenant.state = TenantState::Active;
 
     Ok(())
 }
 
+// If tenant is not found in the repository, return CloudOnly state
+pub fn get_tenant_state(tenantid: ZTenantId) -> TenantState {
+    let m = access_tenants();
+    match m.get(&tenantid) {
+        Some(tenant) => tenant.state,
+        None => TenantState::CloudOnly,
+    }
+}
+
+pub fn set_tenant_state(tenantid: ZTenantId, state: TenantState) -> Result<TenantState> {
+    let mut m = access_tenants();
+    let tenant = m.get_mut(&tenantid);
+
+    match tenant {
+        Some(tenant) => {
+            if state == TenantState::Idle && tenant.state != TenantState::Active {
+                // Only Active tenant can become Idle
+                return Ok(tenant.state);
+            }
+            info!("set_tenant_state: {} -> {}", tenant.state, state);
+            tenant.state = state;
+            Ok(tenant.state)
+        }
+        None => bail!("Tenant not found for tenant {}", tenantid),
+    }
+}
+
 pub fn get_repository_for_tenant(tenantid: ZTenantId) -> Result<Arc<dyn Repository>> {
-    access_repository()
+    let m = access_tenants();
+    let tenant = m
         .get(&tenantid)
-        .map(Arc::clone)
-        .ok_or_else(|| anyhow!("repository not found for tenant name {}", tenantid))
+        .ok_or_else(|| anyhow!("Tenant not found for tenant {}", tenantid))?;
+
+    match &tenant.repo {
+        Some(repo) => Ok(Arc::clone(repo)),
+        None => anyhow::bail!("Repository for tenant {} is not yet valid", tenantid),
+    }
 }
 
 pub fn get_timeline_for_tenant(
@@ -170,13 +250,32 @@ pub fn get_timeline_for_tenant(
         .with_context(|| format!("cannot fetch timeline {}", timelineid))
 }
 
-fn list_tenants() -> Result<Vec<ZTenantId>> {
-    let o = &mut REPOSITORY.lock().unwrap();
-
-    o.iter()
-        .map(|tenant| {
-            let (tenantid, _) = tenant;
+fn list_tenantids() -> Result<Vec<ZTenantId>> {
+    let m = access_tenants();
+    m.iter()
+        .map(|v| {
+            let (tenantid, _) = v;
             Ok(*tenantid)
         })
         .collect()
 }
+
+#[derive(Serialize, Deserialize, Clone)]
+pub struct TenantInfo {
+    #[serde(with = "hex")]
+    pub id: ZTenantId,
+    pub state: TenantState,
+}
+
+pub fn list_tenants() -> Result<Vec<TenantInfo>> {
+    let m = access_tenants();
+    m.iter()
+        .map(|v| {
+            let (id, tenant) = v;
+            Ok(TenantInfo {
+                id: *id,
+                state: tenant.state,
+            })
+        })
+        .collect()
+}

pageserver/src/tenant_threads.rs

@@ -0,0 +1,125 @@
+//! This module contains functions to serve per-tenant background processes,
+//! such as checkpointer and GC
+use crate::tenant_mgr;
+use crate::tenant_mgr::TenantState;
+use crate::CheckpointConfig;
+use crate::PageServerConf;
+use anyhow::Result;
+use lazy_static::lazy_static;
+use std::collections::HashMap;
+use std::sync::Mutex;
+use std::thread::JoinHandle;
+use std::time::Duration;
+use tracing::*;
+use zenith_utils::zid::ZTenantId;
+
+struct TenantHandleEntry {
+    checkpointer_handle: Option<JoinHandle<()>>,
+    gc_handle: Option<JoinHandle<()>>,
+}
+
+// Preserve handles to wait for thread completion
+// at shutdown
+lazy_static! {
+    static ref TENANT_HANDLES: Mutex<HashMap<ZTenantId, TenantHandleEntry>> =
+        Mutex::new(HashMap::new());
+}
+
+pub fn start_tenant_threads(conf: &'static PageServerConf, tenantid: ZTenantId) {
+    //ensure that old threads are stopeed
+    wait_for_tenant_threads_to_stop(tenantid);
+
+    let checkpointer_handle = std::thread::Builder::new()
+        .name("Checkpointer thread".into())
+        .spawn(move || {
+            checkpoint_loop(tenantid, conf).expect("Checkpointer thread died");
+        })
+        .ok();
+
+    let gc_handle = std::thread::Builder::new()
+        .name("GC thread".into())
+        .spawn(move || {
+            gc_loop(tenantid, conf).expect("GC thread died");
+        })
+        .ok();
+
+    // TODO handle thread errors if any
+
+    let mut handles = TENANT_HANDLES.lock().unwrap();
+    let h = TenantHandleEntry {
+        checkpointer_handle,
+        gc_handle,
+    };
+
+    handles.insert(tenantid, h);
+}
+
+pub fn wait_for_tenant_threads_to_stop(tenantid: ZTenantId) {
+    let mut handles = TENANT_HANDLES.lock().unwrap();
+    if let Some(h) = handles.get_mut(&tenantid) {
+        h.checkpointer_handle.take().map(JoinHandle::join);
+        trace!("checkpointer for tenant {} has stopped", tenantid);
+        h.gc_handle.take().map(JoinHandle::join);
+        trace!("gc for tenant {} has stopped", tenantid);
+    }
+    handles.remove(&tenantid);
+}
+
+///
+/// Checkpointer thread's main loop
+///
+fn checkpoint_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result<()> {
+    loop {
+        if tenant_mgr::get_tenant_state(tenantid) != TenantState::Active {
+            break;
+        }
+
+        std::thread::sleep(conf.checkpoint_period);
+        trace!("checkpointer thread for tenant {} waking up", tenantid);
+
+        // checkpoint timelines that have accumulated more than CHECKPOINT_DISTANCE
+        // bytes of WAL since last checkpoint.
+        let repo = tenant_mgr::get_repository_for_tenant(tenantid)?;
+        repo.checkpoint_iteration(CheckpointConfig::Distance(conf.checkpoint_distance))?;
+    }
+
+    trace!(
+        "checkpointer thread stopped for tenant {} state is {}",
+        tenantid,
+        tenant_mgr::get_tenant_state(tenantid)
+    );
+    Ok(())
+}
+
+///
+/// GC thread's main loop
+///
+fn gc_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result<()> {
+    loop {
+        if tenant_mgr::get_tenant_state(tenantid) != TenantState::Active {
+            break;
+        }
+
+        trace!("gc thread for tenant {} waking up", tenantid);
+
+        // Garbage collect old files that are not needed for PITR anymore
+        if conf.gc_horizon > 0 {
+            let repo = tenant_mgr::get_repository_for_tenant(tenantid)?;
+            repo.gc_iteration(None, conf.gc_horizon, false).unwrap();
+        }
+
+        // TODO Write it in more adequate way using
+        // condvar.wait_timeout() or something
+        let mut sleep_time = conf.gc_period.as_secs();
+        while sleep_time > 0 && tenant_mgr::get_tenant_state(tenantid) == TenantState::Active {
+            sleep_time -= 1;
+            std::thread::sleep(Duration::from_secs(1));
+        }
+    }
+    trace!(
+        "GC thread stopped for tenant {} state is {}",
+        tenantid,
+        tenant_mgr::get_tenant_state(tenantid)
+    );
+    Ok(())
+}

pageserver/src/walreceiver.rs

@@ -284,12 +284,14 @@ fn walreceiver_main(
         if let Some(last_lsn) = status_update {
             // TODO: More thought should go into what values are sent here.
             let last_lsn = PgLsn::from(u64::from(last_lsn));
-            let write_lsn = last_lsn;
+            // We are using disk consistent LSN as `write_lsn`, i.e. LSN at which page server
+            // may guarantee persistence of all received data. Safekeeper is not free to remove
+            // WAL preceding `write_lsn`: it should not be requested by this page server.
+            let write_lsn = PgLsn::from(u64::from(timeline.get_disk_consistent_lsn()));
             let flush_lsn = last_lsn;
             let apply_lsn = PgLsn::from(0);
             let ts = SystemTime::now();
             const NO_REPLY: u8 = 0;
-
             physical_stream.standby_status_update(write_lsn, flush_lsn, apply_lsn, ts, NO_REPLY)?;
         }
 

pageserver/src/walredo.rs

@@ -153,6 +153,13 @@ struct WalRedoRequest {
     records: Vec<(Lsn, WALRecord)>,
 }
 
+impl WalRedoRequest {
+    // Can this request be served by zenith redo funcitons
+    // or we need to pass it to wal-redo postgres process?
+    fn can_apply_in_zenith(&self) -> bool {
+        !matches!(self.rel, RelishTag::Relation(_))
+    }
+}
 /// An error happened in WAL redo
 #[derive(Debug, thiserror::Error)]
 pub enum WalRedoError {
@@ -161,6 +168,8 @@ pub enum WalRedoError {
 
     #[error("cannot perform WAL redo now")]
     InvalidState,
+    #[error("cannot perform WAL redo for this request")]
+    InvalidRequest,
 }
 
 ///
@@ -182,7 +191,6 @@ impl WalRedoManager for PostgresRedoManager {
         records: Vec<(Lsn, WALRecord)>,
     ) -> Result<Bytes, WalRedoError> {
         let start_time;
-        let lock_time;
         let end_time;
 
         let request = WalRedoRequest {
@@ -194,9 +202,16 @@ impl WalRedoManager for PostgresRedoManager {
         };
 
         start_time = Instant::now();
-        let result = {
+        let result;
+
+        if request.can_apply_in_zenith() {
+            result = self.handle_apply_request_zenith(&request);
+
+            end_time = Instant::now();
+            WAL_REDO_TIME.observe(end_time.duration_since(start_time).as_secs_f64());
+        } else {
             let mut process_guard = self.process.lock().unwrap();
-            lock_time = Instant::now();
+            let lock_time = Instant::now();
 
             // launch the WAL redo process on first use
             if process_guard.is_none() {
@@ -207,13 +222,14 @@ impl WalRedoManager for PostgresRedoManager {
             }
             let process = process_guard.as_mut().unwrap();
 
-            self.runtime
-                .block_on(self.handle_apply_request(process, &request))
-        };
-        end_time = Instant::now();
+            result = self
+                .runtime
+                .block_on(self.handle_apply_request_postgres(process, &request));
 
-        WAL_REDO_WAIT_TIME.observe(lock_time.duration_since(start_time).as_secs_f64());
-        WAL_REDO_TIME.observe(end_time.duration_since(lock_time).as_secs_f64());
+            WAL_REDO_WAIT_TIME.observe(lock_time.duration_since(start_time).as_secs_f64());
+            end_time = Instant::now();
+            WAL_REDO_TIME.observe(end_time.duration_since(lock_time).as_secs_f64());
+        }
 
         result
     }
@@ -242,13 +258,47 @@ impl PostgresRedoManager {
     }
 
     ///
-    /// Process one request for WAL redo.
+    /// Process one request for WAL redo using wal-redo postgres
     ///
-    async fn handle_apply_request(
+    async fn handle_apply_request_postgres(
         &self,
         process: &mut PostgresRedoProcess,
         request: &WalRedoRequest,
     ) -> Result<Bytes, WalRedoError> {
+        let blknum = request.blknum;
+        let lsn = request.lsn;
+        let base_img = request.base_img.clone();
+        let records = &request.records;
+        let nrecords = records.len();
+
+        let start = Instant::now();
+
+        let apply_result: Result<Bytes, Error>;
+
+        if let RelishTag::Relation(rel) = request.rel {
+            // Relational WAL records are applied using wal-redo-postgres
+            let buf_tag = BufferTag { rel, blknum };
+            apply_result = process.apply_wal_records(buf_tag, base_img, records).await;
+
+            let duration = start.elapsed();
+
+            debug!(
+                "postgres applied {} WAL records in {} ms to reconstruct page image at LSN {}",
+                nrecords,
+                duration.as_millis(),
+                lsn
+            );
+
+            apply_result.map_err(WalRedoError::IoError)
+        } else {
+            Err(WalRedoError::InvalidRequest)
+        }
+    }
+
+    ///
+    /// Process one request for WAL redo using custom zenith code
+    ///
+    fn handle_apply_request_zenith(&self, request: &WalRedoRequest) -> Result<Bytes, WalRedoError> {
         let rel = request.rel;
         let blknum = request.blknum;
         let lsn = request.lsn;
@@ -260,178 +310,158 @@ impl PostgresRedoManager {
         let start = Instant::now();
 
         let apply_result: Result<Bytes, Error>;
-        if let RelishTag::Relation(rel) = rel {
-            // Relational WAL records are applied using wal-redo-postgres
-            let buf_tag = BufferTag { rel, blknum };
-            apply_result = process.apply_wal_records(buf_tag, base_img, records).await;
+
+        // Non-relational WAL records are handled here, with custom code that has the
+        // same effects as the corresponding Postgres WAL redo function.
+        const ZERO_PAGE: [u8; 8192] = [0u8; 8192];
+        let mut page = BytesMut::new();
+        if let Some(fpi) = base_img {
+            // If full-page image is provided, then use it...
+            page.extend_from_slice(&fpi[..]);
         } else {
-            // Non-relational WAL records are handled here, with custom code that has the
-            // same effects as the corresponding Postgres WAL redo function.
-            const ZERO_PAGE: [u8; 8192] = [0u8; 8192];
-            let mut page = BytesMut::new();
-            if let Some(fpi) = base_img {
-                // If full-page image is provided, then use it...
-                page.extend_from_slice(&fpi[..]);
-            } else {
-                // otherwise initialize page with zeros
-                page.extend_from_slice(&ZERO_PAGE);
+            // otherwise initialize page with zeros
+            page.extend_from_slice(&ZERO_PAGE);
+        }
+        // Apply all collected WAL records
+        for (_lsn, record) in records {
+            let mut buf = record.rec.clone();
+
+            WAL_REDO_RECORD_COUNTER.inc();
+
+            // 1. Parse XLogRecord struct
+            // FIXME: refactor to avoid code duplication.
+            let xlogrec = XLogRecord::from_bytes(&mut buf);
+
+            //move to main data
+            // TODO probably, we should store some records in our special format
+            // to avoid this weird parsing on replay
+            let skip = (record.main_data_offset - pg_constants::SIZEOF_XLOGRECORD) as usize;
+            if buf.remaining() > skip {
+                buf.advance(skip);
             }
-            // Apply all collected WAL records
-            for (_lsn, record) in records {
-                let mut buf = record.rec.clone();
 
-                WAL_REDO_RECORD_COUNTER.inc();
-
-                // 1. Parse XLogRecord struct
-                // FIXME: refactor to avoid code duplication.
-                let xlogrec = XLogRecord::from_bytes(&mut buf);
-
-                //move to main data
-                // TODO probably, we should store some records in our special format
-                // to avoid this weird parsing on replay
-                let skip = (record.main_data_offset - pg_constants::SIZEOF_XLOGRECORD) as usize;
-                if buf.remaining() > skip {
-                    buf.advance(skip);
-                }
-
-                if xlogrec.xl_rmid == pg_constants::RM_XACT_ID {
-                    // Transaction manager stuff
-                    let rec_segno = match rel {
-                        RelishTag::Slru { slru, segno } => {
-                            assert!(
-                                slru == SlruKind::Clog,
-                                "Not valid XACT relish tag {:?}",
-                                rel
+            if xlogrec.xl_rmid == pg_constants::RM_XACT_ID {
+                // Transaction manager stuff
+                let rec_segno = match rel {
+                    RelishTag::Slru { slru, segno } => {
+                        assert!(
+                            slru == SlruKind::Clog,
+                            "Not valid XACT relish tag {:?}",
+                            rel
+                        );
+                        segno
+                    }
+                    _ => panic!("Not valid XACT relish tag {:?}", rel),
+                };
+                let parsed_xact =
+                    XlXactParsedRecord::decode(&mut buf, xlogrec.xl_xid, xlogrec.xl_info);
+                if parsed_xact.info == pg_constants::XLOG_XACT_COMMIT
+                    || parsed_xact.info == pg_constants::XLOG_XACT_COMMIT_PREPARED
+                {
+                    transaction_id_set_status(
+                        parsed_xact.xid,
+                        pg_constants::TRANSACTION_STATUS_COMMITTED,
+                        &mut page,
+                    );
+                    for subxact in &parsed_xact.subxacts {
+                        let pageno = *subxact as u32 / pg_constants::CLOG_XACTS_PER_PAGE;
+                        let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
+                        let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
+                        // only update xids on the requested page
+                        if rec_segno == segno && blknum == rpageno {
+                            transaction_id_set_status(
+                                *subxact,
+                                pg_constants::TRANSACTION_STATUS_COMMITTED,
+                                &mut page,
                             );
-                            segno
-                        }
-                        _ => panic!("Not valid XACT relish tag {:?}", rel),
-                    };
-                    let parsed_xact =
-                        XlXactParsedRecord::decode(&mut buf, xlogrec.xl_xid, xlogrec.xl_info);
-                    if parsed_xact.info == pg_constants::XLOG_XACT_COMMIT
-                        || parsed_xact.info == pg_constants::XLOG_XACT_COMMIT_PREPARED
-                    {
-                        transaction_id_set_status(
-                            parsed_xact.xid,
-                            pg_constants::TRANSACTION_STATUS_COMMITTED,
-                            &mut page,
-                        );
-                        for subxact in &parsed_xact.subxacts {
-                            let pageno = *subxact as u32 / pg_constants::CLOG_XACTS_PER_PAGE;
-                            let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
-                            let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
-                            // only update xids on the requested page
-                            if rec_segno == segno && blknum == rpageno {
-                                transaction_id_set_status(
-                                    *subxact,
-                                    pg_constants::TRANSACTION_STATUS_COMMITTED,
-                                    &mut page,
-                                );
-                            }
-                        }
-                    } else if parsed_xact.info == pg_constants::XLOG_XACT_ABORT
-                        || parsed_xact.info == pg_constants::XLOG_XACT_ABORT_PREPARED
-                    {
-                        transaction_id_set_status(
-                            parsed_xact.xid,
-                            pg_constants::TRANSACTION_STATUS_ABORTED,
-                            &mut page,
-                        );
-                        for subxact in &parsed_xact.subxacts {
-                            let pageno = *subxact as u32 / pg_constants::CLOG_XACTS_PER_PAGE;
-                            let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
-                            let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
-                            // only update xids on the requested page
-                            if rec_segno == segno && blknum == rpageno {
-                                transaction_id_set_status(
-                                    *subxact,
-                                    pg_constants::TRANSACTION_STATUS_ABORTED,
-                                    &mut page,
-                                );
-                            }
                         }
                     }
-                } else if xlogrec.xl_rmid == pg_constants::RM_MULTIXACT_ID {
-                    // Multixact operations
-                    let info = xlogrec.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
-                    if info == pg_constants::XLOG_MULTIXACT_CREATE_ID {
-                        let xlrec = XlMultiXactCreate::decode(&mut buf);
-                        if let RelishTag::Slru {
-                            slru,
-                            segno: rec_segno,
-                        } = rel
-                        {
-                            if slru == SlruKind::MultiXactMembers {
-                                for i in 0..xlrec.nmembers {
-                                    let pageno =
-                                        i / pg_constants::MULTIXACT_MEMBERS_PER_PAGE as u32;
-                                    let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
-                                    let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
-                                    if segno == rec_segno && rpageno == blknum {
-                                        // update only target block
-                                        let offset = xlrec.moff + i;
-                                        let memberoff = mx_offset_to_member_offset(offset);
-                                        let flagsoff = mx_offset_to_flags_offset(offset);
-                                        let bshift = mx_offset_to_flags_bitshift(offset);
-                                        let mut flagsval =
-                                            LittleEndian::read_u32(&page[flagsoff..flagsoff + 4]);
-                                        flagsval &= !(((1
-                                            << pg_constants::MXACT_MEMBER_BITS_PER_XACT)
-                                            - 1)
+                } else if parsed_xact.info == pg_constants::XLOG_XACT_ABORT
+                    || parsed_xact.info == pg_constants::XLOG_XACT_ABORT_PREPARED
+                {
+                    transaction_id_set_status(
+                        parsed_xact.xid,
+                        pg_constants::TRANSACTION_STATUS_ABORTED,
+                        &mut page,
+                    );
+                    for subxact in &parsed_xact.subxacts {
+                        let pageno = *subxact as u32 / pg_constants::CLOG_XACTS_PER_PAGE;
+                        let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
+                        let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
+                        // only update xids on the requested page
+                        if rec_segno == segno && blknum == rpageno {
+                            transaction_id_set_status(
+                                *subxact,
+                                pg_constants::TRANSACTION_STATUS_ABORTED,
+                                &mut page,
+                            );
+                        }
+                    }
+                }
+            } else if xlogrec.xl_rmid == pg_constants::RM_MULTIXACT_ID {
+                // Multixact operations
+                let info = xlogrec.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
+                if info == pg_constants::XLOG_MULTIXACT_CREATE_ID {
+                    let xlrec = XlMultiXactCreate::decode(&mut buf);
+                    if let RelishTag::Slru {
+                        slru,
+                        segno: rec_segno,
+                    } = rel
+                    {
+                        if slru == SlruKind::MultiXactMembers {
+                            for i in 0..xlrec.nmembers {
+                                let pageno = i / pg_constants::MULTIXACT_MEMBERS_PER_PAGE as u32;
+                                let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
+                                let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
+                                if segno == rec_segno && rpageno == blknum {
+                                    // update only target block
+                                    let offset = xlrec.moff + i;
+                                    let memberoff = mx_offset_to_member_offset(offset);
+                                    let flagsoff = mx_offset_to_flags_offset(offset);
+                                    let bshift = mx_offset_to_flags_bitshift(offset);
+                                    let mut flagsval =
+                                        LittleEndian::read_u32(&page[flagsoff..flagsoff + 4]);
+                                    flagsval &=
+                                        !(((1 << pg_constants::MXACT_MEMBER_BITS_PER_XACT) - 1)
                                             << bshift);
-                                        flagsval |= xlrec.members[i as usize].status << bshift;
-                                        LittleEndian::write_u32(
-                                            &mut page[flagsoff..flagsoff + 4],
-                                            flagsval,
-                                        );
-                                        LittleEndian::write_u32(
-                                            &mut page[memberoff..memberoff + 4],
-                                            xlrec.members[i as usize].xid,
-                                        );
-                                    }
+                                    flagsval |= xlrec.members[i as usize].status << bshift;
+                                    LittleEndian::write_u32(
+                                        &mut page[flagsoff..flagsoff + 4],
+                                        flagsval,
+                                    );
+                                    LittleEndian::write_u32(
+                                        &mut page[memberoff..memberoff + 4],
+                                        xlrec.members[i as usize].xid,
+                                    );
                                 }
-                            } else {
-                                // Multixact offsets SLRU
-                                let offs = (xlrec.mid
-                                    % pg_constants::MULTIXACT_OFFSETS_PER_PAGE as u32
-                                    * 4) as usize;
-                                LittleEndian::write_u32(&mut page[offs..offs + 4], xlrec.moff);
                             }
                         } else {
-                            panic!();
+                            // Multixact offsets SLRU
+                            let offs = (xlrec.mid % pg_constants::MULTIXACT_OFFSETS_PER_PAGE as u32
+                                * 4) as usize;
+                            LittleEndian::write_u32(&mut page[offs..offs + 4], xlrec.moff);
                         }
                     } else {
                         panic!();
                     }
+                } else {
+                    panic!();
                 }
             }
-
-            apply_result = Ok::<Bytes, Error>(page.freeze());
         }
 
+        apply_result = Ok::<Bytes, Error>(page.freeze());
+
         let duration = start.elapsed();
 
-        let result: Result<Bytes, WalRedoError>;
-
         debug!(
-            "applied {} WAL records in {} ms to reconstruct page image at LSN {}",
+            "zenith applied {} WAL records in {} ms to reconstruct page image at LSN {}",
             nrecords,
             duration.as_millis(),
             lsn
         );
 
-        if let Err(e) = apply_result {
-            error!("could not apply WAL records: {:#}", e);
-            result = Err(WalRedoError::IoError(e));
-        } else {
-            let img = apply_result.unwrap();
-
-            result = Ok(img);
-        }
-
-        // The caller is responsible for sending the response
-        result
+        apply_result.map_err(WalRedoError::IoError)
     }
 }
 

postgres_ffi/src/xlog_utils.rs

@@ -9,7 +9,6 @@
 
 use crate::pg_constants;
 use crate::CheckPoint;
-use crate::ControlFileData;
 use crate::FullTransactionId;
 use crate::XLogLongPageHeaderData;
 use crate::XLogPageHeaderData;
@@ -18,8 +17,8 @@ use crate::XLOG_PAGE_MAGIC;
 
 use anyhow::{bail, Result};
 use byteorder::{ByteOrder, LittleEndian};
+use bytes::BytesMut;
 use bytes::{Buf, Bytes};
-use bytes::{BufMut, BytesMut};
 use crc32c::*;
 use log::*;
 use std::cmp::max;
@@ -410,27 +409,25 @@ impl CheckPoint {
 }
 
 //
-// Generate new WAL segment with single XLOG_CHECKPOINT_SHUTDOWN record.
+// Generate new, empty WAL segment.
 // We need this segment to start compute node.
-// In order to minimize changes in Postgres core, we prefer to
-// provide WAL segment from which is can extract checkpoint record in standard way,
-// rather then implement some alternative mechanism.
 //
-pub fn generate_wal_segment(pg_control: &ControlFileData) -> Bytes {
+pub fn generate_wal_segment(segno: u64, system_id: u64) -> Bytes {
     let mut seg_buf = BytesMut::with_capacity(pg_constants::WAL_SEGMENT_SIZE as usize);
 
+    let pageaddr = XLogSegNoOffsetToRecPtr(segno, 0, pg_constants::WAL_SEGMENT_SIZE);
     let hdr = XLogLongPageHeaderData {
         std: {
             XLogPageHeaderData {
                 xlp_magic: XLOG_PAGE_MAGIC as u16,
                 xlp_info: pg_constants::XLP_LONG_HEADER,
                 xlp_tli: 1, // FIXME: always use Postgres timeline 1
-                xlp_pageaddr: pg_control.checkPoint - XLOG_SIZE_OF_XLOG_LONG_PHD as u64,
+                xlp_pageaddr: pageaddr,
                 xlp_rem_len: 0,
                 ..Default::default() // Put 0 in padding fields.
             }
         },
-        xlp_sysid: pg_control.system_identifier,
+        xlp_sysid: system_id,
         xlp_seg_size: pg_constants::WAL_SEGMENT_SIZE as u32,
         xlp_xlog_blcksz: XLOG_BLCKSZ as u32,
     };
@@ -438,36 +435,6 @@ pub fn generate_wal_segment(pg_control: &ControlFileData) -> Bytes {
     let hdr_bytes = hdr.encode();
     seg_buf.extend_from_slice(&hdr_bytes);
 
-    let rec_hdr = XLogRecord {
-        xl_tot_len: (XLOG_SIZE_OF_XLOG_RECORD
-            + SIZE_OF_XLOG_RECORD_DATA_HEADER_SHORT
-            + SIZEOF_CHECKPOINT) as u32,
-        xl_xid: 0, //0 is for InvalidTransactionId
-        xl_prev: 0,
-        xl_info: pg_constants::XLOG_CHECKPOINT_SHUTDOWN,
-        xl_rmid: pg_constants::RM_XLOG_ID,
-        xl_crc: 0,
-        ..Default::default() // Put 0 in padding fields.
-    };
-
-    let mut rec_shord_hdr_bytes = BytesMut::new();
-    rec_shord_hdr_bytes.put_u8(pg_constants::XLR_BLOCK_ID_DATA_SHORT);
-    rec_shord_hdr_bytes.put_u8(SIZEOF_CHECKPOINT as u8);
-
-    let rec_bytes = rec_hdr.encode();
-    let checkpoint_bytes = pg_control.checkPointCopy.encode();
-
-    //calculate record checksum
-    let mut crc = 0;
-    crc = crc32c_append(crc, &rec_shord_hdr_bytes[..]);
-    crc = crc32c_append(crc, &checkpoint_bytes[..]);
-    crc = crc32c_append(crc, &rec_bytes[0..XLOG_RECORD_CRC_OFFS]);
-
-    seg_buf.extend_from_slice(&rec_bytes[0..XLOG_RECORD_CRC_OFFS]);
-    seg_buf.put_u32_le(crc);
-    seg_buf.extend_from_slice(&rec_shord_hdr_bytes);
-    seg_buf.extend_from_slice(&checkpoint_bytes);
-
     //zero out the rest of the file
     seg_buf.resize(pg_constants::WAL_SEGMENT_SIZE, 0);
     seg_buf.freeze()

proxy/src/cplane_api.rs

@@ -12,7 +12,14 @@ pub struct DatabaseInfo {
     pub port: u16,
     pub dbname: String,
     pub user: String,
-    pub password: String,
+    pub password: Option<String>,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub struct ProxyAuthResult {
+    pub ready: bool,
+    pub error: Option<String>,
+    pub conn_info: Option<DatabaseInfo>,
 }
 
 impl DatabaseInfo {
@@ -24,12 +31,23 @@ impl DatabaseInfo {
             .next()
             .ok_or_else(|| anyhow::Error::msg("cannot resolve at least one SocketAddr"))
     }
+}
 
-    pub fn conn_string(&self) -> String {
-        format!(
-            "dbname={} user={} password={}",
-            self.dbname, self.user, self.password
-        )
+impl From<DatabaseInfo> for tokio_postgres::Config {
+    fn from(db_info: DatabaseInfo) -> Self {
+        let mut config = tokio_postgres::Config::new();
+
+        config
+            .host(&db_info.host)
+            .port(db_info.port)
+            .dbname(&db_info.dbname)
+            .user(&db_info.user);
+
+        if let Some(password) = db_info.password {
+            config.password(password);
+        }
+
+        config
     }
 }
 
@@ -44,22 +62,25 @@ impl CPlaneApi {
         database: &str,
         md5_response: &[u8],
         salt: &[u8; 4],
-    ) -> Result<DatabaseInfo> {
+        psql_session_id: &str,
+    ) -> Result<ProxyAuthResult> {
         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));
+            .append_pair("salt", &hex::encode(salt))
+            .append_pair("psql_session_id", psql_session_id);
 
         println!("cplane request: {}", url.as_str());
 
         let resp = reqwest::blocking::get(url)?;
 
         if resp.status().is_success() {
-            let conn_info: DatabaseInfo = serde_json::from_str(resp.text()?.as_str())?;
-            println!("got conn info: #{:?}", conn_info);
-            Ok(conn_info)
+            let auth_info: ProxyAuthResult = serde_json::from_str(resp.text()?.as_str())?;
+            println!("got auth info: #{:?}", auth_info);
+
+            Ok(auth_info)
         } else {
             bail!("Auth failed")
         }

proxy/src/main.rs

@@ -7,7 +7,7 @@
 ///
 use std::{
     collections::HashMap,
-    net::{SocketAddr, TcpListener},
+    net::SocketAddr,
     sync::{mpsc, Arc, Mutex},
     thread,
 };
@@ -17,6 +17,7 @@ use clap::{App, Arg, ArgMatches};
 
 use cplane_api::DatabaseInfo;
 use rustls::{internal::pemfile, NoClientAuth, ProtocolVersion, ServerConfig};
+use zenith_utils::tcp_listener;
 
 mod cplane_api;
 mod mgmt;
@@ -99,7 +100,7 @@ fn main() -> anyhow::Result<()> {
                 .long("uri")
                 .takes_value(true)
                 .help("redirect unauthenticated users to given uri")
-                .default_value("http://127.0.0.1:3000/psql_session/"),
+                .default_value("http://localhost:3000/psql_session/"),
         )
         .arg(
             Arg::with_name("auth-endpoint")
@@ -107,7 +108,7 @@ fn main() -> anyhow::Result<()> {
                 .long("auth-endpoint")
                 .takes_value(true)
                 .help("redirect unauthenticated users to given uri")
-                .default_value("http://127.0.0.1:3000/authenticate_proxy_request/"),
+                .default_value("http://localhost:3000/authenticate_proxy_request/"),
         )
         .arg(
             Arg::with_name("ssl-key")
@@ -140,23 +141,23 @@ fn main() -> anyhow::Result<()> {
 
     // Check that we can bind to address before further initialization
     println!("Starting proxy on {}", state.conf.proxy_address);
-    let pageserver_listener = TcpListener::bind(state.conf.proxy_address)?;
+    let pageserver_listener = tcp_listener::bind(state.conf.proxy_address)?;
 
     println!("Starting mgmt on {}", state.conf.mgmt_address);
-    let mgmt_listener = TcpListener::bind(state.conf.mgmt_address)?;
+    let mgmt_listener = tcp_listener::bind(state.conf.mgmt_address)?;
 
-    let threads = vec![
+    let threads = [
         // Spawn a thread to listen for connections. It will spawn further threads
         // for each connection.
         thread::Builder::new()
-            .name("Proxy thread".into())
+            .name("Listener thread".into())
             .spawn(move || proxy::thread_main(state, pageserver_listener))?,
         thread::Builder::new()
             .name("Mgmt thread".into())
             .spawn(move || mgmt::thread_main(state, mgmt_listener))?,
     ];
 
-    for t in threads.into_iter() {
+    for t in threads {
         t.join().unwrap()?;
     }
 

proxy/src/proxy.rs

@@ -6,7 +6,6 @@ use anyhow::bail;
 use tokio_postgres::NoTls;
 
 use rand::Rng;
-use std::io::Write;
 use std::{io, sync::mpsc::channel, thread};
 use zenith_utils::postgres_backend::Stream;
 use zenith_utils::postgres_backend::{PostgresBackend, ProtoState};
@@ -28,11 +27,13 @@ pub fn thread_main(
         println!("accepted connection from {}", peer_addr);
         socket.set_nodelay(true).unwrap();
 
-        thread::spawn(move || {
-            if let Err(err) = proxy_conn_main(state, socket) {
-                println!("error: {}", err);
-            }
-        });
+        thread::Builder::new()
+            .name("Proxy thread".into())
+            .spawn(move || {
+                if let Err(err) = proxy_conn_main(state, socket) {
+                    println!("error: {}", err);
+                }
+            })?;
     }
 }
 
@@ -74,8 +75,12 @@ pub fn proxy_conn_main(
     // This will set conn.existing_user and we can decide on next actions
     conn.handle_startup()?;
 
+    let mut psql_session_id_buf = [0u8; 8];
+    rand::thread_rng().fill(&mut psql_session_id_buf);
+    conn.psql_session_id = hex::encode(psql_session_id_buf);
+
     // both scenarious here should end up producing database connection string
-    let db_info = if conn.is_existing_user() {
+    let conn_info = if conn.is_existing_user() {
         conn.handle_existing_user()?
     } else {
         conn.handle_new_user()?
@@ -83,7 +88,7 @@ pub fn proxy_conn_main(
 
     // XXX: move that inside handle_new_user/handle_existing_user to be able to
     // report wrong connection error.
-    proxy_pass(conn.pgb, db_info)
+    proxy_pass(conn.pgb, conn_info)
 }
 
 impl ProxyConnection {
@@ -155,9 +160,25 @@ impl ProxyConnection {
         Ok(())
     }
 
+    // Wait for proxy kick form the console with conninfo
+    fn wait_for_conninfo(&mut self) -> anyhow::Result<DatabaseInfo> {
+        let (tx, rx) = channel::<anyhow::Result<DatabaseInfo>>();
+        let _ = self
+            .state
+            .waiters
+            .lock()
+            .unwrap()
+            .insert(self.psql_session_id.clone(), tx);
+
+        // Wait for web console response
+        // TODO: respond with error to client
+        rx.recv()?
+    }
+
     fn handle_existing_user(&mut self) -> anyhow::Result<DatabaseInfo> {
         // ask password
         rand::thread_rng().fill(&mut self.md5_salt);
+
         self.pgb
             .write_message(&BeMessage::AuthenticationMD5Password(&self.md5_salt))?;
         self.pgb.state = ProtoState::Authentication; // XXX
@@ -180,14 +201,41 @@ impl ProxyConnection {
                 self.database.as_str(),
                 md5_response,
                 &self.md5_salt,
+                &self.psql_session_id,
             ) {
                 Err(e) => {
-                    self.pgb
-                        .write_message(&BeMessage::ErrorResponse(format!("{}", e)))?;
+                    self.pgb.write_message(&BeMessage::ErrorResponse(format!(
+                        "cannot authenticate proxy: {}",
+                        e
+                    )))?;
 
                     bail!("auth failed: {}", e);
                 }
-                Ok(conn_info) => {
+
+                Ok(auth_info) => {
+                    let conn_info = if auth_info.ready {
+                        // Cluster is ready, so just take `conn_info` and respond to the client.
+                        auth_info
+                            .conn_info
+                            .expect("conn_info should be provided with ready cluster")
+                    } else {
+                        match auth_info.error {
+                            Some(e) => {
+                                self.pgb.write_message(&BeMessage::ErrorResponse(format!(
+                                    "cannot authenticate proxy: {}",
+                                    e
+                                )))?;
+
+                                bail!("auth failed: {}", e);
+                            }
+                            None => {
+                                // Cluster exists, but isn't active, await its start and proxy kick
+                                // with `conn_info`.
+                                self.wait_for_conninfo()?
+                            }
+                        }
+                    };
+
                     self.pgb
                         .write_message_noflush(&BeMessage::AuthenticationOk)?;
                     self.pgb
@@ -203,10 +251,6 @@ impl ProxyConnection {
     }
 
     fn handle_new_user(&mut self) -> anyhow::Result<DatabaseInfo> {
-        let mut psql_session_id_buf = [0u8; 8];
-        rand::thread_rng().fill(&mut psql_session_id_buf);
-        self.psql_session_id = hex::encode(psql_session_id_buf);
-
         let hello_message = format!("☀️  Welcome to Zenith!
 
 To proceed with database creation, open the following link:
@@ -225,76 +269,83 @@ databases without opening the browser.
         self.pgb
             .write_message(&BeMessage::NoticeResponse(hello_message))?;
 
-        // await for database creation
-        let (tx, rx) = channel::<anyhow::Result<DatabaseInfo>>();
-        let _ = self
-            .state
-            .waiters
-            .lock()
-            .unwrap()
-            .insert(self.psql_session_id.clone(), tx);
-
-        // Wait for web console response
-        // XXX: respond with error to client
-        let dbinfo = rx.recv()??;
+        // We requested the DB creation from the console. Now wait for conninfo
+        let conn_info = self.wait_for_conninfo()?;
 
         self.pgb.write_message_noflush(&BeMessage::NoticeResponse(
             "Connecting to database.".to_string(),
         ))?;
         self.pgb.write_message(&BeMessage::ReadyForQuery)?;
 
-        Ok(dbinfo)
+        Ok(conn_info)
     }
 }
 
 /// 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 config = db_info.conn_string().parse::<tokio_postgres::Config>()?;
+    let config = tokio_postgres::Config::from(db_info);
     let _ = config.connect_raw(&mut socket, NoTls).await?;
     Ok(socket)
 }
 
 /// Concurrently proxy both directions of the client and server connections
 fn proxy(
-    client_read: ReadStream,
-    client_write: WriteStream,
-    server_read: ReadStream,
-    server_write: WriteStream,
+    (client_read, client_write): (ReadStream, WriteStream),
+    (server_read, server_write): (ReadStream, WriteStream),
 ) -> anyhow::Result<()> {
-    fn do_proxy(mut reader: ReadStream, mut writer: WriteStream) -> io::Result<()> {
-        std::io::copy(&mut reader, &mut writer)?;
-        writer.flush()?;
-        writer.shutdown(std::net::Shutdown::Both)
+    fn do_proxy(mut reader: impl io::Read, mut writer: WriteStream) -> io::Result<u64> {
+        /// FlushWriter will make sure that every message is sent as soon as possible
+        struct FlushWriter<W>(W);
+
+        impl<W: io::Write> io::Write for FlushWriter<W> {
+            fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+                // `std::io::copy` is guaranteed to exit if we return an error,
+                // so we can afford to lose `res` in case `flush` fails
+                let res = self.0.write(buf);
+                if res.is_ok() {
+                    self.0.flush()?;
+                }
+                res
+            }
+
+            fn flush(&mut self) -> io::Result<()> {
+                self.0.flush()
+            }
+        }
+
+        let res = std::io::copy(&mut reader, &mut FlushWriter(&mut writer));
+        writer.shutdown(std::net::Shutdown::Both)?;
+        res
     }
 
     let client_to_server_jh = thread::spawn(move || do_proxy(client_read, server_write));
 
-    let res1 = do_proxy(server_read, client_write);
-    let res2 = client_to_server_jh.join().unwrap();
-    res1?;
-    res2?;
+    do_proxy(server_read, client_write)?;
+    client_to_server_jh.join().unwrap()?;
 
     Ok(())
 }
 
 /// Proxy a client connection to a postgres database
 fn proxy_pass(pgb: PostgresBackend, db_info: DatabaseInfo) -> anyhow::Result<()> {
-    let runtime = tokio::runtime::Builder::new_current_thread()
-        .enable_all()
-        .build()?;
-    let db_stream = runtime.block_on(connect_to_db(db_info))?;
-    let db_stream = db_stream.into_std()?;
-    db_stream.set_nonblocking(false)?;
+    let db_stream = {
+        // We'll get rid of this once migration to async is complete
+        let runtime = tokio::runtime::Builder::new_current_thread()
+            .enable_all()
+            .build()?;
 
-    let db_stream = zenith_utils::sock_split::BidiStream::from_tcp(db_stream);
-    let (db_read, db_write) = db_stream.split();
+        let stream = runtime.block_on(connect_to_db(db_info))?.into_std()?;
+        stream.set_nonblocking(false)?;
+        stream
+    };
 
-    let stream = match pgb.into_stream() {
+    let db = zenith_utils::sock_split::BidiStream::from_tcp(db_stream);
+
+    let client = match pgb.into_stream() {
         Stream::Bidirectional(bidi_stream) => bidi_stream,
         _ => bail!("invalid stream"),
     };
 
-    let (client_read, client_write) = stream.split();
-    proxy(client_read, client_write, db_read, db_write)
+    proxy(client.split(), db.split())
 }

test_runner/Pipfile

@@ -11,12 +11,19 @@ pyjwt = {extras = ["crypto"], version = "*"}
 requests = "*"
 pytest-xdist = "*"
 asyncpg = "*"
+cached-property = "*"
 
 [dev-packages]
-yapf = "*"
+# Behavior may change slightly between versions. These are run continuously,
+# so we pin exact versions to avoid suprising breaks. Update if comfortable.
+yapf = "==0.31.0"
+mypy = "==0.910"
+# Non-pinned packages follow.
+pipenv = "*"
 flake8 = "*"
-mypy = "*"
+types-requests = "*"
+types-psycopg2 = "*"
 
 [requires]
-# we need at least 3.6, but pipenv doesn't allow to say this directly
+# we need at least 3.7, but pipenv doesn't allow to say this directly
 python_version = "3"

test_runner/Pipfile.lock

@@ -1,7 +1,7 @@
 {
     "_meta": {
         "hash": {
-            "sha256": "3cdc048691824d0b93912b6b78a0aa01dc98f278212c1badb0cc2edbd2103c3a"
+            "sha256": "63b72760ef37375186a638066ba0ad5804dbace99ddc503ea654e9749070ab24"
         },
         "pipfile-spec": 6,
         "requires": {
@@ -43,94 +43,108 @@
             "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3, 3.4'",
             "version": "==21.2.0"
         },
+        "cached-property": {
+            "hashes": [
+                "sha256:9fa5755838eecbb2d234c3aa390bd80fbd3ac6b6869109bfc1b499f7bd89a130",
+                "sha256:df4f613cf7ad9a588cc381aaf4a512d26265ecebd5eb9e1ba12f1319eb85a6a0"
+            ],
+            "index": "pypi",
+            "version": "==1.5.2"
+        },
         "certifi": {
             "hashes": [
-                "sha256:2bbf76fd432960138b3ef6dda3dde0544f27cbf8546c458e60baf371917ba9ee",
-                "sha256:50b1e4f8446b06f41be7dd6338db18e0990601dce795c2b1686458aa7e8fa7d8"
+                "sha256:78884e7c1d4b00ce3cea67b44566851c4343c120abd683433ce934a68ea58872",
+                "sha256:d62a0163eb4c2344ac042ab2bdf75399a71a2d8c7d47eac2e2ee91b9d6339569"
             ],
-            "version": "==2021.5.30"
+            "version": "==2021.10.8"
         },
         "cffi": {
             "hashes": [
-                "sha256:06c54a68935738d206570b20da5ef2b6b6d92b38ef3ec45c5422c0ebaf338d4d",
-                "sha256:0c0591bee64e438883b0c92a7bed78f6290d40bf02e54c5bf0978eaf36061771",
-                "sha256:19ca0dbdeda3b2615421d54bef8985f72af6e0c47082a8d26122adac81a95872",
-                "sha256:22b9c3c320171c108e903d61a3723b51e37aaa8c81255b5e7ce102775bd01e2c",
-                "sha256:26bb2549b72708c833f5abe62b756176022a7b9a7f689b571e74c8478ead51dc",
-                "sha256:33791e8a2dc2953f28b8d8d300dde42dd929ac28f974c4b4c6272cb2955cb762",
-                "sha256:3c8d896becff2fa653dc4438b54a5a25a971d1f4110b32bd3068db3722c80202",
-                "sha256:4373612d59c404baeb7cbd788a18b2b2a8331abcc84c3ba40051fcd18b17a4d5",
-                "sha256:487d63e1454627c8e47dd230025780e91869cfba4c753a74fda196a1f6ad6548",
-                "sha256:48916e459c54c4a70e52745639f1db524542140433599e13911b2f329834276a",
-                "sha256:4922cd707b25e623b902c86188aca466d3620892db76c0bdd7b99a3d5e61d35f",
-                "sha256:55af55e32ae468e9946f741a5d51f9896da6b9bf0bbdd326843fec05c730eb20",
-                "sha256:57e555a9feb4a8460415f1aac331a2dc833b1115284f7ded7278b54afc5bd218",
-                "sha256:5d4b68e216fc65e9fe4f524c177b54964af043dde734807586cf5435af84045c",
-                "sha256:64fda793737bc4037521d4899be780534b9aea552eb673b9833b01f945904c2e",
-                "sha256:6d6169cb3c6c2ad50db5b868db6491a790300ade1ed5d1da29289d73bbe40b56",
-                "sha256:7bcac9a2b4fdbed2c16fa5681356d7121ecabf041f18d97ed5b8e0dd38a80224",
-                "sha256:80b06212075346b5546b0417b9f2bf467fea3bfe7352f781ffc05a8ab24ba14a",
-                "sha256:818014c754cd3dba7229c0f5884396264d51ffb87ec86e927ef0be140bfdb0d2",
-                "sha256:8eb687582ed7cd8c4bdbff3df6c0da443eb89c3c72e6e5dcdd9c81729712791a",
-                "sha256:99f27fefe34c37ba9875f224a8f36e31d744d8083e00f520f133cab79ad5e819",
-                "sha256:9f3e33c28cd39d1b655ed1ba7247133b6f7fc16fa16887b120c0c670e35ce346",
-                "sha256:a8661b2ce9694ca01c529bfa204dbb144b275a31685a075ce123f12331be790b",
-                "sha256:a9da7010cec5a12193d1af9872a00888f396aba3dc79186604a09ea3ee7c029e",
-                "sha256:aedb15f0a5a5949ecb129a82b72b19df97bbbca024081ed2ef88bd5c0a610534",
-                "sha256:b315d709717a99f4b27b59b021e6207c64620790ca3e0bde636a6c7f14618abb",
-                "sha256:ba6f2b3f452e150945d58f4badd92310449876c4c954836cfb1803bdd7b422f0",
-                "sha256:c33d18eb6e6bc36f09d793c0dc58b0211fccc6ae5149b808da4a62660678b156",
-                "sha256:c9a875ce9d7fe32887784274dd533c57909b7b1dcadcc128a2ac21331a9765dd",
-                "sha256:c9e005e9bd57bc987764c32a1bee4364c44fdc11a3cc20a40b93b444984f2b87",
-                "sha256:d2ad4d668a5c0645d281dcd17aff2be3212bc109b33814bbb15c4939f44181cc",
-                "sha256:d950695ae4381ecd856bcaf2b1e866720e4ab9a1498cba61c602e56630ca7195",
-                "sha256:e22dcb48709fc51a7b58a927391b23ab37eb3737a98ac4338e2448bef8559b33",
-                "sha256:e8c6a99be100371dbb046880e7a282152aa5d6127ae01783e37662ef73850d8f",
-                "sha256:e9dc245e3ac69c92ee4c167fbdd7428ec1956d4e754223124991ef29eb57a09d",
-                "sha256:eb687a11f0a7a1839719edd80f41e459cc5366857ecbed383ff376c4e3cc6afd",
-                "sha256:eb9e2a346c5238a30a746893f23a9535e700f8192a68c07c0258e7ece6ff3728",
-                "sha256:ed38b924ce794e505647f7c331b22a693bee1538fdf46b0222c4717b42f744e7",
-                "sha256:f0010c6f9d1a4011e429109fda55a225921e3206e7f62a0c22a35344bfd13cca",
-                "sha256:f0c5d1acbfca6ebdd6b1e3eded8d261affb6ddcf2186205518f1428b8569bb99",
-                "sha256:f10afb1004f102c7868ebfe91c28f4a712227fe4cb24974350ace1f90e1febbf",
-                "sha256:f174135f5609428cc6e1b9090f9268f5c8935fddb1b25ccb8255a2d50de6789e",
-                "sha256:f3ebe6e73c319340830a9b2825d32eb6d8475c1dac020b4f0aa774ee3b898d1c",
-                "sha256:f627688813d0a4140153ff532537fbe4afea5a3dffce1f9deb7f91f848a832b5",
-                "sha256:fd4305f86f53dfd8cd3522269ed7fc34856a8ee3709a5e28b2836b2db9d4cd69"
+                "sha256:00c878c90cb53ccfaae6b8bc18ad05d2036553e6d9d1d9dbcf323bbe83854ca3",
+                "sha256:0104fb5ae2391d46a4cb082abdd5c69ea4eab79d8d44eaaf79f1b1fd806ee4c2",
+                "sha256:06c48159c1abed75c2e721b1715c379fa3200c7784271b3c46df01383b593636",
+                "sha256:0808014eb713677ec1292301ea4c81ad277b6cdf2fdd90fd540af98c0b101d20",
+                "sha256:10dffb601ccfb65262a27233ac273d552ddc4d8ae1bf93b21c94b8511bffe728",
+                "sha256:14cd121ea63ecdae71efa69c15c5543a4b5fbcd0bbe2aad864baca0063cecf27",
+                "sha256:17771976e82e9f94976180f76468546834d22a7cc404b17c22df2a2c81db0c66",
+                "sha256:181dee03b1170ff1969489acf1c26533710231c58f95534e3edac87fff06c443",
+                "sha256:23cfe892bd5dd8941608f93348c0737e369e51c100d03718f108bf1add7bd6d0",
+                "sha256:263cc3d821c4ab2213cbe8cd8b355a7f72a8324577dc865ef98487c1aeee2bc7",
+                "sha256:2756c88cbb94231c7a147402476be2c4df2f6078099a6f4a480d239a8817ae39",
+                "sha256:27c219baf94952ae9d50ec19651a687b826792055353d07648a5695413e0c605",
+                "sha256:2a23af14f408d53d5e6cd4e3d9a24ff9e05906ad574822a10563efcef137979a",
+                "sha256:31fb708d9d7c3f49a60f04cf5b119aeefe5644daba1cd2a0fe389b674fd1de37",
+                "sha256:3415c89f9204ee60cd09b235810be700e993e343a408693e80ce7f6a40108029",
+                "sha256:3773c4d81e6e818df2efbc7dd77325ca0dcb688116050fb2b3011218eda36139",
+                "sha256:3b96a311ac60a3f6be21d2572e46ce67f09abcf4d09344c49274eb9e0bf345fc",
+                "sha256:3f7d084648d77af029acb79a0ff49a0ad7e9d09057a9bf46596dac9514dc07df",
+                "sha256:41d45de54cd277a7878919867c0f08b0cf817605e4eb94093e7516505d3c8d14",
+                "sha256:4238e6dab5d6a8ba812de994bbb0a79bddbdf80994e4ce802b6f6f3142fcc880",
+                "sha256:45db3a33139e9c8f7c09234b5784a5e33d31fd6907800b316decad50af323ff2",
+                "sha256:45e8636704eacc432a206ac7345a5d3d2c62d95a507ec70d62f23cd91770482a",
+                "sha256:4958391dbd6249d7ad855b9ca88fae690783a6be9e86df65865058ed81fc860e",
+                "sha256:4a306fa632e8f0928956a41fa8e1d6243c71e7eb59ffbd165fc0b41e316b2474",
+                "sha256:57e9ac9ccc3101fac9d6014fba037473e4358ef4e89f8e181f8951a2c0162024",
+                "sha256:59888172256cac5629e60e72e86598027aca6bf01fa2465bdb676d37636573e8",
+                "sha256:5e069f72d497312b24fcc02073d70cb989045d1c91cbd53979366077959933e0",
+                "sha256:64d4ec9f448dfe041705426000cc13e34e6e5bb13736e9fd62e34a0b0c41566e",
+                "sha256:6dc2737a3674b3e344847c8686cf29e500584ccad76204efea14f451d4cc669a",
+                "sha256:74fdfdbfdc48d3f47148976f49fab3251e550a8720bebc99bf1483f5bfb5db3e",
+                "sha256:75e4024375654472cc27e91cbe9eaa08567f7fbdf822638be2814ce059f58032",
+                "sha256:786902fb9ba7433aae840e0ed609f45c7bcd4e225ebb9c753aa39725bb3e6ad6",
+                "sha256:8b6c2ea03845c9f501ed1313e78de148cd3f6cad741a75d43a29b43da27f2e1e",
+                "sha256:91d77d2a782be4274da750752bb1650a97bfd8f291022b379bb8e01c66b4e96b",
+                "sha256:91ec59c33514b7c7559a6acda53bbfe1b283949c34fe7440bcf917f96ac0723e",
+                "sha256:920f0d66a896c2d99f0adbb391f990a84091179542c205fa53ce5787aff87954",
+                "sha256:a5263e363c27b653a90078143adb3d076c1a748ec9ecc78ea2fb916f9b861962",
+                "sha256:abb9a20a72ac4e0fdb50dae135ba5e77880518e742077ced47eb1499e29a443c",
+                "sha256:c2051981a968d7de9dd2d7b87bcb9c939c74a34626a6e2f8181455dd49ed69e4",
+                "sha256:c21c9e3896c23007803a875460fb786118f0cdd4434359577ea25eb556e34c55",
+                "sha256:c2502a1a03b6312837279c8c1bd3ebedf6c12c4228ddbad40912d671ccc8a962",
+                "sha256:d4d692a89c5cf08a8557fdeb329b82e7bf609aadfaed6c0d79f5a449a3c7c023",
+                "sha256:da5db4e883f1ce37f55c667e5c0de439df76ac4cb55964655906306918e7363c",
+                "sha256:e7022a66d9b55e93e1a845d8c9eba2a1bebd4966cd8bfc25d9cd07d515b33fa6",
+                "sha256:ef1f279350da2c586a69d32fc8733092fd32cc8ac95139a00377841f59a3f8d8",
+                "sha256:f54a64f8b0c8ff0b64d18aa76675262e1700f3995182267998c31ae974fbc382",
+                "sha256:f5c7150ad32ba43a07c4479f40241756145a1f03b43480e058cfd862bf5041c7",
+                "sha256:f6f824dc3bce0edab5f427efcfb1d63ee75b6fcb7282900ccaf925be84efb0fc",
+                "sha256:fd8a250edc26254fe5b33be00402e6d287f562b6a5b2152dec302fa15bb3e997",
+                "sha256:ffaa5c925128e29efbde7301d8ecaf35c8c60ffbcd6a1ffd3a552177c8e5e796"
             ],
-            "version": "==1.14.6"
+            "version": "==1.15.0"
         },
         "charset-normalizer": {
             "hashes": [
-                "sha256:5d209c0a931f215cee683b6445e2d77677e7e75e159f78def0db09d68fafcaa6",
-                "sha256:5ec46d183433dcbd0ab716f2d7f29d8dee50505b3fdb40c6b985c7c4f5a3591f"
+                "sha256:e019de665e2bcf9c2b64e2e5aa025fa991da8720daa3c1138cadd2fd1856aed0",
+                "sha256:f7af805c321bfa1ce6714c51f254e0d5bb5e5834039bc17db7ebe3a4cec9492b"
             ],
             "markers": "python_version >= '3'",
-            "version": "==2.0.6"
+            "version": "==2.0.7"
         },
         "cryptography": {
             "hashes": [
-                "sha256:0a7dcbcd3f1913f664aca35d47c1331fce738d44ec34b7be8b9d332151b0b01e",
-                "sha256:1eb7bb0df6f6f583dd8e054689def236255161ebbcf62b226454ab9ec663746b",
-                "sha256:21ca464b3a4b8d8e86ba0ee5045e103a1fcfac3b39319727bc0fc58c09c6aff7",
-                "sha256:34dae04a0dce5730d8eb7894eab617d8a70d0c97da76b905de9efb7128ad7085",
-                "sha256:3520667fda779eb788ea00080124875be18f2d8f0848ec00733c0ec3bb8219fc",
-                "sha256:3c4129fc3fdc0fa8e40861b5ac0c673315b3c902bbdc05fc176764815b43dd1d",
-                "sha256:3fa3a7ccf96e826affdf1a0a9432be74dc73423125c8f96a909e3835a5ef194a",
-                "sha256:5b0fbfae7ff7febdb74b574055c7466da334a5371f253732d7e2e7525d570498",
-                "sha256:695104a9223a7239d155d7627ad912953b540929ef97ae0c34c7b8bf30857e89",
-                "sha256:8695456444f277af73a4877db9fc979849cd3ee74c198d04fc0776ebc3db52b9",
-                "sha256:94cc5ed4ceaefcbe5bf38c8fba6a21fc1d365bb8fb826ea1688e3370b2e24a1c",
-                "sha256:94fff993ee9bc1b2440d3b7243d488c6a3d9724cc2b09cdb297f6a886d040ef7",
-                "sha256:9965c46c674ba8cc572bc09a03f4c649292ee73e1b683adb1ce81e82e9a6a0fb",
-                "sha256:a00cf305f07b26c351d8d4e1af84ad7501eca8a342dedf24a7acb0e7b7406e14",
-                "sha256:a305600e7a6b7b855cd798e00278161b681ad6e9b7eca94c721d5f588ab212af",
-                "sha256:cd65b60cfe004790c795cc35f272e41a3df4631e2fb6b35aa7ac6ef2859d554e",
-                "sha256:d2a6e5ef66503da51d2110edf6c403dc6b494cc0082f85db12f54e9c5d4c3ec5",
-                "sha256:d9ec0e67a14f9d1d48dd87a2531009a9b251c02ea42851c060b25c782516ff06",
-                "sha256:f44d141b8c4ea5eb4dbc9b3ad992d45580c1d22bf5e24363f2fbf50c2d7ae8a7"
+                "sha256:07bb7fbfb5de0980590ddfc7f13081520def06dc9ed214000ad4372fb4e3c7f6",
+                "sha256:18d90f4711bf63e2fb21e8c8e51ed8189438e6b35a6d996201ebd98a26abbbe6",
+                "sha256:1ed82abf16df40a60942a8c211251ae72858b25b7421ce2497c2eb7a1cee817c",
+                "sha256:22a38e96118a4ce3b97509443feace1d1011d0571fae81fc3ad35f25ba3ea999",
+                "sha256:2d69645f535f4b2c722cfb07a8eab916265545b3475fdb34e0be2f4ee8b0b15e",
+                "sha256:4a2d0e0acc20ede0f06ef7aa58546eee96d2592c00f450c9acb89c5879b61992",
+                "sha256:54b2605e5475944e2213258e0ab8696f4f357a31371e538ef21e8d61c843c28d",
+                "sha256:7075b304cd567694dc692ffc9747f3e9cb393cc4aa4fb7b9f3abd6f5c4e43588",
+                "sha256:7b7ceeff114c31f285528ba8b390d3e9cfa2da17b56f11d366769a807f17cbaa",
+                "sha256:7eba2cebca600a7806b893cb1d541a6e910afa87e97acf2021a22b32da1df52d",
+                "sha256:928185a6d1ccdb816e883f56ebe92e975a262d31cc536429041921f8cb5a62fd",
+                "sha256:9933f28f70d0517686bd7de36166dda42094eac49415459d9bdf5e7df3e0086d",
+                "sha256:a688ebcd08250eab5bb5bca318cc05a8c66de5e4171a65ca51db6bd753ff8953",
+                "sha256:abb5a361d2585bb95012a19ed9b2c8f412c5d723a9836418fab7aaa0243e67d2",
+                "sha256:c10c797ac89c746e488d2ee92bd4abd593615694ee17b2500578b63cad6b93a8",
+                "sha256:ced40344e811d6abba00295ced98c01aecf0c2de39481792d87af4fa58b7b4d6",
+                "sha256:d57e0cdc1b44b6cdf8af1d01807db06886f10177469312fbde8f44ccbb284bc9",
+                "sha256:d99915d6ab265c22873f1b4d6ea5ef462ef797b4140be4c9d8b179915e0985c6",
+                "sha256:eb80e8a1f91e4b7ef8b33041591e6d89b2b8e122d787e87eeb2b08da71bb16ad",
+                "sha256:ebeddd119f526bcf323a89f853afb12e225902a24d29b55fe18dd6fcb2838a76"
             ],
-            "version": "==3.4.8"
+            "version": "==35.0.0"
         },
         "execnet": {
             "hashes": [
@@ -142,11 +156,19 @@
         },
         "idna": {
             "hashes": [
-                "sha256:14475042e284991034cb48e06f6851428fb14c4dc953acd9be9a5e95c7b6dd7a",
-                "sha256:467fbad99067910785144ce333826c71fb0e63a425657295239737f7ecd125f3"
+                "sha256:84d9dd047ffa80596e0f246e2eab0b391788b0503584e8945f2368256d2735ff",
+                "sha256:9d643ff0a55b762d5cdb124b8eaa99c66322e2157b69160bc32796e824360e6d"
             ],
             "markers": "python_version >= '3'",
-            "version": "==3.2"
+            "version": "==3.3"
+        },
+        "importlib-metadata": {
+            "hashes": [
+                "sha256:b618b6d2d5ffa2f16add5697cf57a46c76a56229b0ed1c438322e4e95645bd15",
+                "sha256:f284b3e11256ad1e5d03ab86bb2ccd6f5339688ff17a4d797a0fe7df326f23b1"
+            ],
+            "markers": "python_version < '3.8'",
+            "version": "==4.8.1"
         },
         "iniconfig": {
             "hashes": [
@@ -207,11 +229,11 @@
                 "crypto"
             ],
             "hashes": [
-                "sha256:934d73fbba91b0483d3857d1aff50e96b2a892384ee2c17417ed3203f173fca1",
-                "sha256:fba44e7898bbca160a2b2b501f492824fc8382485d3a6f11ba5d0c1937ce6130"
+                "sha256:b888b4d56f06f6dcd777210c334e69c737be74755d3e5e9ee3fe67dc18a0ee41",
+                "sha256:e0c4bb8d9f0af0c7f5b1ec4c5036309617d03d56932877f2f7a0beeb5318322f"
             ],
             "index": "pypi",
-            "version": "==2.1.0"
+            "version": "==2.3.0"
         },
         "pyparsing": {
             "hashes": [
@@ -272,21 +294,67 @@
         },
         "urllib3": {
             "hashes": [
-                "sha256:39fb8672126159acb139a7718dd10806104dec1e2f0f6c88aab05d17df10c8d4",
-                "sha256:f57b4c16c62fa2760b7e3d97c35b255512fb6b59a259730f36ba32ce9f8e342f"
+                "sha256:4987c65554f7a2dbf30c18fd48778ef124af6fab771a377103da0585e2336ece",
+                "sha256:c4fdf4019605b6e5423637e01bc9fe4daef873709a7973e195ceba0a62bbc844"
             ],
             "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3, 3.4' and python_version < '4'",
-            "version": "==1.26.6"
+            "version": "==1.26.7"
+        },
+        "zipp": {
+            "hashes": [
+                "sha256:71c644c5369f4a6e07636f0aa966270449561fcea2e3d6747b8d23efaa9d7832",
+                "sha256:9fe5ea21568a0a70e50f273397638d39b03353731e6cbbb3fd8502a33fec40bc"
+            ],
+            "markers": "python_version >= '3.6'",
+            "version": "==3.6.0"
         }
     },
     "develop": {
+        "backports.entry-points-selectable": {
+            "hashes": [
+                "sha256:988468260ec1c196dab6ae1149260e2f5472c9110334e5d51adcb77867361f6a",
+                "sha256:a6d9a871cde5e15b4c4a53e3d43ba890cc6861ec1332c9c2428c92f977192acc"
+            ],
+            "markers": "python_version >= '2.7'",
+            "version": "==1.1.0"
+        },
+        "certifi": {
+            "hashes": [
+                "sha256:78884e7c1d4b00ce3cea67b44566851c4343c120abd683433ce934a68ea58872",
+                "sha256:d62a0163eb4c2344ac042ab2bdf75399a71a2d8c7d47eac2e2ee91b9d6339569"
+            ],
+            "version": "==2021.10.8"
+        },
+        "distlib": {
+            "hashes": [
+                "sha256:c8b54e8454e5bf6237cc84c20e8264c3e991e824ef27e8f1e81049867d861e31",
+                "sha256:d982d0751ff6eaaab5e2ec8e691d949ee80eddf01a62eaa96ddb11531fe16b05"
+            ],
+            "version": "==0.3.3"
+        },
+        "filelock": {
+            "hashes": [
+                "sha256:2b5eb3589e7fdda14599e7eb1a50e09b4cc14f34ed98b8ba56d33bfaafcbef2f",
+                "sha256:34a9f35f95c441e7b38209775d6e0337f9a3759f3565f6c5798f19618527c76f"
+            ],
+            "markers": "python_version >= '3.6'",
+            "version": "==3.3.1"
+        },
         "flake8": {
             "hashes": [
-                "sha256:07528381786f2a6237b061f6e96610a4167b226cb926e2aa2b6b1d78057c576b",
-                "sha256:bf8fd333346d844f616e8d47905ef3a3384edae6b4e9beb0c5101e25e3110907"
+                "sha256:479b1304f72536a55948cb40a32dce8bb0ffe3501e26eaf292c7e60eb5e0428d",
+                "sha256:806e034dda44114815e23c16ef92f95c91e4c71100ff52813adf7132a6ad870d"
             ],
             "index": "pypi",
-            "version": "==3.9.2"
+            "version": "==4.0.1"
+        },
+        "importlib-metadata": {
+            "hashes": [
+                "sha256:b618b6d2d5ffa2f16add5697cf57a46c76a56229b0ed1c438322e4e95645bd15",
+                "sha256:f284b3e11256ad1e5d03ab86bb2ccd6f5339688ff17a4d797a0fe7df326f23b1"
+            ],
+            "markers": "python_version < '3.8'",
+            "version": "==4.8.1"
         },
         "mccabe": {
             "hashes": [
@@ -331,21 +399,45 @@
             ],
             "version": "==0.4.3"
         },
+        "pipenv": {
+            "hashes": [
+                "sha256:05958fadcd70b2de6a27542fcd2bd72dd5c59c6d35307fdac3e06361fb06e30e",
+                "sha256:d180f5be4775c552fd5e69ae18a9d6099d9dafb462efe54f11c72cb5f4d5e977"
+            ],
+            "index": "pypi",
+            "version": "==2021.5.29"
+        },
+        "platformdirs": {
+            "hashes": [
+                "sha256:367a5e80b3d04d2428ffa76d33f124cf11e8fff2acdaa9b43d545f5c7d661ef2",
+                "sha256:8868bbe3c3c80d42f20156f22e7131d2fb321f5bc86a2a345375c6481a67021d"
+            ],
+            "markers": "python_version >= '3.6'",
+            "version": "==2.4.0"
+        },
         "pycodestyle": {
             "hashes": [
-                "sha256:514f76d918fcc0b55c6680472f0a37970994e07bbb80725808c17089be302068",
-                "sha256:c389c1d06bf7904078ca03399a4816f974a1d590090fecea0c63ec26ebaf1cef"
+                "sha256:720f8b39dde8b293825e7ff02c475f3077124006db4f440dcbc9a20b76548a20",
+                "sha256:eddd5847ef438ea1c7870ca7eb78a9d47ce0cdb4851a5523949f2601d0cbbe7f"
             ],
-            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3'",
-            "version": "==2.7.0"
+            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3, 3.4'",
+            "version": "==2.8.0"
         },
         "pyflakes": {
             "hashes": [
-                "sha256:7893783d01b8a89811dd72d7dfd4d84ff098e5eed95cfa8905b22bbffe52efc3",
-                "sha256:f5bc8ecabc05bb9d291eb5203d6810b49040f6ff446a756326104746cc00c1db"
+                "sha256:05a85c2872edf37a4ed30b0cce2f6093e1d0581f8c19d7393122da7e25b2b24c",
+                "sha256:3bb3a3f256f4b7968c9c788781e4ff07dce46bdf12339dcda61053375426ee2e"
             ],
             "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3'",
-            "version": "==2.3.1"
+            "version": "==2.4.0"
+        },
+        "six": {
+            "hashes": [
+                "sha256:1e61c37477a1626458e36f7b1d82aa5c9b094fa4802892072e49de9c60c4c926",
+                "sha256:8abb2f1d86890a2dfb989f9a77cfcfd3e47c2a354b01111771326f8aa26e0254"
+            ],
+            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3'",
+            "version": "==1.16.0"
         },
         "toml": {
             "hashes": [
@@ -355,6 +447,58 @@
             "markers": "python_version >= '2.6' and python_version not in '3.0, 3.1, 3.2, 3.3'",
             "version": "==0.10.2"
         },
+        "typed-ast": {
+            "hashes": [
+                "sha256:01ae5f73431d21eead5015997ab41afa53aa1fbe252f9da060be5dad2c730ace",
+                "sha256:067a74454df670dcaa4e59349a2e5c81e567d8d65458d480a5b3dfecec08c5ff",
+                "sha256:0fb71b8c643187d7492c1f8352f2c15b4c4af3f6338f21681d3681b3dc31a266",
+                "sha256:1b3ead4a96c9101bef08f9f7d1217c096f31667617b58de957f690c92378b528",
+                "sha256:2068531575a125b87a41802130fa7e29f26c09a2833fea68d9a40cf33902eba6",
+                "sha256:209596a4ec71d990d71d5e0d312ac935d86930e6eecff6ccc7007fe54d703808",
+                "sha256:2c726c276d09fc5c414693a2de063f521052d9ea7c240ce553316f70656c84d4",
+                "sha256:398e44cd480f4d2b7ee8d98385ca104e35c81525dd98c519acff1b79bdaac363",
+                "sha256:52b1eb8c83f178ab787f3a4283f68258525f8d70f778a2f6dd54d3b5e5fb4341",
+                "sha256:5feca99c17af94057417d744607b82dd0a664fd5e4ca98061480fd8b14b18d04",
+                "sha256:7538e495704e2ccda9b234b82423a4038f324f3a10c43bc088a1636180f11a41",
+                "sha256:760ad187b1041a154f0e4d0f6aae3e40fdb51d6de16e5c99aedadd9246450e9e",
+                "sha256:777a26c84bea6cd934422ac2e3b78863a37017618b6e5c08f92ef69853e765d3",
+                "sha256:95431a26309a21874005845c21118c83991c63ea800dd44843e42a916aec5899",
+                "sha256:9ad2c92ec681e02baf81fdfa056fe0d818645efa9af1f1cd5fd6f1bd2bdfd805",
+                "sha256:9c6d1a54552b5330bc657b7ef0eae25d00ba7ffe85d9ea8ae6540d2197a3788c",
+                "sha256:aee0c1256be6c07bd3e1263ff920c325b59849dc95392a05f258bb9b259cf39c",
+                "sha256:af3d4a73793725138d6b334d9d247ce7e5f084d96284ed23f22ee626a7b88e39",
+                "sha256:b36b4f3920103a25e1d5d024d155c504080959582b928e91cb608a65c3a49e1a",
+                "sha256:b9574c6f03f685070d859e75c7f9eeca02d6933273b5e69572e5ff9d5e3931c3",
+                "sha256:bff6ad71c81b3bba8fa35f0f1921fb24ff4476235a6e94a26ada2e54370e6da7",
+                "sha256:c190f0899e9f9f8b6b7863debfb739abcb21a5c054f911ca3596d12b8a4c4c7f",
+                "sha256:c907f561b1e83e93fad565bac5ba9c22d96a54e7ea0267c708bffe863cbe4075",
+                "sha256:cae53c389825d3b46fb37538441f75d6aecc4174f615d048321b716df2757fb0",
+                "sha256:dd4a21253f42b8d2b48410cb31fe501d32f8b9fbeb1f55063ad102fe9c425e40",
+                "sha256:dde816ca9dac1d9c01dd504ea5967821606f02e510438120091b84e852367428",
+                "sha256:f2362f3cb0f3172c42938946dbc5b7843c2a28aec307c49100c8b38764eb6927",
+                "sha256:f328adcfebed9f11301eaedfa48e15bdece9b519fb27e6a8c01aa52a17ec31b3",
+                "sha256:f8afcf15cc511ada719a88e013cec87c11aff7b91f019295eb4530f96fe5ef2f",
+                "sha256:fb1bbeac803adea29cedd70781399c99138358c26d05fcbd23c13016b7f5ec65"
+            ],
+            "markers": "python_version < '3.8'",
+            "version": "==1.4.3"
+        },
+        "types-psycopg2": {
+            "hashes": [
+                "sha256:77ed80f2668582654623e04fb3d741ecce93effcc39c929d7e02f4a917a538ce",
+                "sha256:98a6e0e9580cd7eb4bd4d20f7c7063d154b2589a2b90c0ce4e3ca6085cde77c6"
+            ],
+            "index": "pypi",
+            "version": "==2.9.1"
+        },
+        "types-requests": {
+            "hashes": [
+                "sha256:b279284e51f668e38ee12d9665e4d789089f532dc2a0be4a1508ca0efd98ba9e",
+                "sha256:ba1d108d512e294b6080c37f6ae7cb2a2abf527560e2b671d1786c1fc46b541a"
+            ],
+            "index": "pypi",
+            "version": "==2.25.11"
+        },
         "typing-extensions": {
             "hashes": [
                 "sha256:49f75d16ff11f1cd258e1b988ccff82a3ca5570217d7ad8c5f48205dd99a677e",
@@ -364,6 +508,22 @@
             "index": "pypi",
             "version": "==3.10.0.2"
         },
+        "virtualenv": {
+            "hashes": [
+                "sha256:10062e34c204b5e4ec5f62e6ef2473f8ba76513a9a617e873f1f8fb4a519d300",
+                "sha256:bcc17f0b3a29670dd777d6f0755a4c04f28815395bca279cdcb213b97199a6b8"
+            ],
+            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3, 3.4'",
+            "version": "==20.8.1"
+        },
+        "virtualenv-clone": {
+            "hashes": [
+                "sha256:418ee935c36152f8f153c79824bb93eaf6f0f7984bae31d3f48f350b9183501a",
+                "sha256:44d5263bceed0bac3e1424d64f798095233b64def1c5689afa43dc3223caf5b0"
+            ],
+            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3'",
+            "version": "==0.5.7"
+        },
         "yapf": {
             "hashes": [
                 "sha256:408fb9a2b254c302f49db83c59f9aa0b4b0fd0ec25be3a5c51181327922ff63d",
@@ -371,6 +531,14 @@
             ],
             "index": "pypi",
             "version": "==0.31.0"
+        },
+        "zipp": {
+            "hashes": [
+                "sha256:71c644c5369f4a6e07636f0aa966270449561fcea2e3d6747b8d23efaa9d7832",
+                "sha256:9fe5ea21568a0a70e50f273397638d39b03353731e6cbbb3fd8502a33fec40bc"
+            ],
+            "markers": "python_version >= '3.6'",
+            "version": "==3.6.0"
         }
     }
 }

test_runner/README.md

@@ -3,10 +3,13 @@
 This directory contains integration tests.
 
 Prerequisites:
-- Python 3.6 or later
+- Python 3.7 or later
+    - Development headers may also be needed to build `psycopg2` from source.
+    - Python 3.7 is recommended if you want to update tests.
 - Dependencies: install them via `pipenv install`. Note that Debian/Ubuntu
   packages are stale, as it commonly happens, so manual installation is not
   recommended.
+  Exact version of `pipenv` is not important unless you change dependencies.
   Run `pipenv shell` to activate the venv or use `pipenv run` to run a single
   command in the venv, e.g. `pipenv run pytest`.
 - Zenith and Postgres binaries
@@ -53,8 +56,8 @@ Useful environment variables:
 should go.
 `TEST_SHARED_FIXTURES`: Try to re-use a single pageserver for all the tests.
 
-Let stdout and stderr go to the terminal instead of capturing them:
-`pytest -s ...`
+Let stdout, stderr and `INFO` log messages go to the terminal instead of capturing them:
+`pytest -s --log-cli-level=INFO ...`
 (Note many tests capture subprocess outputs separately, so this may not
 show much.)
 
@@ -62,44 +65,87 @@ Exit after the first test failure:
 `pytest -x ...`
 (there are many more pytest options; run `pytest -h` to see them.)
 
+### Writing a test
 
-### Building new tests
+Every test needs a Zenith Environment, or ZenithEnv to operate in. A Zenith Environment
+is like a little cloud-in-a-box, and consists of a Pageserver, 0-N Safekeepers, and
+compute Postgres nodes. The connections between them can be configured to use JWT
+authentication tokens, and some other configuration options can be tweaked too.
 
-The tests make heavy use of pytest fixtures. You can read about how they work here: https://docs.pytest.org/en/stable/fixture.html
+The easiest way to get access to a Zenith Environment is by using the `zenith_simple_env`
+fixture. The 'simple' env may be shared across multiple tests, so don't shut down the nodes
+or make other destructive changes in that environment. Also don't assume that
+there are no tenants or branches or data in the cluster. For convenience, there is a
+branch called `empty`, though. The convention is to create a test-specific branch of
+that and load any test data there, instead of the 'main' branch.
 
-Essentially, this means that each time you see a fixture named as an input parameter, the function with that name will be run and passed as a parameter to the function.
-
-So this code:
+For more complicated cases, you can build a custom Zenith Environment, with the `zenith_env`
+fixture:
 
 ```python
-def test_something(zenith_cli, pg_bin):
-    pass
+def test_foobar(zenith_env_builder: ZenithEnvBuilder):
+    # Prescribe the environment.
+    # We want to have 3 safekeeper nodes, and use JWT authentication in the
+    # connections to the page server
+    zenith_env_builder.num_safekeepers = 3
+    zenith_env_builder.set_pageserver_auth(True)
+
+    # Now create the environment. This initializes the repository, and starts
+    # up the page server and the safekeepers
+    env = zenith_env_builder.init()
+
+    # Run the test
+    ...
 ```
 
-... will run the fixtures called `zenith_cli` and `pg_bin` and deliver those results to the test function.
+For more information about pytest fixtures, see https://docs.pytest.org/en/stable/fixture.html
 
-Fixtures can't be imported using the normal python syntax. Instead, use this:
+At the end of a test, all the nodes in the environment are automatically stopped, so you
+don't need to worry about cleaning up. Logs and test data are preserved for the analysis,
+in a directory under `../test_output/<testname>`
 
-```python
-pytest_plugins = ("fixtures.something")
+### Before submitting a patch
+#### Obligatory checks
+Install dev dependencies via `pipenv --python 3.7 install --dev` (better)
+or `pipenv install --dev` (if you don't have Python 3.7 and don't need to change dependencies).
+
+We force code formatting via yapf and type hints via mypy.
+Run the following commands in the `test_runner/` directory:
+
+```bash
+pipenv run yapf -ri .  # All code is reformatted
+pipenv run mypy .  # Ensure there are no typing errors
 ```
 
-That will make all the fixtures in the `fixtures/something.py` file available.
-
-Anything that's likely to be used in multiple tests should be built into a fixture.
-
-Note that fixtures can clean up after themselves if they use the `yield` syntax.
-Cleanup will happen even if the test fails (raises an unhandled exception).
-Python destructors, e.g. `__del__()` aren't recommended for cleanup.
-
-
-### Code quality
-
-Before submitting a patch, please consider:
-
+#### Advisable actions
 * Writing a couple of docstrings to clarify the reasoning behind a new test.
 * Running `flake8` (or a linter of your choice, e.g. `pycodestyle`) and fixing possible defects, if any.
-* Formatting the code with `yapf -r -i .` (TODO: implement an opt-in pre-commit hook for that).
-* (Optional) Typechecking the code with `mypy .`. Currently this mostly affects `fixtures/zenith_fixtures.py`.
+* Adding more type hints to your code to avoid `Any`, especially:
+  * For fixture parameters, they are not automatically deduced.
+  * For function arguments and return values.
 
-The tools can be installed with `pipenv install --dev`.
+#### Changing dependencies
+You have to update `Pipfile.lock` if you have changed `Pipfile`:
+
+```bash
+pipenv --python 3.7 install --dev  # Re-create venv for Python 3.7 and install recent pipenv inside
+pipenv run pipenv --version  # Should be at least 2021.5.29
+pipenv run pipenv lock  # Regenerate Pipfile.lock
+```
+
+As the minimal supported version is Python 3.7 and we use it in CI,
+you have to use a Python 3.7 environment when updating `Pipfile.lock`.
+Otherwise some back-compatibility packages will be missing.
+
+It is also important to run recent `pipenv`.
+Older versions remove markers from `Pipfile.lock`.
+
+If you don't have Python 3.7, you should install it and its headers (for `psycopg2`)
+separately, e.g.:
+
+```bash
+# In Ubuntu
+sudo add-apt-repository ppa:deadsnakes/ppa
+sudo apt update
+sudo apt install python3.7 python3.7-dev
+```

test_runner/batch_others/test_auth.py

@@ -1,21 +1,22 @@
-
 from contextlib import closing
 from typing import Iterator
 from uuid import uuid4
 import psycopg2
-from fixtures.zenith_fixtures import PortDistributor, Postgres, ZenithCli, ZenithPageserver, PgBin
+from fixtures.zenith_fixtures import ZenithEnvBuilder
 import pytest
 
-
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
-def test_pageserver_auth(pageserver_auth_enabled: ZenithPageserver):
-    ps = pageserver_auth_enabled
+def test_pageserver_auth(zenith_env_builder: ZenithEnvBuilder):
+    zenith_env_builder.pageserver_auth_enabled = True
+    env = zenith_env_builder.init()
 
-    tenant_token = ps.auth_keys.generate_tenant_token(ps.initial_tenant)
-    invalid_tenant_token = ps.auth_keys.generate_tenant_token(uuid4().hex)
-    management_token = ps.auth_keys.generate_management_token()
+    ps = env.pageserver
+
+    tenant_token = env.auth_keys.generate_tenant_token(env.initial_tenant)
+    invalid_tenant_token = env.auth_keys.generate_tenant_token(uuid4().hex)
+    management_token = env.auth_keys.generate_management_token()
 
     # this does not invoke auth check and only decodes jwt and checks it for validity
     # check both tokens
@@ -23,57 +24,41 @@ def test_pageserver_auth(pageserver_auth_enabled: ZenithPageserver):
     ps.safe_psql("status", password=management_token)
 
     # tenant can create branches
-    ps.safe_psql(f"branch_create {ps.initial_tenant} new1 main", password=tenant_token)
+    ps.safe_psql(f"branch_create {env.initial_tenant} new1 main", password=tenant_token)
     # console can create branches for tenant
-    ps.safe_psql(f"branch_create {ps.initial_tenant} new2 main", password=management_token)
+    ps.safe_psql(f"branch_create {env.initial_tenant} new2 main", password=management_token)
 
     # fail to create branch using token with different tenantid
     with pytest.raises(psycopg2.DatabaseError, match='Tenant id mismatch. Permission denied'):
-        ps.safe_psql(f"branch_create {ps.initial_tenant} new2 main", password=invalid_tenant_token)
+        ps.safe_psql(f"branch_create {env.initial_tenant} new2 main", password=invalid_tenant_token)
 
     # create tenant using management token
     ps.safe_psql(f"tenant_create {uuid4().hex}", password=management_token)
 
     # fail to create tenant using tenant token
-    with pytest.raises(psycopg2.DatabaseError, match='Attempt to access management api with tenant scope. Permission denied'):
+    with pytest.raises(
+            psycopg2.DatabaseError,
+            match='Attempt to access management api with tenant scope. Permission denied'):
         ps.safe_psql(f"tenant_create {uuid4().hex}", password=tenant_token)
 
 
 @pytest.mark.parametrize('with_wal_acceptors', [False, True])
-def test_compute_auth_to_pageserver(
-    zenith_cli: ZenithCli,
-    wa_factory,
-    pageserver_auth_enabled: ZenithPageserver,
-    repo_dir: str,
-    with_wal_acceptors: bool,
-    pg_bin: PgBin,
-    port_distributor: PortDistributor,
-):
-    ps = pageserver_auth_enabled
-    # since we are in progress of refactoring protocols between compute safekeeper and page server
-    # use hardcoded management token in safekeeper
-    management_token = ps.auth_keys.generate_management_token()
+def test_compute_auth_to_pageserver(zenith_env_builder: ZenithEnvBuilder, with_wal_acceptors: bool):
+    zenith_env_builder.pageserver_auth_enabled = True
+    if with_wal_acceptors:
+        zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
 
     branch = f"test_compute_auth_to_pageserver{with_wal_acceptors}"
-    zenith_cli.run(["branch", branch, "empty"])
-    if with_wal_acceptors:
-        wa_factory.start_n_new(3, management_token)
+    env.zenith_cli(["branch", branch, "main"])
 
-    with Postgres(
-        zenith_cli=zenith_cli,
-        repo_dir=repo_dir,
-        pg_bin=pg_bin,
-        tenant_id=ps.initial_tenant,
-        port=port_distributor.get_port(),
-    ).create_start(
-        branch,
-        wal_acceptors=wa_factory.get_connstrs() if with_wal_acceptors else None,
-    ) as pg:
-        with closing(pg.connect()) as conn:
-            with conn.cursor() as cur:
-                # we rely upon autocommit after each statement
-                # as waiting for acceptors happens there
-                cur.execute('CREATE TABLE t(key int primary key, value text)')
-                cur.execute("INSERT INTO t SELECT generate_series(1,100000), 'payload'")
-                cur.execute('SELECT sum(key) FROM t')
-                assert cur.fetchone() == (5000050000, )
+    pg = env.postgres.create_start(branch)
+
+    with closing(pg.connect()) as conn:
+        with conn.cursor() as cur:
+            # we rely upon autocommit after each statement
+            # as waiting for acceptors happens there
+            cur.execute('CREATE TABLE t(key int primary key, value text)')
+            cur.execute("INSERT INTO t SELECT generate_series(1,100000), 'payload'")
+            cur.execute('SELECT sum(key) FROM t')
+            assert cur.fetchone() == (5000050000, )

test_runner/batch_others/test_branch_behind.py

@@ -1,5 +1,5 @@
 import subprocess
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -8,11 +8,12 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 #
 # Create a couple of branches off the main branch, at a historical point in time.
 #
-def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin):
+def test_branch_behind(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
     # Branch at the point where only 100 rows were inserted
-    zenith_cli.run(["branch", "test_branch_behind", "empty"])
+    env.zenith_cli(["branch", "test_branch_behind", "empty"])
 
-    pgmain = postgres.create_start('test_branch_behind')
+    pgmain = env.postgres.create_start('test_branch_behind')
     log.info("postgres is running on 'test_branch_behind' branch")
 
     main_pg_conn = pgmain.connect()
@@ -40,7 +41,7 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     log.info(f'LSN after 200100 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])
+    env.zenith_cli(["branch", "test_branch_behind_hundred", "test_branch_behind@" + lsn_a])
 
     # Insert many more rows. This generates enough WAL to fill a few segments.
     main_cur.execute('''
@@ -55,10 +56,10 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     log.info(f'LSN after 400100 rows: {lsn_c}')
 
     # Branch at the point where only 200100 rows were inserted
-    zenith_cli.run(["branch", "test_branch_behind_more", "test_branch_behind@" + lsn_b])
+    env.zenith_cli(["branch", "test_branch_behind_more", "test_branch_behind@" + lsn_b])
 
-    pg_hundred = postgres.create_start("test_branch_behind_hundred")
-    pg_more = postgres.create_start("test_branch_behind_more")
+    pg_hundred = env.postgres.create_start("test_branch_behind_hundred")
+    pg_more = env.postgres.create_start("test_branch_behind_more")
 
     # On the 'hundred' branch, we should see only 100 rows
     hundred_pg_conn = pg_hundred.connect()
@@ -79,14 +80,17 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
     # Check bad lsn's for branching
 
     # branch at segment boundary
-    zenith_cli.run(["branch", "test_branch_segment_boundary", "test_branch_behind@0/3000000"])
-    pg = postgres.create_start("test_branch_segment_boundary")
+    env.zenith_cli(["branch", "test_branch_segment_boundary", "test_branch_behind@0/3000000"])
+    pg = env.postgres.create_start("test_branch_segment_boundary")
     cur = pg.connect().cursor()
     cur.execute('SELECT 1')
     assert cur.fetchone() == (1, )
 
     # branch at pre-initdb lsn
+    #
+    # FIXME: This works currently, but probably shouldn't be allowed
     try:
-        zenith_cli.run(["branch", "test_branch_preinitdb", "test_branch_behind@0/42"])
+        env.zenith_cli(["branch", "test_branch_preinitdb", "test_branch_behind@0/42"])
+        # FIXME: assert false, "branch with invalid LSN should have failed"
     except subprocess.CalledProcessError:
         log.info("Branch creation with pre-initdb LSN failed (as expected)")

test_runner/batch_others/test_clog_truncate.py

@@ -3,7 +3,7 @@ import os
 
 from contextlib import closing
 
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -12,19 +12,23 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 #
 # Test compute node start after clog truncation
 #
-def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin):
+def test_clog_truncate(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_clog_truncate", "empty"])
+    env.zenith_cli(["branch", "test_clog_truncate", "empty"])
 
     # set agressive autovacuum to make sure that truncation will happen
     config = [
-        'autovacuum_max_workers=10', 'autovacuum_vacuum_threshold=0',
-        'autovacuum_vacuum_insert_threshold=0', 'autovacuum_vacuum_cost_delay=0',
-        'autovacuum_vacuum_cost_limit=10000', 'autovacuum_naptime =1s',
+        'autovacuum_max_workers=10',
+        'autovacuum_vacuum_threshold=0',
+        'autovacuum_vacuum_insert_threshold=0',
+        'autovacuum_vacuum_cost_delay=0',
+        'autovacuum_vacuum_cost_limit=10000',
+        'autovacuum_naptime =1s',
         'autovacuum_freeze_max_age=100000'
     ]
 
-    pg = postgres.create_start('test_clog_truncate', config_lines=config)
+    pg = env.postgres.create_start('test_clog_truncate', config_lines=config)
     log.info('postgres is running on test_clog_truncate branch')
 
     # Install extension containing function needed for test
@@ -61,10 +65,10 @@ def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
 
     # create new branch after clog truncation and start a compute node on it
     log.info(f'create branch at lsn_after_truncation {lsn_after_truncation}')
-    zenith_cli.run(
+    env.zenith_cli(
         ["branch", "test_clog_truncate_new", "test_clog_truncate@" + lsn_after_truncation])
 
-    pg2 = postgres.create_start('test_clog_truncate_new')
+    pg2 = env.postgres.create_start('test_clog_truncate_new')
     log.info('postgres is running on test_clog_truncate_new branch')
 
     # check that new node doesn't contain truncated segment

test_runner/batch_others/test_config.py

@@ -1,6 +1,6 @@
 from contextlib import closing
 
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -9,12 +9,13 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 #
 # Test starting Postgres with custom options
 #
-def test_config(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin):
+def test_config(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_config", "empty"])
+    env.zenith_cli(["branch", "test_config", "empty"])
 
     # change config
-    pg = postgres.create_start('test_config', config_lines=['log_min_messages=debug1'])
+    pg = env.postgres.create_start('test_config', config_lines=['log_min_messages=debug1'])
     log.info('postgres is running on test_config branch')
 
     with closing(pg.connect()) as conn:

test_runner/batch_others/test_createdropdb.py

@@ -2,7 +2,7 @@ import os
 import pathlib
 
 from contextlib import closing
-from fixtures.zenith_fixtures import ZenithPageserver, PostgresFactory, ZenithCli, check_restored_datadir_content
+from fixtures.zenith_fixtures import ZenithEnv, check_restored_datadir_content
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -11,15 +11,11 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 #
 # Test CREATE DATABASE when there have been relmapper changes
 #
-def test_createdb(
-    zenith_cli: ZenithCli,
-    pageserver: ZenithPageserver,
-    postgres: PostgresFactory,
-    pg_bin,
-):
-    zenith_cli.run(["branch", "test_createdb", "empty"])
+def test_createdb(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    env.zenith_cli(["branch", "test_createdb", "empty"])
 
-    pg = postgres.create_start('test_createdb')
+    pg = env.postgres.create_start('test_createdb')
     log.info("postgres is running on 'test_createdb' branch")
 
     with closing(pg.connect()) as conn:
@@ -33,27 +29,23 @@ def test_createdb(
             lsn = cur.fetchone()[0]
 
     # Create a branch
-    zenith_cli.run(["branch", "test_createdb2", "test_createdb@" + lsn])
+    env.zenith_cli(["branch", "test_createdb2", "test_createdb@" + lsn])
 
-    pg2 = postgres.create_start('test_createdb2')
+    pg2 = env.postgres.create_start('test_createdb2')
 
     # Test that you can connect to the new database on both branches
     for db in (pg, pg2):
         db.connect(dbname='foodb').close()
 
+
 #
 # Test DROP DATABASE
 #
-def test_dropdb(
-    zenith_cli: ZenithCli,
-    pageserver: ZenithPageserver,
-    postgres: PostgresFactory,
-    pg_bin,
-    test_output_dir
-):
-    zenith_cli.run(["branch", "test_dropdb", "empty"])
+def test_dropdb(zenith_simple_env: ZenithEnv, test_output_dir):
+    env = zenith_simple_env
+    env.zenith_cli(["branch", "test_dropdb", "empty"])
 
-    pg = postgres.create_start('test_dropdb')
+    pg = env.postgres.create_start('test_dropdb')
     log.info("postgres is running on 'test_dropdb' branch")
 
     with closing(pg.connect()) as conn:
@@ -66,7 +58,6 @@ def test_dropdb(
             cur.execute("SELECT oid FROM pg_database WHERE datname='foodb';")
             dboid = cur.fetchone()[0]
 
-
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
             cur.execute('DROP DATABASE foodb')
@@ -76,28 +67,29 @@ def test_dropdb(
             cur.execute('SELECT pg_current_wal_insert_lsn()')
             lsn_after_drop = cur.fetchone()[0]
 
-
     # Create two branches before and after database drop.
-    zenith_cli.run(["branch", "test_before_dropdb", "test_dropdb@" + lsn_before_drop])
-    pg_before = postgres.create_start('test_before_dropdb')
+    env.zenith_cli(["branch", "test_before_dropdb", "test_dropdb@" + lsn_before_drop])
+    pg_before = env.postgres.create_start('test_before_dropdb')
 
-    zenith_cli.run(["branch", "test_after_dropdb", "test_dropdb@" + lsn_after_drop])
-    pg_after = postgres.create_start('test_after_dropdb')
+    env.zenith_cli(["branch", "test_after_dropdb", "test_dropdb@" + lsn_after_drop])
+    pg_after = env.postgres.create_start('test_after_dropdb')
 
     # Test that database exists on the branch before drop
     pg_before.connect(dbname='foodb').close()
 
     # Test that database subdir exists on the branch before drop
+    assert pg_before.pgdata_dir
     dbpath = pathlib.Path(pg_before.pgdata_dir) / 'base' / str(dboid)
     log.info(dbpath)
 
     assert os.path.isdir(dbpath) == True
 
     # Test that database subdir doesn't exist on the branch after drop
+    assert pg_after.pgdata_dir
     dbpath = pathlib.Path(pg_after.pgdata_dir) / 'base' / str(dboid)
     log.info(dbpath)
 
     assert os.path.isdir(dbpath) == False
 
     # Check that we restore the content of the datadir correctly
-    check_restored_datadir_content(zenith_cli, test_output_dir, pg, pageserver.service_port.pg)
+    check_restored_datadir_content(test_output_dir, env, pg)

test_runner/batch_others/test_createuser.py

@@ -1,6 +1,6 @@
 from contextlib import closing
 
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -9,10 +9,11 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 #
 # Test CREATE USER to check shared catalog restore
 #
-def test_createuser(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin):
-    zenith_cli.run(["branch", "test_createuser", "empty"])
+def test_createuser(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    env.zenith_cli(["branch", "test_createuser", "empty"])
 
-    pg = postgres.create_start('test_createuser')
+    pg = env.postgres.create_start('test_createuser')
     log.info("postgres is running on 'test_createuser' branch")
 
     with closing(pg.connect()) as conn:
@@ -26,9 +27,9 @@ def test_createuser(zenith_cli, pageserver: ZenithPageserver, postgres: Postgres
             lsn = cur.fetchone()[0]
 
     # Create a branch
-    zenith_cli.run(["branch", "test_createuser2", "test_createuser@" + lsn])
+    env.zenith_cli(["branch", "test_createuser2", "test_createuser@" + lsn])
 
-    pg2 = postgres.create_start('test_createuser2')
+    pg2 = env.postgres.create_start('test_createuser2')
 
     # Test that you can connect to new branch as a new user
     assert pg2.safe_psql('select current_user', username='testuser') == [('testuser', )]

test_runner/batch_others/test_multixact.py

@@ -1,4 +1,4 @@
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver, check_restored_datadir_content
+from fixtures.zenith_fixtures import ZenithEnv, check_restored_datadir_content
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -10,11 +10,11 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 # it only checks next_multixact_id field in restored pg_control,
 # since we don't have functions to check multixact internals.
 #
-def test_multixact(pageserver: ZenithPageserver, postgres: PostgresFactory,
-                    pg_bin, zenith_cli, base_dir, test_output_dir):
+def test_multixact(zenith_simple_env: ZenithEnv, test_output_dir):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_multixact", "empty"])
-    pg = postgres.create_start('test_multixact')
+    env.zenith_cli(["branch", "test_multixact", "empty"])
+    pg = env.postgres.create_start('test_multixact')
 
     log.info("postgres is running on 'test_multixact' branch")
     pg_conn = pg.connect()
@@ -53,8 +53,8 @@ def test_multixact(pageserver: ZenithPageserver, postgres: PostgresFactory,
     assert int(next_multixact_id) > int(next_multixact_id_old)
 
     # Branch at this point
-    zenith_cli.run(["branch", "test_multixact_new", "test_multixact@" + lsn])
-    pg_new = postgres.create_start('test_multixact_new')
+    env.zenith_cli(["branch", "test_multixact_new", "test_multixact@" + lsn])
+    pg_new = env.postgres.create_start('test_multixact_new')
 
     log.info("postgres is running on 'test_multixact_new' branch")
     pg_new_conn = pg_new.connect()
@@ -67,4 +67,4 @@ def test_multixact(pageserver: ZenithPageserver, postgres: PostgresFactory,
     assert next_multixact_id_new == next_multixact_id
 
     # Check that we restore the content of the datadir correctly
-    check_restored_datadir_content(zenith_cli, test_output_dir, pg_new, pageserver.service_port.pg)
+    check_restored_datadir_content(test_output_dir, env, pg_new)

test_runner/batch_others/test_old_request_lsn.py

@@ -1,10 +1,11 @@
 from contextlib import closing
 
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
+
 #
 # Test where Postgres generates a lot of WAL, and it's garbage collected away, but
 # no pages are evicted so that Postgres uses an old LSN in a GetPage request.
@@ -15,10 +16,11 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 # just a hint that the page hasn't been modified since that LSN, and the page
 # server should return the latest page version regardless of the LSN.
 #
-def test_old_request_lsn(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin):
+def test_old_request_lsn(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_old_request_lsn", "empty"])
-    pg = postgres.create_start('test_old_request_lsn')
+    env.zenith_cli(["branch", "test_old_request_lsn", "empty"])
+    pg = env.postgres.create_start('test_old_request_lsn')
     log.info('postgres is running on test_old_request_lsn branch')
 
     pg_conn = pg.connect()
@@ -28,7 +30,7 @@ def test_old_request_lsn(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
     cur.execute("SHOW zenith.zenith_timeline")
     timeline = cur.fetchone()[0]
 
-    psconn = pageserver.connect()
+    psconn = env.pageserver.connect()
     pscur = psconn.cursor()
 
     # Create table, and insert some rows. Make it big enough that it doesn't fit in
@@ -47,20 +49,20 @@ def test_old_request_lsn(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
         from pg_settings where name = 'shared_buffers'
     ''')
     row = cur.fetchone()
-    log.info(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');
+    cur.execute('VACUUM foo')
 
     # Make a lot of updates on a single row, generating a lot of WAL. Trigger
     # garbage collections so that the page server will remove old page versions.
     for i in range(10):
-        pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+        pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
         for j in range(100):
-            cur.execute('UPDATE foo SET val = val + 1 WHERE id = 1;');
+            cur.execute('UPDATE foo SET val = val + 1 WHERE id = 1;')
 
     # All (or at least most of) the updates should've been on the same page, so
     # that we haven't had to evict any dirty pages for a long time. Now run
     # a query that sends GetPage@LSN requests with the old LSN.
-    cur.execute("SELECT COUNT(*), SUM(val) FROM foo");
+    cur.execute("SELECT COUNT(*), SUM(val) FROM foo")
     assert cur.fetchone() == (100000, 101000)

test_runner/batch_others/test_pageserver_api.py

@@ -3,25 +3,28 @@ from uuid import uuid4
 import pytest
 import psycopg2
 import requests
-from fixtures.zenith_fixtures import ZenithPageserver, ZenithPageserverHttpClient
+from fixtures.zenith_fixtures import ZenithEnv, ZenithEnvBuilder, ZenithPageserverHttpClient
+from typing import cast
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
-def test_status_psql(pageserver):
-    assert pageserver.safe_psql('status') == [
+def test_status_psql(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    assert env.pageserver.safe_psql('status') == [
         ('hello world', ),
     ]
 
 
-def test_branch_list_psql(pageserver: ZenithPageserver, zenith_cli):
+def test_branch_list_psql(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_branch_list_main", "empty"])
+    env.zenith_cli(["branch", "test_branch_list_main", "empty"])
 
-    conn = pageserver.connect()
+    conn = env.pageserver.connect()
     cur = conn.cursor()
 
-    cur.execute(f'branch_list {pageserver.initial_tenant}')
+    cur.execute(f'branch_list {env.initial_tenant}')
     branches = json.loads(cur.fetchone()[0])
     # Filter out branches created by other tests
     branches = [x for x in branches if x['name'].startswith('test_branch_list')]
@@ -34,10 +37,10 @@ def test_branch_list_psql(pageserver: ZenithPageserver, zenith_cli):
     assert 'ancestor_lsn' in branches[0]
 
     # Create another branch, and start Postgres on it
-    zenith_cli.run(['branch', 'test_branch_list_experimental', 'test_branch_list_main'])
-    zenith_cli.run(['pg', 'create', 'test_branch_list_experimental'])
+    env.zenith_cli(['branch', 'test_branch_list_experimental', 'test_branch_list_main'])
+    env.zenith_cli(['pg', 'create', 'test_branch_list_experimental'])
 
-    cur.execute(f'branch_list {pageserver.initial_tenant}')
+    cur.execute(f'branch_list {env.initial_tenant}')
     new_branches = json.loads(cur.fetchone()[0])
     # Filter out branches created by other tests
     new_branches = [x for x in new_branches if x['name'].startswith('test_branch_list')]
@@ -53,18 +56,22 @@ def test_branch_list_psql(pageserver: ZenithPageserver, zenith_cli):
     conn.close()
 
 
-def test_tenant_list_psql(pageserver: ZenithPageserver, zenith_cli):
-    res = zenith_cli.run(["tenant", "list"])
-    res.check_returncode()
-    tenants = res.stdout.splitlines()
-    assert tenants == [pageserver.initial_tenant]
+def test_tenant_list_psql(zenith_env_builder: ZenithEnvBuilder):
+    # don't use zenith_simple_env, because there might be other tenants there,
+    # left over from other tests.
+    env = zenith_env_builder.init()
 
-    conn = pageserver.connect()
+    res = env.zenith_cli(["tenant", "list"])
+    res.check_returncode()
+    tenants = sorted(map(lambda t: t.split()[0], res.stdout.splitlines()))
+    assert tenants == [env.initial_tenant]
+
+    conn = env.pageserver.connect()
     cur = conn.cursor()
 
     # check same tenant cannot be created twice
-    with pytest.raises(psycopg2.DatabaseError, match=f'tenant {pageserver.initial_tenant} already exists'):
-        cur.execute(f'tenant_create {pageserver.initial_tenant}')
+    with pytest.raises(psycopg2.DatabaseError, match=f'tenant {env.initial_tenant} already exists'):
+        cur.execute(f'tenant_create {env.initial_tenant}')
 
     # create one more tenant
     tenant1 = uuid4().hex
@@ -73,20 +80,20 @@ def test_tenant_list_psql(pageserver: ZenithPageserver, zenith_cli):
     cur.execute('tenant_list')
 
     # compare tenants list
-    new_tenants = sorted(json.loads(cur.fetchone()[0]))
-    assert sorted([pageserver.initial_tenant, tenant1]) == new_tenants
+    new_tenants = sorted(map(lambda t: cast(str, t['id']), json.loads(cur.fetchone()[0])))
+    assert sorted([env.initial_tenant, tenant1]) == new_tenants
 
 
 def check_client(client: ZenithPageserverHttpClient, initial_tenant: str):
     client.check_status()
 
     # check initial tenant is there
-    assert initial_tenant in set(client.tenant_list())
+    assert initial_tenant in {t['id'] for t in client.tenant_list()}
 
     # create new tenant and check it is also there
     tenant_id = uuid4()
     client.tenant_create(tenant_id)
-    assert tenant_id.hex in set(client.tenant_list())
+    assert tenant_id.hex in {t['id'] for t in client.tenant_list()}
 
     # create branch
     branch_name = uuid4().hex
@@ -96,11 +103,17 @@ def check_client(client: ZenithPageserverHttpClient, initial_tenant: str):
     assert branch_name in {b['name'] for b in client.branch_list(tenant_id)}
 
 
-def test_pageserver_http_api_client(pageserver: ZenithPageserver):
-    client = pageserver.http_client()
-    check_client(client, pageserver.initial_tenant)
+def test_pageserver_http_api_client(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    client = env.pageserver.http_client()
+    check_client(client, env.initial_tenant)
 
 
-def test_pageserver_http_api_client_auth_enabled(pageserver_auth_enabled: ZenithPageserver):
-    client = pageserver_auth_enabled.http_client(auth_token=pageserver_auth_enabled.auth_keys.generate_management_token())
-    check_client(client, pageserver_auth_enabled.initial_tenant)
+def test_pageserver_http_api_client_auth_enabled(zenith_env_builder: ZenithEnvBuilder):
+    zenith_env_builder.pageserver_auth_enabled = True
+    env = zenith_env_builder.init()
+
+    management_token = env.auth_keys.generate_management_token()
+
+    client = env.pageserver.http_client(auth_token=management_token)
+    check_client(client, env.initial_tenant)

test_runner/batch_others/test_pageserver_restart.py

@@ -4,22 +4,22 @@ import time
 
 from contextlib import closing
 from multiprocessing import Process, Value
-from fixtures.zenith_fixtures import WalAcceptorFactory, ZenithPageserver, PostgresFactory
+from fixtures.zenith_fixtures import ZenithEnvBuilder
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
+
 # Check that dead minority doesn't prevent the commits: execute insert n_inserts
 # times, with fault_probability chance of getting a wal acceptor down or up
 # along the way. 2 of 3 are always alive, so the work keeps going.
-def test_pageserver_restart(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, wa_factory: WalAcceptorFactory):
-
+def test_pageserver_restart(zenith_env_builder: ZenithEnvBuilder):
     # One safekeeper is enough for this test.
-    wa_factory.start_n_new(1)
+    zenith_env_builder.num_safekeepers = 1
+    env = zenith_env_builder.init()
 
-    zenith_cli.run(["branch", "test_pageserver_restart", "empty"])
-    pg = postgres.create_start('test_pageserver_restart',
-                               wal_acceptors=wa_factory.get_connstrs())
+    env.zenith_cli(["branch", "test_pageserver_restart", "main"])
+    pg = env.postgres.create_start('test_pageserver_restart')
 
     pg_conn = pg.connect()
     cur = pg_conn.cursor()
@@ -41,14 +41,14 @@ def test_pageserver_restart(zenith_cli, pageserver: ZenithPageserver, postgres:
         from pg_settings where name = 'shared_buffers'
     ''')
     row = cur.fetchone()
-    log.info(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])
 
     # Stop and restart pageserver. This is a more or less graceful shutdown, although
     # the page server doesn't currently have a shutdown routine so there's no difference
     # between stopping and crashing.
-    pageserver.stop();
-    pageserver.start();
+    env.pageserver.stop()
+    env.pageserver.start()
 
     # Stopping the pageserver breaks the connection from the postgres backend to
     # the page server, and causes the next query on the connection to fail. Start a new
@@ -62,6 +62,5 @@ def test_pageserver_restart(zenith_cli, pageserver: ZenithPageserver, postgres:
     assert cur.fetchone() == (100000, )
 
     # Stop the page server by force, and restart it
-    pageserver.stop();
-    pageserver.start();
-
+    env.pageserver.stop()
+    env.pageserver.start()

test_runner/batch_others/test_pgbench.py

@@ -1,14 +1,15 @@
-from fixtures.zenith_fixtures import PostgresFactory
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
-def test_pgbench(postgres: PostgresFactory, pg_bin, zenith_cli):
+def test_pgbench(zenith_simple_env: ZenithEnv, pg_bin):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_pgbench", "empty"])
+    env.zenith_cli(["branch", "test_pgbench", "empty"])
 
-    pg = postgres.create_start('test_pgbench')
+    pg = env.postgres.create_start('test_pgbench')
     log.info("postgres is running on 'test_pgbench' branch")
 
     connstr = pg.connstr()

test_runner/batch_others/test_readonly_node.py

@@ -0,0 +1,91 @@
+import subprocess
+from fixtures.zenith_fixtures import ZenithEnv
+
+pytest_plugins = ("fixtures.zenith_fixtures")
+
+
+#
+# Create read-only compute nodes, anchored at historical points in time.
+#
+# This is very similar to the 'test_branch_behind' test, but instead of
+# creating branches, creates read-only nodes.
+#
+def test_readonly_node(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    env.zenith_cli(["branch", "test_readonly_node", "empty"])
+
+    pgmain = env.postgres.create_start('test_readonly_node')
+    print("postgres is running on 'test_readonly_node' branch")
+
+    main_pg_conn = pgmain.connect()
+    main_cur = main_pg_conn.cursor()
+
+    # Create table, and insert the first 100 rows
+    main_cur.execute('CREATE TABLE foo (t text)')
+    main_cur.execute('''
+        INSERT INTO foo
+            SELECT 'long string to consume some space' || g
+            FROM generate_series(1, 100) g
+    ''')
+    main_cur.execute('SELECT pg_current_wal_insert_lsn()')
+    lsn_a = main_cur.fetchone()[0]
+    print('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
+    ''')
+    main_cur.execute('SELECT pg_current_wal_insert_lsn()')
+    lsn_b = main_cur.fetchone()[0]
+    print('LSN after 200100 rows: ' + lsn_b)
+
+    # Insert many 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
+    ''')
+
+    main_cur.execute('SELECT pg_current_wal_insert_lsn()')
+    lsn_c = main_cur.fetchone()[0]
+    print('LSN after 400100 rows: ' + lsn_c)
+
+    # Create first read-only node at the point where only 100 rows were inserted
+    pg_hundred = env.postgres.create_start("test_readonly_node_hundred",
+                                           branch=f'test_readonly_node@{lsn_a}')
+
+    # And another at the point where 200100 rows were inserted
+    pg_more = env.postgres.create_start("test_readonly_node_more",
+                                        branch=f'test_readonly_node@{lsn_b}')
+
+    # On the 'hundred' node, we should see only 100 rows
+    hundred_pg_conn = pg_hundred.connect()
+    hundred_cur = hundred_pg_conn.cursor()
+    hundred_cur.execute('SELECT count(*) FROM foo')
+    assert hundred_cur.fetchone() == (100, )
+
+    # On the 'more' node, we should see 100200 rows
+    more_pg_conn = pg_more.connect()
+    more_cur = more_pg_conn.cursor()
+    more_cur.execute('SELECT count(*) FROM foo')
+    assert more_cur.fetchone() == (200100, )
+
+    # All the rows are visible on the main branch
+    main_cur.execute('SELECT count(*) FROM foo')
+    assert main_cur.fetchone() == (400100, )
+
+    # Check creating a node at segment boundary
+    pg = env.postgres.create_start("test_branch_segment_boundary",
+                                   branch="test_readonly_node@0/3000000")
+    cur = pg.connect().cursor()
+    cur.execute('SELECT 1')
+    assert cur.fetchone() == (1, )
+
+    # Create node at pre-initdb lsn
+    try:
+        env.zenith_cli(["pg", "start", "test_branch_preinitdb", "test_readonly_node@0/42"])
+        assert False, "compute node startup with invalid LSN should have failed"
+    except Exception:
+        print("Node creation with pre-initdb LSN failed (as expected)")

test_runner/batch_others/test_restart_compute.py

@@ -1,7 +1,7 @@
 import pytest
 
 from contextlib import closing
-from fixtures.zenith_fixtures import ZenithPageserver, PostgresFactory
+from fixtures.zenith_fixtures import ZenithEnvBuilder
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -11,23 +11,15 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 # Test restarting and recreating a postgres instance
 #
 @pytest.mark.parametrize('with_wal_acceptors', [False, True])
-def test_restart_compute(
-        zenith_cli,
-        pageserver: ZenithPageserver,
-        postgres: PostgresFactory,
-        pg_bin,
-        wa_factory,
-        with_wal_acceptors: bool,
-    ):
-    wal_acceptor_connstrs = None
-    zenith_cli.run(["branch", "test_restart_compute", "empty"])
-
+def test_restart_compute(zenith_env_builder: ZenithEnvBuilder, with_wal_acceptors: bool):
+    zenith_env_builder.pageserver_auth_enabled = True
     if with_wal_acceptors:
-        wa_factory.start_n_new(3)
-        wal_acceptor_connstrs = wa_factory.get_connstrs()
+        zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
 
-    pg = postgres.create_start('test_restart_compute',
-                               wal_acceptors=wal_acceptor_connstrs)
+    env.zenith_cli(["branch", "test_restart_compute", "main"])
+
+    pg = env.postgres.create_start('test_restart_compute')
     log.info("postgres is running on 'test_restart_compute' branch")
 
     with closing(pg.connect()) as conn:
@@ -40,9 +32,7 @@ def test_restart_compute(
             log.info(f"res = {r}")
 
     # Remove data directory and restart
-    pg.stop_and_destroy().create_start('test_restart_compute',
-                                       wal_acceptors=wal_acceptor_connstrs)
-
+    pg.stop_and_destroy().create_start('test_restart_compute')
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
@@ -61,8 +51,7 @@ def test_restart_compute(
             log.info(f"res = {r}")
 
     # Again remove data directory and restart
-    pg.stop_and_destroy().create_start('test_restart_compute',
-                                       wal_acceptors=wal_acceptor_connstrs)
+    pg.stop_and_destroy().create_start('test_restart_compute')
 
     # That select causes lots of FPI's and increases probability of wakeepers
     # lagging behind after query completion
@@ -76,8 +65,7 @@ def test_restart_compute(
             log.info(f"res = {r}")
 
     # And again remove data directory and restart
-    pg.stop_and_destroy().create_start('test_restart_compute',
-                                       wal_acceptors=wal_acceptor_connstrs)
+    pg.stop_and_destroy().create_start('test_restart_compute')
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:

test_runner/batch_others/test_snapfiles_gc.py

@@ -1,14 +1,20 @@
 from contextlib import closing
 import psycopg2.extras
 import time
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
+
 def print_gc_result(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))
+    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))
 
 
 #
@@ -17,14 +23,15 @@ def print_gc_result(row):
 # This test is pretty tightly coupled with the current implementation of layered
 # storage, in layered_repository.rs.
 #
-def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
-    zenith_cli.run(["branch", "test_layerfiles_gc", "empty"])
-    pg = postgres.create_start('test_layerfiles_gc')
+def test_layerfiles_gc(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    env.zenith_cli(["branch", "test_layerfiles_gc", "empty"])
+    pg = env.postgres.create_start('test_layerfiles_gc')
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
-            with closing(pageserver.connect()) as psconn:
-                with psconn.cursor(cursor_factory = psycopg2.extras.DictCursor) as pscur:
+            with closing(env.pageserver.connect()) as psconn:
+                with psconn.cursor(cursor_factory=psycopg2.extras.DictCursor) as pscur:
 
                     # Get the timeline ID of our branch. We need it for the 'do_gc' command
                     cur.execute("SHOW zenith.zenith_timeline")
@@ -34,9 +41,9 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     cur.execute("CREATE TABLE foo(x integer)")
                     cur.execute("INSERT INTO foo VALUES (1)")
 
-                    cur.execute("select relfilenode from pg_class where oid = 'foo'::regclass");
-                    row = cur.fetchone();
-                    log.info(f"relfilenode is {row[0]}");
+                    cur.execute("select relfilenode from pg_class where oid = 'foo'::regclass")
+                    row = cur.fetchone()
+                    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
@@ -52,11 +59,12 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     cur.execute("DELETE FROM foo")
 
                     log.info("Running GC before test")
-                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
-                    print_gc_result(row);
+                    print_gc_result(row)
                     # remember the number of files
-                    layer_relfiles_remain = row['layer_relfiles_total'] - row['layer_relfiles_removed']
+                    layer_relfiles_remain = (row['layer_relfiles_total'] -
+                                             row['layer_relfiles_removed'])
                     assert layer_relfiles_remain > 0
 
                     # Insert a row and run GC. Checkpoint should freeze the layer
@@ -64,9 +72,9 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     # removing the old image and delta layer.
                     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")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
-                    print_gc_result(row);
+                    print_gc_result(row)
                     assert row['layer_relfiles_total'] == layer_relfiles_remain + 2
                     assert row['layer_relfiles_removed'] == 2
                     assert row['layer_relfiles_dropped'] == 0
@@ -78,9 +86,9 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     cur.execute("INSERT INTO foo VALUES (2)")
                     cur.execute("INSERT INTO foo VALUES (3)")
 
-                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
-                    print_gc_result(row);
+                    print_gc_result(row)
                     assert row['layer_relfiles_total'] == layer_relfiles_remain + 2
                     assert row['layer_relfiles_removed'] == 2
                     assert row['layer_relfiles_dropped'] == 0
@@ -90,18 +98,18 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                     cur.execute("INSERT INTO foo VALUES (2)")
                     cur.execute("INSERT INTO foo VALUES (3)")
 
-                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
-                    print_gc_result(row);
+                    print_gc_result(row)
                     assert row['layer_relfiles_total'] == layer_relfiles_remain + 2
                     assert row['layer_relfiles_removed'] == 2
                     assert row['layer_relfiles_dropped'] == 0
 
                     # Run GC again, with no changes in the database. Should not remove anything.
                     log.info("Run GC again, with nothing to do")
-                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
-                    print_gc_result(row);
+                    print_gc_result(row)
                     assert row['layer_relfiles_total'] == layer_relfiles_remain
                     assert row['layer_relfiles_removed'] == 0
                     assert row['layer_relfiles_dropped'] == 0
@@ -109,12 +117,12 @@ 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
                     #
-                    log.info("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")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
                     row = pscur.fetchone()
-                    print_gc_result(row);
+                    print_gc_result(row)
 
                     # We still cannot remove the latest layers
                     # because they serve as tombstones for earlier layers.

test_runner/batch_others/test_tenants.py

@@ -2,39 +2,41 @@ from contextlib import closing
 
 import pytest
 
-from fixtures.zenith_fixtures import (
-    TenantFactory,
-    ZenithCli,
-    PostgresFactory,
-)
+from fixtures.zenith_fixtures import ZenithEnvBuilder
 
 
 @pytest.mark.parametrize('with_wal_acceptors', [False, True])
-def test_tenants_normal_work(
-    zenith_cli: ZenithCli,
-    tenant_factory: TenantFactory,
-    postgres: PostgresFactory,
-    wa_factory,
-    with_wal_acceptors: bool,
-):
-    """Tests tenants with and without wal acceptors"""
-    tenant_1 = tenant_factory.create()
-    tenant_2 = tenant_factory.create()
-
-    zenith_cli.run(["branch", f"test_tenants_normal_work_with_wal_acceptors{with_wal_acceptors}", "main", f"--tenantid={tenant_1}"])
-    zenith_cli.run(["branch", f"test_tenants_normal_work_with_wal_acceptors{with_wal_acceptors}", "main", f"--tenantid={tenant_2}"])
+def test_tenants_normal_work(zenith_env_builder: ZenithEnvBuilder, with_wal_acceptors: bool):
     if with_wal_acceptors:
-        wa_factory.start_n_new(3)
+        zenith_env_builder.num_safekeepers = 3
 
-    pg_tenant1 = postgres.create_start(
+    env = zenith_env_builder.init()
+    """Tests tenants with and without wal acceptors"""
+    tenant_1 = env.create_tenant()
+    tenant_2 = env.create_tenant()
+
+    env.zenith_cli([
+        "branch",
         f"test_tenants_normal_work_with_wal_acceptors{with_wal_acceptors}",
+        "main",
+        f"--tenantid={tenant_1}"
+    ])
+    env.zenith_cli([
+        "branch",
+        f"test_tenants_normal_work_with_wal_acceptors{with_wal_acceptors}",
+        "main",
+        f"--tenantid={tenant_2}"
+    ])
+
+    pg_tenant1 = env.postgres.create_start(
+        f"test_tenants_normal_work_with_wal_acceptors{with_wal_acceptors}",
+        None,  # branch name, None means same as node name
         tenant_1,
-        wal_acceptors=wa_factory.get_connstrs() if with_wal_acceptors else None,
     )
-    pg_tenant2 = postgres.create_start(
+    pg_tenant2 = env.postgres.create_start(
         f"test_tenants_normal_work_with_wal_acceptors{with_wal_acceptors}",
+        None,  # branch name, None means same as node name
         tenant_2,
-        wal_acceptors=wa_factory.get_connstrs() if with_wal_acceptors else None,
     )
 
     for pg in [pg_tenant1, pg_tenant2]:
@@ -45,4 +47,4 @@ def test_tenants_normal_work(
                 cur.execute("CREATE TABLE t(key int primary key, value text)")
                 cur.execute("INSERT INTO t SELECT generate_series(1,100000), 'payload'")
                 cur.execute("SELECT sum(key) FROM t")
-                assert cur.fetchone() == (5000050000,)
+                assert cur.fetchone() == (5000050000, )

test_runner/batch_others/test_timeline_size.py

@@ -1,20 +1,20 @@
 from contextlib import closing
 from uuid import UUID
 import psycopg2.extras
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
-def test_timeline_size(
-    zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin
-):
-    # Branch at the point where only 100 rows were inserted
-    zenith_cli.run(["branch", "test_timeline_size", "empty"])
 
-    client = pageserver.http_client()
-    res = client.branch_detail(UUID(pageserver.initial_tenant), "test_timeline_size")
+def test_timeline_size(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    # Branch at the point where only 100 rows were inserted
+    env.zenith_cli(["branch", "test_timeline_size", "empty"])
+
+    client = env.pageserver.http_client()
+    res = client.branch_detail(UUID(env.initial_tenant), "test_timeline_size")
     assert res["current_logical_size"] == res["current_logical_size_non_incremental"]
 
-    pgmain = postgres.create_start("test_timeline_size")
+    pgmain = env.postgres.create_start("test_timeline_size")
     log.info("postgres is running on 'test_timeline_size' branch")
 
     with closing(pgmain.connect()) as conn:
@@ -23,17 +23,15 @@ def test_timeline_size(
 
             # Create table, and insert the first 100 rows
             cur.execute("CREATE TABLE foo (t text)")
-            cur.execute(
-                """
+            cur.execute("""
                 INSERT INTO foo
                     SELECT 'long string to consume some space' || g
                     FROM generate_series(1, 10) g
-            """
-            )
+            """)
 
-            res = client.branch_detail(UUID(pageserver.initial_tenant), "test_timeline_size")
+            res = client.branch_detail(UUID(env.initial_tenant), "test_timeline_size")
             assert res["current_logical_size"] == res["current_logical_size_non_incremental"]
             cur.execute("TRUNCATE foo")
 
-            res = client.branch_detail(UUID(pageserver.initial_tenant), "test_timeline_size")
+            res = client.branch_detail(UUID(env.initial_tenant), "test_timeline_size")
             assert res["current_logical_size"] == res["current_logical_size_non_incremental"]

test_runner/batch_others/test_twophase.py

@@ -1,6 +1,6 @@
 import os
 
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver, PgBin
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -9,10 +9,11 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 #
 # Test branching, when a transaction is in prepared state
 #
-def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin: PgBin):
-    zenith_cli.run(["branch", "test_twophase", "empty"])
+def test_twophase(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    env.zenith_cli(["branch", "test_twophase", "empty"])
 
-    pg = postgres.create_start('test_twophase', config_lines=['max_prepared_transactions=5'])
+    pg = env.postgres.create_start('test_twophase', config_lines=['max_prepared_transactions=5'])
     log.info("postgres is running on 'test_twophase' branch")
 
     conn = pg.connect()
@@ -57,10 +58,10 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
     assert len(twophase_files) == 2
 
     # Create a branch with the transaction in prepared state
-    zenith_cli.run(["branch", "test_twophase_prepared", "test_twophase"])
+    env.zenith_cli(["branch", "test_twophase_prepared", "test_twophase"])
 
     # Start compute on the new branch
-    pg2 = postgres.create_start(
+    pg2 = env.postgres.create_start(
         'test_twophase_prepared',
         config_lines=['max_prepared_transactions=5'],
     )
@@ -79,8 +80,8 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
     cur2.execute("ROLLBACK PREPARED 'insert_two'")
 
     cur2.execute('SELECT * FROM foo')
-    assert cur2.fetchall() == [('one',), ('three',)]
+    assert cur2.fetchall() == [('one', ), ('three', )]
 
     # Only one committed insert is visible on the original branch
     cur.execute('SELECT * FROM foo')
-    assert cur.fetchall() == [('three',)]
+    assert cur.fetchall() == [('three', )]

test_runner/batch_others/test_vm_bits.py

@@ -1,16 +1,19 @@
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
+
 #
 # Test that the VM bit is cleared correctly at a HEAP_DELETE and
 # HEAP_UPDATE record.
 #
-def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin, zenith_cli, base_dir):
+def test_vm_bit_clear(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+
     # Create a branch for us
-    zenith_cli.run(["branch", "test_vm_bit_clear", "empty"])
-    pg = postgres.create_start('test_vm_bit_clear')
+    env.zenith_cli(["branch", "test_vm_bit_clear", "empty"])
+    pg = env.postgres.create_start('test_vm_bit_clear')
 
     log.info("postgres is running on 'test_vm_bit_clear' branch")
     pg_conn = pg.connect()
@@ -33,7 +36,7 @@ def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, p
     cur.execute('UPDATE vmtest_update SET id = 5000 WHERE id = 1')
 
     # Branch at this point, to test that later
-    zenith_cli.run(["branch", "test_vm_bit_clear_new", "test_vm_bit_clear"])
+    env.zenith_cli(["branch", "test_vm_bit_clear_new", "test_vm_bit_clear"])
 
     # Clear the buffer cache, to force the VM page to be re-fetched from
     # the page server
@@ -49,20 +52,19 @@ def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, p
     ''')
 
     cur.execute('SELECT * FROM vmtest_delete WHERE id = 1')
-    assert(cur.fetchall() == []);
+    assert (cur.fetchall() == [])
     cur.execute('SELECT * FROM vmtest_update WHERE id = 1')
-    assert(cur.fetchall() == []);
+    assert (cur.fetchall() == [])
 
     cur.close()
 
-
     # Check the same thing on the branch that we created right after the DELETE
     #
     # As of this writing, the code in smgrwrite() creates a full-page image whenever
     # a dirty VM page is evicted. If the VM bit was not correctly cleared by the
     # earlier WAL record, the full-page image hides the problem. Starting a new
     # server at the right point-in-time avoids that full-page image.
-    pg_new = postgres.create_start('test_vm_bit_clear_new')
+    pg_new = env.postgres.create_start('test_vm_bit_clear_new')
 
     log.info("postgres is running on 'test_vm_bit_clear_new' branch")
     pg_new_conn = pg_new.connect()
@@ -75,6 +77,6 @@ def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, p
     ''')
 
     cur_new.execute('SELECT * FROM vmtest_delete WHERE id = 1')
-    assert(cur_new.fetchall() == []);
+    assert (cur_new.fetchall() == [])
     cur_new.execute('SELECT * FROM vmtest_update WHERE id = 1')
-    assert(cur_new.fetchall() == []);
+    assert (cur_new.fetchall() == [])

test_runner/batch_others/test_wal_acceptor.py

@@ -7,7 +7,7 @@ import uuid
 
 from contextlib import closing
 from multiprocessing import Process, Value
-from fixtures.zenith_fixtures import WalAcceptorFactory, ZenithPageserver, PostgresFactory, PgBin
+from fixtures.zenith_fixtures import PgBin, ZenithEnv, ZenithEnvBuilder
 from fixtures.utils import lsn_to_hex, mkdir_if_needed
 from fixtures.log_helper import log
 
@@ -16,11 +16,13 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 
 # basic test, write something in setup with wal acceptors, ensure that commits
 # succeed and data is written
-def test_normal_work(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, wa_factory):
-    zenith_cli.run(["branch", "test_wal_acceptors_normal_work", "empty"])
-    wa_factory.start_n_new(3)
-    pg = postgres.create_start('test_wal_acceptors_normal_work',
-                               wal_acceptors=wa_factory.get_connstrs())
+def test_normal_work(zenith_env_builder: ZenithEnvBuilder):
+    zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
+
+    env.zenith_cli(["branch", "test_wal_acceptors_normal_work", "main"])
+
+    pg = env.postgres.create_start('test_wal_acceptors_normal_work')
 
     with closing(pg.connect()) as conn:
         with conn.cursor() as cur:
@@ -34,18 +36,19 @@ def test_normal_work(zenith_cli, pageserver: ZenithPageserver, postgres: Postgre
 
 # Run page server and multiple acceptors, and multiple compute nodes running
 # against different timelines.
-def test_many_timelines(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, wa_factory):
-    n_timelines = 2
+def test_many_timelines(zenith_env_builder: ZenithEnvBuilder):
+    zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
 
-    wa_factory.start_n_new(3)
+    n_timelines = 2
 
     branches = ["test_wal_acceptors_many_timelines_{}".format(tlin) for tlin in range(n_timelines)]
 
     # start postgres on each timeline
     pgs = []
     for branch in branches:
-        zenith_cli.run(["branch", branch, "empty"])
-        pgs.append(postgres.create_start(branch, wal_acceptors=wa_factory.get_connstrs()))
+        env.zenith_cli(["branch", branch, "main"])
+        pgs.append(env.postgres.create_start(branch))
 
     # Do everything in different loops to have actions on different timelines
     # interleaved.
@@ -66,16 +69,16 @@ def test_many_timelines(zenith_cli, pageserver: ZenithPageserver, postgres: Post
 # Check that dead minority doesn't prevent the commits: execute insert n_inserts
 # times, with fault_probability chance of getting a wal acceptor down or up
 # along the way. 2 of 3 are always alive, so the work keeps going.
-def test_restarts(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, wa_factory: WalAcceptorFactory):
+def test_restarts(zenith_env_builder: ZenithEnvBuilder):
     fault_probability = 0.01
     n_inserts = 1000
     n_acceptors = 3
 
-    wa_factory.start_n_new(n_acceptors)
+    zenith_env_builder.num_safekeepers = n_acceptors
+    env = zenith_env_builder.init()
 
-    zenith_cli.run(["branch", "test_wal_acceptors_restarts", "empty"])
-    pg = postgres.create_start('test_wal_acceptors_restarts',
-                               wal_acceptors=wa_factory.get_connstrs())
+    env.zenith_cli(["branch", "test_wal_acceptors_restarts", "main"])
+    pg = env.postgres.create_start('test_wal_acceptors_restarts')
 
     # we rely upon autocommit after each statement
     # as waiting for acceptors happens there
@@ -89,7 +92,7 @@ def test_restarts(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
 
         if random.random() <= fault_probability:
             if failed_node is None:
-                failed_node = wa_factory.instances[random.randrange(0, n_acceptors)]
+                failed_node = env.safekeepers[random.randrange(0, n_acceptors)]
                 failed_node.stop()
             else:
                 failed_node.start()
@@ -107,12 +110,12 @@ def delayed_wal_acceptor_start(wa):
 
 
 # When majority of acceptors is offline, commits are expected to be frozen
-def test_unavailability(zenith_cli, postgres: PostgresFactory, wa_factory):
-    wa_factory.start_n_new(2)
+def test_unavailability(zenith_env_builder: ZenithEnvBuilder):
+    zenith_env_builder.num_safekeepers = 2
+    env = zenith_env_builder.init()
 
-    zenith_cli.run(["branch", "test_wal_acceptors_unavailability", "empty"])
-    pg = postgres.create_start('test_wal_acceptors_unavailability',
-                               wal_acceptors=wa_factory.get_connstrs())
+    env.zenith_cli(["branch", "test_wal_acceptors_unavailability", "main"])
+    pg = env.postgres.create_start('test_wal_acceptors_unavailability')
 
     # we rely upon autocommit after each statement
     # as waiting for acceptors happens there
@@ -124,9 +127,9 @@ def test_unavailability(zenith_cli, postgres: PostgresFactory, wa_factory):
     cur.execute("INSERT INTO t values (1, 'payload')")
 
     # shutdown one of two acceptors, that is, majority
-    wa_factory.instances[0].stop()
+    env.safekeepers[0].stop()
 
-    proc = Process(target=delayed_wal_acceptor_start, args=(wa_factory.instances[0], ))
+    proc = Process(target=delayed_wal_acceptor_start, args=(env.safekeepers[0], ))
     proc.start()
 
     start = time.time()
@@ -136,9 +139,9 @@ def test_unavailability(zenith_cli, postgres: PostgresFactory, wa_factory):
     proc.join()
 
     # for the world's balance, do the same with second acceptor
-    wa_factory.instances[1].stop()
+    env.safekeepers[1].stop()
 
-    proc = Process(target=delayed_wal_acceptor_start, args=(wa_factory.instances[1], ))
+    proc = Process(target=delayed_wal_acceptor_start, args=(env.safekeepers[1], ))
     proc.start()
 
     start = time.time()
@@ -177,13 +180,13 @@ def stop_value():
 
 
 # do inserts while concurrently getting up/down subsets of acceptors
-def test_race_conditions(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, wa_factory, stop_value):
+def test_race_conditions(zenith_env_builder: ZenithEnvBuilder, stop_value):
 
-    wa_factory.start_n_new(3)
+    zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
 
-    zenith_cli.run(["branch", "test_wal_acceptors_race_conditions", "empty"])
-    pg = postgres.create_start('test_wal_acceptors_race_conditions',
-                               wal_acceptors=wa_factory.get_connstrs())
+    env.zenith_cli(["branch", "test_wal_acceptors_race_conditions", "main"])
+    pg = env.postgres.create_start('test_wal_acceptors_race_conditions')
 
     # we rely upon autocommit after each statement
     # as waiting for acceptors happens there
@@ -192,7 +195,7 @@ def test_race_conditions(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
 
     cur.execute('CREATE TABLE t(key int primary key, value text)')
 
-    proc = Process(target=xmas_garland, args=(wa_factory.instances, stop_value))
+    proc = Process(target=xmas_garland, args=(env.safekeepers, stop_value))
     proc.start()
 
     for i in range(1000):
@@ -207,7 +210,8 @@ def test_race_conditions(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
 
 class ProposerPostgres:
     """Object for running safekeepers sync with walproposer"""
-    def __init__(self, pgdata_dir: str, pg_bin: PgBin, timeline_id: str, tenant_id: str):
+    def __init__(self, env: ZenithEnv, pgdata_dir: str, pg_bin, timeline_id: str, tenant_id: str):
+        self.env = env
         self.pgdata_dir: str = pgdata_dir
         self.pg_bin: PgBin = pg_bin
         self.timeline_id: str = timeline_id
@@ -253,16 +257,20 @@ class ProposerPostgres:
 
 
 # insert wal in all safekeepers and run sync on proposer
-def test_sync_safekeepers(repo_dir: str, pg_bin: PgBin, wa_factory: WalAcceptorFactory):
-    wa_factory.start_n_new(3)
+def test_sync_safekeepers(zenith_env_builder: ZenithEnvBuilder, pg_bin: PgBin):
+
+    # We don't really need the full environment for this test, just the
+    # safekeepers would be enough.
+    zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
 
     timeline_id = uuid.uuid4().hex
     tenant_id = uuid.uuid4().hex
 
     # write config for proposer
-    pgdata_dir = os.path.join(repo_dir, "proposer_pgdata")
-    pg = ProposerPostgres(pgdata_dir, pg_bin, timeline_id, tenant_id)
-    pg.create_dir_config(wa_factory.get_connstrs())
+    pgdata_dir = os.path.join(env.repo_dir, "proposer_pgdata")
+    pg = ProposerPostgres(env, pgdata_dir, pg_bin, timeline_id, tenant_id)
+    pg.create_dir_config(env.get_safekeeper_connstrs())
 
     # valid lsn, which is not in the segment start, nor in zero segment
     epoch_start_lsn = 0x16B9188  # 0/16B9188
@@ -271,7 +279,7 @@ def test_sync_safekeepers(repo_dir: str, pg_bin: PgBin, wa_factory: WalAcceptorF
     # append and commit WAL
     lsn_after_append = []
     for i in range(3):
-        res = wa_factory.instances[i].append_logical_message(
+        res = env.safekeepers[i].append_logical_message(
             tenant_id,
             timeline_id,
             {
@@ -295,13 +303,15 @@ def test_sync_safekeepers(repo_dir: str, pg_bin: PgBin, wa_factory: WalAcceptorF
     assert all(lsn_after_sync == lsn for lsn in lsn_after_append)
 
 
-def test_timeline_status(zenith_cli, pageserver, postgres, wa_factory: WalAcceptorFactory):
-    wa_factory.start_n_new(1)
+def test_timeline_status(zenith_env_builder: ZenithEnvBuilder):
 
-    zenith_cli.run(["branch", "test_timeline_status", "empty"])
-    pg = postgres.create_start('test_timeline_status', wal_acceptors=wa_factory.get_connstrs())
+    zenith_env_builder.num_safekeepers = 1
+    env = zenith_env_builder.init()
 
-    wa = wa_factory.instances[0]
+    env.zenith_cli(["branch", "test_timeline_status", "main"])
+    pg = env.postgres.create_start('test_timeline_status')
+
+    wa = env.safekeepers[0]
     wa_http_cli = wa.http_client()
     wa_http_cli.check_status()
 
@@ -318,6 +328,5 @@ def test_timeline_status(zenith_cli, pageserver, postgres, wa_factory: WalAccept
     pg.stop().start()
     pg.safe_psql("insert into t values(10)")
 
-    epoch_after_reboot = wa_http_cli.timeline_status(tenant_id,
-                                                     timeline_id).acceptor_epoch
+    epoch_after_reboot = wa_http_cli.timeline_status(tenant_id, timeline_id).acceptor_epoch
     assert epoch_after_reboot > epoch

test_runner/batch_others/test_wal_acceptor_async.py

@@ -1,9 +1,11 @@
 import asyncio
 import asyncpg
 import random
+import time
 
-from fixtures.zenith_fixtures import WalAcceptor, WalAcceptorFactory, ZenithPageserver, PostgresFactory, Postgres
+from fixtures.zenith_fixtures import ZenithEnvBuilder, Postgres, Safekeeper
 from fixtures.log_helper import getLogger
+from fixtures.utils import lsn_from_hex, lsn_to_hex
 from typing import List
 
 log = getLogger('root.wal_acceptor_async')
@@ -19,13 +21,16 @@ class BankClient(object):
     async def initdb(self):
         await self.conn.execute('DROP TABLE IF EXISTS bank_accs')
         await self.conn.execute('CREATE TABLE bank_accs(uid int primary key, amount int)')
-        await self.conn.execute('''
+        await self.conn.execute(
+            '''
             INSERT INTO bank_accs
             SELECT *, $1 FROM generate_series(0, $2)
-        ''', self.init_amount, self.n_accounts - 1)
+        ''',
+            self.init_amount,
+            self.n_accounts - 1)
         await self.conn.execute('DROP TABLE IF EXISTS bank_log')
         await self.conn.execute('CREATE TABLE bank_log(from_uid int, to_uid int, amount int)')
-        
+
         # TODO: Remove when https://github.com/zenithdb/zenith/issues/644 is fixed
         await self.conn.execute('ALTER TABLE bank_accs SET (autovacuum_enabled = false)')
         await self.conn.execute('ALTER TABLE bank_log SET (autovacuum_enabled = false)')
@@ -34,6 +39,7 @@ class BankClient(object):
         row = await self.conn.fetchrow('SELECT sum(amount) AS sum FROM bank_accs')
         assert row['sum'] == self.n_accounts * self.init_amount
 
+
 async def bank_transfer(conn: asyncpg.Connection, from_uid, to_uid, amount):
     # avoid deadlocks by sorting uids
     if from_uid > to_uid:
@@ -42,16 +48,22 @@ async def bank_transfer(conn: asyncpg.Connection, from_uid, to_uid, amount):
     async with conn.transaction():
         await conn.execute(
             'UPDATE bank_accs SET amount = amount + ($1) WHERE uid = $2',
-            amount, to_uid,
+            amount,
+            to_uid,
         )
         await conn.execute(
             'UPDATE bank_accs SET amount = amount - ($1) WHERE uid = $2',
-            amount, from_uid,
+            amount,
+            from_uid,
         )
-        await conn.execute('INSERT INTO bank_log VALUES ($1, $2, $3)',
-            from_uid, to_uid, amount,
+        await conn.execute(
+            'INSERT INTO bank_log VALUES ($1, $2, $3)',
+            from_uid,
+            to_uid,
+            amount,
         )
 
+
 class WorkerStats(object):
     def __init__(self, n_workers):
         self.counters = [0] * n_workers
@@ -92,11 +104,43 @@ async def run_random_worker(stats: WorkerStats, pg: Postgres, worker_id, n_accou
     await pg_conn.close()
 
 
+async def wait_for_lsn(safekeeper: Safekeeper,
+                       tenant_id: str,
+                       timeline_id: str,
+                       wait_lsn: str,
+                       polling_interval=1,
+                       timeout=600):
+    """
+    Poll flush_lsn from safekeeper until it's greater or equal than
+    provided wait_lsn. To do that, timeline_status is fetched from
+    safekeeper every polling_interval seconds.
+    """
+
+    started_at = time.time()
+    client = safekeeper.http_client()
+
+    flush_lsn = client.timeline_status(tenant_id, timeline_id).flush_lsn
+    log.info(
+        f'Safekeeper at port {safekeeper.port.pg} has flush_lsn {flush_lsn}, waiting for lsn {wait_lsn}'
+    )
+
+    while lsn_from_hex(wait_lsn) > lsn_from_hex(flush_lsn):
+        elapsed = time.time() - started_at
+        if elapsed > timeout:
+            raise RuntimeError(
+                f"timed out waiting for safekeeper at port {safekeeper.port.pg} to reach {wait_lsn}, current lsn is {flush_lsn}"
+            )
+
+        await asyncio.sleep(polling_interval)
+        flush_lsn = client.timeline_status(tenant_id, timeline_id).flush_lsn
+        log.debug(f'safekeeper port={safekeeper.port.pg} flush_lsn={flush_lsn} wait_lsn={wait_lsn}')
+
+
 # This test will run several iterations and check progress in each of them.
 # On each iteration 1 acceptor is stopped, and 2 others should allow
 # background workers execute transactions. In the end, state should remain
 # consistent.
-async def run_restarts_under_load(pg: Postgres, acceptors: List[WalAcceptor], n_workers=10):
+async def run_restarts_under_load(pg: Postgres, acceptors: List[Safekeeper], n_workers=10):
     n_accounts = 100
     init_amount = 100000
     max_transfer = 100
@@ -104,6 +148,9 @@ async def run_restarts_under_load(pg: Postgres, acceptors: List[WalAcceptor], n_
     iterations = 6
 
     pg_conn = await pg.connect_async()
+    tenant_id = await pg_conn.fetchval("show zenith.zenith_tenant")
+    timeline_id = await pg_conn.fetchval("show zenith.zenith_timeline")
+
     bank = BankClient(pg_conn, n_accounts=n_accounts, init_amount=init_amount)
     # create tables and initial balances
     await bank.initdb()
@@ -114,19 +161,19 @@ async def run_restarts_under_load(pg: Postgres, acceptors: List[WalAcceptor], n_
         worker = run_random_worker(stats, pg, worker_id, bank.n_accounts, max_transfer)
         workers.append(asyncio.create_task(worker))
 
-
     for it in range(iterations):
-        victim = acceptors[it % len(acceptors)]
+        victim_idx = it % len(acceptors)
+        victim = acceptors[victim_idx]
         victim.stop()
 
-        # Wait till previous victim recovers so it is ready for the next
-        # iteration by making any writing xact.
-        conn = await pg.connect_async()
-        await conn.execute(
-            'UPDATE bank_accs SET amount = amount WHERE uid = 1',
-            timeout=120
-        )
-        await conn.close()
+        flush_lsn = await pg_conn.fetchval('SELECT pg_current_wal_flush_lsn()')
+        flush_lsn = lsn_to_hex(flush_lsn)
+        log.info(f'Postgres flush_lsn {flush_lsn}')
+
+        # Wait until alive safekeepers catch up with postgres
+        for idx, safekeeper in enumerate(acceptors):
+            if idx != victim_idx:
+                await wait_for_lsn(safekeeper, tenant_id, timeline_id, flush_lsn)
 
         stats.reset()
         await asyncio.sleep(period_time)
@@ -145,16 +192,14 @@ async def run_restarts_under_load(pg: Postgres, acceptors: List[WalAcceptor], n_
 
 
 # restart acceptors one by one, while executing and validating bank transactions
-def test_restarts_under_load(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory,
-                             wa_factory: WalAcceptorFactory):
+def test_restarts_under_load(zenith_env_builder: ZenithEnvBuilder):
+    zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
 
-    wa_factory.start_n_new(3)
+    env.zenith_cli(["branch", "test_wal_acceptors_restarts_under_load", "main"])
+    pg = env.postgres.create_start('test_wal_acceptors_restarts_under_load')
 
-    zenith_cli.run(["branch", "test_wal_acceptors_restarts_under_load", "empty"])
-    pg = postgres.create_start('test_wal_acceptors_restarts_under_load',
-                               wal_acceptors=wa_factory.get_connstrs())
-
-    asyncio.run(run_restarts_under_load(pg, wa_factory.instances))
+    asyncio.run(run_restarts_under_load(pg, env.safekeepers))
 
     # TODO: Remove when https://github.com/zenithdb/zenith/issues/644 is fixed
     pg.stop()

test_runner/batch_others/test_zenith_cli.py

@@ -1,97 +1,107 @@
 import json
 import uuid
 
-from fixtures.zenith_fixtures import ZenithCli, ZenithPageserver
+from psycopg2.extensions import cursor as PgCursor
+from fixtures.zenith_fixtures import ZenithEnv
+from typing import cast
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
-def helper_compare_branch_list(page_server_cur, zenith_cli, initial_tenant: str):
+def helper_compare_branch_list(page_server_cur: PgCursor, env: ZenithEnv, initial_tenant: str):
     """
     Compare branches list returned by CLI and directly via API.
     Filters out branches created by other tests.
     """
 
     page_server_cur.execute(f'branch_list {initial_tenant}')
-    branches_api = sorted(map(lambda b: b['name'], json.loads(page_server_cur.fetchone()[0])))
+    branches_api = sorted(
+        map(lambda b: cast(str, b['name']), json.loads(page_server_cur.fetchone()[0])))
     branches_api = [b for b in branches_api if b.startswith('test_cli_') or b in ('empty', 'main')]
 
-    res = zenith_cli.run(["branch"])
+    res = env.zenith_cli(["branch"])
     res.check_returncode()
     branches_cli = sorted(map(lambda b: b.split(':')[-1].strip(), res.stdout.strip().split("\n")))
     branches_cli = [b for b in branches_cli if b.startswith('test_cli_') or b in ('empty', 'main')]
 
-    res = zenith_cli.run(["branch", f"--tenantid={initial_tenant}"])
+    res = env.zenith_cli(["branch", f"--tenantid={initial_tenant}"])
     res.check_returncode()
-    branches_cli_with_tenant_arg = sorted(map(lambda b: b.split(':')[-1].strip(), res.stdout.strip().split("\n")))
-    branches_cli_with_tenant_arg = [b for b in branches_cli if b.startswith('test_cli_') or b in ('empty', 'main')]
+    branches_cli_with_tenant_arg = sorted(
+        map(lambda b: b.split(':')[-1].strip(), res.stdout.strip().split("\n")))
+    branches_cli_with_tenant_arg = [
+        b for b in branches_cli if b.startswith('test_cli_') or b in ('empty', 'main')
+    ]
 
     assert branches_api == branches_cli == branches_cli_with_tenant_arg
 
 
-def test_cli_branch_list(pageserver: ZenithPageserver, zenith_cli):
-    page_server_conn = pageserver.connect()
+def test_cli_branch_list(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    page_server_conn = env.pageserver.connect()
     page_server_cur = page_server_conn.cursor()
 
     # Initial sanity check
-    helper_compare_branch_list(page_server_cur, zenith_cli, pageserver.initial_tenant)
+    helper_compare_branch_list(page_server_cur, env, env.initial_tenant)
 
     # Create a branch for us
-    res = zenith_cli.run(["branch", "test_cli_branch_list_main", "main"])
+    res = env.zenith_cli(["branch", "test_cli_branch_list_main", "empty"])
     assert res.stderr == ''
-    helper_compare_branch_list(page_server_cur, zenith_cli, pageserver.initial_tenant)
+    helper_compare_branch_list(page_server_cur, env, env.initial_tenant)
 
     # Create a nested branch
-    res = zenith_cli.run(["branch", "test_cli_branch_list_nested", "test_cli_branch_list_main"])
+    res = env.zenith_cli(["branch", "test_cli_branch_list_nested", "test_cli_branch_list_main"])
     assert res.stderr == ''
-    helper_compare_branch_list(page_server_cur, zenith_cli, pageserver.initial_tenant)
+    helper_compare_branch_list(page_server_cur, env, env.initial_tenant)
 
     # Check that all new branches are visible via CLI
-    res = zenith_cli.run(["branch"])
+    res = env.zenith_cli(["branch"])
     assert res.stderr == ''
     branches_cli = sorted(map(lambda b: b.split(':')[-1].strip(), res.stdout.strip().split("\n")))
 
     assert 'test_cli_branch_list_main' in branches_cli
     assert 'test_cli_branch_list_nested' in branches_cli
 
-def helper_compare_tenant_list(page_server_cur, zenith_cli: ZenithCli):
-    page_server_cur.execute(f'tenant_list')
-    tenants_api = sorted(json.loads(page_server_cur.fetchone()[0]))
 
-    res = zenith_cli.run(["tenant", "list"])
+def helper_compare_tenant_list(page_server_cur: PgCursor, env: ZenithEnv):
+    page_server_cur.execute(f'tenant_list')
+    tenants_api = sorted(
+        map(lambda t: cast(str, t['id']), json.loads(page_server_cur.fetchone()[0])))
+
+    res = env.zenith_cli(["tenant", "list"])
     assert res.stderr == ''
-    tenants_cli = sorted(res.stdout.splitlines())
+    tenants_cli = sorted(map(lambda t: t.split()[0], res.stdout.splitlines()))
 
     assert tenants_api == tenants_cli
 
 
-def test_cli_tenant_list(pageserver: ZenithPageserver, zenith_cli: ZenithCli):
-    page_server_conn = pageserver.connect()
+def test_cli_tenant_list(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    page_server_conn = env.pageserver.connect()
     page_server_cur = page_server_conn.cursor()
 
     # Initial sanity check
-    helper_compare_tenant_list(page_server_cur, zenith_cli)
+    helper_compare_tenant_list(page_server_cur, env)
 
     # Create new tenant
     tenant1 = uuid.uuid4().hex
-    res = zenith_cli.run(["tenant", "create", tenant1])
+    res = env.zenith_cli(["tenant", "create", tenant1])
     res.check_returncode()
 
     # check tenant1 appeared
-    helper_compare_tenant_list(page_server_cur, zenith_cli)
+    helper_compare_tenant_list(page_server_cur, env)
 
     # Create new tenant
     tenant2 = uuid.uuid4().hex
-    res = zenith_cli.run(["tenant", "create", tenant2])
+    res = env.zenith_cli(["tenant", "create", tenant2])
     res.check_returncode()
 
     # check tenant2 appeared
-    helper_compare_tenant_list(page_server_cur, zenith_cli)
+    helper_compare_tenant_list(page_server_cur, env)
 
-    res = zenith_cli.run(["tenant", "list"])
+    res = env.zenith_cli(["tenant", "list"])
     res.check_returncode()
-    tenants = sorted(res.stdout.splitlines())
+    tenants = sorted(map(lambda t: t.split()[0], res.stdout.splitlines()))
 
-    assert pageserver.initial_tenant in tenants
+    assert env.initial_tenant in tenants
     assert tenant1 in tenants
     assert tenant2 in tenants

test_runner/batch_pg_regress/test_isolation.py

@@ -1,20 +1,20 @@
 import os
 
 from fixtures.utils import mkdir_if_needed
-from fixtures.zenith_fixtures import ZenithPageserver, PostgresFactory
+from fixtures.zenith_fixtures import ZenithEnv, base_dir, pg_distrib_dir
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
-def test_isolation(pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin, zenith_cli, test_output_dir, pg_distrib_dir,
-                   base_dir, capsys):
+def test_isolation(zenith_simple_env: ZenithEnv, test_output_dir, pg_bin, capsys):
+    env = zenith_simple_env
 
     # Create a branch for us
-    zenith_cli.run(["branch", "test_isolation", "empty"])
+    env.zenith_cli(["branch", "test_isolation", "empty"])
 
     # Connect to postgres and create a database called "regression".
     # isolation tests use prepared transactions, so enable them
-    pg = postgres.create_start('test_isolation', config_lines=['max_prepared_transactions=100'])
+    pg = env.postgres.create_start('test_isolation', config_lines=['max_prepared_transactions=100'])
     pg.safe_psql('CREATE DATABASE isolation_regression')
 
     # Create some local directories for pg_isolation_regress to run in.
@@ -38,7 +38,7 @@ def test_isolation(pageserver: ZenithPageserver, postgres: PostgresFactory, pg_b
         '--schedule={}'.format(schedule),
     ]
 
-    env = {
+    env_vars = {
         'PGPORT': str(pg.port),
         'PGUSER': pg.username,
         'PGHOST': pg.host,
@@ -48,4 +48,4 @@ def test_isolation(pageserver: ZenithPageserver, postgres: PostgresFactory, pg_b
     # We don't capture the output. It's not too chatty, and it always
     # logs the exact same data to `regression.out` anyway.
     with capsys.disabled():
-        pg_bin.run(pg_isolation_regress_command, env=env, cwd=runpath)
+        pg_bin.run(pg_isolation_regress_command, env=env_vars, cwd=runpath)

test_runner/batch_pg_regress/test_pg_regress.py

@@ -1,19 +1,19 @@
 import os
 
 from fixtures.utils import mkdir_if_needed
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver, check_restored_datadir_content
+from fixtures.zenith_fixtures import ZenithEnv, check_restored_datadir_content, base_dir, pg_distrib_dir
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
-def test_pg_regress(pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin, zenith_cli, test_output_dir, pg_distrib_dir,
-                    base_dir, capsys):
+def test_pg_regress(zenith_simple_env: ZenithEnv, test_output_dir: str, pg_bin, capsys):
+    env = zenith_simple_env
 
     # Create a branch for us
-    zenith_cli.run(["branch", "test_pg_regress", "empty"])
+    env.zenith_cli(["branch", "test_pg_regress", "empty"])
 
     # Connect to postgres and create a database called "regression".
-    pg = postgres.create_start('test_pg_regress')
+    pg = env.postgres.create_start('test_pg_regress')
     pg.safe_psql('CREATE DATABASE regression')
 
     # Create some local directories for pg_regress to run in.
@@ -38,7 +38,7 @@ def test_pg_regress(pageserver: ZenithPageserver, postgres: PostgresFactory, pg_
         '--inputdir={}'.format(src_path),
     ]
 
-    env = {
+    env_vars = {
         'PGPORT': str(pg.port),
         'PGUSER': pg.username,
         'PGHOST': pg.host,
@@ -48,11 +48,11 @@ def test_pg_regress(pageserver: ZenithPageserver, postgres: PostgresFactory, pg_
     # We don't capture the output. It's not too chatty, and it always
     # logs the exact same data to `regression.out` anyway.
     with capsys.disabled():
-        pg_bin.run(pg_regress_command, env=env, cwd=runpath)
+        pg_bin.run(pg_regress_command, env=env_vars, cwd=runpath)
 
         # checkpoint one more time to ensure that the lsn we get is the latest one
         pg.safe_psql('CHECKPOINT')
         lsn = pg.safe_psql('select pg_current_wal_insert_lsn()')[0][0]
 
         # Check that we restore the content of the datadir correctly
-        check_restored_datadir_content(zenith_cli, test_output_dir, pg, pageserver.service_port.pg)
+        check_restored_datadir_content(test_output_dir, env, pg)

test_runner/batch_pg_regress/test_zenith_regress.py

@@ -1,20 +1,23 @@
 import os
 
 from fixtures.utils import mkdir_if_needed
-from fixtures.zenith_fixtures import PageserverPort, PostgresFactory, check_restored_datadir_content
+from fixtures.zenith_fixtures import (ZenithEnv,
+                                      check_restored_datadir_content,
+                                      base_dir,
+                                      pg_distrib_dir)
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
 
 
-def test_zenith_regress(postgres: PostgresFactory, pg_bin, zenith_cli, test_output_dir, pg_distrib_dir,
-                        base_dir, capsys, pageserver_port: PageserverPort):
+def test_zenith_regress(zenith_simple_env: ZenithEnv, test_output_dir, pg_bin, capsys):
+    env = zenith_simple_env
 
     # Create a branch for us
-    zenith_cli.run(["branch", "test_zenith_regress", "empty"])
+    env.zenith_cli(["branch", "test_zenith_regress", "empty"])
 
     # Connect to postgres and create a database called "regression".
-    pg = postgres.create_start('test_zenith_regress')
+    pg = env.postgres.create_start('test_zenith_regress')
     pg.safe_psql('CREATE DATABASE regression')
 
     # Create some local directories for pg_regress to run in.
@@ -40,7 +43,7 @@ def test_zenith_regress(postgres: PostgresFactory, pg_bin, zenith_cli, test_outp
     ]
 
     log.info(pg_regress_command)
-    env = {
+    env_vars = {
         'PGPORT': str(pg.port),
         'PGUSER': pg.username,
         'PGHOST': pg.host,
@@ -50,11 +53,11 @@ def test_zenith_regress(postgres: PostgresFactory, pg_bin, zenith_cli, test_outp
     # We don't capture the output. It's not too chatty, and it always
     # logs the exact same data to `regression.out` anyway.
     with capsys.disabled():
-        pg_bin.run(pg_regress_command, env=env, cwd=runpath)
+        pg_bin.run(pg_regress_command, env=env_vars, cwd=runpath)
 
         # checkpoint one more time to ensure that the lsn we get is the latest one
         pg.safe_psql('CHECKPOINT')
         lsn = pg.safe_psql('select pg_current_wal_insert_lsn()')[0][0]
 
         # Check that we restore the content of the datadir correctly
-        check_restored_datadir_content(zenith_cli, test_output_dir, pg, pageserver_port.pg)
+        check_restored_datadir_content(test_output_dir, env, pg)

test_runner/fixtures/benchmark_fixture.py

@@ -24,7 +24,6 @@ from typing import Any, Callable, Dict, Iterator, List, Optional, TypeVar, cast
 from typing_extensions import Literal
 
 from .utils import (get_self_dir, mkdir_if_needed, subprocess_capture)
-
 """
 This file contains fixtures for micro-benchmarks.
 
@@ -32,11 +31,11 @@ To use, declare the 'zenbenchmark' fixture in the test function. Run the
 bencmark, and then record the result by calling zenbenchmark.record. For example:
 
 import timeit
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 
 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 
-def test_mybench(postgres: PostgresFactory, pageserver: ZenithPageserver, zenbenchmark):
+def test_mybench(zenith_simple_env: env, zenbenchmark):
 
     # Initialize the test
     ...
@@ -56,15 +55,9 @@ in the test initialization, or measure disk usage after the test query.
 """
 
 
-# All the results are collected in this list, as a tuple:
-# (test_name: str, metric_name: str, metric_value: float, unit: str)
-#
 # TODO: It would perhaps be better to store the results as additional
 # properties in the pytest TestReport objects, to make them visible to
 # other pytest tools.
-global zenbenchmark_results
-zenbenchmark_results = []
-
 class ZenithBenchmarkResults:
     """ An object for recording benchmark results. """
     def __init__(self):
@@ -77,6 +70,11 @@ class ZenithBenchmarkResults:
 
         self.results.append((test_name, metric_name, metric_value, unit))
 
+
+# Will be recreated in each session.
+zenbenchmark_results: ZenithBenchmarkResults = ZenithBenchmarkResults()
+
+
 # Session scope fixture that initializes the results object
 @pytest.fixture(autouse=True, scope='session')
 def zenbenchmark_global(request) -> Iterator[ZenithBenchmarkResults]:
@@ -88,6 +86,7 @@ def zenbenchmark_global(request) -> Iterator[ZenithBenchmarkResults]:
 
     yield zenbenchmark_results
 
+
 class ZenithBenchmarker:
     """
     An object for recording benchmark results. This is created for each test
@@ -103,7 +102,6 @@ class ZenithBenchmarker:
         """
         self.results.record(self.request.node.name, metric_name, metric_value, unit)
 
-
     @contextmanager
     def record_duration(self, metric_name):
         """
@@ -134,8 +132,10 @@ 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+)$', all_metrics,
+        matches = re.search(r'^pageserver_disk_io_bytes{io_operation="write"} (\S+)$',
+                            all_metrics,
                             re.MULTILINE)
+        assert matches
         return int(round(float(matches.group(1))))
 
     def get_peak_mem(self, pageserver) -> int:
@@ -145,8 +145,8 @@ class ZenithBenchmarker:
         # 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+)$', all_metrics,
-                            re.MULTILINE)
+        matches = re.search(r'^pageserver_maxrss_kb (\S+)$', all_metrics, re.MULTILINE)
+        assert matches
         return int(round(float(matches.group(1))))
 
     def get_timeline_size(self, repo_dir: str, tenantid: str, timelineid: str):
@@ -171,7 +171,11 @@ class ZenithBenchmarker:
         yield
         after = self.get_io_writes(pageserver)
 
-        self.results.record(self.request.node.name, metric_name, round((after - before) / (1024 * 1024)), 'MB')
+        self.results.record(self.request.node.name,
+                            metric_name,
+                            round((after - before) / (1024 * 1024)),
+                            'MB')
+
 
 @pytest.fixture(scope='function')
 def zenbenchmark(zenbenchmark_global, request) -> Iterator[ZenithBenchmarker]:
@@ -185,9 +189,7 @@ def zenbenchmark(zenbenchmark_global, request) -> Iterator[ZenithBenchmarker]:
 
 # Hook to print the results at the end
 @pytest.hookimpl(hookwrapper=True)
-def pytest_terminal_summary(
-    terminalreporter: TerminalReporter, exitstatus: int, config: Config
-):
+def pytest_terminal_summary(terminalreporter: TerminalReporter, exitstatus: int, config: Config):
     yield
 
     global zenbenchmark_results

test_runner/fixtures/log_helper.py

@@ -1,6 +1,5 @@
 import logging
 import logging.config
-
 """
 This file configures logging to use in python tests.
 Logs are automatically captured and shown in their
@@ -27,17 +26,19 @@ LOGGING = {
             "level": "INFO"
         },
         "root.wal_acceptor_async": {
-            "level": "INFO" # a lot of logs on DEBUG level
+            "level": "INFO"  # a lot of logs on DEBUG level
         }
     }
 }
 
+
 def getLogger(name='root') -> logging.Logger:
     """Method to get logger for tests.
     
     Should be used to get correctly initialized logger. """
     return logging.getLogger(name)
 
+
 # default logger for tests
 log = getLogger()
 

test_runner/fixtures/utils.py

@@ -4,6 +4,7 @@ import subprocess
 from typing import Any, List
 from fixtures.log_helper import log
 
+
 def get_self_dir() -> str:
     """ Get the path to the directory where this script lives. """
     return os.path.dirname(os.path.abspath(__file__))
@@ -58,6 +59,13 @@ def global_counter() -> int:
     _global_counter += 1
     return _global_counter
 
+
 def lsn_to_hex(num: int) -> str:
     """ Convert lsn from int to standard hex notation. """
     return "{:X}/{:X}".format(num >> 32, num & 0xffffffff)
+
+
+def lsn_from_hex(lsn_hex: str) -> int:
+    """ Convert lsn from hex notation to int. """
+    l, r = lsn_hex.split('/')
+    return (int(l, 16) << 32) + int(r, 16)

test_runner/netstat-script.sh

@@ -1,8 +0,0 @@
-#!/bin/bash
-
-while true; do
-  echo -n "==== CURRENT TIME:" >> /tmp/test_output/netstat.stdout
-  date +"%T.%N" >> /tmp/test_output/netstat.stdout
-  sudo netstat -vpnoa | grep tcp | sort >> /tmp/test_output/netstat.stdout
-  sleep 0.5
-done

test_runner/performance/test_bulk_insert.py

@@ -1,10 +1,11 @@
 import os
 from contextlib import closing
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 
+
 #
 # Run bulk INSERT test.
 #
@@ -15,16 +16,17 @@ pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 # 3. Disk space used
 # 4. Peak memory usage
 #
-def test_bulk_insert(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
+def test_bulk_insert(zenith_simple_env: ZenithEnv, zenbenchmark):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_bulk_insert", "empty"])
+    env.zenith_cli(["branch", "test_bulk_insert", "empty"])
 
-    pg = postgres.create_start('test_bulk_insert')
+    pg = env.postgres.create_start('test_bulk_insert')
     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
-    psconn = pageserver.connect();
+    psconn = env.pageserver.connect()
     pscur = psconn.cursor()
 
     # Get the timeline ID of our branch. We need it for the 'do_gc' command
@@ -36,17 +38,19 @@ def test_bulk_insert(postgres: PostgresFactory, pageserver: ZenithPageserver, pg
             cur.execute("create table huge (i int, j int);")
 
             # Run INSERT, recording the time and I/O it takes
-            with zenbenchmark.record_pageserver_writes(pageserver, 'pageserver_writes'):
+            with zenbenchmark.record_pageserver_writes(env.pageserver, 'pageserver_writes'):
                 with zenbenchmark.record_duration('insert'):
                     cur.execute("insert into huge values (generate_series(1, 5000000), 0);")
 
                     # Flush the layers from memory to disk. This is included in the reported
                     # time and I/O
-                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
 
             # Record peak memory usage
-            zenbenchmark.record("peak_mem", zenbenchmark.get_peak_mem(pageserver) / 1024, 'MB')
+            zenbenchmark.record("peak_mem", zenbenchmark.get_peak_mem(env.pageserver) / 1024, 'MB')
 
             # 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')
+            timeline_size = zenbenchmark.get_timeline_size(env.repo_dir,
+                                                           env.initial_tenant,
+                                                           timeline)
+            zenbenchmark.record('size', timeline_size / (1024 * 1024), 'MB')

test_runner/performance/test_bulk_tenant_create.py

@@ -1,11 +1,7 @@
 import timeit
 import pytest
 
-from fixtures.zenith_fixtures import (
-    TenantFactory,
-    ZenithCli,
-    PostgresFactory,
-)
+from fixtures.zenith_fixtures import ZenithEnvBuilder
 
 pytest_plugins = ("fixtures.benchmark_fixture")
 
@@ -20,34 +16,37 @@ pytest_plugins = ("fixtures.benchmark_fixture")
 @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,
+    zenith_env_builder: ZenithEnvBuilder,
     use_wal_acceptors: str,
     tenants_count: int,
     zenbenchmark,
 ):
     """Measure tenant creation time (with and without wal acceptors)"""
+    if use_wal_acceptors == 'with_wa':
+        zenith_env_builder.num_safekeepers = 3
+    env = zenith_env_builder.init()
 
     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",
+        tenant = env.create_tenant()
+        env.zenith_cli([
+            "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)
+        # FIXME: We used to start new safekeepers here. Did that make sense? Should we do it now?
+        #if use_wal_acceptors == 'with_wa':
+        #    wa_factory.start_n_new(3)
 
-        pg_tenant = postgres.create_start(
+        pg_tenant = env.postgres.create_start(
             f"test_bulk_tenant_create_{tenants_count}_{i}_{use_wal_acceptors}",
+            None,  # branch name, None means same as node name
             tenant,
-            wal_acceptors=wa_factory.get_connstrs() if use_wal_acceptors == 'with_wa' else None,
         )
 
         end = timeit.default_timer()

test_runner/performance/test_gist_build.py

@@ -1,25 +1,27 @@
 import os
 from contextlib import closing
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 
+
 #
 # Test buffering GisT build. It WAL-logs the whole relation, in 32-page chunks.
 # As of this writing, we're duplicate those giant WAL records for each page,
 # which makes the delta layer about 32x larger than it needs to be.
 #
-def test_gist_buffering_build(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
+def test_gist_buffering_build(zenith_simple_env: ZenithEnv, zenbenchmark):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_gist_buffering_build", "empty"])
+    env.zenith_cli(["branch", "test_gist_buffering_build", "empty"])
 
-    pg = postgres.create_start('test_gist_buffering_build')
+    pg = env.postgres.create_start('test_gist_buffering_build')
     log.info("postgres is running on 'test_gist_buffering_build' branch")
 
     # Open a connection directly to the page server that we'll use to force
     # flushing the layers to disk
-    psconn = pageserver.connect();
+    psconn = env.pageserver.connect()
     pscur = psconn.cursor()
 
     # Get the timeline ID of our branch. We need it for the 'do_gc' command
@@ -29,21 +31,27 @@ def test_gist_buffering_build(postgres: PostgresFactory, pageserver: ZenithPages
             timeline = cur.fetchone()[0]
 
             # Create test table.
-            cur.execute("create table gist_point_tbl(id int4, p point)");
-            cur.execute("insert into gist_point_tbl select g, point(g, g) from generate_series(1, 1000000) g;");
+            cur.execute("create table gist_point_tbl(id int4, p point)")
+            cur.execute(
+                "insert into gist_point_tbl select g, point(g, g) from generate_series(1, 1000000) g;"
+            )
 
             # Build the index.
-            with zenbenchmark.record_pageserver_writes(pageserver, 'pageserver_writes'):
+            with zenbenchmark.record_pageserver_writes(env.pageserver, 'pageserver_writes'):
                 with zenbenchmark.record_duration('build'):
-                    cur.execute("create index gist_pointidx2 on gist_point_tbl using gist(p) with (buffering = on)");
+                    cur.execute(
+                        "create index gist_pointidx2 on gist_point_tbl using gist(p) with (buffering = on)"
+                    )
 
                     # Flush the layers from memory to disk. This is included in the reported
                     # time and I/O
-                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 1000000")
+                    pscur.execute(f"do_gc {env.initial_tenant} {timeline} 1000000")
 
             # Record peak memory usage
-            zenbenchmark.record("peak_mem", zenbenchmark.get_peak_mem(pageserver) / 1024, 'MB')
+            zenbenchmark.record("peak_mem", zenbenchmark.get_peak_mem(env.pageserver) / 1024, 'MB')
 
             # 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')
+            timeline_size = zenbenchmark.get_timeline_size(env.repo_dir,
+                                                           env.initial_tenant,
+                                                           timeline)
+            zenbenchmark.record('size', timeline_size / (1024 * 1024), 'MB')

test_runner/performance/test_perf_pgbench.py

@@ -1,10 +1,11 @@
 import os
 from contextlib import closing
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 
+
 #
 # Run a very short pgbench test.
 #
@@ -14,16 +15,17 @@ pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 # 2. Time to run 5000 pgbench transactions
 # 3. Disk space used
 #
-def test_pgbench(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
+def test_pgbench(zenith_simple_env: ZenithEnv, pg_bin, zenbenchmark):
+    env = zenith_simple_env
     # Create a branch for us
-    zenith_cli.run(["branch", "test_pgbench_perf", "empty"])
+    env.zenith_cli(["branch", "test_pgbench_perf", "empty"])
 
-    pg = postgres.create_start('test_pgbench_perf')
+    pg = env.postgres.create_start('test_pgbench_perf')
     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
-    psconn = pageserver.connect();
+    psconn = env.pageserver.connect()
     pscur = psconn.cursor()
 
     # Get the timeline ID of our branch. We need it for the 'do_gc' command
@@ -35,13 +37,13 @@ def test_pgbench(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin
     connstr = pg.connstr()
 
     # Initialize pgbench database, recording the time and I/O it takes
-    with zenbenchmark.record_pageserver_writes(pageserver, 'pageserver_writes'):
+    with zenbenchmark.record_pageserver_writes(env.pageserver, 'pageserver_writes'):
         with zenbenchmark.record_duration('init'):
             pg_bin.run_capture(['pgbench', '-s5', '-i', connstr])
 
             # Flush the layers from memory to disk. This is included in the reported
             # time and I/O
-            pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+            pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
 
     # Run pgbench for 5000 transactions
     with zenbenchmark.record_duration('5000_xacts'):
@@ -49,8 +51,8 @@ def test_pgbench(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin
 
     # Flush the layers to disk again. This is *not' included in the reported time,
     # though.
-    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
+    pscur.execute(f"do_gc {env.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')
+    timeline_size = zenbenchmark.get_timeline_size(env.repo_dir, env.initial_tenant, timeline)
+    zenbenchmark.record('size', timeline_size / (1024 * 1024), 'MB')

test_runner/performance/test_write_amplification.py

@@ -12,21 +12,23 @@
 # Amplification problem at its finest.
 import os
 from contextlib import closing
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 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')
+def test_write_amplification(zenith_simple_env: ZenithEnv, zenbenchmark):
+    env = zenith_simple_env
+    # Create a branch for us
+    env.zenith_cli(["branch", "test_write_amplification", "empty"])
+
+    pg = env.postgres.create_start('test_write_amplification')
     log.info("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();
+    psconn = env.pageserver.connect()
     pscur = psconn.cursor()
 
     with closing(pg.connect()) as conn:
@@ -35,7 +37,7 @@ def test_write_amplification(postgres: PostgresFactory, pageserver: ZenithPagese
             cur.execute("SHOW zenith.zenith_timeline")
             timeline = cur.fetchone()[0]
 
-            with zenbenchmark.record_pageserver_writes(pageserver, 'pageserver_writes'):
+            with zenbenchmark.record_pageserver_writes(env.pageserver, 'pageserver_writes'):
                 with zenbenchmark.record_duration('run'):
 
                     # NOTE: Because each iteration updates every table already created,
@@ -68,8 +70,10 @@ def test_write_amplification(postgres: PostgresFactory, pageserver: ZenithPagese
                         # 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")
+                        pscur.execute(f"do_gc {env.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')
+            timeline_size = zenbenchmark.get_timeline_size(env.repo_dir,
+                                                           env.initial_tenant,
+                                                           timeline)
+            zenbenchmark.record('size', timeline_size / (1024 * 1024), 'MB')

test_runner/pytest.ini

@@ -2,3 +2,4 @@
 minversion = 6.0
 log_format = %(asctime)s.%(msecs)-3d %(levelname)s [%(filename)s:%(lineno)d] %(message)s
 log_date_format = %Y-%m-%d %H:%M:%S
+log_cli = true

test_runner/setup.cfg

@@ -10,6 +10,7 @@ max-line-length = 100
 [yapf]
 based_on_style = pep8
 column_limit = 100
+split_all_top_level_comma_separated_values = true
 
 [mypy]
 # some tests don't typecheck when this flag is set
@@ -21,7 +22,11 @@ disallow_untyped_decorators = false
 disallow_untyped_defs = false
 strict = true
 
-[mypy-psycopg2.*]
+[mypy-asyncpg.*]
+# There is some work in progress, though: https://github.com/MagicStack/asyncpg/pull/577
+ignore_missing_imports = true
+
+[mypy-cached_property.*]
 ignore_missing_imports = true
 
 [mypy-pytest.*]

test_runner/test_broken.py

@@ -1,6 +1,7 @@
 import pytest
 import os
 
+from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
 pytest_plugins = ("fixtures.zenith_fixtures")
@@ -19,11 +20,13 @@ run_broken = pytest.mark.skipif(os.environ.get('RUN_BROKEN') is None,
 
 
 @run_broken
-def test_broken(zenith_cli, pageserver, postgres, pg_bin):
-    # Create a branch for us
-    zenith_cli.run(["branch", "test_broken", "empty"])
+def test_broken(zenith_simple_env: ZenithEnv, pg_bin):
+    env = zenith_simple_env
 
-    postgres.create_start("test_broken")
+    # Create a branch for us
+    env.zenith_cli(["branch", "test_broken", "empty"])
+
+    env.postgres.create_start("test_broken")
     log.info('postgres is running')
 
     log.info('THIS NEXT COMMAND WILL FAIL:')

walkeeper/src/bin/safekeeper.rs

@@ -7,17 +7,16 @@ 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 zenith_utils::{logging, tcp_listener};
 
 use walkeeper::defaults::{DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_PG_LISTEN_ADDR};
 use walkeeper::http;
 use walkeeper::s3_offload;
 use walkeeper::wal_service;
-use walkeeper::WalAcceptorConf;
+use walkeeper::SafeKeeperConf;
 
 fn main() -> Result<()> {
     zenith_metrics::set_common_metrics_prefix("safekeeper");
@@ -54,7 +53,7 @@ fn main() -> Result<()> {
             Arg::with_name("ttl")
                 .long("ttl")
                 .takes_value(true)
-                .help("interval for keeping WAL as walkeeper node, after which them will be uploaded to S3 and removed locally"),
+                .help("interval for keeping WAL at safekeeper node, after which them will be uploaded to S3 and removed locally"),
         )
         .arg(
             Arg::with_name("recall")
@@ -78,8 +77,11 @@ fn main() -> Result<()> {
         )
         .get_matches();
 
-    let mut conf = WalAcceptorConf {
-        data_dir: PathBuf::from("./"),
+    let mut conf = SafeKeeperConf {
+        // Always set to './'. We will chdir into the directory specified on the
+        // command line, so that when the server is running, all paths are relative
+        // to that.
+        workdir: PathBuf::from("./"),
         daemonize: false,
         no_sync: false,
         pageserver_addr: None,
@@ -91,10 +93,8 @@ fn main() -> Result<()> {
     };
 
     if let Some(dir) = arg_matches.value_of("datadir") {
-        conf.data_dir = PathBuf::from(dir);
-
         // change into the data directory.
-        std::env::set_current_dir(&conf.data_dir)?;
+        std::env::set_current_dir(PathBuf::from(dir))?;
     }
 
     if arg_matches.is_present("no-sync") {
@@ -125,20 +125,19 @@ fn main() -> Result<()> {
         conf.recall_period = Some(humantime::parse_duration(recall)?);
     }
 
-    start_wal_acceptor(conf)
+    start_safekeeper(conf)
 }
 
-fn start_wal_acceptor(conf: WalAcceptorConf) -> Result<()> {
-    let log_filename = conf.data_dir.join("safekeeper.log");
-    let log_file = logging::init(log_filename, conf.daemonize)?;
+fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
+    let log_file = logging::init("safekeeper.log", conf.daemonize)?;
 
-    let http_listener = TcpListener::bind(conf.listen_http_addr.clone()).map_err(|e| {
+    let http_listener = tcp_listener::bind(conf.listen_http_addr.clone()).map_err(|e| {
         error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
         e
     })?;
 
     info!("Starting safekeeper on {}", conf.listen_pg_addr);
-    let pg_listener = TcpListener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
+    let pg_listener = tcp_listener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
         error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
         e
     })?;

walkeeper/src/http/routes.rs

@@ -10,7 +10,7 @@ use zenith_utils::lsn::Lsn;
 use crate::safekeeper::AcceptorState;
 use crate::timeline::CreateControlFile;
 use crate::timeline::GlobalTimelines;
-use crate::WalAcceptorConf;
+use crate::SafeKeeperConf;
 use zenith_utils::http::endpoint;
 use zenith_utils::http::error::ApiError;
 use zenith_utils::http::json::json_response;
@@ -22,9 +22,9 @@ async fn status_handler(_: Request<Body>) -> Result<Response<Body>, ApiError> {
     Ok(json_response(StatusCode::OK, "")?)
 }
 
-fn get_conf(request: &Request<Body>) -> &WalAcceptorConf {
+fn get_conf(request: &Request<Body>) -> &SafeKeeperConf {
     request
-        .data::<Arc<WalAcceptorConf>>()
+        .data::<Arc<SafeKeeperConf>>()
         .expect("unknown state type")
         .as_ref()
 }
@@ -49,6 +49,8 @@ struct TimelineStatus {
     commit_lsn: Lsn,
     #[serde(serialize_with = "display_serialize")]
     truncate_lsn: Lsn,
+    #[serde(serialize_with = "display_serialize")]
+    flush_lsn: Lsn,
 }
 
 /// Report info about timeline.
@@ -64,6 +66,7 @@ async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body
     )
     .map_err(ApiError::from_err)?;
     let sk_state = tli.get_info();
+    let (flush_lsn, _) = tli.get_end_of_wal();
 
     let status = TimelineStatus {
         tenant_id,
@@ -71,12 +74,13 @@ async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body
         acceptor_state: sk_state.acceptor_state,
         commit_lsn: sk_state.commit_lsn,
         truncate_lsn: sk_state.truncate_lsn,
+        flush_lsn,
     };
     Ok(json_response(StatusCode::OK, status)?)
 }
 
 /// Safekeeper http router.
-pub fn make_router(conf: WalAcceptorConf) -> RouterBuilder<hyper::Body, ApiError> {
+pub fn make_router(conf: SafeKeeperConf) -> RouterBuilder<hyper::Body, ApiError> {
     let router = endpoint::make_router();
     router
         .data(Arc::new(conf))

walkeeper/src/lib.rs

@@ -23,8 +23,15 @@ pub mod defaults {
 }
 
 #[derive(Debug, Clone)]
-pub struct WalAcceptorConf {
-    pub data_dir: PathBuf,
+pub struct SafeKeeperConf {
+    // Repository directory, relative to current working directory.
+    // Normally, the safekeeper changes the current working directory
+    // to the repository, and 'workdir' is always '.'. But we don't do
+    // that during unit testing, because the current directory is global
+    // to the process but different unit tests work on different
+    // data directories to avoid clashing with each other.
+    pub workdir: PathBuf,
+
     pub daemonize: bool,
     pub no_sync: bool,
     pub listen_pg_addr: String,

walkeeper/src/receive_wal.rs

@@ -16,7 +16,7 @@ use crate::safekeeper::ProposerAcceptorMessage;
 
 use crate::send_wal::SendWalHandler;
 use crate::timeline::TimelineTools;
-use crate::WalAcceptorConf;
+use crate::SafeKeeperConf;
 use zenith_utils::connstring::connection_host_port;
 use zenith_utils::postgres_backend::PostgresBackend;
 use zenith_utils::pq_proto::{BeMessage, FeMessage};
@@ -33,7 +33,7 @@ pub struct ReceiveWalConn<'pg> {
 /// Periodically request pageserver to call back.
 /// If pageserver already has replication channel, it will just ignore this request
 ///
-fn request_callback(conf: WalAcceptorConf, timelineid: ZTimelineId, tenantid: ZTenantId) {
+fn request_callback(conf: SafeKeeperConf, timelineid: ZTimelineId, tenantid: ZTenantId) {
     let ps_addr = conf.pageserver_addr.unwrap();
     let ps_connstr = format!(
         "postgresql://no_user:{}@{}/no_db",

walkeeper/src/replication.rs

@@ -2,7 +2,7 @@
 //! with the "START_REPLICATION" message.
 
 use crate::send_wal::SendWalHandler;
-use crate::timeline::{Timeline, TimelineTools};
+use crate::timeline::{ReplicaState, Timeline, TimelineTools};
 use anyhow::{anyhow, Context, Result};
 use bytes::Bytes;
 use log::*;
@@ -20,7 +20,7 @@ use std::{str, thread};
 use zenith_utils::bin_ser::BeSer;
 use zenith_utils::lsn::Lsn;
 use zenith_utils::postgres_backend::PostgresBackend;
-use zenith_utils::pq_proto::{BeMessage, FeMessage, XLogDataBody};
+use zenith_utils::pq_proto::{BeMessage, FeMessage, WalSndKeepAlive, XLogDataBody};
 use zenith_utils::sock_split::ReadStream;
 
 pub const END_REPLICATION_MARKER: Lsn = Lsn::MAX;
@@ -32,7 +32,7 @@ const STANDBY_STATUS_UPDATE_TAG_BYTE: u8 = b'r';
 type FullTransactionId = u64;
 
 /// Hot standby feedback received from replica
-#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+#[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize)]
 pub struct HotStandbyFeedback {
     pub ts: TimestampTz,
     pub xmin: FullTransactionId,
@@ -49,6 +49,16 @@ impl HotStandbyFeedback {
     }
 }
 
+/// Standby status update
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct StandbyReply {
+    pub write_lsn: Lsn, // disk consistent lSN
+    pub flush_lsn: Lsn, // LSN committedby quorum
+    pub apply_lsn: Lsn, // not used
+    pub reply_ts: TimestampTz,
+    pub reply_requested: bool,
+}
+
 /// A network connection that's speaking the replication protocol.
 pub struct ReplicationConn {
     /// This is an `Option` because we will spawn a background thread that will
@@ -56,16 +66,15 @@ pub struct ReplicationConn {
     stream_in: Option<ReadStream>,
 }
 
-// TODO: move this to crate::timeline when there's more users
-// TODO: design a proper Timeline mock api
-trait HsFeedbackSubscriber {
-    fn add_hs_feedback(&self, _feedback: HotStandbyFeedback) {}
+/// Scope guard to unregister replication connection from timeline
+struct ReplicationConnGuard {
+    replica: usize, // replica internal ID assigned by timeline
+    timeline: Arc<Timeline>,
 }
 
-impl HsFeedbackSubscriber for Arc<Timeline> {
-    #[inline(always)]
-    fn add_hs_feedback(&self, feedback: HotStandbyFeedback) {
-        Timeline::add_hs_feedback(self, feedback);
+impl Drop for ReplicationConnGuard {
+    fn drop(&mut self) {
+        self.timeline.update_replica_state(self.replica, None);
     }
 }
 
@@ -79,26 +88,33 @@ impl ReplicationConn {
 
     /// Handle incoming messages from the network.
     /// This is spawned into the background by `handle_start_replication`.
-    fn background_thread(
-        mut stream_in: impl Read,
-        subscriber: impl HsFeedbackSubscriber,
-    ) -> Result<()> {
+    fn background_thread(mut stream_in: impl Read, timeline: Arc<Timeline>) -> Result<()> {
+        let mut state = ReplicaState::new();
+        let replica = timeline.add_replica(state);
+        let _guard = ReplicationConnGuard {
+            replica,
+            timeline: timeline.clone(),
+        };
         // Wait for replica's feedback.
         while let Some(msg) = FeMessage::read(&mut stream_in)? {
             match &msg {
                 FeMessage::CopyData(m) => {
                     // There's two possible data messages that the client is supposed to send here:
-                    // `HotStandbyFeedback` and `StandbyStatusUpdate`. We only handle hot standby
-                    // feedback.
+                    // `HotStandbyFeedback` and `StandbyStatusUpdate`.
 
                     match m.first().cloned() {
                         Some(HOT_STANDBY_FEEDBACK_TAG_BYTE) => {
                             // Note: deserializing is on m[1..] because we skip the tag byte.
-                            let feedback = HotStandbyFeedback::des(&m[1..])
+                            state.hs_feedback = HotStandbyFeedback::des(&m[1..])
                                 .context("failed to deserialize HotStandbyFeedback")?;
-                            subscriber.add_hs_feedback(feedback);
+                            timeline.update_replica_state(replica, Some(state));
+                        }
+                        Some(STANDBY_STATUS_UPDATE_TAG_BYTE) => {
+                            let reply = StandbyReply::des(&m[1..])
+                                .context("failed to deserialize StandbyReply")?;
+                            state.disk_consistent_lsn = reply.write_lsn;
+                            timeline.update_replica_state(replica, Some(state));
                         }
-                        Some(STANDBY_STATUS_UPDATE_TAG_BYTE) => (),
                         _ => warn!("unexpected message {:?}", msg),
                     }
                 }
@@ -187,7 +203,7 @@ impl ReplicationConn {
         // switch to copy
         pgb.write_message(&BeMessage::CopyBothResponse)?;
 
-        let mut end_pos: Lsn;
+        let mut end_pos = Lsn(0);
         let mut wal_file: Option<File> = None;
 
         loop {
@@ -202,7 +218,18 @@ impl ReplicationConn {
             } else {
                 /* normal mode */
                 let timeline = swh.timeline.get();
-                end_pos = timeline.wait_for_lsn(start_pos);
+                if let Some(lsn) = timeline.wait_for_lsn(start_pos) {
+                    end_pos = lsn
+                } else {
+                    // timeout expired: request pageserver status
+                    pgb.write_message(&BeMessage::KeepAlive(WalSndKeepAlive {
+                        sent_ptr: end_pos.0,
+                        timestamp: get_current_timestamp(),
+                        request_reply: true,
+                    }))
+                    .context("Failed to send KeepAlive message")?;
+                    continue;
+                }
             }
             if end_pos == END_REPLICATION_MARKER {
                 break;
@@ -216,7 +243,7 @@ impl ReplicationConn {
                     let segno = start_pos.segment_number(wal_seg_size);
                     let wal_file_name = XLogFileName(timeline, segno, wal_seg_size);
                     let timeline_id = swh.timeline.get().timelineid.to_string();
-                    let wal_file_path = swh.conf.data_dir.join(timeline_id).join(wal_file_name);
+                    let wal_file_path = swh.conf.workdir.join(timeline_id).join(wal_file_name);
                     Self::open_wal_file(&wal_file_path)?
                 }
             };
@@ -257,18 +284,3 @@ impl ReplicationConn {
         Ok(())
     }
 }
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    // A no-op impl for tests
-    impl HsFeedbackSubscriber for () {}
-
-    #[test]
-    fn test_replication_conn_background_thread_eof() {
-        // Test that background_thread recognizes EOF
-        let stream: &[u8] = &[];
-        ReplicationConn::background_thread(stream, ()).unwrap();
-    }
-}

walkeeper/src/s3_offload.rs

@@ -18,9 +18,9 @@ use tokio::runtime;
 use tokio::time::sleep;
 use walkdir::WalkDir;
 
-use crate::WalAcceptorConf;
+use crate::SafeKeeperConf;
 
-pub fn thread_main(conf: WalAcceptorConf) {
+pub fn thread_main(conf: SafeKeeperConf) {
     // Create a new thread pool
     //
     // FIXME: keep it single-threaded for now, make it easier to debug with gdb,
@@ -42,7 +42,7 @@ async fn offload_files(
     bucket: &Bucket,
     listing: &HashSet<String>,
     dir_path: &Path,
-    conf: &WalAcceptorConf,
+    conf: &SafeKeeperConf,
 ) -> Result<u64> {
     let horizon = SystemTime::now() - conf.ttl.unwrap();
     let mut n: u64 = 0;
@@ -54,7 +54,7 @@ async fn offload_files(
             && IsXLogFileName(entry.file_name().to_str().unwrap())
             && entry.metadata().unwrap().created().unwrap() <= horizon
         {
-            let relpath = path.strip_prefix(&conf.data_dir).unwrap();
+            let relpath = path.strip_prefix(&conf.workdir).unwrap();
             let s3path = String::from("walarchive/") + relpath.to_str().unwrap();
             if !listing.contains(&s3path) {
                 let mut file = File::open(&path)?;
@@ -70,7 +70,7 @@ async fn offload_files(
     Ok(n)
 }
 
-async fn main_loop(conf: &WalAcceptorConf) -> Result<()> {
+async fn main_loop(conf: &SafeKeeperConf) -> Result<()> {
     let region = Region::Custom {
         region: env::var("S3_REGION").unwrap(),
         endpoint: env::var("S3_ENDPOINT").unwrap(),
@@ -97,7 +97,7 @@ async fn main_loop(conf: &WalAcceptorConf) -> Result<()> {
             .flat_map(|b| b.contents.iter().map(|o| o.key.clone()))
             .collect();
 
-        let n = offload_files(&bucket, &listing, &conf.data_dir, conf).await?;
+        let n = offload_files(&bucket, &listing, &conf.workdir, conf).await?;
         info!("Offload {} files to S3", n);
         sleep(conf.ttl.unwrap()).await;
     }

walkeeper/src/safekeeper.rs

@@ -19,7 +19,10 @@ 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_metrics::{
+    register_gauge_vec, register_histogram_vec, Gauge, GaugeVec, Histogram, HistogramVec,
+    DISK_WRITE_SECONDS_BUCKETS,
+};
 use zenith_utils::bin_ser::LeSer;
 use zenith_utils::lsn::Lsn;
 use zenith_utils::pq_proto::SystemId;
@@ -191,6 +194,8 @@ pub struct AppendResponse {
     // We report back our awareness about which WAL is committed, as this is
     // a criterion for walproposer --sync mode exit
     pub commit_lsn: Lsn,
+    // Min disk consistent lsn of pageservers (portion of WAL applied and written to the disk by pageservers)
+    pub disk_consistent_lsn: Lsn,
     pub hs_feedback: HotStandbyFeedback,
 }
 
@@ -297,11 +302,27 @@ lazy_static! {
         &["ztli"]
     )
     .expect("Failed to register safekeeper_commit_lsn gauge vec");
+    static ref WRITE_WAL_BYTES: HistogramVec = register_histogram_vec!(
+        "safekeeper_write_wal_bytes",
+        "Bytes written to WAL in a single request, grouped by timeline",
+        &["timeline_id"],
+        vec![1.0, 10.0, 100.0, 1024.0, 8192.0, 128.0 * 1024.0, 1024.0 * 1024.0, 10.0 * 1024.0 * 1024.0]
+    )
+    .expect("Failed to register safekeeper_write_wal_bytes histogram vec");
+    static ref WRITE_WAL_SECONDS: HistogramVec = register_histogram_vec!(
+        "safekeeper_write_wal_seconds",
+        "Seconds spent writing and syncing WAL to a disk in a single request, grouped by timeline",
+        &["timeline_id"],
+        DISK_WRITE_SECONDS_BUCKETS.to_vec()
+    )
+    .expect("Failed to register safekeeper_write_wal_seconds histogram vec");
 }
 
 struct SafeKeeperMetrics {
     flush_lsn: Gauge,
     commit_lsn: Gauge,
+    write_wal_bytes: Histogram,
+    write_wal_seconds: Histogram,
 }
 
 impl SafeKeeperMetrics {
@@ -310,6 +331,8 @@ impl SafeKeeperMetrics {
         SafeKeeperMetrics {
             flush_lsn: FLUSH_LSN_GAUGE.with_label_values(&[&ztli_str]),
             commit_lsn: COMMIT_LSN_GAUGE.with_label_values(&[&ztli_str]),
+            write_wal_bytes: WRITE_WAL_BYTES.with_label_values(&[&ztli_str]),
+            write_wal_seconds: WRITE_WAL_SECONDS.with_label_values(&[&ztli_str]),
         }
     }
 
@@ -317,6 +340,8 @@ impl SafeKeeperMetrics {
         SafeKeeperMetrics {
             flush_lsn: FLUSH_LSN_GAUGE.with_label_values(&["n/a"]),
             commit_lsn: COMMIT_LSN_GAUGE.with_label_values(&["n/a"]),
+            write_wal_bytes: WRITE_WAL_BYTES.with_label_values(&["n/a"]),
+            write_wal_seconds: WRITE_WAL_SECONDS.with_label_values(&["n/a"]),
         }
     }
 }
@@ -458,6 +483,7 @@ where
                 epoch: self.s.acceptor_state.epoch,
                 commit_lsn: Lsn(0),
                 flush_lsn: Lsn(0),
+                disk_consistent_lsn: Lsn(0),
                 hs_feedback: HotStandbyFeedback::empty(),
             };
             return Ok(AcceptorProposerMessage::AppendResponse(resp));
@@ -469,8 +495,14 @@ where
         // do the job
         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)?;
+            self.metrics
+                .write_wal_bytes
+                .observe(msg.wal_data.len() as f64);
+            {
+                let _timer = self.metrics.write_wal_seconds.start_timer();
+                self.storage
+                    .write_wal(&self.s.server, msg.h.begin_lsn, &msg.wal_data)?;
+            }
 
             // figure out last record's end lsn for reporting (if we got the
             // whole record)
@@ -567,6 +599,7 @@ where
             epoch: self.s.acceptor_state.epoch,
             flush_lsn: self.flush_lsn,
             commit_lsn: self.s.commit_lsn,
+            disk_consistent_lsn: Lsn(0),
             // will be filled by caller code to avoid bothering safekeeper
             hs_feedback: HotStandbyFeedback::empty(),
         };

walkeeper/src/send_wal.rs

@@ -1,4 +1,4 @@
-//! Part of WAL acceptor pretending to be Postgres, streaming xlog to
+//! Part of Safekeeper pretending to be Postgres, streaming xlog to
 //! pageserver/any other consumer.
 //!
 
@@ -6,7 +6,7 @@ use crate::json_ctrl::handle_json_ctrl;
 use crate::receive_wal::ReceiveWalConn;
 use crate::replication::ReplicationConn;
 use crate::timeline::{Timeline, TimelineTools};
-use crate::WalAcceptorConf;
+use crate::SafeKeeperConf;
 use anyhow::{anyhow, bail, Result};
 use bytes::Bytes;
 use std::str::FromStr;
@@ -20,7 +20,7 @@ use crate::timeline::CreateControlFile;
 
 /// Handler for streaming WAL from acceptor
 pub struct SendWalHandler {
-    pub conf: WalAcceptorConf,
+    pub conf: SafeKeeperConf,
     /// assigned application name
     pub appname: Option<String>,
     pub tenantid: Option<ZTenantId>,
@@ -85,7 +85,7 @@ impl postgres_backend::Handler for SendWalHandler {
 }
 
 impl SendWalHandler {
-    pub fn new(conf: WalAcceptorConf) -> Self {
+    pub fn new(conf: SafeKeeperConf) -> Self {
         SendWalHandler {
             conf,
             appname: None,

walkeeper/src/timeline.rs

@@ -11,9 +11,10 @@ use std::collections::HashMap;
 use std::fs::{self, File, OpenOptions};
 use std::io::{Seek, SeekFrom, Write};
 use std::sync::{Arc, Condvar, Mutex};
+use std::time::Duration;
+use zenith_metrics::{register_histogram_vec, Histogram, HistogramVec, DISK_WRITE_SECONDS_BUCKETS};
 use zenith_utils::bin_ser::LeSer;
 use zenith_utils::lsn::Lsn;
-
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 use crate::replication::{HotStandbyFeedback, END_REPLICATION_MARKER};
@@ -21,10 +22,39 @@ use crate::safekeeper::{
     AcceptorProposerMessage, ProposerAcceptorMessage, SafeKeeper, SafeKeeperState, ServerInfo,
     Storage, SK_FORMAT_VERSION, SK_MAGIC,
 };
-use crate::WalAcceptorConf;
+use crate::SafeKeeperConf;
 use postgres_ffi::xlog_utils::{XLogFileName, XLOG_BLCKSZ};
 
 const CONTROL_FILE_NAME: &str = "safekeeper.control";
+const POLL_STATE_TIMEOUT: Duration = Duration::from_secs(1);
+
+/// Replica status: host standby feedback + disk consistent lsn
+#[derive(Debug, Clone, Copy)]
+pub struct ReplicaState {
+    /// combined disk_consistent_lsn of pageservers
+    pub disk_consistent_lsn: Lsn,
+    /// combined hot standby feedback from all replicas
+    pub hs_feedback: HotStandbyFeedback,
+}
+
+impl Default for ReplicaState {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl ReplicaState {
+    pub fn new() -> ReplicaState {
+        ReplicaState {
+            disk_consistent_lsn: Lsn(u64::MAX),
+            hs_feedback: HotStandbyFeedback {
+                ts: 0,
+                xmin: u64::MAX,
+                catalog_xmin: u64::MAX,
+            },
+        }
+    }
+}
 
 /// Shared state associated with database instance (tenant)
 struct SharedState {
@@ -33,8 +63,8 @@ struct SharedState {
     /// For receiving-sending wal cooperation
     /// quorum commit LSN we've notified walsenders about
     notified_commit_lsn: Lsn,
-    /// combined hot standby feedback from all replicas
-    hs_feedback: HotStandbyFeedback,
+    /// State of replicas
+    replicas: Vec<Option<ReplicaState>>,
 }
 
 // A named boolean.
@@ -44,23 +74,70 @@ pub enum CreateControlFile {
     False,
 }
 
+lazy_static! {
+    static ref PERSIST_SYNC_CONTROL_FILE_SECONDS: HistogramVec = register_histogram_vec!(
+        "safekeeper_persist_sync_control_file_seconds",
+        "Seconds to persist and sync control file, grouped by timeline",
+        &["timeline_id"],
+        DISK_WRITE_SECONDS_BUCKETS.to_vec()
+    )
+    .expect("Failed to register safekeeper_persist_sync_control_file_seconds histogram vec");
+    static ref PERSIST_NOSYNC_CONTROL_FILE_SECONDS: HistogramVec = register_histogram_vec!(
+        "safekeeper_persist_nosync_control_file_seconds",
+        "Seconds to persist and sync control file, grouped by timeline",
+        &["timeline_id"],
+        DISK_WRITE_SECONDS_BUCKETS.to_vec()
+    )
+    .expect("Failed to register safekeeper_persist_nosync_control_file_seconds histogram vec");
+}
+
 impl SharedState {
+    /// Get combined stateof all alive replicas
+    pub fn get_replicas_state(&self) -> ReplicaState {
+        let mut acc = ReplicaState::new();
+        for state in self.replicas.iter().flatten() {
+            acc.hs_feedback.ts = max(acc.hs_feedback.ts, state.hs_feedback.ts);
+            acc.hs_feedback.xmin = min(acc.hs_feedback.xmin, state.hs_feedback.xmin);
+            acc.hs_feedback.catalog_xmin =
+                min(acc.hs_feedback.catalog_xmin, state.hs_feedback.catalog_xmin);
+            acc.disk_consistent_lsn = Lsn::min(acc.disk_consistent_lsn, state.disk_consistent_lsn);
+        }
+        acc
+    }
+
+    /// Assign new replica ID. We choose first empty cell in the replicas vector
+    /// or extend the vector if there are not free items.
+    pub fn add_replica(&mut self, state: ReplicaState) -> usize {
+        if let Some(pos) = self.replicas.iter().position(|r| r.is_none()) {
+            self.replicas[pos] = Some(state);
+            return pos;
+        }
+        let pos = self.replicas.len();
+        self.replicas.push(Some(state));
+        pos
+    }
+
     /// Restore SharedState from control file. Locks the control file along the
     /// way to prevent running more than one instance of safekeeper on the same
     /// data dir.
     /// If create=false and file doesn't exist, bails out.
     fn create_restore(
-        conf: &WalAcceptorConf,
+        conf: &SafeKeeperConf,
         timelineid: ZTimelineId,
         create: CreateControlFile,
     ) -> Result<Self> {
         let (cf, state) = SharedState::load_control_file(conf, timelineid, create)?;
+        let timelineid_str = format!("{}", timelineid);
         let storage = FileStorage {
             control_file: cf,
             conf: conf.clone(),
+            persist_sync_control_file_seconds: PERSIST_SYNC_CONTROL_FILE_SECONDS
+                .with_label_values(&[&timelineid_str]),
+            persist_nosync_control_file_seconds: PERSIST_NOSYNC_CONTROL_FILE_SECONDS
+                .with_label_values(&[&timelineid_str]),
         };
         let (flush_lsn, tli) = if state.server.wal_seg_size != 0 {
-            let wal_dir = conf.data_dir.join(format!("{}", timelineid));
+            let wal_dir = conf.workdir.join(format!("{}", timelineid));
             find_end_of_wal(
                 &wal_dir,
                 state.server.wal_seg_size as usize,
@@ -74,30 +151,19 @@ impl SharedState {
         Ok(Self {
             notified_commit_lsn: Lsn(0),
             sk: SafeKeeper::new(Lsn(flush_lsn), tli, storage, state),
-            hs_feedback: HotStandbyFeedback {
-                ts: 0,
-                xmin: u64::MAX,
-                catalog_xmin: u64::MAX,
-            },
+            replicas: Vec::new(),
         })
     }
 
-    /// Accumulate hot standby feedbacks from replicas
-    pub fn add_hs_feedback(&mut self, feedback: HotStandbyFeedback) {
-        self.hs_feedback.xmin = min(self.hs_feedback.xmin, feedback.xmin);
-        self.hs_feedback.catalog_xmin = min(self.hs_feedback.catalog_xmin, feedback.catalog_xmin);
-        self.hs_feedback.ts = max(self.hs_feedback.ts, feedback.ts);
-    }
-
     /// Fetch and lock control file (prevent running more than one instance of safekeeper)
     /// If create=false and file doesn't exist, bails out.
     fn load_control_file(
-        conf: &WalAcceptorConf,
+        conf: &SafeKeeperConf,
         timelineid: ZTimelineId,
         create: CreateControlFile,
     ) -> Result<(File, SafeKeeperState)> {
         let control_file_path = conf
-            .data_dir
+            .workdir
             .join(timelineid.to_string())
             .join(CONTROL_FILE_NAME);
         info!(
@@ -178,20 +244,27 @@ impl Timeline {
         }
     }
 
-    /// Wait for an LSN to be committed.
+    /// Timed wait for an LSN to be committed.
     ///
     /// Returns the last committed LSN, which will be at least
-    /// as high as the LSN waited for.
+    /// as high as the LSN waited for, or None if timeout expired.
     ///
-    pub fn wait_for_lsn(&self, lsn: Lsn) -> Lsn {
+    pub fn wait_for_lsn(&self, lsn: Lsn) -> Option<Lsn> {
         let mut shared_state = self.mutex.lock().unwrap();
         loop {
             let commit_lsn = shared_state.notified_commit_lsn;
             // This must be `>`, not `>=`.
             if commit_lsn > lsn {
-                return commit_lsn;
+                return Some(commit_lsn);
             }
-            shared_state = self.cond.wait(shared_state).unwrap();
+            let result = self
+                .cond
+                .wait_timeout(shared_state, POLL_STATE_TIMEOUT)
+                .unwrap();
+            if result.1.timed_out() {
+                return None;
+            }
+            shared_state = result.0
         }
     }
 
@@ -219,9 +292,11 @@ impl Timeline {
             // commit_lsn if we are catching up safekeeper.
             commit_lsn = shared_state.sk.commit_lsn;
 
-            // if this is AppendResponse, fill in proper hot standby feedback
+            // if this is AppendResponse, fill in proper hot standby feedback and disk consistent lsn
             if let AcceptorProposerMessage::AppendResponse(ref mut resp) = rmsg {
-                resp.hs_feedback = shared_state.hs_feedback.clone();
+                let state = shared_state.get_replicas_state();
+                resp.hs_feedback = state.hs_feedback;
+                resp.disk_consistent_lsn = state.disk_consistent_lsn;
             }
         }
         // Ping wal sender that new data might be available.
@@ -233,15 +308,14 @@ impl Timeline {
         self.mutex.lock().unwrap().sk.s.clone()
     }
 
-    // Accumulate hot standby feedbacks from replicas
-    pub fn add_hs_feedback(&self, feedback: HotStandbyFeedback) {
+    pub fn add_replica(&self, state: ReplicaState) -> usize {
         let mut shared_state = self.mutex.lock().unwrap();
-        shared_state.add_hs_feedback(feedback);
+        shared_state.add_replica(state)
     }
 
-    pub fn get_hs_feedback(&self) -> HotStandbyFeedback {
-        let shared_state = self.mutex.lock().unwrap();
-        shared_state.hs_feedback.clone()
+    pub fn update_replica_state(&self, id: usize, state: Option<ReplicaState>) {
+        let mut shared_state = self.mutex.lock().unwrap();
+        shared_state.replicas[id] = state;
     }
 
     pub fn get_end_of_wal(&self) -> (Lsn, u32) {
@@ -254,7 +328,7 @@ impl Timeline {
 pub trait TimelineTools {
     fn set(
         &mut self,
-        conf: &WalAcceptorConf,
+        conf: &SafeKeeperConf,
         tenant_id: ZTenantId,
         timeline_id: ZTimelineId,
         create: CreateControlFile,
@@ -266,7 +340,7 @@ pub trait TimelineTools {
 impl TimelineTools for Option<Arc<Timeline>> {
     fn set(
         &mut self,
-        conf: &WalAcceptorConf,
+        conf: &SafeKeeperConf,
         tenant_id: ZTenantId,
         timeline_id: ZTimelineId,
         create: CreateControlFile,
@@ -295,7 +369,7 @@ impl GlobalTimelines {
     /// Get a timeline with control file loaded from the global TIMELINES map.
     /// If control file doesn't exist and create=false, bails out.
     pub fn get(
-        conf: &WalAcceptorConf,
+        conf: &SafeKeeperConf,
         tenant_id: ZTenantId,
         timeline_id: ZTimelineId,
         create: CreateControlFile,
@@ -324,11 +398,19 @@ impl GlobalTimelines {
 #[derive(Debug)]
 struct FileStorage {
     control_file: File,
-    conf: WalAcceptorConf,
+    conf: SafeKeeperConf,
+    persist_sync_control_file_seconds: Histogram,
+    persist_nosync_control_file_seconds: Histogram,
 }
 
 impl Storage for FileStorage {
     fn persist(&mut self, s: &SafeKeeperState, sync: bool) -> Result<()> {
+        let _timer = if sync {
+            &self.persist_sync_control_file_seconds
+        } else {
+            &self.persist_nosync_control_file_seconds
+        }
+        .start_timer();
         self.control_file.seek(SeekFrom::Start(0))?;
         s.ser_into(&mut self.control_file)?;
         if sync {
@@ -368,12 +450,12 @@ impl Storage for FileStorage {
             let wal_file_name = XLogFileName(server.tli, segno, wal_seg_size);
             let wal_file_path = self
                 .conf
-                .data_dir
+                .workdir
                 .join(ztli.to_string())
                 .join(wal_file_name.clone());
             let wal_file_partial_path = self
                 .conf
-                .data_dir
+                .workdir
                 .join(ztli.to_string())
                 .join(wal_file_name.clone() + ".partial");
 

walkeeper/src/wal_service.rs

@@ -8,11 +8,11 @@ use std::net::{TcpListener, TcpStream};
 use std::thread;
 
 use crate::send_wal::SendWalHandler;
-use crate::WalAcceptorConf;
+use crate::SafeKeeperConf;
 use zenith_utils::postgres_backend::{AuthType, PostgresBackend};
 
 /// Accept incoming TCP connections and spawn them into a background thread.
-pub fn thread_main(conf: WalAcceptorConf, listener: TcpListener) -> Result<()> {
+pub fn thread_main(conf: SafeKeeperConf, listener: TcpListener) -> Result<()> {
     loop {
         match listener.accept() {
             Ok((socket, peer_addr)) => {
@@ -31,7 +31,7 @@ pub fn thread_main(conf: WalAcceptorConf, listener: TcpListener) -> Result<()> {
 
 /// This is run by `thread_main` above, inside a background thread.
 ///
-fn handle_socket(socket: TcpStream, conf: WalAcceptorConf) -> Result<()> {
+fn handle_socket(socket: TcpStream, conf: SafeKeeperConf) -> Result<()> {
     socket.set_nodelay(true)?;
 
     let mut conn_handler = SendWalHandler::new(conf);

zenith/Cargo.toml

@@ -15,6 +15,7 @@ postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbf
 # FIXME: 'pageserver' is needed for BranchInfo. Refactor
 pageserver = { path = "../pageserver" }
 control_plane = { path = "../control_plane" }
+walkeeper = { path = "../walkeeper" }
 postgres_ffi = { path = "../postgres_ffi" }
 zenith_utils = { path = "../zenith_utils" }
 workspace_hack = { path = "../workspace_hack" }

zenith/src/main.rs

@@ -1,19 +1,54 @@
-use anyhow::anyhow;
+use anyhow::{anyhow, bail};
 use anyhow::{Context, Result};
 use clap::{App, AppSettings, Arg, ArgMatches, SubCommand};
 use control_plane::compute::ComputeControlPlane;
 use control_plane::local_env;
+use control_plane::local_env::LocalEnv;
+use control_plane::safekeeper::SafekeeperNode;
 use control_plane::storage::PageServerNode;
-use pageserver::defaults::{DEFAULT_HTTP_LISTEN_PORT, DEFAULT_PG_LISTEN_PORT};
+use pageserver::defaults::{
+    DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
+    DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
+};
 use std::collections::HashMap;
 use std::process::exit;
 use std::str::FromStr;
-use zenith_utils::auth::{encode_from_key_path, Claims, Scope};
+use walkeeper::defaults::{
+    DEFAULT_HTTP_LISTEN_PORT as DEFAULT_SAFEKEEPER_HTTP_PORT,
+    DEFAULT_PG_LISTEN_PORT as DEFAULT_SAFEKEEPER_PG_PORT,
+};
+use zenith_utils::auth::{Claims, Scope};
 use zenith_utils::postgres_backend::AuthType;
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 use pageserver::branches::BranchInfo;
 
+// Default name of a safekeeper node, if not specified on the command line.
+const DEFAULT_SAFEKEEPER_NAME: &str = "single";
+
+fn default_conf() -> String {
+    format!(
+        r#"
+# Default built-in configuration, defined in main.rs
+[pageserver]
+pg_port = {pageserver_pg_port}
+http_port = {pageserver_http_port}
+auth_type = '{pageserver_auth_type}'
+
+[[safekeepers]]
+name = '{safekeeper_name}'
+pg_port = {safekeeper_pg_port}
+http_port = {safekeeper_http_port}
+"#,
+        pageserver_pg_port = DEFAULT_PAGESERVER_PG_PORT,
+        pageserver_http_port = DEFAULT_PAGESERVER_HTTP_PORT,
+        pageserver_auth_type = AuthType::Trust,
+        safekeeper_name = DEFAULT_SAFEKEEPER_NAME,
+        safekeeper_pg_port = DEFAULT_SAFEKEEPER_PG_PORT,
+        safekeeper_http_port = DEFAULT_SAFEKEEPER_HTTP_PORT,
+    )
+}
+
 ///
 /// Branches tree element used as a value in the HashMap.
 ///
@@ -32,12 +67,21 @@ struct BranchTreeEl {
 //   * Providing CLI api to the pageserver
 //   * TODO: export/import to/from usual postgres
 fn main() -> Result<()> {
-    let timeline_arg = Arg::with_name("timeline")
-        .short("n")
+    let pg_node_arg = Arg::with_name("node")
         .index(1)
-        .help("Timeline name")
+        .help("Node name")
         .required(true);
 
+    let safekeeper_node_arg = Arg::with_name("node")
+        .index(1)
+        .help("Node name")
+        .required(false);
+
+    let timeline_arg = Arg::with_name("timeline")
+        .index(2)
+        .help("Branch name or a point-in time specification")
+        .required(false);
+
     let tenantid_arg = Arg::with_name("tenantid")
         .long("tenantid")
         .help("Tenant id. Represented as a hexadecimal string 32 symbols length")
@@ -49,29 +93,25 @@ fn main() -> Result<()> {
         .required(false)
         .value_name("port");
 
+    let stop_mode_arg = Arg::with_name("stop-mode")
+        .short("m")
+        .takes_value(true)
+        .possible_values(&["fast", "immediate"])
+        .help("If 'immediate', don't flush repository data at shutdown")
+        .required(false)
+        .value_name("stop-mode");
+
     let matches = App::new("Zenith CLI")
         .setting(AppSettings::ArgRequiredElseHelp)
         .subcommand(
             SubCommand::with_name("init")
                 .about("Initialize a new Zenith repository")
                 .arg(
-                    Arg::with_name("pageserver-pg-port")
-                        .long("pageserver-pg-port")
+                    Arg::with_name("config")
+                        .long("config")
                         .required(false)
-                        .value_name("pageserver-pg-port"),
+                        .value_name("config"),
                 )
-                .arg(
-                    Arg::with_name("pageserver-http-port")
-                        .long("pageserver-http-port")
-                        .required(false)
-                        .value_name("pageserver-http-port"),
-                )
-                .arg(
-                    Arg::with_name("enable-auth")
-                        .long("enable-auth")
-                        .takes_value(false)
-                        .help("Enable authentication using ZenithJWT")
-                ),
         )
         .subcommand(
             SubCommand::with_name("branch")
@@ -86,15 +126,35 @@ fn main() -> Result<()> {
             .subcommand(SubCommand::with_name("list"))
             .subcommand(SubCommand::with_name("create").arg(Arg::with_name("tenantid").required(false).index(1)))
         )
-        .subcommand(SubCommand::with_name("status"))
-        .subcommand(SubCommand::with_name("start").about("Start local pageserver"))
-        .subcommand(SubCommand::with_name("stop").about("Stop local pageserver")
-                    .arg(Arg::with_name("immediate")
-                    .help("Don't flush repository data at shutdown")
-                    .required(false)
-                    )
+        .subcommand(
+            SubCommand::with_name("pageserver")
+                .setting(AppSettings::ArgRequiredElseHelp)
+                .about("Manage pageserver")
+                .subcommand(SubCommand::with_name("status"))
+                .subcommand(SubCommand::with_name("start").about("Start local pageserver"))
+                .subcommand(SubCommand::with_name("stop").about("Stop local pageserver")
+                            .arg(stop_mode_arg.clone()))
+                .subcommand(SubCommand::with_name("restart").about("Restart local pageserver"))
+        )
+        .subcommand(
+            SubCommand::with_name("safekeeper")
+                .setting(AppSettings::ArgRequiredElseHelp)
+                .about("Manage safekeepers")
+                .subcommand(SubCommand::with_name("start")
+                            .about("Start local safekeeper")
+                            .arg(safekeeper_node_arg.clone())
+                )
+                .subcommand(SubCommand::with_name("stop")
+                            .about("Stop local safekeeper")
+                            .arg(safekeeper_node_arg.clone())
+                            .arg(stop_mode_arg.clone())
+                )
+                .subcommand(SubCommand::with_name("restart")
+                            .about("Restart local safekeeper")
+                            .arg(safekeeper_node_arg.clone())
+                            .arg(stop_mode_arg.clone())
+                )
         )
-        .subcommand(SubCommand::with_name("restart").about("Restart local pageserver"))
         .subcommand(
             SubCommand::with_name("pg")
                 .setting(AppSettings::ArgRequiredElseHelp)
@@ -102,7 +162,10 @@ fn main() -> Result<()> {
                 .subcommand(SubCommand::with_name("list").arg(tenantid_arg.clone()))
                 .subcommand(SubCommand::with_name("create")
                     .about("Create a postgres compute node")
-                    .arg(timeline_arg.clone()).arg(tenantid_arg.clone()).arg(port_arg.clone())
+                    .arg(pg_node_arg.clone())
+                    .arg(timeline_arg.clone())
+                    .arg(tenantid_arg.clone())
+                    .arg(port_arg.clone())
                     .arg(
                         Arg::with_name("config-only")
                             .help("Don't do basebackup, create compute node with only config files")
@@ -111,13 +174,13 @@ fn main() -> Result<()> {
                     ))
                 .subcommand(SubCommand::with_name("start")
                     .about("Start a postgres compute node.\n This command actually creates new node from scratch, but preserves existing config files")
-                    .arg(
-                        timeline_arg.clone()
-                    ).arg(
-                        tenantid_arg.clone()
-                    ).arg(port_arg.clone()))
+                    .arg(pg_node_arg.clone())
+                    .arg(timeline_arg.clone())
+                    .arg(tenantid_arg.clone())
+                    .arg(port_arg.clone()))
                 .subcommand(
                     SubCommand::with_name("stop")
+                        .arg(pg_node_arg.clone())
                         .arg(timeline_arg.clone())
                         .arg(tenantid_arg.clone())
                         .arg(
@@ -129,116 +192,49 @@ fn main() -> Result<()> {
                     )
 
         )
+        .subcommand(
+            SubCommand::with_name("start")
+                .about("Start page server and safekeepers")
+        )
+        .subcommand(
+            SubCommand::with_name("stop")
+                .about("Stop page server and safekeepers")
+                .arg(stop_mode_arg.clone())
+        )
         .get_matches();
 
-    // Create config file
-    if let ("init", Some(init_match)) = matches.subcommand() {
-        let tenantid = ZTenantId::generate();
-        let pageserver_pg_port = match init_match.value_of("pageserver-pg-port") {
-            Some(v) => v.parse()?,
-            None => DEFAULT_PG_LISTEN_PORT,
-        };
-        let pageserver_http_port = match init_match.value_of("pageserver-http-port") {
-            Some(v) => v.parse()?,
-            None => DEFAULT_HTTP_LISTEN_PORT,
+    let (sub_name, sub_args) = matches.subcommand();
+    let sub_args = sub_args.expect("no subcommand");
+
+    // Check for 'zenith init' command first.
+    let subcmd_result = if sub_name == "init" {
+        handle_init(sub_args)
+    } else {
+        // all other commands need an existing config
+        let env = match LocalEnv::load_config() {
+            Ok(conf) => conf,
+            Err(e) => {
+                eprintln!("Error loading config: {}", e);
+                exit(1);
+            }
         };
 
-        let auth_type = if init_match.is_present("enable-auth") {
-            AuthType::ZenithJWT
-        } else {
-            AuthType::Trust
-        };
-
-        local_env::init(
-            pageserver_pg_port,
-            pageserver_http_port,
-            tenantid,
-            auth_type,
-        )
-        .with_context(|| "Failed to create config file")?;
+        match sub_name {
+            "tenant" => handle_tenant(sub_args, &env),
+            "branch" => handle_branch(sub_args, &env),
+            "start" => handle_start_all(sub_args, &env),
+            "stop" => handle_stop_all(sub_args, &env),
+            "pageserver" => handle_pageserver(sub_args, &env),
+            "pg" => handle_pg(sub_args, &env),
+            "safekeeper" => handle_safekeeper(sub_args, &env),
+            _ => bail!("unexpected subcommand {}", sub_name),
+        }
+    };
+    if let Err(e) = subcmd_result {
+        eprintln!("command failed: {}", e);
+        exit(1);
     }
 
-    // all other commands would need config
-    let env = match local_env::load_config() {
-        Ok(conf) => conf,
-        Err(e) => {
-            eprintln!("Error loading config: {}", e);
-            exit(1);
-        }
-    };
-
-    match matches.subcommand() {
-        ("init", Some(init_match)) => {
-            let pageserver = PageServerNode::from_env(&env);
-            if let Err(e) = pageserver.init(
-                Some(&env.tenantid.to_string()),
-                init_match.is_present("enable-auth"),
-            ) {
-                eprintln!("pageserver init failed: {}", e);
-                exit(1);
-            }
-        }
-        ("tenant", Some(args)) => {
-            if let Err(e) = handle_tenant(args, &env) {
-                eprintln!("tenant command failed: {}", e);
-                exit(1);
-            }
-        }
-
-        ("branch", Some(sub_args)) => {
-            if let Err(e) = handle_branch(sub_args, &env) {
-                eprintln!("branch command failed: {}", e);
-                exit(1);
-            }
-        }
-
-        ("start", Some(_sub_m)) => {
-            let pageserver = PageServerNode::from_env(&env);
-
-            if let Err(e) = pageserver.start() {
-                eprintln!("pageserver start failed: {}", e);
-                exit(1);
-            }
-        }
-
-        ("stop", Some(stop_match)) => {
-            let pageserver = PageServerNode::from_env(&env);
-
-            let immediate = stop_match.is_present("immediate");
-
-            if let Err(e) = pageserver.stop(immediate) {
-                eprintln!("pageserver stop failed: {}", e);
-                exit(1);
-            }
-        }
-
-        ("restart", Some(_sub_m)) => {
-            let pageserver = PageServerNode::from_env(&env);
-
-            //TODO what shutdown strategy should we use here?
-            if let Err(e) = pageserver.stop(false) {
-                eprintln!("pageserver stop failed: {}", e);
-                exit(1);
-            }
-
-            if let Err(e) = pageserver.start() {
-                eprintln!("pageserver start failed: {}", e);
-                exit(1);
-            }
-        }
-
-        ("status", Some(_sub_m)) => {}
-
-        ("pg", Some(pg_match)) => {
-            if let Err(e) = handle_pg(pg_match, &env) {
-                eprintln!("pg operation failed: {:?}", e);
-                exit(1);
-            }
-        }
-
-        _ => {}
-    };
-
     Ok(())
 }
 
@@ -373,12 +369,52 @@ fn get_branch_infos(
     Ok(branch_infos)
 }
 
+// Helper function to parse --tenantid option, or get the default from config file
+fn get_tenantid(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<ZTenantId> {
+    if let Some(tenantid_cmd) = sub_match.value_of("tenantid") {
+        Ok(ZTenantId::from_str(tenantid_cmd)?)
+    } else if let Some(tenantid_conf) = env.default_tenantid {
+        Ok(tenantid_conf)
+    } else {
+        bail!("No tenantid. Use --tenantid, or set 'default_tenantid' in the config file");
+    }
+}
+
+fn handle_init(init_match: &ArgMatches) -> Result<()> {
+    // Create config file
+    let toml_file: String = if let Some(config_path) = init_match.value_of("config") {
+        // load and parse the file
+        std::fs::read_to_string(std::path::Path::new(config_path))
+            .with_context(|| format!("Could not read configuration file \"{}\"", config_path))?
+    } else {
+        // Built-in default config
+        default_conf()
+    };
+
+    let mut env = LocalEnv::create_config(&toml_file)
+        .with_context(|| "Failed to create zenith configuration")?;
+    env.init()
+        .with_context(|| "Failed to initialize zenith repository")?;
+
+    // Call 'pageserver init'.
+    let pageserver = PageServerNode::from_env(&env);
+    if let Err(e) = pageserver.init(
+        // default_tenantid was generated by the `env.init()` call above
+        Some(&env.default_tenantid.unwrap().to_string()),
+    ) {
+        eprintln!("pageserver init failed: {}", e);
+        exit(1);
+    }
+
+    Ok(())
+}
+
 fn handle_tenant(tenant_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     let pageserver = PageServerNode::from_env(env);
     match tenant_match.subcommand() {
         ("list", Some(_)) => {
-            for tenant in pageserver.tenant_list()? {
-                println!("{}", tenant);
+            for t in pageserver.tenant_list()? {
+                println!("{} {}", t.id, t.state);
             }
         }
         ("create", Some(create_match)) => {
@@ -390,7 +426,10 @@ fn handle_tenant(tenant_match: &ArgMatches, env: &local_env::LocalEnv) -> Result
             pageserver.tenant_create(tenantid)?;
             println!("tenant successfully created on the pageserver");
         }
-        _ => {}
+
+        (sub_name, _) => {
+            bail!("Unexpected tenant subcommand '{}'", sub_name)
+        }
     }
     Ok(())
 }
@@ -398,22 +437,18 @@ fn handle_tenant(tenant_match: &ArgMatches, env: &local_env::LocalEnv) -> Result
 fn handle_branch(branch_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     let pageserver = PageServerNode::from_env(env);
 
+    let tenantid = get_tenantid(branch_match, env)?;
+
     if let Some(branchname) = branch_match.value_of("branchname") {
         let startpoint_str = branch_match
             .value_of("start-point")
             .ok_or_else(|| anyhow!("Missing start-point"))?;
-        let tenantid: ZTenantId = branch_match
-            .value_of("tenantid")
-            .map_or(Ok(env.tenantid), |value| value.parse())?;
         let branch = pageserver.branch_create(branchname, startpoint_str, &tenantid)?;
         println!(
             "Created branch '{}' at {:?} for tenant: {}",
             branch.name, branch.latest_valid_lsn, tenantid,
         );
     } else {
-        let tenantid: ZTenantId = branch_match
-            .value_of("tenantid")
-            .map_or(Ok(env.tenantid), |value| value.parse())?;
         // No arguments, list branches for tenant
         let branches = pageserver.branch_list(&tenantid)?;
         print_branches_tree(branches)?;
@@ -423,74 +458,79 @@ fn handle_branch(branch_match: &ArgMatches, env: &local_env::LocalEnv) -> Result
 }
 
 fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+    let (sub_name, sub_args) = pg_match.subcommand();
+    let sub_args = sub_args.expect("no pg subcommand");
+
     let mut cplane = ComputeControlPlane::load(env.clone())?;
 
-    match pg_match.subcommand() {
-        ("list", Some(list_match)) => {
-            let tenantid: ZTenantId = list_match
-                .value_of("tenantid")
-                .map_or(Ok(env.tenantid), |value| value.parse())?;
+    // All subcommands take an optional --tenantid option
+    let tenantid = get_tenantid(sub_args, env)?;
+
+    match sub_name {
+        "list" => {
             let branch_infos = get_branch_infos(env, &tenantid).unwrap_or_else(|e| {
                 eprintln!("Failed to load branch info: {}", e);
                 HashMap::new()
             });
 
-            println!("BRANCH\tADDRESS\t\tLSN\t\tSTATUS");
-            for ((_, timeline_name), node) in cplane
+            println!("NODE\tADDRESS\t\tBRANCH\tLSN\t\tSTATUS");
+            for ((_, node_name), node) in cplane
                 .nodes
                 .iter()
                 .filter(|((node_tenantid, _), _)| node_tenantid == &tenantid)
             {
+                // FIXME: This shows the LSN at the end of the timeline. It's not the
+                // right thing to do for read-only nodes that might be anchored at an
+                // older point in time, or following but lagging behind the primary.
+                let lsn_str = branch_infos
+                    .get(&node.timelineid)
+                    .map(|bi| bi.latest_valid_lsn.to_string())
+                    .unwrap_or_else(|| "?".to_string());
+
                 println!(
-                    "{}\t{}\t{}\t{}",
-                    timeline_name,
+                    "{}\t{}\t{}\t{}\t{}",
+                    node_name,
                     node.address,
-                    branch_infos
-                        .get(&node.timelineid)
-                        .map(|bi| bi.latest_valid_lsn.to_string())
-                        .unwrap_or_else(|| "?".to_string()),
+                    node.timelineid, // FIXME: resolve human-friendly branch name
+                    lsn_str,
                     node.status(),
                 );
             }
         }
-        ("create", Some(create_match)) => {
-            let tenantid: ZTenantId = create_match
-                .value_of("tenantid")
-                .map_or(Ok(env.tenantid), |value| value.parse())?;
-            let timeline_name = create_match.value_of("timeline").unwrap_or("main");
+        "create" => {
+            let node_name = sub_args.value_of("node").unwrap_or("main");
+            let timeline_name = sub_args.value_of("timeline").unwrap_or(node_name);
 
-            let port: Option<u16> = match create_match.value_of("port") {
+            let port: Option<u16> = match sub_args.value_of("port") {
                 Some(p) => Some(p.parse()?),
                 None => None,
             };
-            cplane.new_node(tenantid, timeline_name, port)?;
+            cplane.new_node(tenantid, node_name, timeline_name, port)?;
         }
-        ("start", Some(start_match)) => {
-            let tenantid: ZTenantId = start_match
-                .value_of("tenantid")
-                .map_or(Ok(env.tenantid), |value| value.parse())?;
-            let timeline_name = start_match.value_of("timeline").unwrap_or("main");
+        "start" => {
+            let node_name = sub_args.value_of("node").unwrap_or("main");
+            let timeline_name = sub_args.value_of("timeline");
 
-            let port: Option<u16> = match start_match.value_of("port") {
+            let port: Option<u16> = match sub_args.value_of("port") {
                 Some(p) => Some(p.parse()?),
                 None => None,
             };
 
-            let node = cplane.nodes.get(&(tenantid, timeline_name.to_owned()));
+            let node = cplane.nodes.get(&(tenantid, node_name.to_owned()));
 
-            let auth_token = if matches!(env.auth_type, AuthType::ZenithJWT) {
+            let auth_token = if matches!(env.pageserver.auth_type, AuthType::ZenithJWT) {
                 let claims = Claims::new(Some(tenantid), Scope::Tenant);
-                Some(encode_from_key_path(&claims, &env.private_key_path)?)
+
+                Some(env.generate_auth_token(&claims)?)
             } else {
                 None
             };
 
-            println!(
-                "Starting {} postgres on timeline {}...",
-                if node.is_some() { "existing" } else { "new" },
-                timeline_name
-            );
             if let Some(node) = node {
+                if timeline_name.is_some() {
+                    println!("timeline name ignored because node exists already");
+                }
+                println!("Starting existing postgres {}...", node_name);
                 node.start(&auth_token)?;
             } else {
                 // when used with custom port this results in non obvious behaviour
@@ -498,26 +538,170 @@ fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
                 // start --port X
                 // stop
                 // start <-- will also use port X even without explicit port argument
-                let node = cplane.new_node(tenantid, timeline_name, port)?;
+                let timeline_name = timeline_name.unwrap_or(node_name);
+                println!(
+                    "Starting new postgres {} on {}...",
+                    node_name, timeline_name
+                );
+                let node = cplane.new_node(tenantid, node_name, timeline_name, port)?;
                 node.start(&auth_token)?;
             }
         }
-        ("stop", Some(stop_match)) => {
-            let timeline_name = stop_match.value_of("timeline").unwrap_or("main");
-            let destroy = stop_match.is_present("destroy");
-            let tenantid: ZTenantId = stop_match
-                .value_of("tenantid")
-                .map_or(Ok(env.tenantid), |value| value.parse())?;
+        "stop" => {
+            let node_name = sub_args.value_of("node").unwrap_or("main");
+            let destroy = sub_args.is_present("destroy");
 
             let node = cplane
                 .nodes
-                .get(&(tenantid, timeline_name.to_owned()))
-                .ok_or_else(|| anyhow!("postgres {} is not found", timeline_name))?;
+                .get(&(tenantid, node_name.to_owned()))
+                .ok_or_else(|| anyhow!("postgres {} is not found", node_name))?;
             node.stop(destroy)?;
         }
 
-        _ => {}
+        _ => {
+            bail!("Unexpected pg subcommand '{}'", sub_name)
+        }
     }
 
     Ok(())
 }
+
+fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+    let pageserver = PageServerNode::from_env(env);
+
+    match sub_match.subcommand() {
+        ("start", Some(_sub_m)) => {
+            if let Err(e) = pageserver.start() {
+                eprintln!("pageserver start failed: {}", e);
+                exit(1);
+            }
+        }
+
+        ("stop", Some(stop_match)) => {
+            let immediate = stop_match.value_of("stop-mode") == Some("immediate");
+
+            if let Err(e) = pageserver.stop(immediate) {
+                eprintln!("pageserver stop failed: {}", e);
+                exit(1);
+            }
+        }
+
+        ("restart", Some(_sub_m)) => {
+            //TODO what shutdown strategy should we use here?
+            if let Err(e) = pageserver.stop(false) {
+                eprintln!("pageserver stop failed: {}", e);
+                exit(1);
+            }
+
+            if let Err(e) = pageserver.start() {
+                eprintln!("pageserver start failed: {}", e);
+                exit(1);
+            }
+        }
+
+        (sub_name, _) => {
+            bail!("Unexpected pageserver subcommand '{}'", sub_name)
+        }
+    }
+    Ok(())
+}
+
+fn get_safekeeper(env: &local_env::LocalEnv, name: &str) -> Result<SafekeeperNode> {
+    if let Some(node) = env.safekeepers.iter().find(|node| node.name == name) {
+        Ok(SafekeeperNode::from_env(env, node))
+    } else {
+        bail!("could not find safekeeper '{}'", name)
+    }
+}
+
+fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+    let (sub_name, sub_args) = sub_match.subcommand();
+    let sub_args = sub_args.expect("no safekeeper subcommand");
+
+    // All the commands take an optional safekeeper name argument
+    let node_name = sub_args.value_of("node").unwrap_or(DEFAULT_SAFEKEEPER_NAME);
+    let safekeeper = get_safekeeper(env, node_name)?;
+
+    match sub_name {
+        "start" => {
+            if let Err(e) = safekeeper.start() {
+                eprintln!("safekeeper start failed: {}", e);
+                exit(1);
+            }
+        }
+
+        "stop" => {
+            let immediate = sub_args.value_of("stop-mode") == Some("immediate");
+
+            if let Err(e) = safekeeper.stop(immediate) {
+                eprintln!("safekeeper stop failed: {}", e);
+                exit(1);
+            }
+        }
+
+        "restart" => {
+            let immediate = sub_args.value_of("stop-mode") == Some("immediate");
+
+            if let Err(e) = safekeeper.stop(immediate) {
+                eprintln!("safekeeper stop failed: {}", e);
+                exit(1);
+            }
+
+            if let Err(e) = safekeeper.start() {
+                eprintln!("safekeeper start failed: {}", e);
+                exit(1);
+            }
+        }
+
+        _ => {
+            bail!("Unexpected safekeeper subcommand '{}'", sub_name)
+        }
+    }
+    Ok(())
+}
+
+fn handle_start_all(_sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+    let pageserver = PageServerNode::from_env(env);
+
+    // Postgres nodes are not started automatically
+
+    if let Err(e) = pageserver.start() {
+        eprintln!("pageserver start failed: {}", e);
+        exit(1);
+    }
+
+    for node in env.safekeepers.iter() {
+        let safekeeper = SafekeeperNode::from_env(env, node);
+        if let Err(e) = safekeeper.start() {
+            eprintln!("safekeeper '{}' start failed: {}", safekeeper.name, e);
+            exit(1);
+        }
+    }
+    Ok(())
+}
+
+fn handle_stop_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+    let immediate = sub_match.value_of("stop-mode") == Some("immediate");
+
+    let pageserver = PageServerNode::from_env(env);
+
+    // Stop all compute nodes
+    let cplane = ComputeControlPlane::load(env.clone())?;
+    for (_k, node) in cplane.nodes {
+        if let Err(e) = node.stop(false) {
+            eprintln!("postgres stop failed: {}", e);
+        }
+    }
+
+    if let Err(e) = pageserver.stop(immediate) {
+        eprintln!("pageserver stop failed: {}", e);
+    }
+
+    for node in env.safekeepers.iter() {
+        let safekeeper = SafekeeperNode::from_env(env, node);
+        if let Err(e) = safekeeper.stop(immediate) {
+            eprintln!("safekeeper '{}' stop failed: {}", safekeeper.name, e);
+        }
+    }
+    Ok(())
+}

zenith_metrics/src/lib.rs

@@ -74,6 +74,10 @@ lazy_static! {
     .expect("Failed to register maxrss_kb int gauge");
 }
 
+pub const DISK_WRITE_SECONDS_BUCKETS: &[f64] = &[
+    0.000_050, 0.000_100, 0.000_500, 0.001, 0.003, 0.005, 0.01, 0.05, 0.1, 0.3, 0.5,
+];
+
 // Records I/O stats in a "cross-platform" way.
 // Compiles both on macOS and Linux, but current macOS implementation always returns 0 as values for I/O stats.
 // An alternative is to read procfs (`/proc/[pid]/io`) which does not work under macOS at all, hence abandoned.

zenith_utils/Cargo.toml

@@ -11,7 +11,6 @@ byteorder = "1.4.3"
 bytes = "1.0.1"
 hyper = { version = "0.14.7", features = ["full"] }
 lazy_static = "1.4.0"
-log = "0.4.14"
 postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
 routerify = "2"
 serde = { version = "1.0", features = ["derive"] }
@@ -19,8 +18,8 @@ serde_json = "1"
 thiserror = "1.0"
 tokio = "1.11"
 tracing = "0.1"
-tracing-log = "0.1"
-tracing-subscriber = "0.2"
+tracing-subscriber = { version = "0.3", features = ["env-filter"] }
+nix = "0.23.0"
 
 zenith_metrics = { path = "../zenith_metrics" }
 workspace_hack = { path = "../workspace_hack" }

zenith_utils/src/auth.rs

@@ -104,8 +104,8 @@ impl JwtAuth {
     }
 
     pub fn from_key_path(key_path: &Path) -> Result<Self> {
-        let public_key = fs::read_to_string(key_path)?;
-        Ok(Self::new(DecodingKey::from_rsa_pem(public_key.as_bytes())?))
+        let public_key = fs::read(key_path)?;
+        Ok(Self::new(DecodingKey::from_rsa_pem(&public_key)?))
     }
 
     pub fn decode(&self, token: &str) -> Result<TokenData<Claims>> {
@@ -114,8 +114,7 @@ impl JwtAuth {
 }
 
 // this function is used only for testing purposes in CLI e g generate tokens during init
-pub fn encode_from_key_path(claims: &Claims, key_path: &Path) -> Result<String> {
-    let key_data = fs::read_to_string(key_path)?;
-    let key = EncodingKey::from_rsa_pem(key_data.as_bytes())?;
+pub fn encode_from_key_file(claims: &Claims, key_data: &[u8]) -> Result<String> {
+    let key = EncodingKey::from_rsa_pem(key_data)?;
     Ok(encode(&Header::new(JWT_ALGORITHM), claims, &key)?)
 }