set pageserver id in dockerfile

This adds node id parameter to pageserver configuration. Also I use a
simple builder to construct pageserver config struct to avoid setting
node id to some temporary invalid value. Some of the changes in test
fixtures are needed to split init and start operations for envrionment.

.circleci/config.yml

@@ -1,28 +1,28 @@
 version: 2.1
 
 executors:
-  zenith-build-executor:
+  zenith-xlarge-executor:
     resource_class: xlarge
     docker:
-      - image: cimg/rust:1.56.1
-  zenith-python-executor:
+      # NB: when changed, do not forget to update rust image tag in all Dockerfiles
+      - image: zimg/rust:1.56
+  zenith-executor:
     docker:
-      - image: cimg/python:3.7.10  # Oldest available 3.7 with Ubuntu 20.04 (for GLIBC and Rust) at CirlceCI
+      - image: zimg/rust:1.56
 
 jobs:
   check-codestyle-rust:
-    executor: zenith-build-executor
+    executor: zenith-xlarge-executor
     steps:
       - checkout
       - run:
           name: rustfmt
           when: always
-          command: |
-            cargo fmt --all -- --check
+          command: cargo fmt --all -- --check
 
   # A job to build postgres
   build-postgres:
-    executor: zenith-build-executor
+    executor: zenith-xlarge-executor
     parameters:
       build_type:
         type: enum
@@ -37,8 +37,7 @@ jobs:
         # Note this works even though the submodule hasn't been checkout out yet.
       - run:
           name: Get postgres cache key
-          command: |
-            git rev-parse HEAD:vendor/postgres > /tmp/cache-key-postgres
+          command: git rev-parse HEAD:vendor/postgres > /tmp/cache-key-postgres
 
       - restore_cache:
           name: Restore postgres cache
@@ -46,15 +45,6 @@ jobs:
             # Restore ONLY if the rev key matches exactly
             - v04-postgres-cache-<< parameters.build_type >>-{{ checksum "/tmp/cache-key-postgres" }}
 
-        # FIXME We could cache our own docker container, instead of installing packages every time.
-      - run:
-          name: apt install dependencies
-          command: |
-            if [ ! -e tmp_install/bin/postgres ]; then
-              sudo apt update
-              sudo apt install build-essential libreadline-dev zlib1g-dev flex bison libseccomp-dev
-            fi
-
         # Build postgres if the restore_cache didn't find a build.
         # `make` can't figure out whether the cache is valid, since
         # it only compares file timestamps.
@@ -64,7 +54,8 @@ jobs:
             if [ ! -e tmp_install/bin/postgres ]; then
               # "depth 1" saves some time by not cloning the whole repo
               git submodule update --init --depth 1
-              make postgres -j8
+              # bail out on any warnings
+              COPT='-Werror' mold -run make postgres -j$(nproc)
             fi
 
       - save_cache:
@@ -75,7 +66,7 @@ jobs:
 
   # A job to build zenith rust code
   build-zenith:
-    executor: zenith-build-executor
+    executor: zenith-xlarge-executor
     parameters:
       build_type:
         type: enum
@@ -83,12 +74,6 @@ jobs:
     environment:
       BUILD_TYPE: << parameters.build_type >>
     steps:
-      - run:
-          name: apt install dependencies
-          command: |
-            sudo apt update
-            sudo apt install libssl-dev clang
-
         # Checkout the git repo (without submodules)
       - checkout
 
@@ -126,7 +111,7 @@ jobs:
             fi
 
             export CARGO_INCREMENTAL=0
-            "${cov_prefix[@]}" cargo build $CARGO_FLAGS --bins --tests
+            "${cov_prefix[@]}" mold -run cargo build $CARGO_FLAGS --bins --tests
 
       - save_cache:
           name: Save rust cache
@@ -210,6 +195,14 @@ jobs:
           command: |
             cp -a tmp_install /tmp/zenith/pg_install
 
+      - run:
+          name: Merge coverage data
+          command: |
+            # This will speed up workspace uploads
+            if [[ $BUILD_TYPE == "debug" ]]; then
+              scripts/coverage "--profraw-prefix=$CIRCLE_JOB" --dir=/tmp/zenith/coverage merge
+            fi
+
         # Save the rust binaries and coverage data for other jobs in this workflow.
       - persist_to_workspace:
           root: /tmp/zenith
@@ -217,23 +210,30 @@ jobs:
             - "*"
 
   check-codestyle-python:
-    executor: zenith-python-executor
+    executor: zenith-executor
     steps:
       - checkout
+      - restore_cache:
+          keys:
+            - v1-python-deps-{{ checksum "poetry.lock" }}
       - run:
           name: Install deps
-          command: pipenv --python 3.7 install --dev
+          command: ./scripts/pysync
+      - save_cache:
+          key: v1-python-deps-{{ checksum "poetry.lock" }}
+          paths:
+            - /home/circleci/.cache/pypoetry/virtualenvs
       - run:
           name: Run yapf to ensure code format
           when: always
-          command: pipenv run yapf --recursive --diff .
+          command: poetry run yapf --recursive --diff .
       - run:
           name: Run mypy to check types
           when: always
-          command: pipenv run mypy .
+          command: poetry run mypy .
 
   run-pytest:
-    executor: zenith-python-executor
+    executor: zenith-executor
     parameters:
       # pytest args to specify the tests to run.
       #
@@ -272,9 +272,16 @@ jobs:
           condition: << parameters.needs_postgres_source >>
           steps:
             - run: git submodule update --init --depth 1
+      - restore_cache:
+          keys:
+            - v1-python-deps-{{ checksum "poetry.lock" }}
       - run:
           name: Install deps
-          command: pipenv --python 3.7 install
+          command: ./scripts/pysync
+      - save_cache:
+          key: v1-python-deps-{{ checksum "poetry.lock" }}
+          paths:
+            - /home/circleci/.cache/pypoetry/virtualenvs
       - run:
           name: Run pytest
           # pytest doesn't output test logs in real time, so CI job may fail with
@@ -291,6 +298,7 @@ jobs:
             - PLATFORM: zenith-local-ci
           command: |
             PERF_REPORT_DIR="$(realpath test_runner/perf-report-local)"
+            rm -rf $PERF_REPORT_DIR
 
             TEST_SELECTION="test_runner/<< parameters.test_selection >>"
             EXTRA_PARAMS="<< parameters.extra_params >>"
@@ -326,7 +334,7 @@ jobs:
             # -n4 uses four processes to run tests via pytest-xdist
             # -s is not used to prevent pytest from capturing output, because tests are running
             # in parallel and logs are mixed between different tests
-            "${cov_prefix[@]}" pipenv run pytest \
+            "${cov_prefix[@]}" ./scripts/pytest \
               --junitxml=$TEST_OUTPUT/junit.xml \
               --tb=short \
               --verbose \
@@ -335,7 +343,6 @@ jobs:
 
             if << parameters.save_perf_report >>; then
               if [[ $CIRCLE_BRANCH == "main" ]]; then
-                # TODO: reuse scripts/git-upload
                 export REPORT_FROM="$PERF_REPORT_DIR"
                 export REPORT_TO=local
                 scripts/generate_and_push_perf_report.sh
@@ -356,6 +363,13 @@ jobs:
       # The store_test_results step tells CircleCI where to find the junit.xml file.
       - store_test_results:
           path: /tmp/test_output
+      - run:
+          name: Merge coverage data
+          command: |
+            # This will speed up workspace uploads
+            if [[ $BUILD_TYPE == "debug" ]]; then
+              scripts/coverage "--profraw-prefix=$CIRCLE_JOB" --dir=/tmp/zenith/coverage merge
+            fi
       # Save coverage data (if any)
       - persist_to_workspace:
           root: /tmp/zenith
@@ -363,7 +377,7 @@ jobs:
             - "*"
 
   coverage-report:
-    executor: zenith-build-executor
+    executor: zenith-xlarge-executor
     steps:
       - attach_workspace:
           at: /tmp/zenith
@@ -375,12 +389,6 @@ jobs:
             # there's no way to clean out old packages, so the cache grows every time something
             # changes.
             - v04-rust-cache-deps-debug-{{ checksum "Cargo.lock" }}
-      - run:
-          name: Install llvm-tools
-          command: |
-            # TODO: install a proper symbol demangler, e.g. rustfilt
-            # TODO: we should embed this into a docker image
-            rustup component add llvm-tools-preview
       - run:
           name: Build coverage report
           command: |
@@ -450,7 +458,7 @@ jobs:
           name: Build and push compute-tools Docker image
           command: |
             echo $DOCKER_PWD | docker login -u $DOCKER_LOGIN --password-stdin
-            docker build -t zenithdb/compute-tools:latest ./compute_tools/
+            docker build -t zenithdb/compute-tools:latest -f Dockerfile.compute-tools .
             docker push zenithdb/compute-tools:latest
       - run:
           name: Init postgres submodule
@@ -571,55 +579,6 @@ jobs:
                 }
               }"
 
-  #
-  #
-  # compute-tools jobs
-  # TODO: unify with main build_and_test pipeline
-  #
-  #
-  compute-tools-test:
-    executor: zenith-build-executor
-    working_directory: ~/repo/compute_tools
-    steps:
-      - checkout:
-          path: ~/repo
-
-      - restore_cache:
-          name: Restore rust cache
-          keys:
-            # 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-debug-{{ checksum "Cargo.lock" }}
-
-      # Build the rust code, including test binaries
-      - run:
-          name: Rust build
-          environment:
-            CARGO_INCREMENTAL: 0
-          command: cargo build --bins --tests
-
-      - save_cache:
-          name: Save rust cache
-          key: v03-rust-cache-deps-debug-{{ checksum "Cargo.lock" }}
-          paths:
-            - ~/.cargo/registry
-            - ~/.cargo/git
-            - target
-
-      # Run Rust formatting checks
-      - run:
-          name: cargo fmt check
-          command: cargo fmt --all -- --check
-
-      # Run Rust linter (clippy)
-      - run:
-          name: cargo clippy check
-          command: cargo clippy --all --all-targets -- -Dwarnings -Drust-2018-idioms
-
-      # Run Rust integration and unittests
-      - run: cargo test
-
 workflows:
   build_and_test:
     jobs:
@@ -639,6 +598,7 @@ workflows:
             - build-postgres-<< matrix.build_type >>
       - run-pytest:
           name: pg_regress-tests-<< matrix.build_type >>
+          context: PERF_TEST_RESULT_CONNSTR
           matrix:
             parameters:
               build_type: ["debug", "release"]
@@ -656,6 +616,7 @@ workflows:
             - build-zenith-<< matrix.build_type >>
       - run-pytest:
           name: benchmarks
+          context: PERF_TEST_RESULT_CONNSTR
           build_type: release
           test_selection: performance
           run_in_parallel: false
@@ -668,7 +629,6 @@ workflows:
           requires:
             # TODO: consider adding more
             - other-tests-debug
-      - compute-tools-test
       - docker-image:
           # Context gives an ability to login
           context: Docker Hub
@@ -691,7 +651,6 @@ workflows:
           requires:
             - pg_regress-tests-release
             - other-tests-release
-            - compute-tools-test
       - deploy-staging:
           # Context gives an ability to login
           context: Docker Hub

.circleci/proxy.staging.yaml

@@ -5,9 +5,23 @@ settings:
   authEndpoint: "https://console.stage.zenith.tech/authenticate_proxy_request/"
   uri: "https://console.stage.zenith.tech/psql_session/"
 
+# -- Additional labels for zenith-proxy pods
+podLabels:
+  zenith_service: proxy
+  zenith_env: staging
+  zenith_region: us-east-1
+  zenith_region_slug: virginia
+
 exposedService:
   annotations:
     service.beta.kubernetes.io/aws-load-balancer-type: external
     service.beta.kubernetes.io/aws-load-balancer-nlb-target-type: ip
     service.beta.kubernetes.io/aws-load-balancer-scheme: internet-facing
     external-dns.alpha.kubernetes.io/hostname: start.stage.zenith.tech
+
+metrics:
+  enabled: true
+  serviceMonitor:
+    enabled: true
+    selector:
+      release: kube-prometheus-stack

.github/workflows/benchmarking.yml

@@ -3,7 +3,7 @@ name: benchmarking
 on:
   # uncomment to run on push for debugging your PR
   # push:
-  #   branches: [ mybranch ]
+  #   branches: [ your branch ]
   schedule:
     # * is a special character in YAML so you have to quote this string
     #          ┌───────────── minute (0 - 59)
@@ -36,20 +36,20 @@ jobs:
     # see https://github.com/actions/setup-python/issues/162
     # and probably https://github.com/actions/setup-python/issues/162#issuecomment-865387976 in particular
     # so the simplest solution to me is to use already installed system python and spin virtualenvs for job runs.
-    # there is Python 3.7.10 already installed on the machine so use it to install pipenv and then use pipenv's virtuealenvs
-    - name: Install pipenv & deps
+    # there is Python 3.7.10 already installed on the machine so use it to install poetry and then use poetry's virtuealenvs
+    - name: Install poetry & deps
       run: |
-        python3 -m pip install --upgrade pipenv wheel
-        # since pip/pipenv caches are reused there shouldn't be any troubles with install every time
-        pipenv install
+        python3 -m pip install --upgrade poetry wheel
+        # since pip/poetry caches are reused there shouldn't be any troubles with install every time
+        ./scripts/pysync
 
     - name: Show versions
       run: |
         echo Python
         python3 --version
-        pipenv run python3 --version
+        poetry run python3 --version
         echo Pipenv
-        pipenv --version
+        poetry --version
         echo Pgbench
         $PG_BIN/pgbench --version
 
@@ -89,11 +89,15 @@ jobs:
         BENCHMARK_CONNSTR: "${{ secrets.BENCHMARK_STAGING_CONNSTR }}"
         REMOTE_ENV: "1" # indicate to test harness that we do not have zenith binaries locally
       run: |
+        # just to be sure that no data was cached on self hosted runner
+        # since it might generate duplicates when calling ingest_perf_test_result.py
+        rm -rf perf-report-staging
         mkdir -p perf-report-staging
-        pipenv run pytest test_runner/performance/ -v -m "remote_cluster" --skip-interfering-proc-check --out-dir perf-report-staging
+        ./scripts/pytest test_runner/performance/ -v -m "remote_cluster" --skip-interfering-proc-check --out-dir perf-report-staging
 
     - name: Submit result
       env:
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
+        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
       run: |
         REPORT_FROM=$(realpath perf-report-staging) REPORT_TO=staging scripts/generate_and_push_perf_report.sh

Cargo.toml

@@ -1,5 +1,6 @@
 [workspace]
 members = [
+    "compute_tools",
     "control_plane",
     "pageserver",
     "postgres_ffi",
@@ -15,3 +16,8 @@ members = [
 # This is useful for profiling and, to some extent, debug.
 # Besides, debug info should not affect the performance.
 debug = true
+
+# This is only needed for proxy's tests
+# TODO: we should probably fork tokio-postgres-rustls instead
+[patch.crates-io]
+tokio-postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="2949d98df52587d562986aad155dd4e889e408b7" }

Dockerfile.build

@@ -2,8 +2,9 @@
 # Image with all the required dependencies to build https://github.com/zenithdb/zenith
 # and Postgres from https://github.com/zenithdb/postgres
 # Also includes some rust development and build tools.
+# NB: keep in sync with rust image version in .circle/config.yml
 #
-FROM rust:slim-buster
+FROM rust:1.56.1-slim-buster
 WORKDIR /zenith
 
 # Install postgres and zenith build dependencies

Dockerfile.compute-tools

@@ -0,0 +1,14 @@
+# First transient image to build compute_tools binaries
+# NB: keep in sync with rust image version in .circle/config.yml
+FROM rust:1.56.1-slim-buster AS rust-build
+
+WORKDIR /zenith
+
+COPY . .
+
+RUN cargo build -p compute_tools --release
+
+# Final image that only has one binary
+FROM debian:buster-slim
+
+COPY --from=rust-build /zenith/target/release/zenith_ctl /usr/local/bin/zenith_ctl

Pipfile

@@ -1,30 +0,0 @@
-[[source]]
-url = "https://pypi.python.org/simple"
-verify_ssl = true
-name = "pypi"
-
-[packages]
-pytest = ">=6.0.0"
-typing-extensions = "*"
-pyjwt = {extras = ["crypto"], version = "*"}
-requests = "*"
-pytest-xdist = "*"
-asyncpg = "*"
-cached-property = "*"
-psycopg2-binary = "*"
-jinja2 = "*"
-
-[dev-packages]
-# 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 = "*"
-types-requests = "*"
-types-psycopg2 = "*"
-
-[requires]
-# we need at least 3.7, but pipenv doesn't allow to say this directly
-python_version = "3"

Pipfile.lock

@@ -1,652 +0,0 @@
-{
-    "_meta": {
-        "hash": {
-            "sha256": "c309cb963a7b07ae3d30e9cbf08b495f77bdecc0e5356fc89d133c4fbcb65b2b"
-        },
-        "pipfile-spec": 6,
-        "requires": {
-            "python_version": "3"
-        },
-        "sources": [
-            {
-                "name": "pypi",
-                "url": "https://pypi.python.org/simple",
-                "verify_ssl": true
-            }
-        ]
-    },
-    "default": {
-        "asyncpg": {
-            "hashes": [
-                "sha256:129d501f3d30616afd51eb8d3142ef51ba05374256bd5834cec3ef4956a9b317",
-                "sha256:29ef6ae0a617fc13cc2ac5dc8e9b367bb83cba220614b437af9b67766f4b6b20",
-                "sha256:41704c561d354bef01353835a7846e5606faabbeb846214dfcf666cf53319f18",
-                "sha256:556b0e92e2b75dc028b3c4bc9bd5162ddf0053b856437cf1f04c97f9c6837d03",
-                "sha256:8ff5073d4b654e34bd5eaadc01dc4d68b8a9609084d835acd364cd934190a08d",
-                "sha256:a458fc69051fbb67d995fdda46d75a012b5d6200f91e17d23d4751482640ed4c",
-                "sha256:a7095890c96ba36f9f668eb552bb020dddb44f8e73e932f8573efc613ee83843",
-                "sha256:a738f4807c853623d3f93f0fea11f61be6b0e5ca16ea8aeb42c2c7ee742aa853",
-                "sha256:c4fc0205fe4ddd5aeb3dfdc0f7bafd43411181e1f5650189608e5971cceacff1",
-                "sha256:dd2fa063c3344823487d9ddccb40802f02622ddf8bf8a6cc53885ee7a2c1c0c6",
-                "sha256:ddffcb85227bf39cd1bedd4603e0082b243cf3b14ced64dce506a15b05232b83",
-                "sha256:e36c6806883786b19551bb70a4882561f31135dc8105a59662e0376cf5b2cbc5",
-                "sha256:eed43abc6ccf1dc02e0d0efc06ce46a411362f3358847c6b0ec9a43426f91ece"
-            ],
-            "index": "pypi",
-            "version": "==0.24.0"
-        },
-        "attrs": {
-            "hashes": [
-                "sha256:149e90d6d8ac20db7a955ad60cf0e6881a3f20d37096140088356da6c716b0b1",
-                "sha256:ef6aaac3ca6cd92904cdd0d83f629a15f18053ec84e6432106f7a4d04ae4f5fb"
-            ],
-            "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": [
-                "sha256:78884e7c1d4b00ce3cea67b44566851c4343c120abd683433ce934a68ea58872",
-                "sha256:d62a0163eb4c2344ac042ab2bdf75399a71a2d8c7d47eac2e2ee91b9d6339569"
-            ],
-            "version": "==2021.10.8"
-        },
-        "cffi": {
-            "hashes": [
-                "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": "==1.15.0"
-        },
-        "charset-normalizer": {
-            "hashes": [
-                "sha256:e019de665e2bcf9c2b64e2e5aa025fa991da8720daa3c1138cadd2fd1856aed0",
-                "sha256:f7af805c321bfa1ce6714c51f254e0d5bb5e5834039bc17db7ebe3a4cec9492b"
-            ],
-            "markers": "python_version >= '3'",
-            "version": "==2.0.7"
-        },
-        "cryptography": {
-            "hashes": [
-                "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": "==35.0.0"
-        },
-        "execnet": {
-            "hashes": [
-                "sha256:8f694f3ba9cc92cab508b152dcfe322153975c29bda272e2fd7f3f00f36e47c5",
-                "sha256:a295f7cc774947aac58dde7fdc85f4aa00c42adf5d8f5468fc630c1acf30a142"
-            ],
-            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3, 3.4'",
-            "version": "==1.9.0"
-        },
-        "idna": {
-            "hashes": [
-                "sha256:84d9dd047ffa80596e0f246e2eab0b391788b0503584e8945f2368256d2735ff",
-                "sha256:9d643ff0a55b762d5cdb124b8eaa99c66322e2157b69160bc32796e824360e6d"
-            ],
-            "markers": "python_version >= '3'",
-            "version": "==3.3"
-        },
-        "importlib-metadata": {
-            "hashes": [
-                "sha256:b618b6d2d5ffa2f16add5697cf57a46c76a56229b0ed1c438322e4e95645bd15",
-                "sha256:f284b3e11256ad1e5d03ab86bb2ccd6f5339688ff17a4d797a0fe7df326f23b1"
-            ],
-            "markers": "python_version < '3.8'",
-            "version": "==4.8.1"
-        },
-        "iniconfig": {
-            "hashes": [
-                "sha256:011e24c64b7f47f6ebd835bb12a743f2fbe9a26d4cecaa7f53bc4f35ee9da8b3",
-                "sha256:bc3af051d7d14b2ee5ef9969666def0cd1a000e121eaea580d4a313df4b37f32"
-            ],
-            "version": "==1.1.1"
-        },
-        "jinja2": {
-            "hashes": [
-                "sha256:827a0e32839ab1600d4eb1c4c33ec5a8edfbc5cb42dafa13b81f182f97784b45",
-                "sha256:8569982d3f0889eed11dd620c706d39b60c36d6d25843961f33f77fb6bc6b20c"
-            ],
-            "index": "pypi",
-            "version": "==3.0.2"
-        },
-        "markupsafe": {
-            "hashes": [
-                "sha256:01a9b8ea66f1658938f65b93a85ebe8bc016e6769611be228d797c9d998dd298",
-                "sha256:023cb26ec21ece8dc3907c0e8320058b2e0cb3c55cf9564da612bc325bed5e64",
-                "sha256:0446679737af14f45767963a1a9ef7620189912317d095f2d9ffa183a4d25d2b",
-                "sha256:04635854b943835a6ea959e948d19dcd311762c5c0c6e1f0e16ee57022669194",
-                "sha256:0717a7390a68be14b8c793ba258e075c6f4ca819f15edfc2a3a027c823718567",
-                "sha256:0955295dd5eec6cb6cc2fe1698f4c6d84af2e92de33fbcac4111913cd100a6ff",
-                "sha256:0d4b31cc67ab36e3392bbf3862cfbadac3db12bdd8b02a2731f509ed5b829724",
-                "sha256:10f82115e21dc0dfec9ab5c0223652f7197feb168c940f3ef61563fc2d6beb74",
-                "sha256:168cd0a3642de83558a5153c8bd34f175a9a6e7f6dc6384b9655d2697312a646",
-                "sha256:1d609f577dc6e1aa17d746f8bd3c31aa4d258f4070d61b2aa5c4166c1539de35",
-                "sha256:1f2ade76b9903f39aa442b4aadd2177decb66525062db244b35d71d0ee8599b6",
-                "sha256:20dca64a3ef2d6e4d5d615a3fd418ad3bde77a47ec8a23d984a12b5b4c74491a",
-                "sha256:2a7d351cbd8cfeb19ca00de495e224dea7e7d919659c2841bbb7f420ad03e2d6",
-                "sha256:2d7d807855b419fc2ed3e631034685db6079889a1f01d5d9dac950f764da3dad",
-                "sha256:2ef54abee730b502252bcdf31b10dacb0a416229b72c18b19e24a4509f273d26",
-                "sha256:36bc903cbb393720fad60fc28c10de6acf10dc6cc883f3e24ee4012371399a38",
-                "sha256:37205cac2a79194e3750b0af2a5720d95f786a55ce7df90c3af697bfa100eaac",
-                "sha256:3c112550557578c26af18a1ccc9e090bfe03832ae994343cfdacd287db6a6ae7",
-                "sha256:3dd007d54ee88b46be476e293f48c85048603f5f516008bee124ddd891398ed6",
-                "sha256:4296f2b1ce8c86a6aea78613c34bb1a672ea0e3de9c6ba08a960efe0b0a09047",
-                "sha256:47ab1e7b91c098ab893b828deafa1203de86d0bc6ab587b160f78fe6c4011f75",
-                "sha256:49e3ceeabbfb9d66c3aef5af3a60cc43b85c33df25ce03d0031a608b0a8b2e3f",
-                "sha256:4dc8f9fb58f7364b63fd9f85013b780ef83c11857ae79f2feda41e270468dd9b",
-                "sha256:4efca8f86c54b22348a5467704e3fec767b2db12fc39c6d963168ab1d3fc9135",
-                "sha256:53edb4da6925ad13c07b6d26c2a852bd81e364f95301c66e930ab2aef5b5ddd8",
-                "sha256:5855f8438a7d1d458206a2466bf82b0f104a3724bf96a1c781ab731e4201731a",
-                "sha256:594c67807fb16238b30c44bdf74f36c02cdf22d1c8cda91ef8a0ed8dabf5620a",
-                "sha256:5b6d930f030f8ed98e3e6c98ffa0652bdb82601e7a016ec2ab5d7ff23baa78d1",
-                "sha256:5bb28c636d87e840583ee3adeb78172efc47c8b26127267f54a9c0ec251d41a9",
-                "sha256:60bf42e36abfaf9aff1f50f52644b336d4f0a3fd6d8a60ca0d054ac9f713a864",
-                "sha256:611d1ad9a4288cf3e3c16014564df047fe08410e628f89805e475368bd304914",
-                "sha256:6300b8454aa6930a24b9618fbb54b5a68135092bc666f7b06901f897fa5c2fee",
-                "sha256:63f3268ba69ace99cab4e3e3b5840b03340efed0948ab8f78d2fd87ee5442a4f",
-                "sha256:6557b31b5e2c9ddf0de32a691f2312a32f77cd7681d8af66c2692efdbef84c18",
-                "sha256:693ce3f9e70a6cf7d2fb9e6c9d8b204b6b39897a2c4a1aa65728d5ac97dcc1d8",
-                "sha256:6a7fae0dd14cf60ad5ff42baa2e95727c3d81ded453457771d02b7d2b3f9c0c2",
-                "sha256:6c4ca60fa24e85fe25b912b01e62cb969d69a23a5d5867682dd3e80b5b02581d",
-                "sha256:6fcf051089389abe060c9cd7caa212c707e58153afa2c649f00346ce6d260f1b",
-                "sha256:7d91275b0245b1da4d4cfa07e0faedd5b0812efc15b702576d103293e252af1b",
-                "sha256:89c687013cb1cd489a0f0ac24febe8c7a666e6e221b783e53ac50ebf68e45d86",
-                "sha256:8d206346619592c6200148b01a2142798c989edcb9c896f9ac9722a99d4e77e6",
-                "sha256:905fec760bd2fa1388bb5b489ee8ee5f7291d692638ea5f67982d968366bef9f",
-                "sha256:97383d78eb34da7e1fa37dd273c20ad4320929af65d156e35a5e2d89566d9dfb",
-                "sha256:984d76483eb32f1bcb536dc27e4ad56bba4baa70be32fa87152832cdd9db0833",
-                "sha256:99df47edb6bda1249d3e80fdabb1dab8c08ef3975f69aed437cb69d0a5de1e28",
-                "sha256:9f02365d4e99430a12647f09b6cc8bab61a6564363f313126f775eb4f6ef798e",
-                "sha256:a30e67a65b53ea0a5e62fe23682cfe22712e01f453b95233b25502f7c61cb415",
-                "sha256:ab3ef638ace319fa26553db0624c4699e31a28bb2a835c5faca8f8acf6a5a902",
-                "sha256:aca6377c0cb8a8253e493c6b451565ac77e98c2951c45f913e0b52facdcff83f",
-                "sha256:add36cb2dbb8b736611303cd3bfcee00afd96471b09cda130da3581cbdc56a6d",
-                "sha256:b2f4bf27480f5e5e8ce285a8c8fd176c0b03e93dcc6646477d4630e83440c6a9",
-                "sha256:b7f2d075102dc8c794cbde1947378051c4e5180d52d276987b8d28a3bd58c17d",
-                "sha256:baa1a4e8f868845af802979fcdbf0bb11f94f1cb7ced4c4b8a351bb60d108145",
-                "sha256:be98f628055368795d818ebf93da628541e10b75b41c559fdf36d104c5787066",
-                "sha256:bf5d821ffabf0ef3533c39c518f3357b171a1651c1ff6827325e4489b0e46c3c",
-                "sha256:c47adbc92fc1bb2b3274c4b3a43ae0e4573d9fbff4f54cd484555edbf030baf1",
-                "sha256:cdfba22ea2f0029c9261a4bd07e830a8da012291fbe44dc794e488b6c9bb353a",
-                "sha256:d6c7ebd4e944c85e2c3421e612a7057a2f48d478d79e61800d81468a8d842207",
-                "sha256:d7f9850398e85aba693bb640262d3611788b1f29a79f0c93c565694658f4071f",
-                "sha256:d8446c54dc28c01e5a2dbac5a25f071f6653e6e40f3a8818e8b45d790fe6ef53",
-                "sha256:deb993cacb280823246a026e3b2d81c493c53de6acfd5e6bfe31ab3402bb37dd",
-                "sha256:e0f138900af21926a02425cf736db95be9f4af72ba1bb21453432a07f6082134",
-                "sha256:e9936f0b261d4df76ad22f8fee3ae83b60d7c3e871292cd42f40b81b70afae85",
-                "sha256:f0567c4dc99f264f49fe27da5f735f414c4e7e7dd850cfd8e69f0862d7c74ea9",
-                "sha256:f5653a225f31e113b152e56f154ccbe59eeb1c7487b39b9d9f9cdb58e6c79dc5",
-                "sha256:f826e31d18b516f653fe296d967d700fddad5901ae07c622bb3705955e1faa94",
-                "sha256:f8ba0e8349a38d3001fae7eadded3f6606f0da5d748ee53cc1dab1d6527b9509",
-                "sha256:f9081981fe268bd86831e5c75f7de206ef275defcb82bc70740ae6dc507aee51",
-                "sha256:fa130dd50c57d53368c9d59395cb5526eda596d3ffe36666cd81a44d56e48872"
-            ],
-            "markers": "python_version >= '3.6'",
-            "version": "==2.0.1"
-        },
-        "packaging": {
-            "hashes": [
-                "sha256:096d689d78ca690e4cd8a89568ba06d07ca097e3306a4381635073ca91479966",
-                "sha256:14317396d1e8cdb122989b916fa2c7e9ca8e2be9e8060a6eff75b6b7b4d8a7e0"
-            ],
-            "markers": "python_version >= '3.6'",
-            "version": "==21.2"
-        },
-        "pluggy": {
-            "hashes": [
-                "sha256:4224373bacce55f955a878bf9cfa763c1e360858e330072059e10bad68531159",
-                "sha256:74134bbf457f031a36d68416e1509f34bd5ccc019f0bcc952c7b909d06b37bd3"
-            ],
-            "markers": "python_version >= '3.6'",
-            "version": "==1.0.0"
-        },
-        "psycopg2-binary": {
-            "hashes": [
-                "sha256:0b7dae87f0b729922e06f85f667de7bf16455d411971b2043bbd9577af9d1975",
-                "sha256:0f2e04bd2a2ab54fa44ee67fe2d002bb90cee1c0f1cc0ebc3148af7b02034cbd",
-                "sha256:123c3fb684e9abfc47218d3784c7b4c47c8587951ea4dd5bc38b6636ac57f616",
-                "sha256:1473c0215b0613dd938db54a653f68251a45a78b05f6fc21af4326f40e8360a2",
-                "sha256:14db1752acdd2187d99cb2ca0a1a6dfe57fc65c3281e0f20e597aac8d2a5bd90",
-                "sha256:1e3a362790edc0a365385b1ac4cc0acc429a0c0d662d829a50b6ce743ae61b5a",
-                "sha256:1e85b74cbbb3056e3656f1cc4781294df03383127a8114cbc6531e8b8367bf1e",
-                "sha256:20f1ab44d8c352074e2d7ca67dc00843067788791be373e67a0911998787ce7d",
-                "sha256:24b0b6688b9f31a911f2361fe818492650795c9e5d3a1bc647acbd7440142a4f",
-                "sha256:2f62c207d1740b0bde5c4e949f857b044818f734a3d57f1d0d0edc65050532ed",
-                "sha256:3242b9619de955ab44581a03a64bdd7d5e470cc4183e8fcadd85ab9d3756ce7a",
-                "sha256:35c4310f8febe41f442d3c65066ca93cccefd75013df3d8c736c5b93ec288140",
-                "sha256:4235f9d5ddcab0b8dbd723dca56ea2922b485ea00e1dafacf33b0c7e840b3d32",
-                "sha256:542875f62bc56e91c6eac05a0deadeae20e1730be4c6334d8f04c944fcd99759",
-                "sha256:5ced67f1e34e1a450cdb48eb53ca73b60aa0af21c46b9b35ac3e581cf9f00e31",
-                "sha256:661509f51531ec125e52357a489ea3806640d0ca37d9dada461ffc69ee1e7b6e",
-                "sha256:7360647ea04db2e7dff1648d1da825c8cf68dc5fbd80b8fb5b3ee9f068dcd21a",
-                "sha256:736b8797b58febabb85494142c627bd182b50d2a7ec65322983e71065ad3034c",
-                "sha256:8c13d72ed6af7fd2c8acbd95661cf9477f94e381fce0792c04981a8283b52917",
-                "sha256:988b47ac70d204aed01589ed342303da7c4d84b56c2f4c4b8b00deda123372bf",
-                "sha256:995fc41ebda5a7a663a254a1dcac52638c3e847f48307b5416ee373da15075d7",
-                "sha256:a36c7eb6152ba5467fb264d73844877be8b0847874d4822b7cf2d3c0cb8cdcb0",
-                "sha256:aed4a9a7e3221b3e252c39d0bf794c438dc5453bc2963e8befe9d4cd324dff72",
-                "sha256:aef9aee84ec78af51107181d02fe8773b100b01c5dfde351184ad9223eab3698",
-                "sha256:b0221ca5a9837e040ebf61f48899926b5783668b7807419e4adae8175a31f773",
-                "sha256:b4d7679a08fea64573c969f6994a2631908bb2c0e69a7235648642f3d2e39a68",
-                "sha256:c250a7ec489b652c892e4f0a5d122cc14c3780f9f643e1a326754aedf82d9a76",
-                "sha256:ca86db5b561b894f9e5f115d6a159fff2a2570a652e07889d8a383b5fae66eb4",
-                "sha256:cfc523edecddaef56f6740d7de1ce24a2fdf94fd5e704091856a201872e37f9f",
-                "sha256:d92272c7c16e105788efe2cfa5d680f07e34e0c29b03c1908f8636f55d5f915a",
-                "sha256:da113b70f6ec40e7d81b43d1b139b9db6a05727ab8be1ee559f3a69854a69d34",
-                "sha256:f6fac64a38f6768e7bc7b035b9e10d8a538a9fadce06b983fb3e6fa55ac5f5ce",
-                "sha256:f8559617b1fcf59a9aedba2c9838b5b6aa211ffedecabca412b92a1ff75aac1a",
-                "sha256:fbb42a541b1093385a2d8c7eec94d26d30437d0e77c1d25dae1dcc46741a385e"
-            ],
-            "index": "pypi",
-            "version": "==2.9.1"
-        },
-        "py": {
-            "hashes": [
-                "sha256:21b81bda15b66ef5e1a777a21c4dcd9c20ad3efd0b3f817e7a809035269e1bd3",
-                "sha256:3b80836aa6d1feeaa108e046da6423ab8f6ceda6468545ae8d02d9d58d18818a"
-            ],
-            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3'",
-            "version": "==1.10.0"
-        },
-        "pycparser": {
-            "hashes": [
-                "sha256:2d475327684562c3a96cc71adf7dc8c4f0565175cf86b6d7a404ff4c771f15f0",
-                "sha256:7582ad22678f0fcd81102833f60ef8d0e57288b6b5fb00323d101be910e35705"
-            ],
-            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3'",
-            "version": "==2.20"
-        },
-        "pyjwt": {
-            "extras": [
-                "crypto"
-            ],
-            "hashes": [
-                "sha256:b888b4d56f06f6dcd777210c334e69c737be74755d3e5e9ee3fe67dc18a0ee41",
-                "sha256:e0c4bb8d9f0af0c7f5b1ec4c5036309617d03d56932877f2f7a0beeb5318322f"
-            ],
-            "index": "pypi",
-            "version": "==2.3.0"
-        },
-        "pyparsing": {
-            "hashes": [
-                "sha256:c203ec8783bf771a155b207279b9bccb8dea02d8f0c9e5f8ead507bc3246ecc1",
-                "sha256:ef9d7589ef3c200abe66653d3f1ab1033c3c419ae9b9bdb1240a85b024efc88b"
-            ],
-            "markers": "python_version >= '2.6' and python_version not in '3.0, 3.1, 3.2, 3.3'",
-            "version": "==2.4.7"
-        },
-        "pytest": {
-            "hashes": [
-                "sha256:131b36680866a76e6781d13f101efb86cf674ebb9762eb70d3082b6f29889e89",
-                "sha256:7310f8d27bc79ced999e760ca304d69f6ba6c6649c0b60fb0e04a4a77cacc134"
-            ],
-            "index": "pypi",
-            "version": "==6.2.5"
-        },
-        "pytest-forked": {
-            "hashes": [
-                "sha256:6aa9ac7e00ad1a539c41bec6d21011332de671e938c7637378ec9710204e37ca",
-                "sha256:dc4147784048e70ef5d437951728825a131b81714b398d5d52f17c7c144d8815"
-            ],
-            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3, 3.4'",
-            "version": "==1.3.0"
-        },
-        "pytest-xdist": {
-            "hashes": [
-                "sha256:7b61ebb46997a0820a263553179d6d1e25a8c50d8a8620cd1aa1e20e3be99168",
-                "sha256:89b330316f7fc475f999c81b577c2b926c9569f3d397ae432c0c2e2496d61ff9"
-            ],
-            "index": "pypi",
-            "version": "==2.4.0"
-        },
-        "requests": {
-            "hashes": [
-                "sha256:6c1246513ecd5ecd4528a0906f910e8f0f9c6b8ec72030dc9fd154dc1a6efd24",
-                "sha256:b8aa58f8cf793ffd8782d3d8cb19e66ef36f7aba4353eec859e74678b01b07a7"
-            ],
-            "index": "pypi",
-            "version": "==2.26.0"
-        },
-        "toml": {
-            "hashes": [
-                "sha256:806143ae5bfb6a3c6e736a764057db0e6a0e05e338b5630894a5f779cabb4f9b",
-                "sha256:b3bda1d108d5dd99f4a20d24d9c348e91c4db7ab1b749200bded2f839ccbe68f"
-            ],
-            "markers": "python_version >= '2.6' and python_version not in '3.0, 3.1, 3.2, 3.3'",
-            "version": "==0.10.2"
-        },
-        "typing-extensions": {
-            "hashes": [
-                "sha256:49f75d16ff11f1cd258e1b988ccff82a3ca5570217d7ad8c5f48205dd99a677e",
-                "sha256:d8226d10bc02a29bcc81df19a26e56a9647f8b0a6d4a83924139f4a8b01f17b7",
-                "sha256:f1d25edafde516b146ecd0613dabcc61409817af4766fbbcfb8d1ad4ec441a34"
-            ],
-            "index": "pypi",
-            "version": "==3.10.0.2"
-        },
-        "urllib3": {
-            "hashes": [
-                "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": "==1.26.7"
-        },
-        "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": [
-                "sha256:7afc856f74fa7006a289fd10fa840e1eebd8bbff6bffb69c26c54a0512ea8cf8",
-                "sha256:bb2a1c717df74c48a2d00ed625e5a66f8572a3a30baacb7657add1d7bac4097b"
-            ],
-            "markers": "python_version >= '3.6'",
-            "version": "==3.3.2"
-        },
-        "flake8": {
-            "hashes": [
-                "sha256:479b1304f72536a55948cb40a32dce8bb0ffe3501e26eaf292c7e60eb5e0428d",
-                "sha256:806e034dda44114815e23c16ef92f95c91e4c71100ff52813adf7132a6ad870d"
-            ],
-            "index": "pypi",
-            "version": "==4.0.1"
-        },
-        "importlib-metadata": {
-            "hashes": [
-                "sha256:b618b6d2d5ffa2f16add5697cf57a46c76a56229b0ed1c438322e4e95645bd15",
-                "sha256:f284b3e11256ad1e5d03ab86bb2ccd6f5339688ff17a4d797a0fe7df326f23b1"
-            ],
-            "markers": "python_version < '3.8'",
-            "version": "==4.8.1"
-        },
-        "mccabe": {
-            "hashes": [
-                "sha256:ab8a6258860da4b6677da4bd2fe5dc2c659cff31b3ee4f7f5d64e79735b80d42",
-                "sha256:dd8d182285a0fe56bace7f45b5e7d1a6ebcbf524e8f3bd87eb0f125271b8831f"
-            ],
-            "version": "==0.6.1"
-        },
-        "mypy": {
-            "hashes": [
-                "sha256:088cd9c7904b4ad80bec811053272986611b84221835e079be5bcad029e79dd9",
-                "sha256:0aadfb2d3935988ec3815952e44058a3100499f5be5b28c34ac9d79f002a4a9a",
-                "sha256:119bed3832d961f3a880787bf621634ba042cb8dc850a7429f643508eeac97b9",
-                "sha256:1a85e280d4d217150ce8cb1a6dddffd14e753a4e0c3cf90baabb32cefa41b59e",
-                "sha256:3c4b8ca36877fc75339253721f69603a9c7fdb5d4d5a95a1a1b899d8b86a4de2",
-                "sha256:3e382b29f8e0ccf19a2df2b29a167591245df90c0b5a2542249873b5c1d78212",
-                "sha256:42c266ced41b65ed40a282c575705325fa7991af370036d3f134518336636f5b",
-                "sha256:53fd2eb27a8ee2892614370896956af2ff61254c275aaee4c230ae771cadd885",
-                "sha256:704098302473cb31a218f1775a873b376b30b4c18229421e9e9dc8916fd16150",
-                "sha256:7df1ead20c81371ccd6091fa3e2878559b5c4d4caadaf1a484cf88d93ca06703",
-                "sha256:866c41f28cee548475f146aa4d39a51cf3b6a84246969f3759cb3e9c742fc072",
-                "sha256:a155d80ea6cee511a3694b108c4494a39f42de11ee4e61e72bc424c490e46457",
-                "sha256:adaeee09bfde366d2c13fe6093a7df5df83c9a2ba98638c7d76b010694db760e",
-                "sha256:b6fb13123aeef4a3abbcfd7e71773ff3ff1526a7d3dc538f3929a49b42be03f0",
-                "sha256:b94e4b785e304a04ea0828759172a15add27088520dc7e49ceade7834275bedb",
-                "sha256:c0df2d30ed496a08de5daed2a9ea807d07c21ae0ab23acf541ab88c24b26ab97",
-                "sha256:c6c2602dffb74867498f86e6129fd52a2770c48b7cd3ece77ada4fa38f94eba8",
-                "sha256:ceb6e0a6e27fb364fb3853389607cf7eb3a126ad335790fa1e14ed02fba50811",
-                "sha256:d9dd839eb0dc1bbe866a288ba3c1afc33a202015d2ad83b31e875b5905a079b6",
-                "sha256:e4dab234478e3bd3ce83bac4193b2ecd9cf94e720ddd95ce69840273bf44f6de",
-                "sha256:ec4e0cd079db280b6bdabdc807047ff3e199f334050db5cbb91ba3e959a67504",
-                "sha256:ecd2c3fe726758037234c93df7e98deb257fd15c24c9180dacf1ef829da5f921",
-                "sha256:ef565033fa5a958e62796867b1df10c40263ea9ded87164d67572834e57a174d"
-            ],
-            "index": "pypi",
-            "version": "==0.910"
-        },
-        "mypy-extensions": {
-            "hashes": [
-                "sha256:090fedd75945a69ae91ce1303b5824f428daf5a028d2f6ab8a299250a846f15d",
-                "sha256:2d82818f5bb3e369420cb3c4060a7970edba416647068eb4c5343488a6c604a8"
-            ],
-            "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": [
-                "sha256:720f8b39dde8b293825e7ff02c475f3077124006db4f440dcbc9a20b76548a20",
-                "sha256:eddd5847ef438ea1c7870ca7eb78a9d47ce0cdb4851a5523949f2601d0cbbe7f"
-            ],
-            "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": [
-                "sha256:05a85c2872edf37a4ed30b0cce2f6093e1d0581f8c19d7393122da7e25b2b24c",
-                "sha256:3bb3a3f256f4b7968c9c788781e4ff07dce46bdf12339dcda61053375426ee2e"
-            ],
-            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3'",
-            "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": [
-                "sha256:806143ae5bfb6a3c6e736a764057db0e6a0e05e338b5630894a5f779cabb4f9b",
-                "sha256:b3bda1d108d5dd99f4a20d24d9c348e91c4db7ab1b749200bded2f839ccbe68f"
-            ],
-            "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",
-                "sha256:d8226d10bc02a29bcc81df19a26e56a9647f8b0a6d4a83924139f4a8b01f17b7",
-                "sha256:f1d25edafde516b146ecd0613dabcc61409817af4766fbbcfb8d1ad4ec441a34"
-            ],
-            "index": "pypi",
-            "version": "==3.10.0.2"
-        },
-        "virtualenv": {
-            "hashes": [
-                "sha256:4b02e52a624336eece99c96e3ab7111f469c24ba226a53ec474e8e787b365814",
-                "sha256:576d05b46eace16a9c348085f7d0dc8ef28713a2cabaa1cf0aea41e8f12c9218"
-            ],
-            "markers": "python_version >= '2.7' and python_version not in '3.0, 3.1, 3.2, 3.3, 3.4'",
-            "version": "==20.10.0"
-        },
-        "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",
-                "sha256:e3a234ba8455fe201eaa649cdac872d590089a18b661e39bbac7020978dd9c2e"
-            ],
-            "index": "pypi",
-            "version": "==0.31.0"
-        },
-        "zipp": {
-            "hashes": [
-                "sha256:71c644c5369f4a6e07636f0aa966270449561fcea2e3d6747b8d23efaa9d7832",
-                "sha256:9fe5ea21568a0a70e50f273397638d39b03353731e6cbbb3fd8502a33fec40bc"
-            ],
-            "markers": "python_version >= '3.6'",
-            "version": "==3.6.0"
-        }
-    }
-}

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] 1.55 or later is also required.
+[Rust] 1.56.1 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 or Python scripts (not required to use the code), install
-Python (3.7 or higher), and install python3 packages using `pipenv install` in the project directory.
+Python (3.7 or higher), and install python3 packages using `./scripts/pysync` (requires poetry) in the project directory.
 
 2. Build zenith and patched postgres
 ```sh
@@ -128,8 +128,7 @@ INSERT 0 1
 ```sh
 git clone --recursive https://github.com/zenithdb/zenith.git
 make # builds also postgres and installs it to ./tmp_install
-cd test_runner
-pipenv run pytest
+./scripts/pytest
 ```
 
 ## Documentation

compute_tools/Cargo.toml

@@ -1,28 +1,19 @@
 [package]
 name = "compute_tools"
 version = "0.1.0"
-authors = ["Alexey Kondratov <kondratov.aleksey@gmail.com>"]
-edition = "2018"
-
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
-
-[workspace]
-# TODO: make it a part of global zenith worksapce
+edition = "2021"
 
 [dependencies]
 libc = "0.2"
 anyhow = "1.0"
 chrono = "0.4"
-clap = "2.33"
-env_logger = "0.8"
+clap = "3.0"
+env_logger = "0.9"
 hyper = { version = "0.14", features = ["full"] }
 log = { version = "0.4", features = ["std", "serde"] }
-postgres = "0.19"
+postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
 regex = "1"
 serde = { version = "1.0", features = ["derive"] }
-serde_json = "1.0"
+serde_json = "1"
 tar = "0.4"
-tokio = { version = "1", features = ["full"] }
-
-[profile.release]
-debug = true
+tokio = { version = "1", features = ["macros", "rt", "rt-multi-thread"] }

compute_tools/Dockerfile

@@ -1,14 +0,0 @@
-# First transient image to build compute_tools binaries
-FROM rust:slim-buster AS rust-build
-
-RUN mkdir /compute_tools
-WORKDIR /compute_tools
-
-COPY . /compute_tools/
-
-RUN cargo build --release
-
-# Final image that only has one binary
-FROM debian:buster-slim
-
-COPY --from=rust-build /compute_tools/target/release/zenith_ctl /usr/local/bin/zenith_ctl

compute_tools/src/bin/zenith_ctl.rs

@@ -27,14 +27,14 @@
 //! ```
 //!
 use std::fs::File;
+use std::panic;
 use std::path::Path;
 use std::process::{exit, Command, ExitStatus};
 use std::sync::{Arc, RwLock};
-use std::{env, panic};
 
-use anyhow::Result;
+use anyhow::{Context, Result};
 use chrono::Utc;
-use libc::{prctl, PR_SET_PDEATHSIG, SIGINT};
+use clap::Arg;
 use log::info;
 use postgres::{Client, NoTls};
 
@@ -70,7 +70,7 @@ fn prepare_pgdata(state: &Arc<RwLock<ComputeState>>) -> Result<()> {
         .expect("tenant id should be provided");
 
     info!(
-        "applying spec for cluster #{}, operation #{}",
+        "starting cluster #{}, operation #{}",
         spec.cluster.cluster_id,
         spec.operation_uuid.as_ref().unwrap()
     );
@@ -80,10 +80,23 @@ fn prepare_pgdata(state: &Arc<RwLock<ComputeState>>) -> Result<()> {
     config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), spec)?;
 
     info!("starting safekeepers syncing");
-    let lsn = sync_safekeepers(&state.pgdata, &state.pgbin)?;
+    let lsn = sync_safekeepers(&state.pgdata, &state.pgbin)
+        .with_context(|| "failed to sync safekeepers")?;
     info!("safekeepers synced at LSN {}", lsn);
 
-    get_basebackup(&state.pgdata, &pageserver_connstr, &tenant, &timeline, &lsn)?;
+    info!(
+        "getting basebackup@{} from pageserver {}",
+        lsn, pageserver_connstr
+    );
+    get_basebackup(&state.pgdata, &pageserver_connstr, &tenant, &timeline, &lsn).with_context(
+        || {
+            format!(
+                "failed to get basebackup@{} from pageserver {}",
+                lsn, pageserver_connstr
+            )
+        },
+    )?;
+
     // Update pg_hba.conf received with basebackup.
     update_pg_hba(pgdata_path)?;
 
@@ -142,51 +155,42 @@ fn run_compute(state: &Arc<RwLock<ComputeState>>) -> Result<ExitStatus> {
 }
 
 fn main() -> Result<()> {
-    // During configuration we are starting Postgres as a child process. If we
-    // fail we do not want to leave it running. PR_SET_PDEATHSIG sets the signal
-    // that will be sent to the child process when the parent dies. NB: this is
-    // cleared for the child of a fork(). SIGINT means fast shutdown for Postgres.
-    // This does not matter much for Docker, where `zenith_ctl` is an entrypoint,
-    // so the whole container will exit if it exits. But could be useful when
-    // `zenith_ctl` is used in e.g. systemd.
-    unsafe {
-        prctl(PR_SET_PDEATHSIG, SIGINT);
-    }
-
     // TODO: re-use `zenith_utils::logging` later
     init_logger(DEFAULT_LOG_LEVEL)?;
 
+    // Env variable is set by `cargo`
+    let version: Option<&str> = option_env!("CARGO_PKG_VERSION");
     let matches = clap::App::new("zenith_ctl")
-        .version("0.1.0")
+        .version(version.unwrap_or("unknown"))
         .arg(
-            clap::Arg::with_name("connstr")
-                .short("C")
+            Arg::new("connstr")
+                .short('C')
                 .long("connstr")
                 .value_name("DATABASE_URL")
                 .required(true),
         )
         .arg(
-            clap::Arg::with_name("pgdata")
-                .short("D")
+            Arg::new("pgdata")
+                .short('D')
                 .long("pgdata")
                 .value_name("DATADIR")
                 .required(true),
         )
         .arg(
-            clap::Arg::with_name("pgbin")
-                .short("b")
+            Arg::new("pgbin")
+                .short('b')
                 .long("pgbin")
                 .value_name("POSTGRES_PATH"),
         )
         .arg(
-            clap::Arg::with_name("spec")
-                .short("s")
+            Arg::new("spec")
+                .short('s')
                 .long("spec")
                 .value_name("SPEC_JSON"),
         )
         .arg(
-            clap::Arg::with_name("spec-path")
-                .short("S")
+            Arg::new("spec-path")
+                .short('S')
                 .long("spec-path")
                 .value_name("SPEC_PATH"),
         )
@@ -212,13 +216,7 @@ fn main() -> Result<()> {
                 let file = File::open(path)?;
                 serde_json::from_reader(file)?
             } else {
-                // Finally, try to fetch it from the env
-                // XXX: not tested well and kept as a backup option for k8s, Docker, etc.
-                // TODO: remove later
-                match env::var("CLUSTER_SPEC") {
-                    Ok(json) => serde_json::from_str(&json)?,
-                    Err(_) => panic!("cluster spec should be provided via --spec, --spec-path or env variable CLUSTER_SPEC")
-                }
+                panic!("cluster spec should be provided via --spec or --spec-path argument");
             }
         }
     };

compute_tools/src/pg_helpers.rs

@@ -5,7 +5,7 @@ use std::process::Command;
 use std::str::FromStr;
 use std::{fs, thread, time};
 
-use anyhow::{anyhow, Result};
+use anyhow::{bail, Result};
 use postgres::{Client, Transaction};
 use serde::Deserialize;
 
@@ -226,7 +226,7 @@ pub fn wait_for_postgres(port: &str, pgdata: &Path) -> Result<()> {
         // but postgres starts listening almost immediately, even if it is not really
         // ready to accept connections).
         if slept >= POSTGRES_WAIT_TIMEOUT {
-            return Err(anyhow!("timed out while waiting for Postgres to start"));
+            bail!("timed out while waiting for Postgres to start");
         }
 
         if pid_path.exists() {

compute_tools/src/zenith.rs

@@ -87,17 +87,19 @@ pub fn sync_safekeepers(pgdata: &str, pgbin: &str) -> Result<String> {
         .args(&["--sync-safekeepers"])
         .env("PGDATA", &pgdata) // we cannot use -D in this mode
         .stdout(Stdio::piped())
-        .stderr(Stdio::piped())
         .spawn()
         .expect("postgres --sync-safekeepers failed to start");
 
+    // `postgres --sync-safekeepers` will print all log output to stderr and
+    // final LSN to stdout. So we pipe only stdout, while stderr will be automatically
+    // redirected to the caller output.
     let sync_output = sync_handle
         .wait_with_output()
         .expect("postgres --sync-safekeepers failed");
     if !sync_output.status.success() {
         anyhow::bail!(
-            "postgres --sync-safekeepers failed: '{}'",
-            String::from_utf8_lossy(&sync_output.stderr)
+            "postgres --sync-safekeepers exited with non-zero status: {}",
+            sync_output.status,
         );
     }
 

control_plane/Cargo.toml

@@ -1,14 +1,11 @@
 [package]
 name = "control_plane"
 version = "0.1.0"
-authors = ["Stas Kelvich <stas@zenith.tech>"]
-edition = "2018"
-
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+edition = "2021"
 
 [dependencies]
 tar = "0.4.33"
-postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
+postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="2949d98df52587d562986aad155dd4e889e408b7" }
 serde = { version = "1.0", features = ["derive"] }
 toml = "0.5"
 lazy_static = "1.4"

control_plane/safekeepers.conf

@@ -1,20 +1,20 @@
 # Page server and three safekeepers.
 [pageserver]
-listen_pg_addr = 'localhost:64000'
-listen_http_addr = 'localhost:9898'
+listen_pg_addr = '127.0.0.1:64000'
+listen_http_addr = '127.0.0.1:9898'
 auth_type = 'Trust'
 
 [[safekeepers]]
-name = 'sk1'
+id = 1
 pg_port = 5454
 http_port = 7676
 
 [[safekeepers]]
-name = 'sk2'
+id = 2
 pg_port = 5455
 http_port = 7677
 
 [[safekeepers]]
-name = 'sk3'
+id = 3
 pg_port = 5456
 http_port = 7678

control_plane/simple.conf

@@ -1,11 +1,11 @@
 # Minimal zenith environment with one safekeeper. This is equivalent to the built-in
 # defaults that you get with no --config
 [pageserver]
-listen_pg_addr = 'localhost:64000'
-listen_http_addr = 'localhost:9898'
+listen_pg_addr = '127.0.0.1:64000'
+listen_http_addr = '127.0.0.1:9898'
 auth_type = 'Trust'
 
 [[safekeepers]]
-name = 'single'
+id = 1
 pg_port = 5454
 http_port = 7676

control_plane/src/compute.rs

@@ -82,15 +82,11 @@ impl ComputeControlPlane {
         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
-        }
+        let lsn = strings
+            .next()
+            .map(Lsn::from_str)
+            .transpose()
+            .context("invalid LSN in point-in-time specification")?;
 
         // Resolve the timeline ID, given the human-readable branch name
         let timeline_id = self
@@ -253,16 +249,16 @@ impl PostgresNode {
         let mut client = self
             .pageserver
             .page_server_psql_client()
-            .with_context(|| "connecting to page server failed")?;
+            .context("connecting to page server failed")?;
 
         let copyreader = client
             .copy_out(sql.as_str())
-            .with_context(|| "page server 'basebackup' command failed")?;
+            .context("page server 'basebackup' command failed")?;
 
         // Read the archive directly from the `CopyOutReader`
         tar::Archive::new(copyreader)
             .unpack(&self.pgdata())
-            .with_context(|| "extracting base backup failed")?;
+            .context("extracting base backup failed")?;
 
         Ok(())
     }
@@ -338,14 +334,26 @@ impl PostgresNode {
         if let Some(lsn) = self.lsn {
             conf.append("recovery_target_lsn", &lsn.to_string());
         }
+
         conf.append_line("");
+        // Configure backpressure
+        // - Replication write lag depends on how fast the walreceiver can process incoming WAL.
+        //   This lag determines latency of get_page_at_lsn. Speed of applying WAL is about 10MB/sec,
+        //   so to avoid expiration of 1 minute timeout, this lag should not be larger than 600MB.
+        //   Actually latency should be much smaller (better if < 1sec). But we assume that recently
+        //   updates pages are not requested from pageserver.
+        // - Replication flush lag depends on speed of persisting data by checkpointer (creation of
+        //   delta/image layers) and advancing disk_consistent_lsn. Safekeepers are able to
+        //   remove/archive WAL only beyond disk_consistent_lsn. Too large a lag can cause long
+        //   recovery time (in case of pageserver crash) and disk space overflow at safekeepers.
+        // - Replication apply lag depends on speed of uploading changes to S3 by uploader thread.
+        //   To be able to restore database in case of pageserver node crash, safekeeper should not
+        //   remove WAL beyond this point. Too large lag can cause space exhaustion in safekeepers
+        //   (if they are not able to upload WAL to S3).
+        conf.append("max_replication_write_lag", "500MB");
+        conf.append("max_replication_flush_lag", "10GB");
 
         if !self.env.safekeepers.is_empty() {
-            // Configure backpressure
-            // In setup with safekeepers apply_lag depends on
-            // speed of data checkpointing on pageserver (see disk_consistent_lsn).
-            conf.append("max_replication_apply_lag", "1500MB");
-
             // Configure the node to connect to the safekeepers
             conf.append("synchronous_standby_names", "walproposer");
 
@@ -358,11 +366,6 @@ impl PostgresNode {
                 .join(",");
             conf.append("wal_acceptors", &wal_acceptors);
         } else {
-            // Configure backpressure
-            // In setup without safekeepers, flush_lag depends on
-            // speed of of data checkpointing on pageserver (see disk_consistent_lsn)
-            conf.append("max_replication_flush_lag", "1500MB");
-
             // We only use setup without safekeepers for tests,
             // and don't care about data durability on pageserver,
             // so set more relaxed synchronous_commit.
@@ -443,7 +446,7 @@ impl PostgresNode {
         if let Some(token) = auth_token {
             cmd.env("ZENITH_AUTH_TOKEN", token);
         }
-        let pg_ctl = cmd.status().with_context(|| "pg_ctl failed")?;
+        let pg_ctl = cmd.status().context("pg_ctl failed")?;
 
         if !pg_ctl.success() {
             anyhow::bail!("pg_ctl failed");

control_plane/src/lib.rs

@@ -9,6 +9,7 @@
 use anyhow::{anyhow, bail, Context, Result};
 use std::fs;
 use std::path::Path;
+use std::process::Command;
 
 pub mod compute;
 pub mod local_env;
@@ -31,3 +32,19 @@ pub fn read_pidfile(pidfile: &Path) -> Result<i32> {
     }
     Ok(pid)
 }
+
+fn fill_rust_env_vars(cmd: &mut Command) -> &mut Command {
+    let cmd = cmd.env_clear().env("RUST_BACKTRACE", "1");
+
+    let var = "LLVM_PROFILE_FILE";
+    if let Some(val) = std::env::var_os(var) {
+        cmd.env(var, val);
+    }
+
+    const RUST_LOG_KEY: &str = "RUST_LOG";
+    if let Ok(rust_log_value) = std::env::var(RUST_LOG_KEY) {
+        cmd.env(RUST_LOG_KEY, rust_log_value)
+    } else {
+        cmd
+    }
+}

control_plane/src/local_env.rs

@@ -12,7 +12,9 @@ use std::path::{Path, PathBuf};
 use std::process::{Command, Stdio};
 use zenith_utils::auth::{encode_from_key_file, Claims, Scope};
 use zenith_utils::postgres_backend::AuthType;
-use zenith_utils::zid::{opt_display_serde, ZTenantId};
+use zenith_utils::zid::{opt_display_serde, ZNodeId, ZTenantId};
+
+use crate::safekeeper::SafekeeperNode;
 
 //
 // This data structures represents zenith CLI config
@@ -62,6 +64,8 @@ pub struct LocalEnv {
 #[derive(Serialize, Deserialize, Clone, Debug)]
 #[serde(default)]
 pub struct PageServerConf {
+    // node id
+    pub id: ZNodeId,
     // Pageserver connection settings
     pub listen_pg_addr: String,
     pub listen_http_addr: String,
@@ -76,6 +80,7 @@ pub struct PageServerConf {
 impl Default for PageServerConf {
     fn default() -> Self {
         Self {
+            id: ZNodeId(0),
             listen_pg_addr: String::new(),
             listen_http_addr: String::new(),
             auth_type: AuthType::Trust,
@@ -87,7 +92,7 @@ impl Default for PageServerConf {
 #[derive(Serialize, Deserialize, Clone, Debug)]
 #[serde(default)]
 pub struct SafekeeperConf {
-    pub name: String,
+    pub id: ZNodeId,
     pub pg_port: u16,
     pub http_port: u16,
     pub sync: bool,
@@ -96,7 +101,7 @@ pub struct SafekeeperConf {
 impl Default for SafekeeperConf {
     fn default() -> Self {
         Self {
-            name: String::new(),
+            id: ZNodeId(0),
             pg_port: 0,
             http_port: 0,
             sync: true,
@@ -136,8 +141,8 @@ impl LocalEnv {
         self.base_data_dir.clone()
     }
 
-    pub fn safekeeper_data_dir(&self, node_name: &str) -> PathBuf {
-        self.base_data_dir.join("safekeepers").join(node_name)
+    pub fn safekeeper_data_dir(&self, data_dir_name: &str) -> PathBuf {
+        self.base_data_dir.join("safekeepers").join(data_dir_name)
     }
 
     /// Create a LocalEnv from a config file.
@@ -251,7 +256,7 @@ impl LocalEnv {
                 .arg("2048")
                 .stdout(Stdio::null())
                 .output()
-                .with_context(|| "failed to generate auth private key")?;
+                .context("failed to generate auth private key")?;
             if !keygen_output.status.success() {
                 bail!(
                     "openssl failed: '{}'",
@@ -270,7 +275,7 @@ impl LocalEnv {
                 .args(&["-out", public_key_path.to_str().unwrap()])
                 .stdout(Stdio::null())
                 .output()
-                .with_context(|| "failed to generate auth private key")?;
+                .context("failed to generate auth private key")?;
             if !keygen_output.status.success() {
                 bail!(
                     "openssl failed: '{}'",
@@ -285,7 +290,7 @@ impl LocalEnv {
         fs::create_dir_all(self.pg_data_dirs_path())?;
 
         for safekeeper in &self.safekeepers {
-            fs::create_dir_all(self.safekeeper_data_dir(&safekeeper.name))?;
+            fs::create_dir_all(SafekeeperNode::datadir_path_by_id(self, safekeeper.id))?;
         }
 
         let mut conf_content = String::new();

control_plane/src/postgresql_conf.rs

@@ -4,7 +4,7 @@
 /// NOTE: This doesn't implement the full, correct postgresql.conf syntax. Just
 /// enough to extract a few settings we need in Zenith, assuming you don't do
 /// funny stuff like include-directives or funny escaping.
-use anyhow::{anyhow, bail, Context, Result};
+use anyhow::{bail, Context, Result};
 use lazy_static::lazy_static;
 use regex::Regex;
 use std::collections::HashMap;
@@ -78,7 +78,7 @@ impl PostgresConf {
         <T as FromStr>::Err: std::error::Error + Send + Sync + 'static,
     {
         self.get(field_name)
-            .ok_or_else(|| anyhow!("could not find '{}' option {}", field_name, context))?
+            .with_context(|| format!("could not find '{}' option {}", field_name, context))?
             .parse::<T>()
             .with_context(|| format!("could not parse '{}' option {}", field_name, context))
     }

control_plane/src/safekeeper.rs

@@ -15,10 +15,11 @@ use reqwest::blocking::{Client, RequestBuilder, Response};
 use reqwest::{IntoUrl, Method};
 use thiserror::Error;
 use zenith_utils::http::error::HttpErrorBody;
+use zenith_utils::zid::ZNodeId;
 
 use crate::local_env::{LocalEnv, SafekeeperConf};
-use crate::read_pidfile;
 use crate::storage::PageServerNode;
+use crate::{fill_rust_env_vars, read_pidfile};
 use zenith_utils::connstring::connection_address;
 
 #[derive(Error, Debug)]
@@ -61,7 +62,7 @@ impl ResponseErrorMessageExt for Response {
 //
 #[derive(Debug)]
 pub struct SafekeeperNode {
-    pub name: String,
+    pub id: ZNodeId,
 
     pub conf: SafekeeperConf,
 
@@ -77,15 +78,15 @@ 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);
+        println!("initializing for sk {} for {}", conf.id, conf.http_port);
 
         SafekeeperNode {
-            name: conf.name.clone(),
+            id: conf.id,
             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),
+            http_base_url: format!("http://127.0.0.1:{}/v1", conf.http_port),
             pageserver,
         }
     }
@@ -93,13 +94,17 @@ impl SafekeeperNode {
     /// 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)
+        format!("postgresql://no_user@127.0.0.1:{}/no_db", port)
             .parse()
             .unwrap()
     }
 
+    pub fn datadir_path_by_id(env: &LocalEnv, sk_id: ZNodeId) -> PathBuf {
+        env.safekeeper_data_dir(format!("sk{}", sk_id).as_ref())
+    }
+
     pub fn datadir_path(&self) -> PathBuf {
-        self.env.safekeeper_data_dir(&self.name)
+        SafekeeperNode::datadir_path_by_id(&self.env, self.id)
     }
 
     pub fn pid_file(&self) -> PathBuf {
@@ -114,26 +119,22 @@ impl SafekeeperNode {
         );
         io::stdout().flush().unwrap();
 
-        let listen_pg = format!("localhost:{}", self.conf.pg_port);
-        let listen_http = format!("localhost:{}", self.conf.http_port);
+        let listen_pg = format!("127.0.0.1:{}", self.conf.pg_port);
+        let listen_http = format!("127.0.0.1:{}", self.conf.http_port);
 
         let mut cmd = Command::new(self.env.safekeeper_bin()?);
-        cmd.args(&["-D", self.datadir_path().to_str().unwrap()])
-            .args(&["--listen-pg", &listen_pg])
-            .args(&["--listen-http", &listen_http])
-            .args(&["--recall", "1 second"])
-            .arg("--daemonize")
-            .env_clear()
-            .env("RUST_BACKTRACE", "1");
+        fill_rust_env_vars(
+            cmd.args(&["-D", self.datadir_path().to_str().unwrap()])
+                .args(&["--id", self.id.to_string().as_ref()])
+                .args(&["--listen-pg", &listen_pg])
+                .args(&["--listen-http", &listen_http])
+                .args(&["--recall", "1 second"])
+                .arg("--daemonize"),
+        );
         if !self.conf.sync {
             cmd.arg("--no-sync");
         }
 
-        let var = "LLVM_PROFILE_FILE";
-        if let Some(val) = std::env::var_os(var) {
-            cmd.env(var, val);
-        }
-
         if !cmd.status()?.success() {
             bail!(
                 "Safekeeper failed to start. See '{}' for details.",
@@ -188,7 +189,7 @@ impl SafekeeperNode {
     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);
+            println!("Safekeeper {} is already stopped", self.id);
             return Ok(());
         }
         let pid = read_pidfile(&pid_file)?;

control_plane/src/storage.rs

@@ -19,7 +19,7 @@ use zenith_utils::postgres_backend::AuthType;
 use zenith_utils::zid::ZTenantId;
 
 use crate::local_env::LocalEnv;
-use crate::read_pidfile;
+use crate::{fill_rust_env_vars, read_pidfile};
 use pageserver::branches::BranchInfo;
 use pageserver::tenant_mgr::TenantInfo;
 use zenith_utils::connstring::connection_address;
@@ -96,46 +96,52 @@ impl PageServerNode {
             .unwrap()
     }
 
-    pub fn init(&self, create_tenant: Option<&str>) -> anyhow::Result<()> {
+    pub fn init(
+        &self,
+        create_tenant: Option<&str>,
+        config_overrides: &[&str],
+    ) -> anyhow::Result<()> {
         let mut cmd = Command::new(self.env.pageserver_bin()?);
-        let var = "LLVM_PROFILE_FILE";
-        if let Some(val) = std::env::var_os(var) {
-            cmd.env(var, val);
-        }
+
+        let id = format!("id={}", self.env.pageserver.id);
 
         // FIXME: the paths should be shell-escaped to handle paths with spaces, quotas etc.
-        let mut args = vec![
-            "--init".to_string(),
-            "-D".to_string(),
-            self.env.base_data_dir.display().to_string(),
-            "-c".to_string(),
-            format!("pg_distrib_dir='{}'", self.env.pg_distrib_dir.display()),
-            "-c".to_string(),
-            format!("auth_type='{}'", self.env.pageserver.auth_type),
-            "-c".to_string(),
-            format!(
-                "listen_http_addr='{}'",
-                self.env.pageserver.listen_http_addr
-            ),
-            "-c".to_string(),
-            format!("listen_pg_addr='{}'", self.env.pageserver.listen_pg_addr),
-        ];
+        let base_data_dir_param = self.env.base_data_dir.display().to_string();
+        let pg_distrib_dir_param =
+            format!("pg_distrib_dir='{}'", self.env.pg_distrib_dir.display());
+        let authg_type_param = format!("auth_type='{}'", self.env.pageserver.auth_type);
+        let listen_http_addr_param = format!(
+            "listen_http_addr='{}'",
+            self.env.pageserver.listen_http_addr
+        );
+        let listen_pg_addr_param =
+            format!("listen_pg_addr='{}'", self.env.pageserver.listen_pg_addr);
+        let mut args = Vec::with_capacity(20);
+
+        args.push("--init");
+        args.extend(["-D", &base_data_dir_param]);
+        args.extend(["-c", &pg_distrib_dir_param]);
+        args.extend(["-c", &authg_type_param]);
+        args.extend(["-c", &listen_http_addr_param]);
+        args.extend(["-c", &listen_pg_addr_param]);
+        args.extend(["-c", &id]);
+
+        for config_override in config_overrides {
+            args.extend(["-c", config_override]);
+        }
 
         if self.env.pageserver.auth_type != AuthType::Trust {
             args.extend([
-                "-c".to_string(),
-                "auth_validation_public_key_path='auth_public_key.pem'".to_string(),
+                "-c",
+                "auth_validation_public_key_path='auth_public_key.pem'",
             ]);
         }
 
         if let Some(tenantid) = create_tenant {
-            args.extend(["--create-tenant".to_string(), tenantid.to_string()])
+            args.extend(["--create-tenant", tenantid])
         }
 
-        let status = cmd
-            .args(args)
-            .env_clear()
-            .env("RUST_BACKTRACE", "1")
+        let status = fill_rust_env_vars(cmd.args(args))
             .status()
             .expect("pageserver init failed");
 
@@ -154,7 +160,7 @@ impl PageServerNode {
         self.repo_path().join("pageserver.pid")
     }
 
-    pub fn start(&self) -> anyhow::Result<()> {
+    pub fn start(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
         print!(
             "Starting pageserver at '{}' in '{}'",
             connection_address(&self.pg_connection_config),
@@ -163,16 +169,16 @@ impl PageServerNode {
         io::stdout().flush().unwrap();
 
         let mut cmd = Command::new(self.env.pageserver_bin()?);
-        cmd.args(&["-D", self.repo_path().to_str().unwrap()])
-            .arg("--daemonize")
-            .env_clear()
-            .env("RUST_BACKTRACE", "1");
 
-        let var = "LLVM_PROFILE_FILE";
-        if let Some(val) = std::env::var_os(var) {
-            cmd.env(var, val);
+        let repo_path = self.repo_path();
+        let mut args = vec!["-D", repo_path.to_str().unwrap()];
+
+        for config_override in config_overrides {
+            args.extend(["-c", config_override]);
         }
 
+        fill_rust_env_vars(cmd.args(&args).arg("--daemonize"));
+
         if !cmd.status()?.success() {
             bail!(
                 "Pageserver failed to start. See '{}' for details.",

docker-entrypoint.sh

@@ -4,7 +4,7 @@ set -eux
 if [ "$1" = 'pageserver' ]; then
     if [ ! -d "/data/tenants" ]; then
         echo "Initializing pageserver data directory"
-        pageserver --init -D /data -c "pg_distrib_dir='/usr/local'"
+        pageserver --init -D /data -c "pg_distrib_dir='/usr/local'" -c "id=10"
     fi
     echo "Staring pageserver at 0.0.0.0:6400"
     pageserver -c "listen_pg_addr='0.0.0.0:6400'" -c "listen_http_addr='0.0.0.0:9898'" -D /data

docs/glossary.md

@@ -2,6 +2,16 @@
 
 ### Authentication
 
+### Backpresssure
+
+Backpressure is used to limit the lag between pageserver and compute node or WAL service.
+
+If compute node or WAL service run far ahead of Page Server,
+the time of serving page requests increases. This may lead to timeout errors.
+
+To tune backpressure limits use `max_replication_write_lag`, `max_replication_flush_lag` and `max_replication_apply_lag` settings.
+When lag between current LSN (pg_current_wal_flush_lsn() at compute node) and minimal write/flush/apply position of replica exceeds the limit
+backends performing writes are blocked until the replica is caught up.
 ### Base image (page image)
 
 ### Basebackup
@@ -76,7 +86,37 @@ The layer map tracks what layers exist for all the relishes in a timeline.
 Zenith repository implementation that keeps data in layers.
 ### LSN
 
+The Log Sequence Number (LSN) is a unique identifier of the WAL record[] in the WAL log.
+The insert position is a byte offset into the logs, increasing monotonically with each new record.
+Internally, an LSN is a 64-bit integer, representing a byte position in the write-ahead log stream.
+It is printed as two hexadecimal numbers of up to 8 digits each, separated by a slash.
+Check also [PostgreSQL doc about pg_lsn type](https://www.postgresql.org/docs/devel/datatype-pg-lsn.html)
+Values can be compared to calculate the volume of WAL data that separates them, so they are used to measure the progress of replication and recovery.
 
+In postgres and Zenith lsns are used to describe certain points in WAL handling.
+
+PostgreSQL LSNs and functions to monitor them:
+* `pg_current_wal_insert_lsn()` - Returns the current write-ahead log insert location.
+* `pg_current_wal_lsn()` - Returns the current write-ahead log write location.
+* `pg_current_wal_flush_lsn()` - Returns the current write-ahead log flush location.
+* `pg_last_wal_receive_lsn()` - Returns the last write-ahead log location that has been received and synced to disk by streaming replication. While streaming replication is in progress this will increase monotonically.
+* `pg_last_wal_replay_lsn ()` - Returns the last write-ahead log location that has been replayed during recovery. If recovery is still in progress this will increase monotonically. 
+[source PostgreSQL documentation](https://www.postgresql.org/docs/devel/functions-admin.html):
+
+Zenith safekeeper LSNs. For more check [walkeeper/README_PROTO.md](/walkeeper/README_PROTO.md)
+* `CommitLSN`: position in WAL confirmed by quorum safekeepers.
+* `RestartLSN`: position in WAL confirmed by all safekeepers.
+* `FlushLSN`: part of WAL persisted to the disk by safekeeper.
+* `VCL`: the largerst LSN for which we can guarantee availablity of all prior records.
+
+Zenith pageserver LSNs:
+* `last_record_lsn` - the end of last processed WAL record.
+* `disk_consistent_lsn` - data is known to be fully flushed and fsync'd to local disk on pageserver up to this LSN.
+* `remote_consistent_lsn` - The last LSN that is synced to remote storage and is guaranteed to survive pageserver crash.
+TODO: use this name consistently in remote storage code. Now `disk_consistent_lsn` is used and meaning depends on the context.
+* `ancestor_lsn` - LSN of the branch point (the LSN at which this branch was created)
+
+TODO: add table that describes mapping between PostgreSQL (compute), safekeeper and pageserver LSNs.
 ### Page (block)
 
 The basic structure used to store relation data. All pages are of the same size.

docs/pageserver-tenant-migration.md

@@ -0,0 +1,22 @@
+## Pageserver tenant migration
+
+### Overview
+
+This feature allows to migrate a timeline from one pageserver to another by utilizing remote storage capability.
+
+### Migration process
+
+Pageserver implements two new http handlers: timeline attach and timeline detach.
+Timeline migration is performed in a following way:
+1. Timeline attach is called on a target pageserver. This asks pageserver to download latest checkpoint uploaded to s3.
+2. For now it is necessary to manually initialize replication stream via callmemaybe call so target pageserver initializes replication from safekeeper (it is desired to avoid this and initialize replication directly in attach handler, but this requires some refactoring (probably [#997](https://github.com/zenithdb/zenith/issues/997)/[#1049](https://github.com/zenithdb/zenith/issues/1049))
+3. Replication state can be tracked via timeline detail pageserver call.
+4. Compute node should be restarted with new pageserver connection string. Issue with multiple compute nodes for one timeline is handled on the safekeeper consensus level. So this is not a problem here.Currently responsibility for rescheduling the compute with updated config lies on external coordinator (console).
+5. Timeline is detached from old pageserver. On disk data is removed.
+
+
+### Implementation details
+
+Now safekeeper needs to track which pageserver it is replicating to. This introduces complications into replication code:
+* We need to distinguish different pageservers (now this is done by connection string which is imperfect and is covered here: https://github.com/zenithdb/zenith/issues/1105). Callmemaybe subscription management also needs to track that (this is already implemented).
+* We need to track which pageserver is the primary. This is needed to avoid reconnections to non primary pageservers. Because we shouldn't reconnect to them when they decide to stop their walreceiver. I e this can appear when there is a load on the compute and we are trying to detach timeline from old pageserver. In this case callmemaybe will try to reconnect to it because replication termination condition is not met (page server with active compute could never catch up to the latest lsn, so there is always some wal tail)

docs/settings.md

@@ -147,6 +147,10 @@ bucket_name = 'some-sample-bucket'
 # Name of the region where the bucket is located at
 bucket_region = 'eu-north-1'
 
+# A "subfolder" in the bucket, to use the same bucket separately by multiple pageservers at once.
+# Optional, pageserver uses entire bucket if the prefix is not specified.
+prefix_in_bucket = '/some/prefix/'
+
 # Access key to connect to the bucket ("login" part of the credentials)
 access_key_id = 'SOMEKEYAAAAASADSAH*#'
 

docs/sourcetree.md

@@ -87,31 +87,29 @@ so manual installation of dependencies is not recommended.
 A single virtual environment with all dependencies is described in the single `Pipfile`.
 
 ### Prerequisites
-- Install Python 3.7 (the minimal supported version)
-    - Later version (e.g. 3.8) is ok if you don't write Python code
-    - You can install Python 3.7 separately, e.g.:
+- Install Python 3.7 (the minimal supported version) or greater.
+    - Our setup with poetry should work with newer python versions too. So feel free to open an issue with a `c/test-runner` label if something doesnt work as expected.
+    - If you have some trouble with other version you can resolve it by installing Python 3.7 separately, via pyenv or via system package manager e.g.:
       ```bash
       # In Ubuntu
       sudo add-apt-repository ppa:deadsnakes/ppa
       sudo apt update
       sudo apt install python3.7
       ```
-- Install `pipenv`
-    - Exact version of `pipenv` is not important, you can use Debian/Ubuntu package `pipenv`.
-- Install dependencies via either
-  * `pipenv --python 3.7 install --dev` if you will write Python code, or
-  * `pipenv install` if you only want to run Python scripts and don't have Python 3.7.
+- Install `poetry`
+    - Exact version of `poetry` is not important, see installation instructions available at poetry's [website](https://python-poetry.org/docs/#installation)`.
+- Install dependencies via `./scripts/pysync`. Note that CI uses Python 3.7 so if you have different version some linting tools can yield different result locally vs in the CI.
 
-Run `pipenv shell` to activate the virtual environment.
-Alternatively, use `pipenv run` to run a single command in the venv, e.g. `pipenv run pytest`.
+Run `poetry shell` to activate the virtual environment.
+Alternatively, use `poetry run` to run a single command in the venv, e.g. `poetry run pytest`.
 
 ### Obligatory checks
 We force code formatting via `yapf` and type hints via `mypy`.
 Run the following commands in the repository's root (next to `setup.cfg`):
 
 ```bash
-pipenv run yapf -ri .  # All code is reformatted
-pipenv run mypy .  # Ensure there are no typing errors
+poetry run yapf -ri .  # All code is reformatted
+poetry run mypy .  # Ensure there are no typing errors
 ```
 
 **WARNING**: do not run `mypy` from a directory other than the root of the repository.
@@ -123,17 +121,6 @@ Also consider:
 * Adding more type hints to your code to avoid `Any`.
 
 ### Changing dependencies
-You have to update `Pipfile.lock` if you have changed `Pipfile`:
+To add new package or change an existing one you can use `poetry add` or `poetry update` or edit `pyproject.toml` manually. Do not forget to run `poetry lock` in the latter case.
 
-```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`.
+More details are available in poetry's [documentation](https://python-poetry.org/docs/).

pageserver/Cargo.toml

@@ -1,8 +1,7 @@
 [package]
 name = "pageserver"
 version = "0.1.0"
-authors = ["Stas Kelvich <stas@zenith.tech>"]
-edition = "2018"
+edition = "2021"
 
 [dependencies]
 bookfile = { git = "https://github.com/zenithdb/bookfile.git", branch="generic-readext" }
@@ -15,16 +14,15 @@ futures = "0.3.13"
 hyper = "0.14"
 lazy_static = "1.4.0"
 log = "0.4.14"
-clap = "2.33.0"
+clap = "3.0"
 daemonize = "0.4.1"
 tokio = { version = "1.11", features = ["process", "sync", "macros", "fs", "rt", "io-util", "time"] }
-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" }
-tokio-postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
+postgres-types = { git = "https://github.com/zenithdb/rust-postgres.git", rev="2949d98df52587d562986aad155dd4e889e408b7" }
+postgres-protocol = { git = "https://github.com/zenithdb/rust-postgres.git", rev="2949d98df52587d562986aad155dd4e889e408b7" }
+postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="2949d98df52587d562986aad155dd4e889e408b7" }
+tokio-postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="2949d98df52587d562986aad155dd4e889e408b7" }
 tokio-stream = "0.1.8"
-routerify = "2"
-anyhow = "1.0"
+anyhow = { version = "1.0", features = ["backtrace"] }
 crc32c = "0.6.0"
 thiserror = "1.0"
 hex = { version = "0.4.3", features = ["serde"] }
@@ -32,7 +30,7 @@ tar = "0.4.33"
 humantime = "2.1.0"
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1"
-toml_edit = { version = "0.12", features = ["easy"] }
+toml_edit = { version = "0.13", features = ["easy"] }
 scopeguard = "1.1.0"
 async-trait = "0.1"
 const_format = "0.2.21"
@@ -42,8 +40,8 @@ signal-hook = "0.3.10"
 url = "2"
 nix = "0.23"
 once_cell = "1.8.0"
-parking_lot = "0.11.2"
 crossbeam-utils = "0.8.5"
+fail = "0.5.0"
 
 rust-s3 = { version = "0.28", default-features = false, features = ["no-verify-ssl", "tokio-rustls-tls"] }
 async-compression = {version = "0.3", features = ["zstd", "tokio"]}

pageserver/README.md

@@ -129,13 +129,13 @@ 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 [backup readme](./src/remote_storage/README.md) and corresponding Rust file docs.
+Implementation details are covered in the [backup readme](./src/remote_storage/README.md) and corresponding Rust file docs, parameters documentation can be found at [settings docs](../docs/settings.md).
 
 The backup service is disabled by default and can be enabled to interact with a single remote storage.
 
 CLI examples:
 * Local FS: `${PAGESERVER_BIN} -c "remote_storage={local_path='/some/local/path/'}"`
-* AWS S3  : `${PAGESERVER_BIN} -c "remote_storage={bucket_name='some-sample-bucket',bucket_region='eu-north-1',access_key_id='SOMEKEYAAAAASADSAH*#',secret_access_key='SOMEsEcReTsd292v'}"`
+* AWS S3  : `${PAGESERVER_BIN} -c "remote_storage={bucket_name='some-sample-bucket',bucket_region='eu-north-1', prefix_in_bucket='/test_prefix/',access_key_id='SOMEKEYAAAAASADSAH*#',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.
@@ -154,6 +154,7 @@ or
 [remote_storage]
 bucket_name = 'some-sample-bucket'
 bucket_region = 'eu-north-1'
+prefix_in_bucket = '/test_prefix/'
 access_key_id = 'SOMEKEYAAAAASADSAH*#'
 secret_access_key = 'SOMEsEcReTsd292v'
 ```

pageserver/src/bin/dump_layerfile.rs

@@ -13,7 +13,7 @@ fn main() -> Result<()> {
         .about("Dump contents of one layer file, for debugging")
         .version(GIT_VERSION)
         .arg(
-            Arg::with_name("path")
+            Arg::new("path")
                 .help("Path to file to dump")
                 .required(true)
                 .index(1),

pageserver/src/bin/pageserver.rs

@@ -1,6 +1,6 @@
 //! Main entry point for the Page Server executable.
 
-use std::{env, path::Path, str::FromStr, thread};
+use std::{env, path::Path, str::FromStr};
 use tracing::*;
 use zenith_utils::{auth::JwtAuth, logging, postgres_backend::AuthType, tcp_listener, GIT_VERSION};
 
@@ -12,7 +12,9 @@ use daemonize::Daemonize;
 use pageserver::{
     branches,
     config::{defaults::*, PageServerConf},
-    http, page_cache, page_service, remote_storage, tenant_mgr, virtual_file, LOG_FILE_NAME,
+    http, page_cache, page_service, remote_storage, tenant_mgr, thread_mgr,
+    thread_mgr::ThreadKind,
+    virtual_file, LOG_FILE_NAME,
 };
 use zenith_utils::http::endpoint;
 use zenith_utils::postgres_backend;
@@ -25,27 +27,27 @@ fn main() -> Result<()> {
         .about("Materializes WAL stream to pages and serves them to the postgres")
         .version(GIT_VERSION)
         .arg(
-            Arg::with_name("daemonize")
-                .short("d")
+            Arg::new("daemonize")
+                .short('d')
                 .long("daemonize")
                 .takes_value(false)
                 .help("Run in the background"),
         )
         .arg(
-            Arg::with_name("init")
+            Arg::new("init")
                 .long("init")
                 .takes_value(false)
                 .help("Initialize pageserver repo"),
         )
         .arg(
-            Arg::with_name("workdir")
-                .short("D")
+            Arg::new("workdir")
+                .short('D')
                 .long("workdir")
                 .takes_value(true)
                 .help("Working directory for the pageserver"),
         )
         .arg(
-            Arg::with_name("create-tenant")
+            Arg::new("create-tenant")
                 .long("create-tenant")
                 .takes_value(true)
                 .help("Create tenant during init")
@@ -53,13 +55,13 @@ fn main() -> Result<()> {
         )
         // See `settings.md` for more details on the extra configuration patameters pageserver can process
         .arg(
-            Arg::with_name("config-option")
-                .short("c")
+            Arg::new("config-override")
+                .short('c')
                 .takes_value(true)
                 .number_of_values(1)
-                .multiple(true)
-                .help("Additional configuration options or overrides of the ones from the toml config file.
-                Any option has to be a valid toml document, example: `-c \"foo='hey'\"` `-c \"foo={value=1}\"`"),
+                .multiple_occurrences(true)
+                .help("Additional configuration overrides of the ones from the toml config file (or new ones to add there).
+                Any option has to be a valid toml document, example: `-c=\"foo='hey'\"` `-c=\"foo={value=1}\"`"),
         )
         .get_matches();
 
@@ -105,7 +107,7 @@ fn main() -> Result<()> {
     };
 
     // Process any extra options given with -c
-    if let Some(values) = arg_matches.values_of("config-option") {
+    if let Some(values) = arg_matches.values_of("config-override") {
         for option_line in values {
             let doc = toml_edit::Document::from_str(option_line).with_context(|| {
                 format!(
@@ -113,7 +115,14 @@ fn main() -> Result<()> {
                     option_line
                 )
             })?;
+
             for (key, item) in doc.iter() {
+                if key == "id" {
+                    anyhow::ensure!(
+                        init,
+                        "node id can only be set during pageserver init and cannot be overridden"
+                    );
+                }
                 toml.insert(key, item.clone());
             }
         }
@@ -169,7 +178,7 @@ fn start_pageserver(conf: &'static PageServerConf, daemonize: bool) -> Result<()
     );
     let pageserver_listener = tcp_listener::bind(conf.listen_pg_addr.clone())?;
 
-    // XXX: Don't spawn any threads before daemonizing!
+    // NB: Don't spawn any threads before daemonizing!
     if daemonize {
         info!("daemonizing...");
 
@@ -195,15 +204,9 @@ fn start_pageserver(conf: &'static PageServerConf, daemonize: bool) -> Result<()
     }
 
     let signals = signals::install_shutdown_handlers()?;
-    let mut threads = Vec::new();
-
     let sync_startup = remote_storage::start_local_timeline_sync(conf)
         .context("Failed to set up local files sync with external storage")?;
 
-    if let Some(handle) = sync_startup.sync_loop_handle {
-        threads.push(handle);
-    }
-
     // Initialize tenant manager.
     tenant_mgr::set_timeline_states(conf, sync_startup.initial_timeline_states);
 
@@ -220,25 +223,27 @@ fn start_pageserver(conf: &'static PageServerConf, daemonize: bool) -> Result<()
 
     // Spawn a new thread for the http endpoint
     // bind before launching separate thread so the error reported before startup exits
-    let cloned = auth.clone();
-    threads.push(
-        thread::Builder::new()
-            .name("http_endpoint_thread".into())
-            .spawn(move || {
-                let router = http::make_router(conf, cloned);
-                endpoint::serve_thread_main(router, http_listener)
-            })?,
-    );
+    let auth_cloned = auth.clone();
+    thread_mgr::spawn(
+        ThreadKind::HttpEndpointListener,
+        None,
+        None,
+        "http_endpoint_thread",
+        move || {
+            let router = http::make_router(conf, auth_cloned);
+            endpoint::serve_thread_main(router, http_listener, thread_mgr::shutdown_watcher())
+        },
+    )?;
 
-    // Spawn a thread to listen for connections. It will spawn further threads
+    // Spawn a thread to listen for libpq connections. It will spawn further threads
     // for each connection.
-    threads.push(
-        thread::Builder::new()
-            .name("Page Service thread".into())
-            .spawn(move || {
-                page_service::thread_main(conf, auth, pageserver_listener, conf.auth_type)
-            })?,
-    );
+    thread_mgr::spawn(
+        ThreadKind::LibpqEndpointListener,
+        None,
+        None,
+        "libpq endpoint thread",
+        move || page_service::thread_main(conf, auth, pageserver_listener, conf.auth_type),
+    )?;
 
     signals.handle(|signal| match signal {
         Signal::Quit => {
@@ -254,20 +259,38 @@ fn start_pageserver(conf: &'static PageServerConf, daemonize: bool) -> Result<()
                 "Got {}. Terminating gracefully in fast shutdown mode",
                 signal.name()
             );
-
-            postgres_backend::set_pgbackend_shutdown_requested();
-            tenant_mgr::shutdown_all_tenants()?;
-            endpoint::shutdown();
-
-            for handle in std::mem::take(&mut threads) {
-                handle
-                    .join()
-                    .expect("thread panicked")
-                    .expect("thread exited with an error");
-            }
-
-            info!("Shut down successfully completed");
-            std::process::exit(0);
+            shutdown_pageserver();
+            unreachable!()
         }
     })
 }
+
+fn shutdown_pageserver() {
+    // Shut down the libpq endpoint thread. This prevents new connections from
+    // being accepted.
+    thread_mgr::shutdown_threads(Some(ThreadKind::LibpqEndpointListener), None, None);
+
+    // Shut down any page service threads.
+    postgres_backend::set_pgbackend_shutdown_requested();
+    thread_mgr::shutdown_threads(Some(ThreadKind::PageRequestHandler), None, None);
+
+    // Shut down all the tenants. This flushes everything to disk and kills
+    // the checkpoint and GC threads.
+    tenant_mgr::shutdown_all_tenants();
+
+    // Stop syncing with remote storage.
+    //
+    // FIXME: Does this wait for the sync thread to finish syncing what's queued up?
+    // Should it?
+    thread_mgr::shutdown_threads(Some(ThreadKind::StorageSync), None, None);
+
+    // Shut down the HTTP endpoint last, so that you can still check the server's
+    // status while it's shutting down.
+    thread_mgr::shutdown_threads(Some(ThreadKind::HttpEndpointListener), None, None);
+
+    // There should be nothing left, but let's be sure
+    thread_mgr::shutdown_threads(None, None, None);
+
+    info!("Shut down successfully completed");
+    std::process::exit(0);
+}

pageserver/src/bin/pageserver_zst.rs

@@ -0,0 +1,334 @@
+//! A CLI helper to deal with remote storage (S3, usually) blobs as archives.
+//! See [`compression`] for more details about the archives.
+
+use std::{collections::BTreeSet, path::Path};
+
+use anyhow::{bail, ensure, Context};
+use clap::{App, Arg};
+use pageserver::{
+    layered_repository::metadata::{TimelineMetadata, METADATA_FILE_NAME},
+    remote_storage::compression,
+};
+use tokio::{fs, io};
+use zenith_utils::GIT_VERSION;
+
+const LIST_SUBCOMMAND: &str = "list";
+const ARCHIVE_ARG_NAME: &str = "archive";
+
+const EXTRACT_SUBCOMMAND: &str = "extract";
+const TARGET_DIRECTORY_ARG_NAME: &str = "target_directory";
+
+const CREATE_SUBCOMMAND: &str = "create";
+const SOURCE_DIRECTORY_ARG_NAME: &str = "source_directory";
+
+#[tokio::main(flavor = "current_thread")]
+async fn main() -> anyhow::Result<()> {
+    let arg_matches = App::new("pageserver zst blob [un]compressor utility")
+        .version(GIT_VERSION)
+        .subcommands(vec![
+            App::new(LIST_SUBCOMMAND)
+                .about("List the archive contents")
+                .arg(
+                    Arg::new(ARCHIVE_ARG_NAME)
+                        .required(true)
+                        .takes_value(true)
+                        .help("An archive to list the contents of"),
+                ),
+            App::new(EXTRACT_SUBCOMMAND)
+                .about("Extracts the archive into the directory")
+                .arg(
+                    Arg::new(ARCHIVE_ARG_NAME)
+                        .required(true)
+                        .takes_value(true)
+                        .help("An archive to extract"),
+                )
+                .arg(
+                    Arg::new(TARGET_DIRECTORY_ARG_NAME)
+                        .required(false)
+                        .takes_value(true)
+                        .help("A directory to extract the archive into. Optional, will use the current directory if not specified"),
+                ),
+            App::new(CREATE_SUBCOMMAND)
+                .about("Creates an archive with the contents of a directory (only the first level files are taken, metadata file has to be present in the same directory)")
+                .arg(
+                    Arg::new(SOURCE_DIRECTORY_ARG_NAME)
+                        .required(true)
+                        .takes_value(true)
+                        .help("A directory to use for creating the archive"),
+                )
+                .arg(
+                    Arg::new(TARGET_DIRECTORY_ARG_NAME)
+                        .required(false)
+                        .takes_value(true)
+                        .help("A directory to create the archive in. Optional, will use the current directory if not specified"),
+                ),
+        ])
+        .get_matches();
+
+    let subcommand_name = match arg_matches.subcommand_name() {
+        Some(name) => name,
+        None => bail!("No subcommand specified"),
+    };
+
+    let subcommand_matches = match arg_matches.subcommand_matches(subcommand_name) {
+        Some(matches) => matches,
+        None => bail!(
+            "No subcommand arguments were recognized for subcommand '{}'",
+            subcommand_name
+        ),
+    };
+
+    let target_dir = Path::new(
+        subcommand_matches
+            .value_of(TARGET_DIRECTORY_ARG_NAME)
+            .unwrap_or("./"),
+    );
+
+    match subcommand_name {
+        LIST_SUBCOMMAND => {
+            let archive = match subcommand_matches.value_of(ARCHIVE_ARG_NAME) {
+                Some(archive) => Path::new(archive),
+                None => bail!("No '{}' argument is specified", ARCHIVE_ARG_NAME),
+            };
+            list_archive(archive).await
+        }
+        EXTRACT_SUBCOMMAND => {
+            let archive = match subcommand_matches.value_of(ARCHIVE_ARG_NAME) {
+                Some(archive) => Path::new(archive),
+                None => bail!("No '{}' argument is specified", ARCHIVE_ARG_NAME),
+            };
+            extract_archive(archive, target_dir).await
+        }
+        CREATE_SUBCOMMAND => {
+            let source_dir = match subcommand_matches.value_of(SOURCE_DIRECTORY_ARG_NAME) {
+                Some(source) => Path::new(source),
+                None => bail!("No '{}' argument is specified", SOURCE_DIRECTORY_ARG_NAME),
+            };
+            create_archive(source_dir, target_dir).await
+        }
+        unknown => bail!("Unknown subcommand {}", unknown),
+    }
+}
+
+async fn list_archive(archive: &Path) -> anyhow::Result<()> {
+    let archive = archive.canonicalize().with_context(|| {
+        format!(
+            "Failed to get the absolute path for the archive path '{}'",
+            archive.display()
+        )
+    })?;
+    ensure!(
+        archive.is_file(),
+        "Path '{}' is not an archive file",
+        archive.display()
+    );
+    println!("Listing an archive at path '{}'", archive.display());
+    let archive_name = match archive.file_name().and_then(|name| name.to_str()) {
+        Some(name) => name,
+        None => bail!(
+            "Failed to get the archive name from the path '{}'",
+            archive.display()
+        ),
+    };
+
+    let archive_bytes = fs::read(&archive)
+        .await
+        .context("Failed to read the archive bytes")?;
+
+    let header = compression::read_archive_header(archive_name, &mut archive_bytes.as_slice())
+        .await
+        .context("Failed to read the archive header")?;
+
+    let empty_path = Path::new("");
+    println!("-------------------------------");
+
+    let longest_path_in_archive = header
+        .files
+        .iter()
+        .filter_map(|file| Some(file.subpath.as_path(empty_path).to_str()?.len()))
+        .max()
+        .unwrap_or_default()
+        .max(METADATA_FILE_NAME.len());
+
+    for regular_file in &header.files {
+        println!(
+            "File: {:width$} uncompressed size: {} bytes",
+            regular_file.subpath.as_path(empty_path).display(),
+            regular_file.size,
+            width = longest_path_in_archive,
+        )
+    }
+    println!(
+        "File: {:width$} uncompressed size: {} bytes",
+        METADATA_FILE_NAME,
+        header.metadata_file_size,
+        width = longest_path_in_archive,
+    );
+    println!("-------------------------------");
+
+    Ok(())
+}
+
+async fn extract_archive(archive: &Path, target_dir: &Path) -> anyhow::Result<()> {
+    let archive = archive.canonicalize().with_context(|| {
+        format!(
+            "Failed to get the absolute path for the archive path '{}'",
+            archive.display()
+        )
+    })?;
+    ensure!(
+        archive.is_file(),
+        "Path '{}' is not an archive file",
+        archive.display()
+    );
+    let archive_name = match archive.file_name().and_then(|name| name.to_str()) {
+        Some(name) => name,
+        None => bail!(
+            "Failed to get the archive name from the path '{}'",
+            archive.display()
+        ),
+    };
+
+    if !target_dir.exists() {
+        fs::create_dir_all(target_dir).await.with_context(|| {
+            format!(
+                "Failed to create the target dir at path '{}'",
+                target_dir.display()
+            )
+        })?;
+    }
+    let target_dir = target_dir.canonicalize().with_context(|| {
+        format!(
+            "Failed to get the absolute path for the target dir path '{}'",
+            target_dir.display()
+        )
+    })?;
+    ensure!(
+        target_dir.is_dir(),
+        "Path '{}' is not a directory",
+        target_dir.display()
+    );
+    let mut dir_contents = fs::read_dir(&target_dir)
+        .await
+        .context("Failed to list the target directory contents")?;
+    let dir_entry = dir_contents
+        .next_entry()
+        .await
+        .context("Failed to list the target directory contents")?;
+    ensure!(
+        dir_entry.is_none(),
+        "Target directory '{}' is not empty",
+        target_dir.display()
+    );
+
+    println!(
+        "Extracting an archive at path '{}' into directory '{}'",
+        archive.display(),
+        target_dir.display()
+    );
+
+    let mut archive_file = fs::File::open(&archive).await.with_context(|| {
+        format!(
+            "Failed to get the archive name from the path '{}'",
+            archive.display()
+        )
+    })?;
+    let header = compression::read_archive_header(archive_name, &mut archive_file)
+        .await
+        .context("Failed to read the archive header")?;
+    compression::uncompress_with_header(&BTreeSet::new(), &target_dir, header, &mut archive_file)
+        .await
+        .context("Failed to extract the archive")
+}
+
+async fn create_archive(source_dir: &Path, target_dir: &Path) -> anyhow::Result<()> {
+    let source_dir = source_dir.canonicalize().with_context(|| {
+        format!(
+            "Failed to get the absolute path for the source dir path '{}'",
+            source_dir.display()
+        )
+    })?;
+    ensure!(
+        source_dir.is_dir(),
+        "Path '{}' is not a directory",
+        source_dir.display()
+    );
+
+    if !target_dir.exists() {
+        fs::create_dir_all(target_dir).await.with_context(|| {
+            format!(
+                "Failed to create the target dir at path '{}'",
+                target_dir.display()
+            )
+        })?;
+    }
+    let target_dir = target_dir.canonicalize().with_context(|| {
+        format!(
+            "Failed to get the absolute path for the target dir path '{}'",
+            target_dir.display()
+        )
+    })?;
+    ensure!(
+        target_dir.is_dir(),
+        "Path '{}' is not a directory",
+        target_dir.display()
+    );
+
+    println!(
+        "Compressing directory '{}' and creating resulting archive in directory '{}'",
+        source_dir.display(),
+        target_dir.display()
+    );
+
+    let mut metadata_file_contents = None;
+    let mut files_co_archive = Vec::new();
+
+    let mut source_dir_contents = fs::read_dir(&source_dir)
+        .await
+        .context("Failed to read the source directory contents")?;
+
+    while let Some(source_dir_entry) = source_dir_contents
+        .next_entry()
+        .await
+        .context("Failed to read a source dir entry")?
+    {
+        let entry_path = source_dir_entry.path();
+        if entry_path.is_file() {
+            if entry_path.file_name().and_then(|name| name.to_str()) == Some(METADATA_FILE_NAME) {
+                let metadata_bytes = fs::read(entry_path)
+                    .await
+                    .context("Failed to read metata file bytes in the source dir")?;
+                metadata_file_contents = Some(
+                    TimelineMetadata::from_bytes(&metadata_bytes)
+                        .context("Failed to parse metata file contents in the source dir")?,
+                );
+            } else {
+                files_co_archive.push(entry_path);
+            }
+        }
+    }
+
+    let metadata = match metadata_file_contents {
+        Some(metadata) => metadata,
+        None => bail!(
+            "No metadata file found in the source dir '{}', cannot create the archive",
+            source_dir.display()
+        ),
+    };
+
+    let _ = compression::archive_files_as_stream(
+        &source_dir,
+        files_co_archive.iter(),
+        &metadata,
+        move |mut archive_streamer, archive_name| async move {
+            let archive_target = target_dir.join(&archive_name);
+            let mut archive_file = fs::File::create(&archive_target).await?;
+            io::copy(&mut archive_streamer, &mut archive_file).await?;
+            Ok(archive_target)
+        },
+    )
+    .await
+    .context("Failed to create an archive")?;
+
+    Ok(())
+}

pageserver/src/bin/update_metadata.rs

@@ -14,20 +14,20 @@ fn main() -> Result<()> {
         .about("Dump or update metadata file")
         .version(GIT_VERSION)
         .arg(
-            Arg::with_name("path")
+            Arg::new("path")
                 .help("Path to metadata file")
                 .required(true),
         )
         .arg(
-            Arg::with_name("disk_lsn")
-                .short("d")
+            Arg::new("disk_lsn")
+                .short('d')
                 .long("disk_lsn")
                 .takes_value(true)
                 .help("Replace disk constistent lsn"),
         )
         .arg(
-            Arg::with_name("prev_lsn")
-                .short("p")
+            Arg::new("prev_lsn")
+                .short('p')
                 .long("prev_lsn")
                 .takes_value(true)
                 .help("Previous record LSN"),

pageserver/src/branches.rs

@@ -4,7 +4,7 @@
 // TODO: move all paths construction to conf impl
 //
 
-use anyhow::{anyhow, bail, Context, Result};
+use anyhow::{bail, Context, Result};
 use postgres_ffi::ControlFileData;
 use serde::{Deserialize, Serialize};
 use std::{
@@ -45,14 +45,16 @@ impl BranchInfo {
         repo: &Arc<dyn Repository>,
         include_non_incremental_logical_size: bool,
     ) -> Result<Self> {
-        let name = path
-            .as_ref()
-            .file_name()
-            .unwrap()
-            .to_str()
-            .unwrap()
-            .to_string();
-        let timeline_id = std::fs::read_to_string(path)?.parse::<ZTimelineId>()?;
+        let path = path.as_ref();
+        let name = path.file_name().unwrap().to_string_lossy().to_string();
+        let timeline_id = std::fs::read_to_string(path)
+            .with_context(|| {
+                format!(
+                    "Failed to read branch file contents at path '{}'",
+                    path.display()
+                )
+            })?
+            .parse::<ZTimelineId>()?;
 
         let timeline = match repo.get_timeline(timeline_id)? {
             RepositoryTimeline::Local(local_entry) => local_entry,
@@ -116,7 +118,7 @@ pub fn init_pageserver(conf: &'static PageServerConf, create_tenant: Option<&str
     if let Some(tenantid) = create_tenant {
         let tenantid = ZTenantId::from_str(tenantid)?;
         println!("initializing tenantid {}", tenantid);
-        create_repo(conf, tenantid, dummy_redo_mgr).with_context(|| "failed to create repo")?;
+        create_repo(conf, tenantid, dummy_redo_mgr).context("failed to create repo")?;
     }
     crashsafe_dir::create_dir_all(conf.tenants_path())?;
 
@@ -195,7 +197,7 @@ fn run_initdb(conf: &'static PageServerConf, initdbpath: &Path) -> Result<()> {
         .env("DYLD_LIBRARY_PATH", conf.pg_lib_dir().to_str().unwrap())
         .stdout(Stdio::null())
         .output()
-        .with_context(|| "failed to execute initdb")?;
+        .context("failed to execute initdb")?;
     if !initdb_output.status.success() {
         anyhow::bail!(
             "initdb failed: '{}'",
@@ -306,7 +308,7 @@ pub(crate) fn create_branch(
     let timeline = repo
         .get_timeline(startpoint.timelineid)?
         .local_timeline()
-        .ok_or_else(|| anyhow!("Cannot branch off the timeline that's not present locally"))?;
+        .context("Cannot branch off the timeline that's not present locally")?;
     if startpoint.lsn == Lsn(0) {
         // Find end of WAL on the old timeline
         let end_of_wal = timeline.get_last_record_lsn();
@@ -322,12 +324,13 @@ pub(crate) fn create_branch(
         timeline.wait_lsn(startpoint.lsn)?;
     }
     startpoint.lsn = startpoint.lsn.align();
-    if timeline.get_start_lsn() > startpoint.lsn {
+    if timeline.get_ancestor_lsn() > startpoint.lsn {
+        // can we safely just branch from the ancestor instead?
         anyhow::bail!(
-            "invalid startpoint {} for the branch {}: less than timeline start {}",
+            "invalid startpoint {} for the branch {}: less than timeline ancestor lsn {:?}",
             startpoint.lsn,
             branchname,
-            timeline.get_start_lsn()
+            timeline.get_ancestor_lsn()
         );
     }
 
@@ -381,14 +384,11 @@ fn parse_point_in_time(
     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
-    }
+    let lsn = strings
+        .next()
+        .map(Lsn::from_str)
+        .transpose()
+        .context("invalid LSN in point-in-time specification")?;
 
     // Check if it's a tag
     if lsn.is_none() {

pageserver/src/config.rs

@@ -4,11 +4,11 @@
 //! file, or on the command line.
 //! See also `settings.md` for better description on every parameter.
 
-use anyhow::{anyhow, bail, ensure, Context, Result};
+use anyhow::{bail, ensure, Context, Result};
 use toml_edit;
 use toml_edit::{Document, Item};
 use zenith_utils::postgres_backend::AuthType;
-use zenith_utils::zid::{ZTenantId, ZTimelineId};
+use zenith_utils::zid::{ZNodeId, ZTenantId, ZTimelineId};
 
 use std::convert::TryInto;
 use std::env;
@@ -72,6 +72,10 @@ pub mod defaults {
 
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct PageServerConf {
+    // Identifier of that particular pageserver so e g safekeepers
+    // can safely distinguish different pageservers
+    pub id: ZNodeId,
+
     /// Example (default): 127.0.0.1:64000
     pub listen_pg_addr: String,
     /// Example (default): 127.0.0.1:9898
@@ -106,6 +110,184 @@ pub struct PageServerConf {
     pub remote_storage_config: Option<RemoteStorageConfig>,
 }
 
+// use dedicated enum for builder to better indicate the intention
+// and avoid possible confusion with nested options
+pub enum BuilderValue<T> {
+    Set(T),
+    NotSet,
+}
+
+impl<T> BuilderValue<T> {
+    pub fn ok_or<E>(self, err: E) -> Result<T, E> {
+        match self {
+            Self::Set(v) => Ok(v),
+            Self::NotSet => Err(err),
+        }
+    }
+}
+
+// needed to simplify config construction
+struct PageServerConfigBuilder {
+    listen_pg_addr: BuilderValue<String>,
+
+    listen_http_addr: BuilderValue<String>,
+
+    checkpoint_distance: BuilderValue<u64>,
+    checkpoint_period: BuilderValue<Duration>,
+
+    gc_horizon: BuilderValue<u64>,
+    gc_period: BuilderValue<Duration>,
+    superuser: BuilderValue<String>,
+
+    page_cache_size: BuilderValue<usize>,
+    max_file_descriptors: BuilderValue<usize>,
+
+    workdir: BuilderValue<PathBuf>,
+
+    pg_distrib_dir: BuilderValue<PathBuf>,
+
+    auth_type: BuilderValue<AuthType>,
+
+    //
+    auth_validation_public_key_path: BuilderValue<Option<PathBuf>>,
+    remote_storage_config: BuilderValue<Option<RemoteStorageConfig>>,
+
+    id: BuilderValue<ZNodeId>,
+}
+
+impl Default for PageServerConfigBuilder {
+    fn default() -> Self {
+        use self::BuilderValue::*;
+        use defaults::*;
+        Self {
+            listen_pg_addr: Set(DEFAULT_PG_LISTEN_ADDR.to_string()),
+            listen_http_addr: Set(DEFAULT_HTTP_LISTEN_ADDR.to_string()),
+            checkpoint_distance: Set(DEFAULT_CHECKPOINT_DISTANCE),
+            checkpoint_period: Set(humantime::parse_duration(DEFAULT_CHECKPOINT_PERIOD)
+                .expect("cannot parse default checkpoint period")),
+            gc_horizon: Set(DEFAULT_GC_HORIZON),
+            gc_period: Set(humantime::parse_duration(DEFAULT_GC_PERIOD)
+                .expect("cannot parse default gc period")),
+            superuser: Set(DEFAULT_SUPERUSER.to_string()),
+            page_cache_size: Set(DEFAULT_PAGE_CACHE_SIZE),
+            max_file_descriptors: Set(DEFAULT_MAX_FILE_DESCRIPTORS),
+            workdir: Set(PathBuf::new()),
+            pg_distrib_dir: Set(env::current_dir()
+                .expect("cannot access current directory")
+                .join("tmp_install")),
+            auth_type: Set(AuthType::Trust),
+            auth_validation_public_key_path: Set(None),
+            remote_storage_config: Set(None),
+            id: NotSet,
+        }
+    }
+}
+
+impl PageServerConfigBuilder {
+    pub fn listen_pg_addr(&mut self, listen_pg_addr: String) {
+        self.listen_pg_addr = BuilderValue::Set(listen_pg_addr)
+    }
+
+    pub fn listen_http_addr(&mut self, listen_http_addr: String) {
+        self.listen_http_addr = BuilderValue::Set(listen_http_addr)
+    }
+
+    pub fn checkpoint_distance(&mut self, checkpoint_distance: u64) {
+        self.checkpoint_distance = BuilderValue::Set(checkpoint_distance)
+    }
+
+    pub fn checkpoint_period(&mut self, checkpoint_period: Duration) {
+        self.checkpoint_period = BuilderValue::Set(checkpoint_period)
+    }
+
+    pub fn gc_horizon(&mut self, gc_horizon: u64) {
+        self.gc_horizon = BuilderValue::Set(gc_horizon)
+    }
+
+    pub fn gc_period(&mut self, gc_period: Duration) {
+        self.gc_period = BuilderValue::Set(gc_period)
+    }
+
+    pub fn superuser(&mut self, superuser: String) {
+        self.superuser = BuilderValue::Set(superuser)
+    }
+
+    pub fn page_cache_size(&mut self, page_cache_size: usize) {
+        self.page_cache_size = BuilderValue::Set(page_cache_size)
+    }
+
+    pub fn max_file_descriptors(&mut self, max_file_descriptors: usize) {
+        self.max_file_descriptors = BuilderValue::Set(max_file_descriptors)
+    }
+
+    pub fn workdir(&mut self, workdir: PathBuf) {
+        self.workdir = BuilderValue::Set(workdir)
+    }
+
+    pub fn pg_distrib_dir(&mut self, pg_distrib_dir: PathBuf) {
+        self.pg_distrib_dir = BuilderValue::Set(pg_distrib_dir)
+    }
+
+    pub fn auth_type(&mut self, auth_type: AuthType) {
+        self.auth_type = BuilderValue::Set(auth_type)
+    }
+
+    pub fn auth_validation_public_key_path(
+        &mut self,
+        auth_validation_public_key_path: Option<PathBuf>,
+    ) {
+        self.auth_validation_public_key_path = BuilderValue::Set(auth_validation_public_key_path)
+    }
+
+    pub fn remote_storage_config(&mut self, remote_storage_config: Option<RemoteStorageConfig>) {
+        self.remote_storage_config = BuilderValue::Set(remote_storage_config)
+    }
+
+    pub fn id(&mut self, node_id: ZNodeId) {
+        self.id = BuilderValue::Set(node_id)
+    }
+
+    pub fn build(self) -> Result<PageServerConf> {
+        Ok(PageServerConf {
+            listen_pg_addr: self
+                .listen_pg_addr
+                .ok_or(anyhow::anyhow!("missing listen_pg_addr"))?,
+            listen_http_addr: self
+                .listen_http_addr
+                .ok_or(anyhow::anyhow!("missing listen_http_addr"))?,
+            checkpoint_distance: self
+                .checkpoint_distance
+                .ok_or(anyhow::anyhow!("missing checkpoint_distance"))?,
+            checkpoint_period: self
+                .checkpoint_period
+                .ok_or(anyhow::anyhow!("missing checkpoint_period"))?,
+            gc_horizon: self
+                .gc_horizon
+                .ok_or(anyhow::anyhow!("missing gc_horizon"))?,
+            gc_period: self.gc_period.ok_or(anyhow::anyhow!("missing gc_period"))?,
+            superuser: self.superuser.ok_or(anyhow::anyhow!("missing superuser"))?,
+            page_cache_size: self
+                .page_cache_size
+                .ok_or(anyhow::anyhow!("missing page_cache_size"))?,
+            max_file_descriptors: self
+                .max_file_descriptors
+                .ok_or(anyhow::anyhow!("missing max_file_descriptors"))?,
+            workdir: self.workdir.ok_or(anyhow::anyhow!("missing workdir"))?,
+            pg_distrib_dir: self
+                .pg_distrib_dir
+                .ok_or(anyhow::anyhow!("missing pg_distrib_dir"))?,
+            auth_type: self.auth_type.ok_or(anyhow::anyhow!("missing auth_type"))?,
+            auth_validation_public_key_path: self
+                .auth_validation_public_key_path
+                .ok_or(anyhow::anyhow!("missing auth_validation_public_key_path"))?,
+            remote_storage_config: self
+                .remote_storage_config
+                .ok_or(anyhow::anyhow!("missing remote_storage_config"))?,
+            id: self.id.ok_or(anyhow::anyhow!("missing id"))?,
+        })
+    }
+}
+
 /// External backup storage configuration, enough for creating a client for that storage.
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct RemoteStorageConfig {
@@ -135,6 +317,8 @@ pub struct S3Config {
     pub bucket_name: String,
     /// The region where the bucket is located at.
     pub bucket_region: String,
+    /// A "subfolder" in the bucket, to use the same bucket separately by multiple pageservers at once.
+    pub prefix_in_bucket: Option<String>,
     /// "Login" to use when connecting to bucket.
     /// Can be empty for cases like AWS k8s IAM
     /// where we can allow certain pods to connect
@@ -142,6 +326,13 @@ pub struct S3Config {
     pub access_key_id: Option<String>,
     /// "Password" to use when connecting to bucket.
     pub secret_access_key: Option<String>,
+    /// A base URL to send S3 requests to.
+    /// By default, the endpoint is derived from a region name, assuming it's
+    /// an AWS S3 region name, erroring on wrong region name.
+    /// Endpoint provides a way to support other S3 flavors and their regions.
+    ///
+    /// Example: `http://127.0.0.1:5000`
+    pub endpoint: Option<String>,
 }
 
 impl std::fmt::Debug for S3Config {
@@ -149,6 +340,7 @@ impl std::fmt::Debug for S3Config {
         f.debug_struct("S3Config")
             .field("bucket_name", &self.bucket_name)
             .field("bucket_region", &self.bucket_region)
+            .field("prefix_in_bucket", &self.prefix_in_bucket)
             .finish()
     }
 }
@@ -211,57 +403,39 @@ impl PageServerConf {
     ///
     /// This leaves any options not present in the file in the built-in defaults.
     pub fn parse_and_validate(toml: &Document, workdir: &Path) -> Result<Self> {
-        use defaults::*;
-
-        let mut conf = PageServerConf {
-            workdir: workdir.to_path_buf(),
-
-            listen_pg_addr: DEFAULT_PG_LISTEN_ADDR.to_string(),
-            listen_http_addr: DEFAULT_HTTP_LISTEN_ADDR.to_string(),
-            checkpoint_distance: DEFAULT_CHECKPOINT_DISTANCE,
-            checkpoint_period: humantime::parse_duration(DEFAULT_CHECKPOINT_PERIOD)?,
-            gc_horizon: DEFAULT_GC_HORIZON,
-            gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)?,
-            page_cache_size: DEFAULT_PAGE_CACHE_SIZE,
-            max_file_descriptors: DEFAULT_MAX_FILE_DESCRIPTORS,
-
-            pg_distrib_dir: PathBuf::new(),
-            auth_validation_public_key_path: None,
-            auth_type: AuthType::Trust,
-
-            remote_storage_config: None,
-
-            superuser: DEFAULT_SUPERUSER.to_string(),
-        };
+        let mut builder = PageServerConfigBuilder::default();
+        builder.workdir(workdir.to_owned());
 
         for (key, item) in toml.iter() {
             match key {
-                "listen_pg_addr" => conf.listen_pg_addr = parse_toml_string(key, item)?,
-                "listen_http_addr" => conf.listen_http_addr = parse_toml_string(key, item)?,
-                "checkpoint_distance" => conf.checkpoint_distance = parse_toml_u64(key, item)?,
-                "checkpoint_period" => conf.checkpoint_period = parse_toml_duration(key, item)?,
-                "gc_horizon" => conf.gc_horizon = parse_toml_u64(key, item)?,
-                "gc_period" => conf.gc_period = parse_toml_duration(key, item)?,
-                "initial_superuser_name" => conf.superuser = parse_toml_string(key, item)?,
-                "page_cache_size" => conf.page_cache_size = parse_toml_u64(key, item)? as usize,
+                "listen_pg_addr" => builder.listen_pg_addr(parse_toml_string(key, item)?),
+                "listen_http_addr" => builder.listen_http_addr(parse_toml_string(key, item)?),
+                "checkpoint_distance" => builder.checkpoint_distance(parse_toml_u64(key, item)?),
+                "checkpoint_period" => builder.checkpoint_period(parse_toml_duration(key, item)?),
+                "gc_horizon" => builder.gc_horizon(parse_toml_u64(key, item)?),
+                "gc_period" => builder.gc_period(parse_toml_duration(key, item)?),
+                "initial_superuser_name" => builder.superuser(parse_toml_string(key, item)?),
+                "page_cache_size" => builder.page_cache_size(parse_toml_u64(key, item)? as usize),
                 "max_file_descriptors" => {
-                    conf.max_file_descriptors = parse_toml_u64(key, item)? as usize
+                    builder.max_file_descriptors(parse_toml_u64(key, item)? as usize)
                 }
                 "pg_distrib_dir" => {
-                    conf.pg_distrib_dir = PathBuf::from(parse_toml_string(key, item)?)
+                    builder.pg_distrib_dir(PathBuf::from(parse_toml_string(key, item)?))
                 }
-                "auth_validation_public_key_path" => {
-                    conf.auth_validation_public_key_path =
-                        Some(PathBuf::from(parse_toml_string(key, item)?))
-                }
-                "auth_type" => conf.auth_type = parse_toml_auth_type(key, item)?,
+                "auth_validation_public_key_path" => builder.auth_validation_public_key_path(Some(
+                    PathBuf::from(parse_toml_string(key, item)?),
+                )),
+                "auth_type" => builder.auth_type(parse_toml_auth_type(key, item)?),
                 "remote_storage" => {
-                    conf.remote_storage_config = Some(Self::parse_remote_storage_config(item)?)
+                    builder.remote_storage_config(Some(Self::parse_remote_storage_config(item)?))
                 }
+                "id" => builder.id(ZNodeId(parse_toml_u64(key, item)?)),
                 _ => bail!("unrecognized pageserver option '{}'", key),
             }
         }
 
+        let mut conf = builder.build().context("invalid config")?;
+
         if conf.auth_type == AuthType::ZenithJWT {
             let auth_validation_public_key_path = conf
                 .auth_validation_public_key_path
@@ -275,9 +449,6 @@ impl PageServerConf {
             );
         }
 
-        if conf.pg_distrib_dir == PathBuf::new() {
-            conf.pg_distrib_dir = env::current_dir()?.join("tmp_install")
-        };
         if !conf.pg_distrib_dir.join("bin/postgres").exists() {
             bail!(
                 "Can't find postgres binary at {}",
@@ -303,9 +474,7 @@ impl PageServerConf {
                 })
                 .ok()
                 .and_then(NonZeroUsize::new)
-                .ok_or_else(|| {
-                    anyhow!("'max_concurrent_sync' must be a non-zero positive integer")
-                })?
+                .context("'max_concurrent_sync' must be a non-zero positive integer")?
         } else {
             NonZeroUsize::new(defaults::DEFAULT_REMOTE_STORAGE_MAX_CONCURRENT_SYNC).unwrap()
         };
@@ -318,7 +487,7 @@ impl PageServerConf {
                 })
                 .ok()
                 .and_then(NonZeroU32::new)
-                .ok_or_else(|| anyhow!("'max_sync_errors' must be a non-zero positive integer"))?
+                .context("'max_sync_errors' must be a non-zero positive integer")?
         } else {
             NonZeroU32::new(defaults::DEFAULT_REMOTE_STORAGE_MAX_SYNC_ERRORS).unwrap()
         };
@@ -332,18 +501,30 @@ impl PageServerConf {
                 bail!("'bucket_name' option is mandatory if 'bucket_region' is given ")
             }
             (None, Some(bucket_name), Some(bucket_region)) => RemoteStorageKind::AwsS3(S3Config {
-                bucket_name: bucket_name.as_str().unwrap().to_string(),
-                bucket_region: bucket_region.as_str().unwrap().to_string(),
+                bucket_name: parse_toml_string("bucket_name", bucket_name)?,
+                bucket_region: parse_toml_string("bucket_region", bucket_region)?,
                 access_key_id: toml
                     .get("access_key_id")
-                    .map(|x| x.as_str().unwrap().to_string()),
+                    .map(|access_key_id| parse_toml_string("access_key_id", access_key_id))
+                    .transpose()?,
                 secret_access_key: toml
                     .get("secret_access_key")
-                    .map(|x| x.as_str().unwrap().to_string()),
+                    .map(|secret_access_key| {
+                        parse_toml_string("secret_access_key", secret_access_key)
+                    })
+                    .transpose()?,
+                prefix_in_bucket: toml
+                    .get("prefix_in_bucket")
+                    .map(|prefix_in_bucket| parse_toml_string("prefix_in_bucket", prefix_in_bucket))
+                    .transpose()?,
+                endpoint: toml
+                    .get("endpoint")
+                    .map(|endpoint| parse_toml_string("endpoint", endpoint))
+                    .transpose()?,
             }),
-            (Some(local_path), None, None) => {
-                RemoteStorageKind::LocalFs(PathBuf::from(local_path.as_str().unwrap()))
-            }
+            (Some(local_path), None, None) => RemoteStorageKind::LocalFs(PathBuf::from(
+                parse_toml_string("local_path", local_path)?,
+            )),
             (Some(_), Some(_), _) => bail!("local_path and bucket_name are mutually exclusive"),
         };
 
@@ -362,6 +543,7 @@ impl PageServerConf {
     #[cfg(test)]
     pub fn dummy_conf(repo_dir: PathBuf) -> Self {
         PageServerConf {
+            id: ZNodeId(0),
             checkpoint_distance: defaults::DEFAULT_CHECKPOINT_DISTANCE,
             checkpoint_period: Duration::from_secs(10),
             gc_horizon: defaults::DEFAULT_GC_HORIZON,
@@ -385,7 +567,7 @@ impl PageServerConf {
 fn parse_toml_string(name: &str, item: &Item) -> Result<String> {
     let s = item
         .as_str()
-        .ok_or_else(|| anyhow!("configure option {} is not a string", name))?;
+        .with_context(|| format!("configure option {} is not a string", name))?;
     Ok(s.to_string())
 }
 
@@ -394,7 +576,7 @@ fn parse_toml_u64(name: &str, item: &Item) -> Result<u64> {
     // for our use, though.
     let i: i64 = item
         .as_integer()
-        .ok_or_else(|| anyhow!("configure option {} is not an integer", name))?;
+        .with_context(|| format!("configure option {} is not an integer", name))?;
     if i < 0 {
         bail!("configure option {} cannot be negative", name);
     }
@@ -404,7 +586,7 @@ fn parse_toml_u64(name: &str, item: &Item) -> Result<u64> {
 fn parse_toml_duration(name: &str, item: &Item) -> Result<Duration> {
     let s = item
         .as_str()
-        .ok_or_else(|| anyhow!("configure option {} is not a string", name))?;
+        .with_context(|| format!("configure option {} is not a string", name))?;
 
     Ok(humantime::parse_duration(s)?)
 }
@@ -412,7 +594,7 @@ fn parse_toml_duration(name: &str, item: &Item) -> Result<Duration> {
 fn parse_toml_auth_type(name: &str, item: &Item) -> Result<AuthType> {
     let v = item
         .as_str()
-        .ok_or_else(|| anyhow!("configure option {} is not a string", name))?;
+        .with_context(|| format!("configure option {} is not a string", name))?;
     AuthType::from_str(v)
 }
 
@@ -441,15 +623,16 @@ max_file_descriptors = 333
 
 # initial superuser role name to use when creating a new tenant
 initial_superuser_name = 'zzzz'
+id = 10
 
-    "#;
+"#;
 
     #[test]
     fn parse_defaults() -> anyhow::Result<()> {
         let tempdir = tempdir()?;
         let (workdir, pg_distrib_dir) = prepare_fs(&tempdir)?;
         // we have to create dummy pathes to overcome the validation errors
-        let config_string = format!("pg_distrib_dir='{}'", pg_distrib_dir.display());
+        let config_string = format!("pg_distrib_dir='{}'\nid=10", pg_distrib_dir.display());
         let toml = config_string.parse()?;
 
         let parsed_config =
@@ -460,6 +643,7 @@ initial_superuser_name = 'zzzz'
         assert_eq!(
             parsed_config,
             PageServerConf {
+                id: ZNodeId(10),
                 listen_pg_addr: defaults::DEFAULT_PG_LISTEN_ADDR.to_string(),
                 listen_http_addr: defaults::DEFAULT_HTTP_LISTEN_ADDR.to_string(),
                 checkpoint_distance: defaults::DEFAULT_CHECKPOINT_DISTANCE,
@@ -501,6 +685,7 @@ initial_superuser_name = 'zzzz'
         assert_eq!(
             parsed_config,
             PageServerConf {
+                id: ZNodeId(10),
                 listen_pg_addr: "127.0.0.1:64000".to_string(),
                 listen_http_addr: "127.0.0.1:9898".to_string(),
                 checkpoint_distance: 111,
@@ -585,8 +770,10 @@ pg_distrib_dir='{}'
 
         let bucket_name = "some-sample-bucket".to_string();
         let bucket_region = "eu-north-1".to_string();
+        let prefix_in_bucket = "test_prefix".to_string();
         let access_key_id = "SOMEKEYAAAAASADSAH*#".to_string();
         let secret_access_key = "SOMEsEcReTsd292v".to_string();
+        let endpoint = "http://localhost:5000".to_string();
         let max_concurrent_sync = NonZeroUsize::new(111).unwrap();
         let max_sync_errors = NonZeroU32::new(222).unwrap();
 
@@ -597,13 +784,15 @@ max_concurrent_sync = {}
 max_sync_errors = {}
 bucket_name = '{}'
 bucket_region = '{}'
+prefix_in_bucket = '{}'
 access_key_id = '{}'
-secret_access_key = '{}'"#,
-                max_concurrent_sync, max_sync_errors, bucket_name, bucket_region, access_key_id, secret_access_key
+secret_access_key = '{}'
+endpoint = '{}'"#,
+                max_concurrent_sync, max_sync_errors, bucket_name, bucket_region, prefix_in_bucket, access_key_id, secret_access_key, endpoint
             ),
             format!(
-                "remote_storage={{max_concurrent_sync = {}, max_sync_errors = {}, bucket_name='{}', bucket_region='{}', access_key_id='{}', secret_access_key='{}'}}",
-                max_concurrent_sync, max_sync_errors, bucket_name, bucket_region, access_key_id, secret_access_key
+                "remote_storage={{max_concurrent_sync={}, max_sync_errors={}, bucket_name='{}', bucket_region='{}', prefix_in_bucket='{}', access_key_id='{}', secret_access_key='{}', endpoint='{}'}}",
+                max_concurrent_sync, max_sync_errors, bucket_name, bucket_region, prefix_in_bucket, access_key_id, secret_access_key, endpoint
             ),
         ];
 
@@ -637,6 +826,8 @@ pg_distrib_dir='{}'
                         bucket_region: bucket_region.clone(),
                         access_key_id: Some(access_key_id.clone()),
                         secret_access_key: Some(secret_access_key.clone()),
+                        prefix_in_bucket: Some(prefix_in_bucket.clone()),
+                        endpoint: Some(endpoint.clone())
                     }),
                 },
                 "Remote storage config should correctly parse the S3 config"

pageserver/src/http/models.rs

@@ -1,6 +1,7 @@
 use serde::{Deserialize, Serialize};
 
 use crate::ZTenantId;
+use zenith_utils::zid::ZNodeId;
 
 #[derive(Serialize, Deserialize)]
 pub struct BranchCreateRequest {
@@ -15,3 +16,8 @@ pub struct TenantCreateRequest {
     #[serde(with = "hex")]
     pub tenant_id: ZTenantId,
 }
+
+#[derive(Serialize)]
+pub struct StatusResponse {
+    pub id: ZNodeId,
+}

pageserver/src/http/openapi_spec.yml

@@ -17,6 +17,11 @@ paths:
             application/json:
               schema:
                 type: object
+                required:
+                - id
+                properties:
+                  id:
+                    type: integer
   /v1/timeline/{tenant_id}:
     parameters:
       - name: tenant_id
@@ -234,9 +239,7 @@ paths:
           content:
             application/json:
               schema:
-                type: array
-                items:
-                  $ref: "#/components/schemas/BranchInfo"
+                $ref: "#/components/schemas/BranchInfo"
         "400":
           description: Malformed branch create request
           content:
@@ -370,12 +373,15 @@ components:
           format: hex
         ancestor_id:
           type: string
+          format: hex
         ancestor_lsn:
           type: string
         current_logical_size:
           type: integer
         current_logical_size_non_incremental:
           type: integer
+        latest_valid_lsn:
+          type: integer
     TimelineInfo:
       type: object
       required:

pageserver/src/http/routes.rs

@@ -1,10 +1,8 @@
 use std::sync::Arc;
 
-use anyhow::{bail, Context, Result};
-use hyper::header;
+use anyhow::{Context, Result};
 use hyper::StatusCode;
 use hyper::{Body, Request, Response, Uri};
-use routerify::{ext::RequestExt, RouterBuilder};
 use serde::Serialize;
 use tracing::*;
 use zenith_utils::auth::JwtAuth;
@@ -19,12 +17,15 @@ use zenith_utils::http::{
     request::get_request_param,
     request::parse_request_param,
 };
+use zenith_utils::http::{RequestExt, RouterBuilder};
 use zenith_utils::lsn::Lsn;
 use zenith_utils::zid::{opt_display_serde, ZTimelineId};
 
 use super::models::BranchCreateRequest;
+use super::models::StatusResponse;
 use super::models::TenantCreateRequest;
 use crate::branches::BranchInfo;
+use crate::repository::RepositoryTimeline;
 use crate::repository::TimelineSyncState;
 use crate::{branches, config::PageServerConf, tenant_mgr, ZTenantId};
 
@@ -63,12 +64,12 @@ fn get_config(request: &Request<Body>) -> &'static PageServerConf {
 }
 
 // healthcheck handler
-async fn status_handler(_: Request<Body>) -> Result<Response<Body>, ApiError> {
-    Ok(Response::builder()
-        .status(StatusCode::OK)
-        .header(header::CONTENT_TYPE, "application/json")
-        .body(Body::from("{}"))
-        .map_err(ApiError::from_err)?)
+async fn status_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
+    let config = get_config(&request);
+    Ok(json_response(
+        StatusCode::OK,
+        StatusResponse { id: config.id },
+    )?)
 }
 
 async fn branch_create_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -190,18 +191,26 @@ async fn timeline_list_handler(request: Request<Body>) -> Result<Response<Body>,
 }
 
 #[derive(Debug, Serialize)]
-struct TimelineInfo {
-    #[serde(with = "hex")]
-    timeline_id: ZTimelineId,
-    #[serde(with = "hex")]
-    tenant_id: ZTenantId,
-    #[serde(with = "opt_display_serde")]
-    ancestor_timeline_id: Option<ZTimelineId>,
-    last_record_lsn: Lsn,
-    prev_record_lsn: Lsn,
-    start_lsn: Lsn,
-    disk_consistent_lsn: Lsn,
-    timeline_state: Option<TimelineSyncState>,
+#[serde(tag = "type")]
+enum TimelineInfo {
+    Local {
+        #[serde(with = "hex")]
+        timeline_id: ZTimelineId,
+        #[serde(with = "hex")]
+        tenant_id: ZTenantId,
+        #[serde(with = "opt_display_serde")]
+        ancestor_timeline_id: Option<ZTimelineId>,
+        last_record_lsn: Lsn,
+        prev_record_lsn: Lsn,
+        disk_consistent_lsn: Lsn,
+        timeline_state: Option<TimelineSyncState>,
+    },
+    Remote {
+        #[serde(with = "hex")]
+        timeline_id: ZTimelineId,
+        #[serde(with = "hex")]
+        tenant_id: ZTenantId,
+    },
 }
 
 async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -215,19 +224,21 @@ async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body
             info_span!("timeline_detail_handler", tenant = %tenant_id, timeline = %timeline_id)
                 .entered();
         let repo = tenant_mgr::get_repository_for_tenant(tenant_id)?;
-        match repo.get_timeline(timeline_id)?.local_timeline() {
-            None => bail!("Timeline with id {} is not present locally", timeline_id),
-            Some(timeline) => Ok::<_, anyhow::Error>(TimelineInfo {
+        Ok::<_, anyhow::Error>(match repo.get_timeline(timeline_id)?.local_timeline() {
+            None => TimelineInfo::Remote {
+                timeline_id,
+                tenant_id,
+            },
+            Some(timeline) => TimelineInfo::Local {
                 timeline_id,
                 tenant_id,
                 ancestor_timeline_id: timeline.get_ancestor_timeline_id(),
                 disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
                 last_record_lsn: timeline.get_last_record_lsn(),
                 prev_record_lsn: timeline.get_prev_record_lsn(),
-                start_lsn: timeline.get_start_lsn(),
                 timeline_state: repo.get_timeline_state(timeline_id),
-            }),
-        }
+            },
+        })
     })
     .await
     .map_err(ApiError::from_err)??;
@@ -235,6 +246,58 @@ async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body
     Ok(json_response(StatusCode::OK, response_data)?)
 }
 
+async fn timeline_attach_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
+    let tenant_id: ZTenantId = parse_request_param(&request, "tenant_id")?;
+    check_permission(&request, Some(tenant_id))?;
+
+    let timeline_id: ZTimelineId = parse_request_param(&request, "timeline_id")?;
+
+    tokio::task::spawn_blocking(move || {
+        let _enter =
+            info_span!("timeline_attach_handler", tenant = %tenant_id, timeline = %timeline_id)
+                .entered();
+        let repo = tenant_mgr::get_repository_for_tenant(tenant_id)?;
+        match repo.get_timeline(timeline_id)? {
+            RepositoryTimeline::Local(_) => {
+                anyhow::bail!("Timeline with id {} is already local", timeline_id)
+            }
+            RepositoryTimeline::Remote {
+                id: _,
+                disk_consistent_lsn: _,
+            } => {
+                // FIXME (rodionov) get timeline already schedules timeline for download, and duplicate tasks can cause errors
+                //  first should be fixed in https://github.com/zenithdb/zenith/issues/997
+                // TODO (rodionov) change timeline state to awaits download (incapsulate it somewhere in the repo)
+                // TODO (rodionov) can we safely request replication on the timeline before sync is completed? (can be implemented on top of the #997)
+                Ok(())
+            }
+        }
+    })
+    .await
+    .map_err(ApiError::from_err)??;
+
+    Ok(json_response(StatusCode::ACCEPTED, ())?)
+}
+
+async fn timeline_detach_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
+    let tenant_id: ZTenantId = parse_request_param(&request, "tenant_id")?;
+    check_permission(&request, Some(tenant_id))?;
+
+    let timeline_id: ZTimelineId = parse_request_param(&request, "timeline_id")?;
+
+    tokio::task::spawn_blocking(move || {
+        let _enter =
+            info_span!("timeline_detach_handler", tenant = %tenant_id, timeline = %timeline_id)
+                .entered();
+        let repo = tenant_mgr::get_repository_for_tenant(tenant_id)?;
+        repo.detach_timeline(timeline_id)
+    })
+    .await
+    .map_err(ApiError::from_err)??;
+
+    Ok(json_response(StatusCode::OK, ())?)
+}
+
 async fn tenant_list_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
     // check for management permission
     check_permission(&request, None)?;
@@ -255,13 +318,13 @@ async fn tenant_create_handler(mut request: Request<Body>) -> Result<Response<Bo
 
     let request_data: TenantCreateRequest = json_request(&mut request).await?;
 
-    let response_data = tokio::task::spawn_blocking(move || {
+    tokio::task::spawn_blocking(move || {
         let _enter = info_span!("tenant_create", tenant = %request_data.tenant_id).entered();
         tenant_mgr::create_repository_for_tenant(get_config(&request), request_data.tenant_id)
     })
     .await
     .map_err(ApiError::from_err)??;
-    Ok(json_response(StatusCode::CREATED, response_data)?)
+    Ok(json_response(StatusCode::CREATED, ())?)
 }
 
 async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -296,6 +359,14 @@ pub fn make_router(
             "/v1/timeline/:tenant_id/:timeline_id",
             timeline_detail_handler,
         )
+        .post(
+            "/v1/timeline/:tenant_id/:timeline_id/attach",
+            timeline_attach_handler,
+        )
+        .post(
+            "/v1/timeline/:tenant_id/:timeline_id/detach",
+            timeline_detach_handler,
+        )
         .get("/v1/branch/:tenant_id", branch_list_handler)
         .get("/v1/branch/:tenant_id/:branch_name", branch_detail_handler)
         .post("/v1/branch", branch_create_handler)

pageserver/src/import_datadir.rs

@@ -7,7 +7,7 @@ use std::fs::File;
 use std::io::{Read, Seek, SeekFrom};
 use std::path::{Path, PathBuf};
 
-use anyhow::{anyhow, bail, ensure, Result};
+use anyhow::{bail, ensure, Context, Result};
 use bytes::Bytes;
 use tracing::*;
 
@@ -126,7 +126,7 @@ pub fn import_timeline_from_postgres_datadir(
     writer.advance_last_record_lsn(lsn);
 
     // We expect the Postgres server to be shut down cleanly.
-    let pg_control = pg_control.ok_or_else(|| anyhow!("pg_control file not found"))?;
+    let pg_control = pg_control.context("pg_control file not found")?;
     ensure!(
         pg_control.state == DBState_DB_SHUTDOWNED,
         "Postgres cluster was not shut down cleanly"

pageserver/src/layered_repository.rs

@@ -11,7 +11,7 @@
 //! parent timeline, and the last LSN that has been written to disk.
 //!
 
-use anyhow::{anyhow, bail, ensure, Context, Result};
+use anyhow::{bail, ensure, Context, Result};
 use bookfile::Book;
 use bytes::Bytes;
 use lazy_static::lazy_static;
@@ -28,7 +28,7 @@ use std::io::Write;
 use std::ops::{Bound::Included, Deref};
 use std::path::{Path, PathBuf};
 use std::sync::atomic::{self, AtomicBool, AtomicUsize};
-use std::sync::{Arc, Mutex, MutexGuard};
+use std::sync::{Arc, Mutex, MutexGuard, RwLock, RwLockReadGuard};
 use std::time::{Duration, Instant};
 
 use self::metadata::{metadata_path, TimelineMetadata, METADATA_FILE_NAME};
@@ -40,8 +40,8 @@ use crate::repository::{
     BlockNumber, GcResult, Repository, RepositoryTimeline, Timeline, TimelineSyncState,
     TimelineWriter, ZenithWalRecord,
 };
-use crate::tenant_mgr;
-use crate::walreceiver;
+use crate::thread_mgr;
+use crate::virtual_file::VirtualFile;
 use crate::walreceiver::IS_WAL_RECEIVER;
 use crate::walredo::WalRedoManager;
 use crate::CheckpointConfig;
@@ -71,7 +71,6 @@ mod storage_layer;
 use delta_layer::DeltaLayer;
 use ephemeral_file::is_ephemeral_file;
 use filename::{DeltaFileName, ImageFileName};
-use global_layer_map::{LayerId, GLOBAL_LAYER_MAP};
 use image_layer::ImageLayer;
 use inmemory_layer::InMemoryLayer;
 use layer_map::LayerMap;
@@ -127,7 +126,13 @@ pub struct LayeredRepository {
     conf: &'static PageServerConf,
     tenantid: ZTenantId,
     timelines: Mutex<HashMap<ZTimelineId, LayeredTimelineEntry>>,
-
+    // This mutex prevents creation of new timelines during GC.
+    // Adding yet another mutex (in addition to `timelines`) is needed because holding
+    // `timelines` mutex during all GC iteration (especially with enforced checkpoint)
+    // may block for a long time `get_timeline`, `get_timelines_state`,... and other operations
+    // with timelines, which in turn may cause dropping replication connection, expiration of wait_for_lsn
+    // timeout...
+    gc_cs: Mutex<()>,
     walredo_mgr: Arc<dyn WalRedoManager + Send + Sync>,
     /// Makes every timeline to backup their files to remote storage.
     upload_relishes: bool,
@@ -161,7 +166,7 @@ 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::new(Lsn(0), None, None, Lsn(0), Lsn(0), initdb_lsn);
+        let metadata = TimelineMetadata::new(Lsn(0), None, None, Lsn(0), initdb_lsn, initdb_lsn);
         Self::save_metadata(self.conf, timelineid, self.tenantid, &metadata, true)?;
 
         let timeline = LayeredTimeline::new(
@@ -186,6 +191,8 @@ impl Repository for LayeredRepository {
         // We need to hold this lock to prevent GC from starting at the same time. GC scans the directory to learn
         // about timelines, so otherwise a race condition is possible, where we create new timeline and GC
         // concurrently removes data that is needed by the new timeline.
+        let _gc_cs = self.gc_cs.lock().unwrap();
+
         let mut timelines = self.timelines.lock().unwrap();
         let src_timeline = match self.get_or_init_timeline(src, &mut timelines)? {
             LayeredTimelineEntry::Local(timeline) => timeline,
@@ -193,9 +200,10 @@ impl Repository for LayeredRepository {
                 bail!("Cannot branch off the timeline {} that's not local", src)
             }
         };
+        let latest_gc_cutoff_lsn = src_timeline.get_latest_gc_cutoff_lsn();
 
         src_timeline
-            .check_lsn_is_in_scope(start_lsn)
+            .check_lsn_is_in_scope(start_lsn, &latest_gc_cutoff_lsn)
             .context("invalid branch start lsn")?;
 
         let RecordLsn {
@@ -223,7 +231,7 @@ impl Repository for LayeredRepository {
             dst_prev,
             Some(src),
             start_lsn,
-            src_timeline.latest_gc_cutoff_lsn.load(),
+            *src_timeline.latest_gc_cutoff_lsn.read().unwrap(),
             src_timeline.initdb_lsn,
         );
         crashsafe_dir::create_dir_all(self.conf.timeline_path(&dst, &self.tenantid))?;
@@ -277,15 +285,42 @@ impl Repository for LayeredRepository {
         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);
+    // Detaches the timeline from the repository.
+    fn detach_timeline(&self, timeline_id: ZTimelineId) -> Result<()> {
+        let mut timelines = self.timelines.lock().unwrap();
+        match timelines.entry(timeline_id) {
+            Entry::Vacant(_) => {
+                bail!("cannot detach non existing timeline");
+            }
+            Entry::Occupied(mut entry) => {
+                let timeline_entry = entry.get_mut();
 
-        let timelines = self.timelines.lock().unwrap();
-        for (timelineid, timeline) in timelines.iter() {
-            shutdown_timeline(self.tenantid, *timelineid, timeline)?;
-        }
+                let timeline = match timeline_entry {
+                    LayeredTimelineEntry::Remote { .. } => {
+                        bail!("cannot detach remote timeline {}", timeline_id);
+                    }
+                    LayeredTimelineEntry::Local(timeline) => timeline,
+                };
 
+                // TODO (rodionov) keep local state in timeline itself (refactoring related to https://github.com/zenithdb/zenith/issues/997 and #1104)
+
+                // FIXME this is local disk consistent lsn, need to keep the latest succesfully uploaded checkpoint lsn in timeline (metadata?)
+                //  https://github.com/zenithdb/zenith/issues/1104
+                let remote_disk_consistent_lsn = timeline.disk_consistent_lsn.load();
+                // reference to timeline is dropped here
+                entry.insert(LayeredTimelineEntry::Remote {
+                    id: timeline_id,
+                    disk_consistent_lsn: remote_disk_consistent_lsn,
+                });
+            }
+        };
+        // Release the lock to shutdown and remove the files without holding it
+        drop(timelines);
+        // shutdown the timeline (this shuts down the walreceiver)
+        thread_mgr::shutdown_threads(None, Some(self.tenantid), Some(timeline_id));
+
+        // remove timeline files (maybe avoid this for ease of debugging if something goes wrong)
+        fs::remove_dir_all(self.conf.timeline_path(&timeline_id, &self.tenantid))?;
         Ok(())
     }
 
@@ -298,9 +333,13 @@ impl Repository for LayeredRepository {
         timeline_id: ZTimelineId,
         new_state: TimelineSyncState,
     ) -> Result<()> {
+        debug!(
+            "set_timeline_state: timeline_id: {}, new_state: {:?}",
+            timeline_id, new_state
+        );
         let mut timelines_accessor = self.timelines.lock().unwrap();
 
-        let timeline_to_shutdown = match new_state {
+        match new_state {
             TimelineSyncState::Ready(_) => {
                 let reloaded_timeline =
                     self.init_local_timeline(timeline_id, &mut timelines_accessor)?;
@@ -318,12 +357,9 @@ impl Repository for LayeredRepository {
                 },
             ),
         };
+        // NOTE we do not delete local data in case timeline became cloud only, this is performed in detach_timeline
         drop(timelines_accessor);
 
-        if let Some(timeline) = timeline_to_shutdown {
-            shutdown_timeline(self.tenantid, timeline_id, &timeline)?;
-        }
-
         Ok(())
     }
 
@@ -349,30 +385,6 @@ impl Repository for LayeredRepository {
     }
 }
 
-fn shutdown_timeline(
-    tenant_id: ZTenantId,
-    timeline_id: ZTimelineId,
-    timeline: &LayeredTimelineEntry,
-) -> Result<(), anyhow::Error> {
-    match timeline {
-        LayeredTimelineEntry::Local(timeline) => {
-            timeline
-                .upload_relishes
-                .store(false, atomic::Ordering::Relaxed);
-            walreceiver::stop_wal_receiver(timeline_id);
-            trace!("repo shutdown. checkpoint timeline {}", timeline_id);
-            // Do not reconstruct pages to reduce shutdown time
-            timeline.checkpoint(CheckpointConfig::Flush)?;
-            //TODO Wait for walredo process to shutdown too
-        }
-        LayeredTimelineEntry::Remote { .. } => warn!(
-            "Skipping shutdown of a remote timeline {} for tenant {}",
-            timeline_id, tenant_id
-        ),
-    }
-    Ok(())
-}
-
 #[derive(Clone)]
 enum LayeredTimelineEntry {
     Local(Arc<LayeredTimeline>),
@@ -489,6 +501,7 @@ impl LayeredRepository {
             tenantid,
             conf,
             timelines: Mutex::new(HashMap::new()),
+            gc_cs: Mutex::new(()),
             walredo_mgr,
             upload_relishes,
         }
@@ -505,10 +518,10 @@ impl LayeredRepository {
         let _enter = info_span!("saving metadata").entered();
         let path = metadata_path(conf, timelineid, tenantid);
         // use OpenOptions to ensure file presence is consistent with first_save
-        let mut file = OpenOptions::new()
-            .write(true)
-            .create_new(first_save)
-            .open(&path)?;
+        let mut file = VirtualFile::open_with_options(
+            &path,
+            OpenOptions::new().write(true).create_new(first_save),
+        )?;
 
         let metadata_bytes = data.to_bytes().context("Failed to get metadata bytes")?;
 
@@ -575,7 +588,8 @@ impl LayeredRepository {
         let now = Instant::now();
 
         // grab mutex to prevent new timelines from being created here.
-        // TODO: We will hold it for a long time
+        let _gc_cs = self.gc_cs.lock().unwrap();
+
         let mut timelines = self.timelines.lock().unwrap();
 
         // Scan all timelines. For each timeline, remember the timeline ID and
@@ -597,7 +611,7 @@ impl LayeredRepository {
             }
         }
 
-        //Now collect info about branchpoints
+        // Now collect info about branchpoints
         let mut all_branchpoints: BTreeSet<(ZTimelineId, Lsn)> = BTreeSet::new();
         for &timelineid in &timelineids {
             let timeline = match self.get_or_init_timeline(timelineid, &mut timelines)? {
@@ -641,8 +655,10 @@ impl LayeredRepository {
         // Ok, we now know all the branch points.
         // Perform GC for each timeline.
         for timelineid in timelineids {
-            if tenant_mgr::shutdown_requested() {
-                return Ok(totals);
+            if thread_mgr::is_shutdown_requested() {
+                // We were requested to shut down. Stop and return with the progress we
+                // made.
+                break;
             }
 
             // We have already loaded all timelines above
@@ -663,6 +679,7 @@ impl LayeredRepository {
             }
 
             if let Some(cutoff) = timeline.get_last_record_lsn().checked_sub(horizon) {
+                drop(timelines);
                 let branchpoints: Vec<Lsn> = all_branchpoints
                     .range((
                         Included((timelineid, Lsn(0))),
@@ -678,10 +695,10 @@ impl LayeredRepository {
                     timeline.checkpoint(CheckpointConfig::Forced)?;
                     info!("timeline {} checkpoint_before_gc done", timelineid);
                 }
-
                 let result = timeline.gc_timeline(branchpoints, cutoff)?;
 
                 totals += result;
+                timelines = self.timelines.lock().unwrap();
             }
         }
 
@@ -759,8 +776,14 @@ pub struct LayeredTimeline {
     /// to avoid deadlock.
     write_lock: Mutex<()>,
 
+    // Prevent concurrent checkpoints.
+    // Checkpoints are normally performed by one thread. But checkpoint can also be manually requested by admin
+    // (that's used in tests), and shutdown also forces a checkpoint. These forced checkpoints run in a different thread
+    // and could be triggered at the same time as a normal checkpoint.
+    checkpoint_cs: Mutex<()>,
+
     // Needed to ensure that we can't create a branch at a point that was already garbage collected
-    latest_gc_cutoff_lsn: AtomicLsn,
+    latest_gc_cutoff_lsn: RwLock<Lsn>,
 
     // It may change across major versions so for simplicity
     // keep it after running initdb for a timeline.
@@ -804,6 +827,10 @@ impl Timeline for LayeredTimeline {
         Ok(())
     }
 
+    fn get_latest_gc_cutoff_lsn(&self) -> RwLockReadGuard<Lsn> {
+        self.latest_gc_cutoff_lsn.read().unwrap()
+    }
+
     /// Look up given page version.
     fn get_page_at_lsn(&self, rel: RelishTag, rel_blknum: BlockNumber, lsn: Lsn) -> Result<Bytes> {
         if !rel.is_blocky() && rel_blknum != 0 {
@@ -814,14 +841,6 @@ impl Timeline for LayeredTimeline {
             );
         }
         debug_assert!(lsn <= self.get_last_record_lsn());
-        let latest_gc_cutoff_lsn = self.latest_gc_cutoff_lsn.load();
-        // error instead of assert to simplify testing
-        ensure!(
-            lsn >= latest_gc_cutoff_lsn,
-            "tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
-            lsn, latest_gc_cutoff_lsn
-        );
-
         let (seg, seg_blknum) = SegmentTag::from_blknum(rel, rel_blknum);
 
         if let Some((layer, lsn)) = self.get_layer_for_read(seg, lsn)? {
@@ -992,21 +1011,16 @@ impl Timeline for LayeredTimeline {
     ///
     /// Validate lsn against initdb_lsn and latest_gc_cutoff_lsn.
     ///
-    fn check_lsn_is_in_scope(&self, lsn: Lsn) -> Result<()> {
-        let initdb_lsn = self.initdb_lsn;
+    fn check_lsn_is_in_scope(
+        &self,
+        lsn: Lsn,
+        latest_gc_cutoff_lsn: &RwLockReadGuard<Lsn>,
+    ) -> Result<()> {
         ensure!(
-            lsn >= initdb_lsn,
-            "LSN {} is earlier than initdb lsn {}",
-            lsn,
-            initdb_lsn,
-        );
-
-        let latest_gc_cutoff_lsn = self.latest_gc_cutoff_lsn.load();
-        ensure!(
-            lsn >= latest_gc_cutoff_lsn,
+            lsn >= **latest_gc_cutoff_lsn,
             "LSN {} is earlier than latest GC horizon {} (we might've already garbage collected needed data)",
             lsn,
-            latest_gc_cutoff_lsn,
+            **latest_gc_cutoff_lsn,
         );
         Ok(())
     }
@@ -1023,14 +1037,6 @@ impl Timeline for LayeredTimeline {
         self.last_record_lsn.load()
     }
 
-    fn get_start_lsn(&self) -> Lsn {
-        self.ancestor_timeline
-            .as_ref()
-            .and_then(|ancestor_entry| ancestor_entry.local_or_schedule_download(self.tenantid))
-            .map(Timeline::get_start_lsn)
-            .unwrap_or(self.ancestor_lsn)
-    }
-
     fn get_current_logical_size(&self) -> usize {
         self.current_logical_size.load(atomic::Ordering::Acquire) as usize
     }
@@ -1118,8 +1124,9 @@ impl LayeredTimeline {
             upload_relishes: AtomicBool::new(upload_relishes),
 
             write_lock: Mutex::new(()),
+            checkpoint_cs: Mutex::new(()),
 
-            latest_gc_cutoff_lsn: AtomicLsn::from(metadata.latest_gc_cutoff_lsn()),
+            latest_gc_cutoff_lsn: RwLock::new(metadata.latest_gc_cutoff_lsn()),
             initdb_lsn: metadata.initdb_lsn(),
         }
     }
@@ -1145,8 +1152,8 @@ impl LayeredTimeline {
                 // create an ImageLayer struct for each image file.
                 if imgfilename.lsn > disk_consistent_lsn {
                     warn!(
-                        "found future image layer {} on timeline {}",
-                        imgfilename, self.timelineid
+                        "found future image layer {} on timeline {} disk_consistent_lsn is {}",
+                        imgfilename, self.timelineid, disk_consistent_lsn
                     );
 
                     rename_to_backup(direntry.path())?;
@@ -1169,8 +1176,8 @@ impl LayeredTimeline {
                 // before crash.
                 if deltafilename.end_lsn > disk_consistent_lsn + 1 {
                     warn!(
-                        "found future delta layer {} on timeline {}",
-                        deltafilename, self.timelineid
+                        "found future delta layer {} on timeline {} disk_consistent_lsn is {}",
+                        deltafilename, self.timelineid, disk_consistent_lsn
                     );
 
                     rename_to_backup(direntry.path())?;
@@ -1366,7 +1373,7 @@ impl LayeredTimeline {
                     self.tenantid,
                     seg,
                     lsn,
-                    lsn,
+                    last_record_lsn,
                 )?;
             } else {
                 return Ok(open_layer);
@@ -1409,7 +1416,7 @@ impl LayeredTimeline {
                 self.timelineid,
                 self.tenantid,
                 start_lsn,
-                lsn,
+                last_record_lsn,
             )?;
         } else {
             // New relation.
@@ -1420,8 +1427,14 @@ impl LayeredTimeline {
                 lsn
             );
 
-            layer =
-                InMemoryLayer::create(self.conf, self.timelineid, self.tenantid, seg, lsn, lsn)?;
+            layer = InMemoryLayer::create(
+                self.conf,
+                self.timelineid,
+                self.tenantid,
+                seg,
+                lsn,
+                last_record_lsn,
+            )?;
         }
 
         let layer_rc: Arc<InMemoryLayer> = Arc::new(layer);
@@ -1435,7 +1448,10 @@ impl LayeredTimeline {
     ///
     /// NOTE: This has nothing to do with checkpoint in PostgreSQL.
     fn checkpoint_internal(&self, checkpoint_distance: u64, reconstruct_pages: bool) -> Result<()> {
-        let mut write_guard = self.write_lock.lock().unwrap();
+        // Prevent concurrent checkpoints
+        let _checkpoint_cs = self.checkpoint_cs.lock().unwrap();
+
+        let write_guard = self.write_lock.lock().unwrap();
         let mut layers = self.layers.lock().unwrap();
 
         // Bump the generation number in the layer map, so that we can distinguish
@@ -1461,11 +1477,17 @@ impl LayeredTimeline {
         let mut disk_consistent_lsn = last_record_lsn;
 
         let mut layer_paths = Vec::new();
+        let mut freeze_end_lsn = Lsn(0);
+        let mut evicted_layers = Vec::new();
+
+        //
+        // Determine which layers we need to evict and calculate max(latest_lsn)
+        // among those layers.
+        //
         while let Some((oldest_layer_id, oldest_layer, oldest_generation)) =
             layers.peek_oldest_open()
         {
-            let oldest_pending_lsn = oldest_layer.get_oldest_pending_lsn();
-
+            let oldest_lsn = oldest_layer.get_oldest_lsn();
             // Does this layer need freezing?
             //
             // Write out all in-memory layers that contain WAL older than CHECKPOINT_DISTANCE.
@@ -1474,28 +1496,60 @@ impl LayeredTimeline {
             // when we started. We don't want to process layers inserted after we started, to
             // avoid getting into an infinite loop trying to process again entries that we
             // inserted ourselves.
-            let distance = last_record_lsn.widening_sub(oldest_pending_lsn);
-            if distance < 0
+            //
+            // Once we have decided to write out at least one layer, we must also write out
+            // any other layers that contain WAL older than the end LSN of the layers we have
+            // already decided to write out. In other words, we must write out all layers
+            // whose [oldest_lsn, latest_lsn) range overlaps with any of the other layers
+            // that we are writing out. Otherwise, when we advance 'disk_consistent_lsn', it's
+            // ambiguous whether those layers are already durable on disk or not. For example,
+            // imagine that there are two layers in memory that contain page versions in the
+            // following LSN ranges:
+            //
+            // A: 100-150
+            // B: 110-200
+            //
+            // If we flush layer A, we must also flush layer B, because they overlap. If we
+            // flushed only A, and advanced 'disk_consistent_lsn' to 150, we would break the
+            // rule that all WAL older than 'disk_consistent_lsn' are durable on disk, because
+            // B contains some WAL older than 150. On the other hand, if we flushed out A and
+            // advanced 'disk_consistent_lsn' only up to 110, after crash and restart we would
+            // delete the first layer because its end LSN is larger than 110. If we changed
+            // the deletion logic to not delete it, then we would start streaming at 110, and
+            // process again the WAL records in the range 110-150 that are already in layer A,
+            // and the WAL processing code does not cope with that. We solve that dilemma by
+            // insisting that if we write out the first layer, we also write out the second
+            // layer, and advance disk_consistent_lsn all the way up to 200.
+            //
+            let distance = last_record_lsn.widening_sub(oldest_lsn);
+            if (distance < 0
                 || distance < checkpoint_distance.into()
-                || oldest_generation == current_generation
+                || oldest_generation == current_generation)
+                && oldest_lsn >= freeze_end_lsn
+            // this layer intersects with evicted layer and so also need to be evicted
             {
                 info!(
                     "the oldest layer is now {} which is {} bytes behind last_record_lsn",
                     oldest_layer.filename().display(),
                     distance
                 );
-                disk_consistent_lsn = oldest_pending_lsn;
+                disk_consistent_lsn = oldest_lsn;
                 break;
             }
+            let latest_lsn = oldest_layer.get_latest_lsn();
+            if latest_lsn > freeze_end_lsn {
+                freeze_end_lsn = latest_lsn; // calculate max of latest_lsn of the layers we're about to evict
+            }
+            layers.remove_open(oldest_layer_id);
+            evicted_layers.push((oldest_layer_id, oldest_layer));
+        }
 
-            drop(layers);
-            drop(write_guard);
-
-            let mut this_layer_paths = self.evict_layer(oldest_layer_id, reconstruct_pages)?;
-            layer_paths.append(&mut this_layer_paths);
-
-            write_guard = self.write_lock.lock().unwrap();
-            layers = self.layers.lock().unwrap();
+        // Freeze evicted layers
+        for (_evicted_layer_id, evicted_layer) in evicted_layers.iter() {
+            // Mark the layer as no longer accepting writes and record the end_lsn.
+            // This happens in-place, no new layers are created now.
+            evicted_layer.freeze(freeze_end_lsn);
+            layers.insert_historic(evicted_layer.clone());
         }
 
         // Call unload() on all frozen layers, to release memory.
@@ -1508,6 +1562,14 @@ impl LayeredTimeline {
         drop(layers);
         drop(write_guard);
 
+        // Create delta/image layers for evicted layers
+        for (_evicted_layer_id, evicted_layer) in evicted_layers.iter() {
+            let mut this_layer_paths =
+                self.evict_layer(evicted_layer.clone(), reconstruct_pages)?;
+            layer_paths.append(&mut this_layer_paths);
+        }
+
+        // Sync layers
         if !layer_paths.is_empty() {
             // We must fsync the timeline dir to ensure the directory entries for
             // new layer files are durable
@@ -1548,7 +1610,7 @@ impl LayeredTimeline {
                 ondisk_prev_record_lsn,
                 ancestor_timelineid,
                 self.ancestor_lsn,
-                self.latest_gc_cutoff_lsn.load(),
+                *self.latest_gc_cutoff_lsn.read().unwrap(),
                 self.initdb_lsn,
             );
 
@@ -1575,52 +1637,29 @@ impl LayeredTimeline {
         Ok(())
     }
 
-    fn evict_layer(&self, layer_id: LayerId, reconstruct_pages: bool) -> Result<Vec<PathBuf>> {
-        // Mark the layer as no longer accepting writes and record the end_lsn.
-        // This happens in-place, no new layers are created now.
-        // We call `get_last_record_lsn` again, which may be different from the
-        // original load, as we may have released the write lock since then.
-
-        let mut write_guard = self.write_lock.lock().unwrap();
-        let mut layers = self.layers.lock().unwrap();
+    fn evict_layer(
+        &self,
+        layer: Arc<InMemoryLayer>,
+        reconstruct_pages: bool,
+    ) -> Result<Vec<PathBuf>> {
+        let new_historics = layer.write_to_disk(self, reconstruct_pages)?;
 
         let mut layer_paths = Vec::new();
+        let _write_guard = self.write_lock.lock().unwrap();
+        let mut layers = self.layers.lock().unwrap();
 
-        let global_layer_map = GLOBAL_LAYER_MAP.read().unwrap();
-        if let Some(oldest_layer) = global_layer_map.get(&layer_id) {
-            drop(global_layer_map);
-            oldest_layer.freeze(self.get_last_record_lsn());
+        // Finally, replace the frozen in-memory layer with the new on-disk layers
+        layers.remove_historic(layer);
 
-            // The layer is no longer open, update the layer map to reflect this.
-            // We will replace it with on-disk historics below.
-            layers.remove_open(layer_id);
-            layers.insert_historic(oldest_layer.clone());
-
-            // Write the now-frozen layer to disk. That could take a while, so release the lock while do it
-            drop(layers);
-            drop(write_guard);
-
-            let new_historics = oldest_layer.write_to_disk(self, reconstruct_pages)?;
-
-            write_guard = self.write_lock.lock().unwrap();
-            layers = self.layers.lock().unwrap();
-
-            // Finally, replace the frozen in-memory layer with the new on-disk layers
-            layers.remove_historic(oldest_layer);
-
-            // Add the historics to the LayerMap
-            for delta_layer in new_historics.delta_layers {
-                layer_paths.push(delta_layer.path());
-                layers.insert_historic(Arc::new(delta_layer));
-            }
-            for image_layer in new_historics.image_layers {
-                layer_paths.push(image_layer.path());
-                layers.insert_historic(Arc::new(image_layer));
-            }
+        // Add the historics to the LayerMap
+        for delta_layer in new_historics.delta_layers {
+            layer_paths.push(delta_layer.path());
+            layers.insert_historic(Arc::new(delta_layer));
+        }
+        for image_layer in new_historics.image_layers {
+            layer_paths.push(image_layer.path());
+            layers.insert_historic(Arc::new(image_layer));
         }
-        drop(layers);
-        drop(write_guard);
-
         Ok(layer_paths)
     }
 
@@ -1649,12 +1688,14 @@ impl LayeredTimeline {
     pub fn gc_timeline(&self, retain_lsns: Vec<Lsn>, cutoff: Lsn) -> Result<GcResult> {
         let now = Instant::now();
         let mut result: GcResult = Default::default();
+        let disk_consistent_lsn = self.get_disk_consistent_lsn();
+        let _checkpoint_cs = self.checkpoint_cs.lock().unwrap();
 
         let _enter = info_span!("garbage collection", timeline = %self.timelineid, tenant = %self.tenantid, cutoff = %cutoff).entered();
 
         // We need to ensure that no one branches at a point before latest_gc_cutoff_lsn.
         // See branch_timeline() for details.
-        self.latest_gc_cutoff_lsn.store(cutoff);
+        *self.latest_gc_cutoff_lsn.write().unwrap() = cutoff;
 
         info!("GC starting");
 
@@ -1734,7 +1775,12 @@ impl LayeredTimeline {
             }
 
             // 3. Is there a later on-disk layer for this relation?
-            if !l.is_dropped() && !layers.newer_image_layer_exists(l.get_seg_tag(), l.get_end_lsn())
+            if !l.is_dropped()
+                && !layers.newer_image_layer_exists(
+                    l.get_seg_tag(),
+                    l.get_end_lsn(),
+                    disk_consistent_lsn,
+                )
             {
                 info!(
                     "keeping {} {}-{} because it is the latest layer",
@@ -2187,11 +2233,10 @@ impl<'a> TimelineWriter for LayeredTimelineWriter<'a> {
         let oldsize = self
             .tl
             .get_relish_size(rel, self.tl.get_last_record_lsn())?
-            .ok_or_else(|| {
-                anyhow!(
+            .with_context(|| {
+                format!(
                     "attempted to truncate non-existent relish {} at {}",
-                    rel,
-                    lsn
+                    rel, lsn
                 )
             })?;
 
@@ -2314,8 +2359,5 @@ fn rename_to_backup(path: PathBuf) -> anyhow::Result<()> {
         }
     }
 
-    Err(anyhow!(
-        "couldn't find an unused backup number for {:?}",
-        path
-    ))
+    bail!("couldn't find an unused backup number for {:?}", path)
 }

pageserver/src/layered_repository/delta_layer.rs

@@ -169,7 +169,7 @@ impl DeltaLayerInner {
         if let Some((_entry_lsn, entry)) = slice.last() {
             Ok(*entry)
         } else {
-            Err(anyhow::anyhow!("could not find seg size in delta layer"))
+            bail!("could not find seg size in delta layer")
         }
     }
 }

pageserver/src/layered_repository/ephemeral_file.rs

@@ -175,7 +175,10 @@ impl Write for EphemeralFile {
     }
 
     fn flush(&mut self) -> Result<(), std::io::Error> {
-        todo!()
+        // we don't need to flush data:
+        // * we either write input bytes or not, not keeping any intermediate data buffered
+        // * rust unix file `flush` impl does not flush things either, returning `Ok(())`
+        Ok(())
     }
 }
 

pageserver/src/layered_repository/image_layer.rs

@@ -173,7 +173,14 @@ impl Layer for ImageLayer {
                     .as_ref()
                     .unwrap()
                     .chapter_reader(BLOCKY_IMAGES_CHAPTER)?;
-                chapter.read_exact_at(&mut buf, offset)?;
+
+                chapter.read_exact_at(&mut buf, offset).with_context(|| {
+                    format!(
+                        "failed to read page from data file {} at offset {}",
+                        self.filename().display(),
+                        offset
+                    )
+                })?;
 
                 buf
             }

pageserver/src/layered_repository/inmemory_layer.rs

@@ -39,8 +39,20 @@ pub struct InMemoryLayer {
     ///
     start_lsn: Lsn,
 
-    /// LSN of the oldest page version stored in this layer
-    oldest_pending_lsn: Lsn,
+    ///
+    /// LSN of the oldest page version stored in this layer.
+    ///
+    /// This is different from 'start_lsn' in that we enforce that the 'start_lsn'
+    /// of a layer always matches the 'end_lsn' of its predecessor, even if there
+    /// are no page versions until at a later LSN. That way you can detect any
+    /// missing layer files more easily. 'oldest_lsn' is the first page version
+    /// actually stored in this layer. In the range between 'start_lsn' and
+    /// 'oldest_lsn', there are no changes to the segment.
+    /// 'oldest_lsn' is used to adjust 'disk_consistent_lsn' and that is why it should
+    /// point to the beginning of WAL record. This is the other difference with 'start_lsn'
+    /// which points to end of WAL record. This is why 'oldest_lsn' can be smaller than 'start_lsn'.
+    ///
+    oldest_lsn: Lsn,
 
     /// The above fields never change. The parts that do change are in 'inner',
     /// and protected by mutex.
@@ -73,6 +85,14 @@ pub struct InMemoryLayerInner {
     /// a non-blocky rel, 'seg_sizes' is not used and is always empty.
     ///
     seg_sizes: VecMap<Lsn, SegmentBlk>,
+
+    ///
+    /// LSN of the newest page version stored in this layer.
+    ///
+    /// The difference between 'end_lsn' and 'latest_lsn' is the same as between
+    /// 'start_lsn' and 'oldest_lsn'. See comments in 'oldest_lsn'.
+    ///
+    latest_lsn: Lsn,
 }
 
 impl InMemoryLayerInner {
@@ -319,8 +339,13 @@ pub struct LayersOnDisk {
 
 impl InMemoryLayer {
     /// Return the oldest page version that's stored in this layer
-    pub fn get_oldest_pending_lsn(&self) -> Lsn {
-        self.oldest_pending_lsn
+    pub fn get_oldest_lsn(&self) -> Lsn {
+        self.oldest_lsn
+    }
+
+    pub fn get_latest_lsn(&self) -> Lsn {
+        let inner = self.inner.read().unwrap();
+        inner.latest_lsn
     }
 
     ///
@@ -332,7 +357,7 @@ impl InMemoryLayer {
         tenantid: ZTenantId,
         seg: SegmentTag,
         start_lsn: Lsn,
-        oldest_pending_lsn: Lsn,
+        oldest_lsn: Lsn,
     ) -> Result<InMemoryLayer> {
         trace!(
             "initializing new empty InMemoryLayer for writing {} on timeline {} at {}",
@@ -355,13 +380,14 @@ impl InMemoryLayer {
             tenantid,
             seg,
             start_lsn,
-            oldest_pending_lsn,
+            oldest_lsn,
             incremental: false,
             inner: RwLock::new(InMemoryLayerInner {
                 end_lsn: None,
                 dropped: false,
                 page_versions: PageVersions::new(file),
                 seg_sizes,
+                latest_lsn: oldest_lsn,
             }),
         })
     }
@@ -398,6 +424,8 @@ impl InMemoryLayer {
         let mut inner = self.inner.write().unwrap();
 
         inner.assert_writeable();
+        assert!(lsn >= inner.latest_lsn);
+        inner.latest_lsn = lsn;
 
         let old = inner.page_versions.append_or_update_last(blknum, lsn, pv)?;
 
@@ -509,12 +537,11 @@ impl InMemoryLayer {
         timelineid: ZTimelineId,
         tenantid: ZTenantId,
         start_lsn: Lsn,
-        oldest_pending_lsn: Lsn,
+        oldest_lsn: Lsn,
     ) -> Result<InMemoryLayer> {
         let seg = src.get_seg_tag();
 
-        assert!(oldest_pending_lsn.is_aligned());
-        assert!(oldest_pending_lsn >= start_lsn);
+        assert!(oldest_lsn.is_aligned());
 
         trace!(
             "initializing new InMemoryLayer for writing {} on timeline {} at {}",
@@ -538,13 +565,14 @@ impl InMemoryLayer {
             tenantid,
             seg,
             start_lsn,
-            oldest_pending_lsn,
+            oldest_lsn,
             incremental: true,
             inner: RwLock::new(InMemoryLayerInner {
                 end_lsn: None,
                 dropped: false,
                 page_versions: PageVersions::new(file),
                 seg_sizes,
+                latest_lsn: oldest_lsn,
             }),
         })
     }

pageserver/src/layered_repository/layer_map.rs

@@ -40,7 +40,7 @@ pub struct LayerMap {
     /// All the layers keyed by segment tag
     segs: HashMap<SegmentTag, SegEntry>,
 
-    /// All in-memory layers, ordered by 'oldest_pending_lsn' and generation
+    /// All in-memory layers, ordered by 'oldest_lsn' and generation
     /// of each layer. This allows easy access to the in-memory layer that
     /// contains the oldest WAL record.
     open_layers: BinaryHeap<OpenLayerEntry>,
@@ -83,16 +83,16 @@ impl LayerMap {
 
         let layer_id = segentry.update_open(Arc::clone(&layer));
 
-        let oldest_pending_lsn = layer.get_oldest_pending_lsn();
+        let oldest_lsn = layer.get_oldest_lsn();
 
-        // After a crash and restart, 'oldest_pending_lsn' of the oldest in-memory
+        // After a crash and restart, 'oldest_lsn' of the oldest in-memory
         // layer becomes the WAL streaming starting point, so it better not point
         // in the middle of a WAL record.
-        assert!(oldest_pending_lsn.is_aligned());
+        assert!(oldest_lsn.is_aligned());
 
         // Also add it to the binary heap
         let open_layer_entry = OpenLayerEntry {
-            oldest_pending_lsn: layer.get_oldest_pending_lsn(),
+            oldest_lsn: layer.get_oldest_lsn(),
             layer_id,
             generation: self.current_generation,
         };
@@ -191,9 +191,15 @@ impl LayerMap {
     ///
     /// This is used for garbage collection, to determine if an old layer can
     /// be deleted.
-    pub fn newer_image_layer_exists(&self, seg: SegmentTag, lsn: Lsn) -> bool {
+    /// We ignore segments newer than disk_consistent_lsn because they will be removed at restart
+    pub fn newer_image_layer_exists(
+        &self,
+        seg: SegmentTag,
+        lsn: Lsn,
+        disk_consistent_lsn: Lsn,
+    ) -> bool {
         if let Some(segentry) = self.segs.get(&seg) {
-            segentry.newer_image_layer_exists(lsn)
+            segentry.newer_image_layer_exists(lsn, disk_consistent_lsn)
         } else {
             false
         }
@@ -311,13 +317,18 @@ impl SegEntry {
         self.historic.search(lsn)
     }
 
-    pub fn newer_image_layer_exists(&self, lsn: Lsn) -> bool {
+    pub fn newer_image_layer_exists(&self, lsn: Lsn, disk_consistent_lsn: Lsn) -> bool {
         // We only check on-disk layers, because
         // in-memory layers are not durable
 
+        // The end-LSN is exclusive, while disk_consistent_lsn is
+        // inclusive. For example, if disk_consistent_lsn is 100, it is
+        // OK for a delta layer to have end LSN 101, but if the end LSN
+        // is 102, then it might not have been fully flushed to disk
+        // before crash.
         self.historic
             .iter_newer(lsn)
-            .any(|layer| !layer.is_incremental())
+            .any(|layer| !layer.is_incremental() && layer.get_end_lsn() <= disk_consistent_lsn + 1)
     }
 
     // Set new open layer for a SegEntry.
@@ -341,23 +352,23 @@ impl SegEntry {
 }
 
 /// Entry held in LayerMap::open_layers, with boilerplate comparison routines
-/// to implement a min-heap ordered by 'oldest_pending_lsn' and 'generation'
+/// to implement a min-heap ordered by 'oldest_lsn' and 'generation'
 ///
 /// The generation number associated with each entry can be used to distinguish
 /// recently-added entries (i.e after last call to increment_generation()) from older
-/// entries with the same 'oldest_pending_lsn'.
+/// entries with the same 'oldest_lsn'.
 struct OpenLayerEntry {
-    oldest_pending_lsn: Lsn, // copy of layer.get_oldest_pending_lsn()
+    oldest_lsn: Lsn, // copy of layer.get_oldest_lsn()
     generation: u64,
     layer_id: LayerId,
 }
 impl Ord for OpenLayerEntry {
     fn cmp(&self, other: &Self) -> Ordering {
         // BinaryHeap is a max-heap, and we want a min-heap. Reverse the ordering here
-        // to get that. Entries with identical oldest_pending_lsn are ordered by generation
+        // to get that. Entries with identical oldest_lsn are ordered by generation
         other
-            .oldest_pending_lsn
-            .cmp(&self.oldest_pending_lsn)
+            .oldest_lsn
+            .cmp(&self.oldest_lsn)
             .then_with(|| other.generation.cmp(&self.generation))
     }
 }
@@ -426,7 +437,7 @@ mod tests {
         conf: &'static PageServerConf,
         segno: u32,
         start_lsn: Lsn,
-        oldest_pending_lsn: Lsn,
+        oldest_lsn: Lsn,
     ) -> Arc<InMemoryLayer> {
         Arc::new(
             InMemoryLayer::create(
@@ -438,7 +449,7 @@ mod tests {
                     segno,
                 },
                 start_lsn,
-                oldest_pending_lsn,
+                oldest_lsn,
             )
             .unwrap(),
         )

pageserver/src/lib.rs

@@ -11,6 +11,7 @@ pub mod remote_storage;
 pub mod repository;
 pub mod tenant_mgr;
 pub mod tenant_threads;
+pub mod thread_mgr;
 pub mod virtual_file;
 pub mod walingest;
 pub mod walreceiver;

pageserver/src/page_service.rs

@@ -10,16 +10,15 @@
 //     *callmemaybe <zenith timelineid> $url* -- ask pageserver to start walreceiver on $url
 //
 
-use anyhow::{anyhow, bail, ensure, Context, Result};
+use anyhow::{bail, ensure, Context, Result};
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use lazy_static::lazy_static;
 use regex::Regex;
+use std::io;
 use std::net::TcpListener;
 use std::str;
 use std::str::FromStr;
-use std::sync::Arc;
-use std::thread;
-use std::{io, net::TcpStream};
+use std::sync::{Arc, RwLockReadGuard};
 use tracing::*;
 use zenith_metrics::{register_histogram_vec, HistogramVec};
 use zenith_utils::auth::{self, JwtAuth};
@@ -28,17 +27,16 @@ use zenith_utils::lsn::Lsn;
 use zenith_utils::postgres_backend::is_socket_read_timed_out;
 use zenith_utils::postgres_backend::PostgresBackend;
 use zenith_utils::postgres_backend::{self, AuthType};
-use zenith_utils::pq_proto::{
-    BeMessage, FeMessage, RowDescriptor, HELLO_WORLD_ROW, SINGLE_COL_ROWDESC,
-};
+use zenith_utils::pq_proto::{BeMessage, FeMessage, RowDescriptor, SINGLE_COL_ROWDESC};
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
 use crate::basebackup;
-use crate::branches;
 use crate::config::PageServerConf;
 use crate::relish::*;
 use crate::repository::Timeline;
 use crate::tenant_mgr;
+use crate::thread_mgr;
+use crate::thread_mgr::ThreadKind;
 use crate::walreceiver;
 use crate::CheckpointConfig;
 
@@ -189,30 +187,61 @@ pub fn thread_main(
     listener: TcpListener,
     auth_type: AuthType,
 ) -> anyhow::Result<()> {
-    let mut join_handles = Vec::new();
+    listener.set_nonblocking(true)?;
+    let basic_rt = tokio::runtime::Builder::new_current_thread()
+        .enable_io()
+        .build()?;
 
-    while !tenant_mgr::shutdown_requested() {
-        let (socket, peer_addr) = listener.accept()?;
-        debug!("accepted connection from {}", peer_addr);
-        socket.set_nodelay(true).unwrap();
-        let local_auth = auth.clone();
+    let tokio_listener = {
+        let _guard = basic_rt.enter();
+        tokio::net::TcpListener::from_std(listener)
+    }?;
 
-        let handle = thread::Builder::new()
-            .name("serving Page Service thread".into())
-            .spawn(move || {
-                if let Err(err) = page_service_conn_main(conf, local_auth, socket, auth_type) {
-                    error!(%err, "page server thread exited with error");
+    // Wait for a new connection to arrive, or for server shutdown.
+    while let Some(res) = basic_rt.block_on(async {
+        let shutdown_watcher = thread_mgr::shutdown_watcher();
+        tokio::select! {
+            biased;
+
+            _ = shutdown_watcher => {
+                // We were requested to shut down.
+                None
+            }
+
+            res = tokio_listener.accept() => {
+                Some(res)
+            }
+        }
+    }) {
+        match res {
+            Ok((socket, peer_addr)) => {
+                // Connection established. Spawn a new thread to handle it.
+                debug!("accepted connection from {}", peer_addr);
+                let local_auth = auth.clone();
+
+                // PageRequestHandler threads are not associated with any particular
+                // timeline in the thread manager. In practice most connections will
+                // only deal with a particular timeline, but we don't know which one
+                // yet.
+                if let Err(err) = thread_mgr::spawn(
+                    ThreadKind::PageRequestHandler,
+                    None,
+                    None,
+                    "serving Page Service thread",
+                    move || page_service_conn_main(conf, local_auth, socket, auth_type),
+                ) {
+                    // Thread creation failed. Log the error and continue.
+                    error!("could not spawn page service thread: {:?}", err);
                 }
-            })
-            .unwrap();
-
-        join_handles.push(handle);
+            }
+            Err(err) => {
+                // accept() failed. Log the error, and loop back to retry on next connection.
+                error!("accept() failed: {:?}", err);
+            }
+        }
     }
 
-    debug!("page_service loop terminated. wait for connections to cancel");
-    for handle in join_handles.into_iter() {
-        handle.join().unwrap();
-    }
+    debug!("page_service loop terminated");
 
     Ok(())
 }
@@ -220,10 +249,10 @@ pub fn thread_main(
 fn page_service_conn_main(
     conf: &'static PageServerConf,
     auth: Option<Arc<JwtAuth>>,
-    socket: TcpStream,
+    socket: tokio::net::TcpStream,
     auth_type: AuthType,
 ) -> anyhow::Result<()> {
-    // Immediatsely increment the gauge, then create a job to decrement it on thread exit.
+    // Immediately increment the gauge, then create a job to decrement it on thread exit.
     // One of the pros of `defer!` is that this will *most probably*
     // get called, even in presence of panics.
     let gauge = crate::LIVE_CONNECTIONS_COUNT.with_label_values(&["page_service"]);
@@ -232,6 +261,19 @@ fn page_service_conn_main(
         gauge.dec();
     }
 
+    // We use Tokio to accept the connection, but the rest of the code works with a
+    // regular socket. Convert.
+    let socket = socket
+        .into_std()
+        .context("could not convert tokio::net:TcpStream to std::net::TcpStream")?;
+    socket
+        .set_nonblocking(false)
+        .context("could not put socket to blocking mode")?;
+
+    socket
+        .set_nodelay(true)
+        .context("could not set TCP_NODELAY")?;
+
     let mut conn_handler = PageServerHandler::new(conf, auth);
     let pgbackend = PostgresBackend::new(socket, auth_type, None, true)?;
     pgbackend.run(&mut conn_handler)
@@ -286,7 +328,7 @@ impl PageServerHandler {
         /* switch client to COPYBOTH */
         pgb.write_message(&BeMessage::CopyBothResponse)?;
 
-        while !tenant_mgr::shutdown_requested() {
+        while !thread_mgr::is_shutdown_requested() {
             match pgb.read_message() {
                 Ok(message) => {
                     if let Some(message) = message {
@@ -320,7 +362,7 @@ impl PageServerHandler {
                         let response = response.unwrap_or_else(|e| {
                             // print the all details to the log with {:#}, but for the client the
                             // error message is enough
-                            error!("error reading relation or page version: {:#}", e);
+                            error!("error reading relation or page version: {:?}", e);
                             PagestreamBeMessage::Error(PagestreamErrorResponse {
                                 message: e.to_string(),
                             })
@@ -353,7 +395,12 @@ impl PageServerHandler {
     /// In either case, if the page server hasn't received the WAL up to the
     /// requested LSN yet, we will wait for it to arrive. The return value is
     /// the LSN that should be used to look up the page versions.
-    fn wait_or_get_last_lsn(timeline: &dyn Timeline, lsn: Lsn, latest: bool) -> Result<Lsn> {
+    fn wait_or_get_last_lsn(
+        timeline: &dyn Timeline,
+        mut lsn: Lsn,
+        latest: bool,
+        latest_gc_cutoff_lsn: &RwLockReadGuard<Lsn>,
+    ) -> Result<Lsn> {
         if latest {
             // Latest page version was requested. If LSN is given, it is a hint
             // to the page server that there have been no modifications to the
@@ -374,22 +421,26 @@ impl PageServerHandler {
             // walsender completes the authentication and starts streaming the
             // WAL.
             if lsn <= last_record_lsn {
-                Ok(last_record_lsn)
+                lsn = last_record_lsn;
             } else {
                 timeline.wait_lsn(lsn)?;
                 // Since we waited for 'lsn' to arrive, that is now the last
                 // record LSN. (Or close enough for our purposes; the
                 // last-record LSN can advance immediately after we return
                 // anyway)
-                Ok(lsn)
             }
         } else {
             if lsn == Lsn(0) {
                 bail!("invalid LSN(0) in request");
             }
             timeline.wait_lsn(lsn)?;
-            Ok(lsn)
         }
+        ensure!(
+            lsn >= **latest_gc_cutoff_lsn,
+            "tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
+            lsn, **latest_gc_cutoff_lsn
+        );
+        Ok(lsn)
     }
 
     fn handle_get_rel_exists_request(
@@ -400,7 +451,8 @@ impl PageServerHandler {
         let _enter = info_span!("get_rel_exists", rel = %req.rel, req_lsn = %req.lsn).entered();
 
         let tag = RelishTag::Relation(req.rel);
-        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest)?;
+        let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
+        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn)?;
 
         let exists = timeline.get_rel_exists(tag, lsn)?;
 
@@ -416,7 +468,8 @@ impl PageServerHandler {
     ) -> Result<PagestreamBeMessage> {
         let _enter = info_span!("get_nblocks", rel = %req.rel, req_lsn = %req.lsn).entered();
         let tag = RelishTag::Relation(req.rel);
-        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest)?;
+        let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
+        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn)?;
 
         let n_blocks = timeline.get_relish_size(tag, lsn)?;
 
@@ -437,8 +490,16 @@ impl PageServerHandler {
         let _enter = info_span!("get_page", rel = %req.rel, blkno = &req.blkno, req_lsn = %req.lsn)
             .entered();
         let tag = RelishTag::Relation(req.rel);
-        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest)?;
-
+        let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
+        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn)?;
+        /*
+        // Add a 1s delay to some requests. The delayed causes the requests to
+        // hit the race condition from github issue #1047 more easily.
+        use rand::Rng;
+        if rand::thread_rng().gen::<u8>() < 5 {
+            std::thread::sleep(std::time::Duration::from_millis(1000));
+        }
+        */
         let page = timeline.get_page_at_lsn(tag, req.blkno, lsn)?;
 
         Ok(PagestreamBeMessage::GetPage(PagestreamGetPageResponse {
@@ -459,9 +520,10 @@ impl PageServerHandler {
         // check that the timeline exists
         let timeline = tenant_mgr::get_timeline_for_tenant(tenantid, timelineid)
             .context("Cannot handle basebackup request for a remote timeline")?;
+        let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         if let Some(lsn) = lsn {
             timeline
-                .check_lsn_is_in_scope(lsn)
+                .check_lsn_is_in_scope(lsn, &latest_gc_cutoff_lsn)
                 .context("invalid basebackup lsn")?;
         }
 
@@ -579,7 +641,7 @@ impl postgres_backend::Handler for PageServerHandler {
             let re = Regex::new(r"^callmemaybe ([[:xdigit:]]+) ([[:xdigit:]]+) (.*)$").unwrap();
             let caps = re
                 .captures(query_string)
-                .ok_or_else(|| anyhow!("invalid callmemaybe: '{}'", query_string))?;
+                .with_context(|| format!("invalid callmemaybe: '{}'", query_string))?;
 
             let tenantid = ZTenantId::from_str(caps.get(1).unwrap().as_str())?;
             let timelineid = ZTimelineId::from_str(caps.get(2).unwrap().as_str())?;
@@ -594,82 +656,24 @@ impl postgres_backend::Handler for PageServerHandler {
             tenant_mgr::get_timeline_for_tenant(tenantid, timelineid)
                 .context("Failed to fetch local timeline for callmemaybe requests")?;
 
-            walreceiver::launch_wal_receiver(self.conf, timelineid, &connstr, tenantid.to_owned());
+            walreceiver::launch_wal_receiver(self.conf, tenantid, timelineid, &connstr)?;
 
             pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
-        } else if query_string.starts_with("branch_create ") {
-            let err = || anyhow!("invalid branch_create: '{}'", query_string);
-
-            // branch_create <tenantid> <branchname> <startpoint>
-            // TODO lazy static
-            // TODO: escaping, to allow branch names with spaces
-            let re = Regex::new(r"^branch_create ([[:xdigit:]]+) (\S+) ([^\r\n\s;]+)[\r\n\s;]*;?$")
-                .unwrap();
-            let caps = re.captures(query_string).ok_or_else(err)?;
-
-            let tenantid = ZTenantId::from_str(caps.get(1).unwrap().as_str())?;
-            let branchname = caps.get(2).ok_or_else(err)?.as_str().to_owned();
-            let startpoint_str = caps.get(3).ok_or_else(err)?.as_str().to_owned();
-
-            self.check_permission(Some(tenantid))?;
-
-            let _enter =
-                info_span!("branch_create", name = %branchname, tenant = %tenantid).entered();
-
-            let branch =
-                branches::create_branch(self.conf, &branchname, &startpoint_str, &tenantid)?;
-            let branch = serde_json::to_vec(&branch)?;
-
-            pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
-                .write_message_noflush(&BeMessage::DataRow(&[Some(&branch)]))?
-                .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
-        } else if query_string.starts_with("branch_list ") {
-            // branch_list <zenith tenantid as hex string>
-            let re = Regex::new(r"^branch_list ([[:xdigit:]]+)$").unwrap();
-            let caps = re
-                .captures(query_string)
-                .ok_or_else(|| anyhow!("invalid branch_list: '{}'", query_string))?;
-
-            let tenantid = ZTenantId::from_str(caps.get(1).unwrap().as_str())?;
-
-            // 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::tenant_mgr::list_tenants()?;
-            let tenants_buf = serde_json::to_vec(&tenants)?;
-
-            pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
-                .write_message_noflush(&BeMessage::DataRow(&[Some(&tenants_buf)]))?
-                .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
-        } else if query_string.starts_with("tenant_create") {
-            let err = || anyhow!("invalid tenant_create: '{}'", query_string);
-
-            // tenant_create <tenantid>
-            let re = Regex::new(r"^tenant_create ([[:xdigit:]]+)$").unwrap();
-            let caps = re.captures(query_string).ok_or_else(err)?;
-
-            self.check_permission(None)?;
-
-            let tenantid = ZTenantId::from_str(caps.get(1).unwrap().as_str())?;
-
-            tenant_mgr::create_repository_for_tenant(self.conf, tenantid)?;
-
-            pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
-                .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
-        } else if query_string.starts_with("status") {
-            pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
-                .write_message_noflush(&HELLO_WORLD_ROW)?
-                .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
         } else if query_string.to_ascii_lowercase().starts_with("set ") {
             // important because psycopg2 executes "SET datestyle TO 'ISO'"
             // on connect
             pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
+        } else if query_string.starts_with("failpoints ") {
+            let (_, failpoints) = query_string.split_at("failpoints ".len());
+            for failpoint in failpoints.split(';') {
+                if let Some((name, actions)) = failpoint.split_once('=') {
+                    info!("cfg failpoint: {} {}", name, actions);
+                    fail::cfg(name, actions).unwrap();
+                } else {
+                    bail!("Invalid failpoints format");
+                }
+            }
+            pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
         } else if query_string.starts_with("do_gc ") {
             // Run GC immediately on given timeline.
             // FIXME: This is just for tests. See test_runner/batch_others/test_gc.py.
@@ -683,7 +687,7 @@ impl postgres_backend::Handler for PageServerHandler {
 
             let caps = re
                 .captures(query_string)
-                .ok_or_else(|| anyhow!("invalid do_gc: '{}'", query_string))?;
+                .with_context(|| format!("invalid do_gc: '{}'", query_string))?;
 
             let tenantid = ZTenantId::from_str(caps.get(1).unwrap().as_str())?;
             let timelineid = ZTimelineId::from_str(caps.get(2).unwrap().as_str())?;
@@ -767,7 +771,7 @@ impl postgres_backend::Handler for PageServerHandler {
 
             let caps = re
                 .captures(query_string)
-                .ok_or_else(|| anyhow!("invalid checkpoint command: '{}'", query_string))?;
+                .with_context(|| format!("invalid checkpoint command: '{}'", query_string))?;
 
             let tenantid = ZTenantId::from_str(caps.get(1).unwrap().as_str())?;
             let timelineid = ZTimelineId::from_str(caps.get(2).unwrap().as_str())?;

pageserver/src/remote_storage.rs

@@ -5,7 +5,7 @@
 //! There are a few components the storage machinery consists of:
 //! * [`RemoteStorage`] 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
+//!     * [`rust_s3`] uses AWS S3 bucket as an external storage
 //!
 //! * synchronization logic at [`storage_sync`] module that keeps pageserver state (both runtime one and the workdir files) and storage state in sync.
 //! Synchronization internals are split into submodules
@@ -89,13 +89,12 @@ use std::{
     collections::HashMap,
     ffi, fs,
     path::{Path, PathBuf},
-    thread,
 };
 
 use anyhow::{bail, Context};
 use tokio::io;
 use tracing::{error, info};
-use zenith_utils::zid::{ZTenantId, ZTimelineId};
+use zenith_utils::zid::{ZTenantId, ZTenantTimelineId, ZTimelineId};
 
 pub use self::storage_sync::{schedule_timeline_checkpoint_upload, schedule_timeline_download};
 use self::{local_fs::LocalFs, rust_s3::S3};
@@ -105,16 +104,7 @@ use crate::{
     repository::TimelineSyncState,
 };
 
-/// Any timeline has its own id and its own tenant it belongs to,
-/// the sync processes group timelines by both for simplicity.
-#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone, Copy, Hash)]
-pub struct TimelineSyncId(ZTenantId, ZTimelineId);
-
-impl std::fmt::Display for TimelineSyncId {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "(tenant: {}, timeline: {})", self.0, self.1)
-    }
-}
+pub use storage_sync::compression;
 
 /// A structure to combine all synchronization data to share with pageserver after a successful sync loop initialization.
 /// Successful initialization includes a case when sync loop is not started, in which case the startup data is returned still,
@@ -125,8 +115,6 @@ pub struct SyncStartupData {
     /// To reuse the local file scan logic, the timeline states are returned even if no sync loop get started during init:
     /// in this case, no remote files exist and all local timelines with correct metadata files are considered ready.
     pub initial_timeline_states: HashMap<ZTenantId, HashMap<ZTimelineId, TimelineSyncState>>,
-    /// A handle to the sync loop, if it was started from the configuration provided.
-    pub sync_loop_handle: Option<thread::JoinHandle<anyhow::Result<()>>>,
 }
 
 /// Based on the config, initiates the remote storage connection and starts a separate thread
@@ -141,20 +129,27 @@ pub fn start_local_timeline_sync(
 
     match &config.remote_storage_config {
         Some(storage_config) => match &storage_config.storage {
-            RemoteStorageKind::LocalFs(root) => storage_sync::spawn_storage_sync_thread(
-                config,
-                local_timeline_files,
-                LocalFs::new(root.clone(), &config.workdir)?,
-                storage_config.max_concurrent_sync,
-                storage_config.max_sync_errors,
-            ),
-            RemoteStorageKind::AwsS3(s3_config) => storage_sync::spawn_storage_sync_thread(
-                config,
-                local_timeline_files,
-                S3::new(s3_config, &config.workdir)?,
-                storage_config.max_concurrent_sync,
-                storage_config.max_sync_errors,
-            ),
+            RemoteStorageKind::LocalFs(root) => {
+                info!("Using fs root '{}' as a remote storage", root.display());
+                storage_sync::spawn_storage_sync_thread(
+                    config,
+                    local_timeline_files,
+                    LocalFs::new(root.clone(), &config.workdir)?,
+                    storage_config.max_concurrent_sync,
+                    storage_config.max_sync_errors,
+                )
+            },
+            RemoteStorageKind::AwsS3(s3_config) => {
+                info!("Using s3 bucket '{}' in region '{}' as a remote storage, prefix in bucket: '{:?}', bucket endpoint: '{:?}'",
+                    s3_config.bucket_name, s3_config.bucket_region, s3_config.prefix_in_bucket, s3_config.endpoint);
+                storage_sync::spawn_storage_sync_thread(
+                    config,
+                    local_timeline_files,
+                    S3::new(s3_config, &config.workdir)?,
+                    storage_config.max_concurrent_sync,
+                    storage_config.max_sync_errors,
+                )
+            },
         }
         .context("Failed to spawn the storage sync thread"),
         None => {
@@ -163,7 +158,7 @@ pub fn start_local_timeline_sync(
                 ZTenantId,
                 HashMap<ZTimelineId, TimelineSyncState>,
             > = HashMap::new();
-            for (TimelineSyncId(tenant_id, timeline_id), (timeline_metadata, _)) in
+            for (ZTenantTimelineId{tenant_id, timeline_id}, (timeline_metadata, _)) in
                 local_timeline_files
             {
                 initial_timeline_states
@@ -176,7 +171,6 @@ pub fn start_local_timeline_sync(
             }
             Ok(SyncStartupData {
                 initial_timeline_states,
-                sync_loop_handle: None,
             })
         }
     }
@@ -184,7 +178,7 @@ pub fn start_local_timeline_sync(
 
 fn local_tenant_timeline_files(
     config: &'static PageServerConf,
-) -> anyhow::Result<HashMap<TimelineSyncId, (TimelineMetadata, Vec<PathBuf>)>> {
+) -> anyhow::Result<HashMap<ZTenantTimelineId, (TimelineMetadata, Vec<PathBuf>)>> {
     let mut local_tenant_timeline_files = HashMap::new();
     let tenants_dir = config.tenants_path();
     for tenants_dir_entry in fs::read_dir(&tenants_dir)
@@ -205,7 +199,7 @@ fn local_tenant_timeline_files(
                 }
             }
             Err(e) => error!(
-                "Failed to list tenants dir entry {:?} in directory {}, reason: {:#}",
+                "Failed to list tenants dir entry {:?} in directory {}, reason: {:?}",
                 tenants_dir_entry,
                 tenants_dir.display(),
                 e
@@ -219,8 +213,9 @@ fn local_tenant_timeline_files(
 fn collect_timelines_for_tenant(
     config: &'static PageServerConf,
     tenant_path: &Path,
-) -> anyhow::Result<HashMap<TimelineSyncId, (TimelineMetadata, Vec<PathBuf>)>> {
-    let mut timelines: HashMap<TimelineSyncId, (TimelineMetadata, Vec<PathBuf>)> = HashMap::new();
+) -> anyhow::Result<HashMap<ZTenantTimelineId, (TimelineMetadata, Vec<PathBuf>)>> {
+    let mut timelines: HashMap<ZTenantTimelineId, (TimelineMetadata, Vec<PathBuf>)> =
+        HashMap::new();
     let tenant_id = tenant_path
         .file_name()
         .and_then(ffi::OsStr::to_str)
@@ -241,19 +236,22 @@ fn collect_timelines_for_tenant(
                 match collect_timeline_files(&timeline_path) {
                     Ok((timeline_id, metadata, timeline_files)) => {
                         timelines.insert(
-                            TimelineSyncId(tenant_id, timeline_id),
+                            ZTenantTimelineId {
+                                tenant_id,
+                                timeline_id,
+                            },
                             (metadata, timeline_files),
                         );
                     }
                     Err(e) => error!(
-                        "Failed to process timeline dir contents at '{}', reason: {:#}",
+                        "Failed to process timeline dir contents at '{}', reason: {:?}",
                         timeline_path.display(),
                         e
                     ),
                 }
             }
             Err(e) => error!(
-                "Failed to list timelines for entry tenant {}, reason: {:#}",
+                "Failed to list timelines for entry tenant {}, reason: {:?}",
                 tenant_id, e
             ),
         }

pageserver/src/remote_storage/README.md

@@ -70,8 +70,3 @@ on the timeline download, missing remote branch files are downlaoded.
 
 A branch is a per-tenant entity, yet a current implementaion requires synchronizing a timeline first to get the branch files locally.
 Currently, there's no other way to know about the remote branch files, neither the file contents is verified and updated.
-
-* 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/remote_storage/local_fs.rs

@@ -73,7 +73,7 @@ impl RemoteStorage for LocalFs {
     }
 
     async fn list(&self) -> anyhow::Result<Vec<Self::StoragePath>> {
-        Ok(get_all_files(&self.root).await?.into_iter().collect())
+        get_all_files(&self.root).await
     }
 
     async fn upload(

pageserver/src/remote_storage/rust_s3.rs

@@ -1,12 +1,15 @@
 //! AWS S3 storage wrapper around `rust_s3` library.
-//! Currently does not allow multiple pageservers to use the same bucket concurrently: objects are
-//! placed in the root of the bucket.
+//!
+//! Respects `prefix_in_bucket` property from [`S3Config`],
+//! allowing multiple pageservers to independently work with the same S3 bucket, if
+//! their bucket prefixes are both specified and different.
 
 use std::path::{Path, PathBuf};
 
 use anyhow::Context;
 use s3::{bucket::Bucket, creds::Credentials, region::Region};
 use tokio::io::{self, AsyncWriteExt};
+use tracing::debug;
 
 use crate::{
     config::S3Config,
@@ -23,8 +26,26 @@ impl S3ObjectKey {
         &self.0
     }
 
-    fn download_destination(&self, pageserver_workdir: &Path) -> PathBuf {
-        pageserver_workdir.join(self.0.split(S3_FILE_SEPARATOR).collect::<PathBuf>())
+    fn download_destination(
+        &self,
+        pageserver_workdir: &Path,
+        prefix_to_strip: Option<&str>,
+    ) -> PathBuf {
+        let path_without_prefix = match prefix_to_strip {
+            Some(prefix) => self.0.strip_prefix(prefix).unwrap_or_else(|| {
+                panic!(
+                    "Could not strip prefix '{}' from S3 object key '{}'",
+                    prefix, self.0
+                )
+            }),
+            None => &self.0,
+        };
+
+        pageserver_workdir.join(
+            path_without_prefix
+                .split(S3_FILE_SEPARATOR)
+                .collect::<PathBuf>(),
+        )
     }
 }
 
@@ -32,15 +53,27 @@ impl S3ObjectKey {
 pub struct S3 {
     pageserver_workdir: &'static Path,
     bucket: Bucket,
+    prefix_in_bucket: Option<String>,
 }
 
 impl S3 {
     /// Creates the storage, errors if incorrect AWS S3 configuration provided.
     pub fn new(aws_config: &S3Config, pageserver_workdir: &'static Path) -> anyhow::Result<Self> {
-        let region = aws_config
-            .bucket_region
-            .parse::<Region>()
-            .context("Failed to parse the s3 region from config")?;
+        debug!(
+            "Creating s3 remote storage around bucket {}",
+            aws_config.bucket_name
+        );
+        let region = match aws_config.endpoint.clone() {
+            Some(endpoint) => Region::Custom {
+                endpoint,
+                region: aws_config.bucket_region.clone(),
+            },
+            None => aws_config
+                .bucket_region
+                .parse::<Region>()
+                .context("Failed to parse the s3 region from config")?,
+        };
+
         let credentials = Credentials::new(
             aws_config.access_key_id.as_deref(),
             aws_config.secret_access_key.as_deref(),
@@ -49,6 +82,20 @@ impl S3 {
             None,
         )
         .context("Failed to create the s3 credentials")?;
+
+        let prefix_in_bucket = aws_config.prefix_in_bucket.as_deref().map(|prefix| {
+            let mut prefix = prefix;
+            while prefix.starts_with(S3_FILE_SEPARATOR) {
+                prefix = &prefix[1..]
+            }
+
+            let mut prefix = prefix.to_string();
+            while prefix.ends_with(S3_FILE_SEPARATOR) {
+                prefix.pop();
+            }
+            prefix
+        });
+
         Ok(Self {
             bucket: Bucket::new_with_path_style(
                 aws_config.bucket_name.as_str(),
@@ -57,6 +104,7 @@ impl S3 {
             )
             .context("Failed to create the s3 bucket")?,
             pageserver_workdir,
+            prefix_in_bucket,
         })
     }
 }
@@ -67,7 +115,7 @@ impl RemoteStorage for S3 {
 
     fn storage_path(&self, local_path: &Path) -> anyhow::Result<Self::StoragePath> {
         let relative_path = strip_path_prefix(self.pageserver_workdir, local_path)?;
-        let mut key = String::new();
+        let mut key = self.prefix_in_bucket.clone().unwrap_or_default();
         for segment in relative_path {
             key.push(S3_FILE_SEPARATOR);
             key.push_str(&segment.to_string_lossy());
@@ -76,13 +124,14 @@ impl RemoteStorage for S3 {
     }
 
     fn local_path(&self, storage_path: &Self::StoragePath) -> anyhow::Result<PathBuf> {
-        Ok(storage_path.download_destination(self.pageserver_workdir))
+        Ok(storage_path
+            .download_destination(self.pageserver_workdir, self.prefix_in_bucket.as_deref()))
     }
 
     async fn list(&self) -> anyhow::Result<Vec<Self::StoragePath>> {
         let list_response = self
             .bucket
-            .list(String::new(), None)
+            .list(self.prefix_in_bucket.clone().unwrap_or_default(), None)
             .await
             .context("Failed to list s3 objects")?;
 
@@ -225,7 +274,7 @@ mod tests {
 
         assert_eq!(
             local_path,
-            key.download_destination(&repo_harness.conf.workdir),
+            key.download_destination(&repo_harness.conf.workdir, None),
             "Download destination should consist of s3 path joined with the pageserver workdir prefix"
         );
 
@@ -239,14 +288,18 @@ mod tests {
         let segment_1 = "matching";
         let segment_2 = "file";
         let local_path = &repo_harness.conf.workdir.join(segment_1).join(segment_2);
+
+        let storage = dummy_storage(&repo_harness.conf.workdir);
+
         let expected_key = S3ObjectKey(format!(
-            "{SEPARATOR}{}{SEPARATOR}{}",
+            "{}{SEPARATOR}{}{SEPARATOR}{}",
+            storage.prefix_in_bucket.as_deref().unwrap_or_default(),
             segment_1,
             segment_2,
             SEPARATOR = S3_FILE_SEPARATOR,
         ));
 
-        let actual_key = dummy_storage(&repo_harness.conf.workdir)
+        let actual_key = storage
             .storage_path(local_path)
             .expect("Matching path should map to S3 path normally");
         assert_eq!(
@@ -308,18 +361,30 @@ mod tests {
         let timeline_dir = repo_harness.timeline_path(&TIMELINE_ID);
         let relative_timeline_path = timeline_dir.strip_prefix(&repo_harness.conf.workdir)?;
 
-        let s3_key = create_s3_key(&relative_timeline_path.join("not a metadata"));
+        let s3_key = create_s3_key(
+            &relative_timeline_path.join("not a metadata"),
+            storage.prefix_in_bucket.as_deref(),
+        );
         assert_eq!(
-            s3_key.download_destination(&repo_harness.conf.workdir),
+            s3_key.download_destination(
+                &repo_harness.conf.workdir,
+                storage.prefix_in_bucket.as_deref()
+            ),
             storage
                 .local_path(&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 file"
         );
 
-        let s3_key = create_s3_key(&relative_timeline_path.join(METADATA_FILE_NAME));
+        let s3_key = create_s3_key(
+            &relative_timeline_path.join(METADATA_FILE_NAME),
+            storage.prefix_in_bucket.as_deref(),
+        );
         assert_eq!(
-            s3_key.download_destination(&repo_harness.conf.workdir),
+            s3_key.download_destination(
+                &repo_harness.conf.workdir,
+                storage.prefix_in_bucket.as_deref()
+            ),
             storage
                 .local_path(&s3_key)
                 .expect("For a valid input, valid S3 info should be parsed"),
@@ -356,18 +421,18 @@ mod tests {
                 Credentials::anonymous().unwrap(),
             )
             .unwrap(),
+            prefix_in_bucket: Some("dummy_prefix/".to_string()),
         }
     }
 
-    fn create_s3_key(relative_file_path: &Path) -> S3ObjectKey {
-        S3ObjectKey(
-            relative_file_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
-                }),
-        )
+    fn create_s3_key(relative_file_path: &Path, prefix: Option<&str>) -> S3ObjectKey {
+        S3ObjectKey(relative_file_path.iter().fold(
+            prefix.unwrap_or_default().to_string(),
+            |mut path_string, segment| {
+                path_string.push(S3_FILE_SEPARATOR);
+                path_string.push_str(segment.to_str().unwrap());
+                path_string
+            },
+        ))
     }
 }

pageserver/src/remote_storage/storage_sync.rs

@@ -70,7 +70,8 @@
 //!
 //! When pageserver signals shutdown, current sync task gets finished and the loop exists.
 
-mod compression;
+/// Expose the module for a binary CLI tool that deals with the corresponding blobs.
+pub mod compression;
 mod download;
 pub mod index;
 mod upload;
@@ -80,16 +81,19 @@ use std::{
     num::{NonZeroU32, NonZeroUsize},
     path::{Path, PathBuf},
     sync::Arc,
-    thread,
 };
 
 use anyhow::{bail, Context};
 use futures::stream::{FuturesUnordered, StreamExt};
 use lazy_static::lazy_static;
-use tokio::{fs, sync::RwLock};
 use tokio::{
-    sync::mpsc::{self, UnboundedReceiver},
-    time::Instant,
+    fs,
+    runtime::Runtime,
+    sync::{
+        mpsc::{self, UnboundedReceiver},
+        RwLock,
+    },
+    time::{Duration, Instant},
 };
 use tracing::*;
 
@@ -102,11 +106,11 @@ use self::{
     },
     upload::upload_timeline_checkpoint,
 };
-use super::{RemoteStorage, SyncStartupData, TimelineSyncId};
+use super::{RemoteStorage, SyncStartupData, ZTenantTimelineId};
 use crate::{
     config::PageServerConf, layered_repository::metadata::TimelineMetadata,
     remote_storage::storage_sync::compression::read_archive_header, repository::TimelineSyncState,
-    tenant_mgr::set_timeline_states,
+    tenant_mgr::set_timeline_states, thread_mgr, thread_mgr::ThreadKind,
 };
 
 use zenith_metrics::{register_histogram_vec, register_int_gauge, HistogramVec, IntGauge};
@@ -239,13 +243,13 @@ mod sync_queue {
 /// Limited by the number of retries, after certain threshold the failing task gets evicted and the timeline disabled.
 #[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone)]
 pub struct SyncTask {
-    sync_id: TimelineSyncId,
+    sync_id: ZTenantTimelineId,
     retries: u32,
     kind: SyncKind,
 }
 
 impl SyncTask {
-    fn new(sync_id: TimelineSyncId, retries: u32, kind: SyncKind) -> Self {
+    fn new(sync_id: ZTenantTimelineId, retries: u32, kind: SyncKind) -> Self {
         Self {
             sync_id,
             retries,
@@ -304,7 +308,10 @@ pub fn schedule_timeline_checkpoint_upload(
     }
 
     if !sync_queue::push(SyncTask::new(
-        TimelineSyncId(tenant_id, timeline_id),
+        ZTenantTimelineId {
+            tenant_id,
+            timeline_id,
+        },
         0,
         SyncKind::Upload(NewCheckpoint { layers, metadata }),
     )) {
@@ -330,8 +337,15 @@ pub fn schedule_timeline_checkpoint_upload(
 ///
 /// Ensure that the loop is started otherwise the task is never processed.
 pub fn schedule_timeline_download(tenant_id: ZTenantId, timeline_id: ZTimelineId) {
+    debug!(
+        "Scheduling timeline download for tenant {}, timeline {}",
+        tenant_id, timeline_id
+    );
     sync_queue::push(SyncTask::new(
-        TimelineSyncId(tenant_id, timeline_id),
+        ZTenantTimelineId {
+            tenant_id,
+            timeline_id,
+        },
         0,
         SyncKind::Download(TimelineDownload {
             files_to_skip: Arc::new(BTreeSet::new()),
@@ -347,7 +361,7 @@ pub(super) fn spawn_storage_sync_thread<
     S: RemoteStorage<StoragePath = P> + Send + Sync + 'static,
 >(
     conf: &'static PageServerConf,
-    local_timeline_files: HashMap<TimelineSyncId, (TimelineMetadata, Vec<PathBuf>)>,
+    local_timeline_files: HashMap<ZTenantTimelineId, (TimelineMetadata, Vec<PathBuf>)>,
     storage: S,
     max_concurrent_sync: NonZeroUsize,
     max_sync_errors: NonZeroU32,
@@ -369,7 +383,7 @@ pub(super) fn spawn_storage_sync_thread<
             Ok(local_path) => Some(local_path),
             Err(e) => {
                 error!(
-                    "Failed to find local path for remote path {:?}: {:#}",
+                    "Failed to find local path for remote path {:?}: {:?}",
                     remote_path, e
                 );
                 None
@@ -379,9 +393,12 @@ pub(super) fn spawn_storage_sync_thread<
 
     let initial_timeline_states = schedule_first_sync_tasks(&remote_index, local_timeline_files);
 
-    let handle = thread::Builder::new()
-        .name("Remote storage sync thread".to_string())
-        .spawn(move || {
+    thread_mgr::spawn(
+        ThreadKind::StorageSync,
+        None,
+        None,
+        "Remote storage sync thread",
+        move || {
             storage_sync_loop(
                 runtime,
                 conf,
@@ -391,19 +408,25 @@ pub(super) fn spawn_storage_sync_thread<
                 max_concurrent_sync,
                 max_sync_errors,
             )
-        })
-        .context("Failed to spawn remote storage sync thread")?;
+        },
+    )
+    .context("Failed to spawn remote storage sync thread")?;
     Ok(SyncStartupData {
         initial_timeline_states,
-        sync_loop_handle: Some(handle),
     })
 }
 
+enum LoopStep {
+    NewStates(HashMap<ZTenantId, HashMap<ZTimelineId, TimelineSyncState>>),
+    Shutdown,
+}
+
+#[allow(clippy::too_many_arguments)]
 fn storage_sync_loop<
     P: std::fmt::Debug + Send + Sync + 'static,
     S: RemoteStorage<StoragePath = P> + Send + Sync + 'static,
 >(
-    runtime: tokio::runtime::Runtime,
+    runtime: Runtime,
     conf: &'static PageServerConf,
     mut receiver: UnboundedReceiver<SyncTask>,
     index: RemoteTimelineIndex,
@@ -412,23 +435,34 @@ fn storage_sync_loop<
     max_sync_errors: NonZeroU32,
 ) -> anyhow::Result<()> {
     let remote_assets = Arc::new((storage, RwLock::new(index)));
-    while !crate::tenant_mgr::shutdown_requested() {
-        let new_timeline_states = runtime.block_on(
-            loop_step(
-                conf,
-                &mut receiver,
-                Arc::clone(&remote_assets),
-                max_concurrent_sync,
-                max_sync_errors,
-            )
-            .instrument(debug_span!("storage_sync_loop_step")),
-        );
-        // Batch timeline download registration to ensure that the external registration code won't block any running tasks before.
-        set_timeline_states(conf, new_timeline_states);
-        debug!("Sync loop step completed");
+    loop {
+        let loop_step = runtime.block_on(async {
+            tokio::select! {
+                new_timeline_states = loop_step(
+                    conf,
+                    &mut receiver,
+                    Arc::clone(&remote_assets),
+                    max_concurrent_sync,
+                    max_sync_errors,
+                )
+                .instrument(debug_span!("storage_sync_loop_step")) => LoopStep::NewStates(new_timeline_states),
+                _ = thread_mgr::shutdown_watcher() => LoopStep::Shutdown,
+            }
+        });
+
+        match loop_step {
+            LoopStep::NewStates(new_timeline_states) => {
+                // Batch timeline download registration to ensure that the external registration code won't block any running tasks before.
+                set_timeline_states(conf, new_timeline_states);
+                debug!("Sync loop step completed");
+            }
+            LoopStep::Shutdown => {
+                debug!("Shutdown requested, stopping");
+                break;
+            }
+        }
     }
 
-    debug!("Shutdown requested, stopping");
     Ok(())
 }
 
@@ -482,8 +516,8 @@ async fn loop_step<
                 Ok(extra_step) => extra_step,
                 Err(e) => {
                     error!(
-                        "Failed to process storage sync task for tenant {}, timeline {}: {:#}",
-                        sync_id.0, sync_id.1, e
+                        "Failed to process storage sync task for tenant {}, timeline {}: {:?}",
+                        sync_id.tenant_id, sync_id.timeline_id, e
                     );
                     None
                 }
@@ -497,7 +531,10 @@ async fn loop_step<
     while let Some((sync_id, state_update)) = task_batch.next().await {
         debug!("Finished storage sync task for sync id {}", sync_id);
         if let Some(state_update) = state_update {
-            let TimelineSyncId(tenant_id, timeline_id) = sync_id;
+            let ZTenantTimelineId {
+                tenant_id,
+                timeline_id,
+            } = sync_id;
             new_timeline_states
                 .entry(tenant_id)
                 .or_default()
@@ -539,7 +576,7 @@ async fn process_task<
             "Waiting {} seconds before starting the task",
             seconds_to_wait
         );
-        tokio::time::sleep(tokio::time::Duration::from_secs_f64(seconds_to_wait)).await;
+        tokio::time::sleep(Duration::from_secs_f64(seconds_to_wait)).await;
     }
 
     let sync_start = Instant::now();
@@ -591,7 +628,7 @@ async fn process_task<
 
 fn schedule_first_sync_tasks(
     index: &RemoteTimelineIndex,
-    local_timeline_files: HashMap<TimelineSyncId, (TimelineMetadata, Vec<PathBuf>)>,
+    local_timeline_files: HashMap<ZTenantTimelineId, (TimelineMetadata, Vec<PathBuf>)>,
 ) -> HashMap<ZTenantId, HashMap<ZTimelineId, TimelineSyncState>> {
     let mut initial_timeline_statuses: HashMap<ZTenantId, HashMap<ZTimelineId, TimelineSyncState>> =
         HashMap::new();
@@ -602,7 +639,10 @@ fn schedule_first_sync_tasks(
     for (sync_id, (local_metadata, local_files)) in local_timeline_files {
         let local_disk_consistent_lsn = local_metadata.disk_consistent_lsn();
 
-        let TimelineSyncId(tenant_id, timeline_id) = sync_id;
+        let ZTenantTimelineId {
+            tenant_id,
+            timeline_id,
+        } = sync_id;
         match index.timeline_entry(&sync_id) {
             Some(index_entry) => {
                 let timeline_status = compare_local_and_remote_timeline(
@@ -645,10 +685,10 @@ fn schedule_first_sync_tasks(
         }
     }
 
-    let unprocessed_remote_ids = |remote_id: &TimelineSyncId| {
+    let unprocessed_remote_ids = |remote_id: &ZTenantTimelineId| {
         initial_timeline_statuses
-            .get(&remote_id.0)
-            .and_then(|timelines| timelines.get(&remote_id.1))
+            .get(&remote_id.tenant_id)
+            .and_then(|timelines| timelines.get(&remote_id.timeline_id))
             .is_none()
     };
     for unprocessed_remote_id in index
@@ -656,7 +696,10 @@ fn schedule_first_sync_tasks(
         .filter(unprocessed_remote_ids)
         .collect::<Vec<_>>()
     {
-        let TimelineSyncId(cloud_only_tenant_id, cloud_only_timeline_id) = unprocessed_remote_id;
+        let ZTenantTimelineId {
+            tenant_id: cloud_only_tenant_id,
+            timeline_id: cloud_only_timeline_id,
+        } = unprocessed_remote_id;
         match index
             .timeline_entry(&unprocessed_remote_id)
             .and_then(TimelineIndexEntry::disk_consistent_lsn)
@@ -685,7 +728,7 @@ fn schedule_first_sync_tasks(
 
 fn compare_local_and_remote_timeline(
     new_sync_tasks: &mut VecDeque<SyncTask>,
-    sync_id: TimelineSyncId,
+    sync_id: ZTenantTimelineId,
     local_metadata: TimelineMetadata,
     local_files: Vec<PathBuf>,
     remote_entry: &TimelineIndexEntry,
@@ -742,7 +785,7 @@ async fn update_index_description<
 >(
     (storage, index): &(S, RwLock<RemoteTimelineIndex>),
     timeline_dir: &Path,
-    id: TimelineSyncId,
+    id: ZTenantTimelineId,
 ) -> anyhow::Result<RemoteTimeline> {
     let mut index_write = index.write().await;
     let full_index = match index_write.timeline_entry(&id) {
@@ -765,7 +808,7 @@ async fn update_index_description<
                         Ok((archive_id, header_size, header)) => full_index.update_archive_contents(archive_id.0, header, header_size),
                         Err((e, archive_id)) => bail!(
                             "Failed to download archive header for tenant {}, timeline {}, archive for Lsn {}: {}",
-                            id.0, id.1, archive_id.0,
+                            id.tenant_id, id.timeline_id, archive_id.0,
                             e
                         ),
                     }
@@ -843,7 +886,7 @@ mod test_utils {
         timeline_id: ZTimelineId,
         new_upload: NewCheckpoint,
     ) {
-        let sync_id = TimelineSyncId(harness.tenant_id, timeline_id);
+        let sync_id = ZTenantTimelineId::new(harness.tenant_id, timeline_id);
         upload_timeline_checkpoint(
             harness.conf,
             Arc::clone(&remote_assets),
@@ -899,7 +942,7 @@ mod test_utils {
 
     pub async fn expect_timeline(
         index: &RwLock<RemoteTimelineIndex>,
-        sync_id: TimelineSyncId,
+        sync_id: ZTenantTimelineId,
     ) -> RemoteTimeline {
         if let Some(TimelineIndexEntry::Full(remote_timeline)) =
             index.read().await.timeline_entry(&sync_id)
@@ -934,18 +977,18 @@ mod test_utils {
         let mut expected_timeline_entries = BTreeMap::new();
         for sync_id in actual_sync_ids {
             actual_branches.insert(
-                sync_id.1,
+                sync_id.tenant_id,
                 index_read
-                    .branch_files(sync_id.0)
+                    .branch_files(sync_id.tenant_id)
                     .into_iter()
                     .flat_map(|branch_paths| branch_paths.iter())
                     .cloned()
                     .collect::<BTreeSet<_>>(),
             );
             expected_branches.insert(
-                sync_id.1,
+                sync_id.tenant_id,
                 expected_index_with_descriptions
-                    .branch_files(sync_id.0)
+                    .branch_files(sync_id.tenant_id)
                     .into_iter()
                     .flat_map(|branch_paths| branch_paths.iter())
                     .cloned()

pageserver/src/remote_storage/storage_sync/compression.rs

@@ -34,7 +34,7 @@ use std::{
     sync::Arc,
 };
 
-use anyhow::{anyhow, bail, ensure, Context};
+use anyhow::{bail, ensure, Context};
 use async_compression::tokio::bufread::{ZstdDecoder, ZstdEncoder};
 use serde::{Deserialize, Serialize};
 use tokio::{
@@ -211,16 +211,18 @@ pub async fn read_archive_header<A: io::AsyncRead + Send + Sync + Unpin>(
 pub fn parse_archive_name(archive_path: &Path) -> anyhow::Result<(Lsn, u64)> {
     let archive_name = archive_path
         .file_name()
-        .ok_or_else(|| anyhow!("Archive '{}' has no file name", archive_path.display()))?
+        .with_context(|| format!("Archive '{}' has no file name", archive_path.display()))?
         .to_string_lossy();
     let (lsn_str, header_size_str) =
-        archive_name.rsplit_once(ARCHIVE_EXTENSION).ok_or_else(|| {
-            anyhow!(
-                "Archive '{}' has incorrect extension, expected to contain '{}'",
-                archive_path.display(),
-                ARCHIVE_EXTENSION
-            )
-        })?;
+        archive_name
+            .rsplit_once(ARCHIVE_EXTENSION)
+            .with_context(|| {
+                format!(
+                    "Archive '{}' has incorrect extension, expected to contain '{}'",
+                    archive_path.display(),
+                    ARCHIVE_EXTENSION
+                )
+            })?;
     let disk_consistent_lsn = Lsn::from_hex(lsn_str).with_context(|| {
         format!(
             "Archive '{}' has an invalid disk consistent lsn in its extension",
@@ -246,7 +248,7 @@ fn archive_name(disk_consistent_lsn: Lsn, header_size: u64) -> String {
     archive_name
 }
 
-async fn uncompress_with_header(
+pub async fn uncompress_with_header(
     files_to_skip: &BTreeSet<PathBuf>,
     destination_dir: &Path,
     header: ArchiveHeader,
@@ -374,7 +376,7 @@ async fn write_archive_contents(
     }
     let metadata_bytes_written = io::copy(&mut metadata_bytes.as_slice(), &mut archive_input)
         .await
-        .with_context(|| "Failed to add metadata into the archive")?;
+        .context("Failed to add metadata into the archive")?;
     ensure!(
         header.metadata_file_size == metadata_bytes_written,
         "Metadata file was written to the archive incompletely",

pageserver/src/remote_storage/storage_sync/download.rs

@@ -3,7 +3,7 @@
 
 use std::{borrow::Cow, collections::BTreeSet, path::PathBuf, sync::Arc};
 
-use anyhow::{anyhow, ensure, Context};
+use anyhow::{ensure, Context};
 use futures::{stream::FuturesUnordered, StreamExt};
 use tokio::{fs, sync::RwLock};
 use tracing::{debug, error, trace, warn};
@@ -17,7 +17,7 @@ use crate::{
             compression, index::TimelineIndexEntry, sync_queue, tenant_branch_files,
             update_index_description, SyncKind, SyncTask,
         },
-        RemoteStorage, TimelineSyncId,
+        RemoteStorage, ZTenantTimelineId,
     },
 };
 
@@ -52,13 +52,16 @@ pub(super) async fn download_timeline<
 >(
     conf: &'static PageServerConf,
     remote_assets: Arc<(S, RwLock<RemoteTimelineIndex>)>,
-    sync_id: TimelineSyncId,
+    sync_id: ZTenantTimelineId,
     mut download: TimelineDownload,
     retries: u32,
 ) -> DownloadedTimeline {
     debug!("Downloading layers for sync id {}", sync_id);
 
-    let TimelineSyncId(tenant_id, timeline_id) = sync_id;
+    let ZTenantTimelineId {
+        tenant_id,
+        timeline_id,
+    } = sync_id;
     let index_read = remote_assets.1.read().await;
     let remote_timeline = match index_read.timeline_entry(&sync_id) {
         None => {
@@ -80,7 +83,7 @@ pub(super) async fn download_timeline<
                 {
                     Ok(remote_timeline) => Cow::Owned(remote_timeline),
                     Err(e) => {
-                        error!("Failed to download full timeline index: {:#}", e);
+                        error!("Failed to download full timeline index: {:?}", e);
                         return match remote_disk_consistent_lsn {
                             Some(disk_consistent_lsn) => {
                                 sync_queue::push(SyncTask::new(
@@ -110,9 +113,10 @@ pub(super) async fn download_timeline<
         }
     };
 
-    if let Err(e) = download_missing_branches(conf, remote_assets.as_ref(), sync_id.0).await {
+    if let Err(e) = download_missing_branches(conf, remote_assets.as_ref(), sync_id.tenant_id).await
+    {
         error!(
-            "Failed to download missing branches for sync id {}: {:#}",
+            "Failed to download missing branches for sync id {}: {:?}",
             sync_id, e
         );
         sync_queue::push(SyncTask::new(
@@ -150,7 +154,7 @@ pub(super) async fn download_timeline<
             Err(e) => {
                 let archives_left = archives_total - archives_downloaded;
                 error!(
-                    "Failed to download archive {:?} (archives downloaded: {}; archives left: {}) for tenant {} timeline {}, requeueing the download: {:#}",
+                    "Failed to download archive {:?} (archives downloaded: {}; archives left: {}) for tenant {} timeline {}, requeueing the download: {:?}",
                     archive_id, archives_downloaded, archives_left, tenant_id, timeline_id, e
                 );
                 sync_queue::push(SyncTask::new(
@@ -180,7 +184,10 @@ async fn try_download_archive<
     S: RemoteStorage<StoragePath = P> + Send + Sync + 'static,
 >(
     conf: &'static PageServerConf,
-    TimelineSyncId(tenant_id, timeline_id): TimelineSyncId,
+    ZTenantTimelineId {
+        tenant_id,
+        timeline_id,
+    }: ZTenantTimelineId,
     remote_assets: Arc<(S, RwLock<RemoteTimelineIndex>)>,
     remote_timeline: &RemoteTimeline,
     archive_id: ArchiveId,
@@ -189,7 +196,7 @@ async fn try_download_archive<
     debug!("Downloading archive {:?}", archive_id);
     let archive_to_download = remote_timeline
         .archive_data(archive_id)
-        .ok_or_else(|| anyhow!("Archive {:?} not found in remote storage", archive_id))?;
+        .with_context(|| format!("Archive {:?} not found in remote storage", archive_id))?;
     let (archive_header, header_size) = remote_timeline
         .restore_header(archive_id)
         .context("Failed to restore header when downloading an archive")?;
@@ -202,7 +209,7 @@ async fn try_download_archive<
             archive_to_download.disk_consistent_lsn(),
             local_metadata.disk_consistent_lsn()
         ),
-        Err(e) => warn!("Failed to read local metadata file, assuing it's safe to override its with the download. Read: {:#}", e),
+        Err(e) => warn!("Failed to read local metadata file, assuming it's safe to override its with the download. Read: {:#}", e),
     }
     compression::uncompress_file_stream_with_index(
         conf.timeline_path(&timeline_id, &tenant_id),
@@ -307,7 +314,7 @@ async fn download_missing_branches<
         while let Some(download_result) = remote_only_branches_downloads.next().await {
             if let Err(e) = download_result {
                 branch_downloads_failed = true;
-                error!("Failed to download a branch file: {:#}", e);
+                error!("Failed to download a branch file: {:?}", e);
             }
         }
         ensure!(
@@ -343,7 +350,7 @@ mod tests {
     #[tokio::test]
     async fn test_download_timeline() -> anyhow::Result<()> {
         let repo_harness = RepoHarness::create("test_download_timeline")?;
-        let sync_id = TimelineSyncId(repo_harness.tenant_id, TIMELINE_ID);
+        let sync_id = ZTenantTimelineId::new(repo_harness.tenant_id, TIMELINE_ID);
         let storage = LocalFs::new(tempdir()?.path().to_owned(), &repo_harness.conf.workdir)?;
         let index = RwLock::new(RemoteTimelineIndex::try_parse_descriptions_from_paths(
             repo_harness.conf,

pageserver/src/remote_storage/storage_sync/index.rs

@@ -9,7 +9,7 @@ use std::{
     path::{Path, PathBuf},
 };
 
-use anyhow::{anyhow, bail, ensure, Context};
+use anyhow::{bail, ensure, Context};
 use serde::{Deserialize, Serialize};
 use tracing::debug;
 use zenith_utils::{
@@ -22,7 +22,7 @@ use crate::{
     layered_repository::TIMELINES_SEGMENT_NAME,
     remote_storage::{
         storage_sync::compression::{parse_archive_name, FileEntry},
-        TimelineSyncId,
+        ZTenantTimelineId,
     },
 };
 
@@ -53,7 +53,7 @@ impl RelativePath {
 #[derive(Debug, Clone)]
 pub struct RemoteTimelineIndex {
     branch_files: HashMap<ZTenantId, HashSet<RelativePath>>,
-    timeline_files: HashMap<TimelineSyncId, TimelineIndexEntry>,
+    timeline_files: HashMap<ZTenantTimelineId, TimelineIndexEntry>,
 }
 
 impl RemoteTimelineIndex {
@@ -80,19 +80,22 @@ impl RemoteTimelineIndex {
         index
     }
 
-    pub fn timeline_entry(&self, id: &TimelineSyncId) -> Option<&TimelineIndexEntry> {
+    pub fn timeline_entry(&self, id: &ZTenantTimelineId) -> Option<&TimelineIndexEntry> {
         self.timeline_files.get(id)
     }
 
-    pub fn timeline_entry_mut(&mut self, id: &TimelineSyncId) -> Option<&mut TimelineIndexEntry> {
+    pub fn timeline_entry_mut(
+        &mut self,
+        id: &ZTenantTimelineId,
+    ) -> Option<&mut TimelineIndexEntry> {
         self.timeline_files.get_mut(id)
     }
 
-    pub fn add_timeline_entry(&mut self, id: TimelineSyncId, entry: TimelineIndexEntry) {
+    pub fn add_timeline_entry(&mut self, id: ZTenantTimelineId, entry: TimelineIndexEntry) {
         self.timeline_files.insert(id, entry);
     }
 
-    pub fn all_sync_ids(&self) -> impl Iterator<Item = TimelineSyncId> + '_ {
+    pub fn all_sync_ids(&self) -> impl Iterator<Item = ZTenantTimelineId> + '_ {
         self.timeline_files.keys().copied()
     }
 
@@ -214,7 +217,7 @@ impl RemoteTimeline {
         let archive = self
             .checkpoint_archives
             .get(&archive_id)
-            .ok_or_else(|| anyhow!("Archive {:?} not found", archive_id))?;
+            .with_context(|| format!("Archive {:?} not found", archive_id))?;
 
         let mut header_files = Vec::with_capacity(archive.files.len());
         for (expected_archive_position, archive_file) in archive.files.iter().enumerate() {
@@ -226,11 +229,10 @@ impl RemoteTimeline {
                 archive_id,
             );
 
-            let timeline_file = self.timeline_files.get(archive_file).ok_or_else(|| {
-                anyhow!(
+            let timeline_file = self.timeline_files.get(archive_file).with_context(|| {
+                format!(
                     "File with id {:?} not found for archive {:?}",
-                    archive_file,
-                    archive_id
+                    archive_file, archive_id
                 )
             })?;
             header_files.push(timeline_file.clone());
@@ -299,7 +301,7 @@ fn try_parse_index_entry(
         })?
         .iter()
         .next()
-        .ok_or_else(|| anyhow!("Found no tenant id in path '{}'", path.display()))?
+        .with_context(|| format!("Found no tenant id in path '{}'", path.display()))?
         .to_string_lossy()
         .parse::<ZTenantId>()
         .with_context(|| format!("Failed to parse tenant id from path '{}'", path.display()))?;
@@ -321,8 +323,8 @@ fn try_parse_index_entry(
             let mut segments = timelines_subpath.iter();
             let timeline_id = segments
                 .next()
-                .ok_or_else(|| {
-                    anyhow!(
+                .with_context(|| {
+                    format!(
                         "{} directory of tenant {} (path '{}') is not an index entry",
                         TIMELINES_SEGMENT_NAME,
                         tenant_id,
@@ -345,11 +347,14 @@ fn try_parse_index_entry(
 
             let archive_name = path
                 .file_name()
-                .ok_or_else(|| anyhow!("Archive '{}' has no file name", path.display()))?
+                .with_context(|| format!("Archive '{}' has no file name", path.display()))?
                 .to_string_lossy()
                 .to_string();
 
-            let sync_id = TimelineSyncId(tenant_id, timeline_id);
+            let sync_id = ZTenantTimelineId {
+                tenant_id,
+                timeline_id,
+            };
             let timeline_index_entry = index
                 .timeline_files
                 .entry(sync_id)

pageserver/src/remote_storage/storage_sync/upload.rs

@@ -17,7 +17,7 @@ use crate::{
             index::{RemoteTimeline, TimelineIndexEntry},
             sync_queue, tenant_branch_files, update_index_description, SyncKind, SyncTask,
         },
-        RemoteStorage, TimelineSyncId,
+        RemoteStorage, ZTenantTimelineId,
     },
 };
 
@@ -36,14 +36,15 @@ pub(super) async fn upload_timeline_checkpoint<
 >(
     config: &'static PageServerConf,
     remote_assets: Arc<(S, RwLock<RemoteTimelineIndex>)>,
-    sync_id: TimelineSyncId,
+    sync_id: ZTenantTimelineId,
     new_checkpoint: NewCheckpoint,
     retries: u32,
 ) -> Option<bool> {
     debug!("Uploading checkpoint for sync id {}", sync_id);
-    if let Err(e) = upload_missing_branches(config, remote_assets.as_ref(), sync_id.0).await {
+    if let Err(e) = upload_missing_branches(config, remote_assets.as_ref(), sync_id.tenant_id).await
+    {
         error!(
-            "Failed to upload missing branches for sync id {}: {:#}",
+            "Failed to upload missing branches for sync id {}: {:?}",
             sync_id, e
         );
         sync_queue::push(SyncTask::new(
@@ -57,7 +58,10 @@ pub(super) async fn upload_timeline_checkpoint<
 
     let index = &remote_assets.1;
 
-    let TimelineSyncId(tenant_id, timeline_id) = sync_id;
+    let ZTenantTimelineId {
+        tenant_id,
+        timeline_id,
+    } = sync_id;
     let timeline_dir = config.timeline_path(&timeline_id, &tenant_id);
 
     let index_read = index.read().await;
@@ -69,7 +73,7 @@ pub(super) async fn upload_timeline_checkpoint<
             match update_index_description(remote_assets.as_ref(), &timeline_dir, sync_id).await {
                 Ok(remote_timeline) => Some(Cow::Owned(remote_timeline)),
                 Err(e) => {
-                    error!("Failed to download full timeline index: {:#}", e);
+                    error!("Failed to download full timeline index: {:?}", e);
                     sync_queue::push(SyncTask::new(
                         sync_id,
                         retries,
@@ -132,7 +136,7 @@ pub(super) async fn upload_timeline_checkpoint<
         }
         Err(e) => {
             error!(
-                "Failed to upload checkpoint: {:#}, requeueing the upload",
+                "Failed to upload checkpoint: {:?}, requeueing the upload",
                 e
             );
             sync_queue::push(SyncTask::new(
@@ -151,11 +155,14 @@ async fn try_upload_checkpoint<
 >(
     config: &'static PageServerConf,
     remote_assets: Arc<(S, RwLock<RemoteTimelineIndex>)>,
-    sync_id: TimelineSyncId,
+    sync_id: ZTenantTimelineId,
     new_checkpoint: &NewCheckpoint,
     files_to_skip: BTreeSet<PathBuf>,
 ) -> anyhow::Result<(ArchiveHeader, u64)> {
-    let TimelineSyncId(tenant_id, timeline_id) = sync_id;
+    let ZTenantTimelineId {
+        tenant_id,
+        timeline_id,
+    } = sync_id;
     let timeline_dir = config.timeline_path(&timeline_id, &tenant_id);
 
     let files_to_upload = new_checkpoint
@@ -253,7 +260,7 @@ async fn upload_missing_branches<
                 .await
                 .add_branch_file(tenant_id, local_only_branch.clone()),
             Err(e) => {
-                error!("Failed to upload branch file: {:#}", e);
+                error!("Failed to upload branch file: {:?}", e);
                 branch_uploads_failed = true;
             }
         }
@@ -288,7 +295,7 @@ mod tests {
     #[tokio::test]
     async fn reupload_timeline() -> anyhow::Result<()> {
         let repo_harness = RepoHarness::create("reupload_timeline")?;
-        let sync_id = TimelineSyncId(repo_harness.tenant_id, TIMELINE_ID);
+        let sync_id = ZTenantTimelineId::new(repo_harness.tenant_id, TIMELINE_ID);
         let storage = LocalFs::new(tempdir()?.path().to_owned(), &repo_harness.conf.workdir)?;
         let index = RwLock::new(RemoteTimelineIndex::try_parse_descriptions_from_paths(
             repo_harness.conf,
@@ -484,7 +491,7 @@ mod tests {
     #[tokio::test]
     async fn reupload_timeline_rejected() -> anyhow::Result<()> {
         let repo_harness = RepoHarness::create("reupload_timeline_rejected")?;
-        let sync_id = TimelineSyncId(repo_harness.tenant_id, TIMELINE_ID);
+        let sync_id = ZTenantTimelineId::new(repo_harness.tenant_id, TIMELINE_ID);
         let storage = LocalFs::new(tempdir()?.path().to_owned(), &repo_harness.conf.workdir)?;
         let index = RwLock::new(RemoteTimelineIndex::try_parse_descriptions_from_paths(
             repo_harness.conf,

pageserver/src/repository.rs

@@ -7,7 +7,7 @@ use postgres_ffi::{MultiXactId, MultiXactOffset, TransactionId};
 use serde::{Deserialize, Serialize};
 use std::collections::HashSet;
 use std::ops::{AddAssign, Deref};
-use std::sync::Arc;
+use std::sync::{Arc, RwLockReadGuard};
 use std::time::Duration;
 use zenith_utils::lsn::{Lsn, RecordLsn};
 use zenith_utils::zid::ZTimelineId;
@@ -19,7 +19,7 @@ pub type BlockNumber = u32;
 /// A repository corresponds to one .zenith directory. One repository holds multiple
 /// timelines, forked off from the same initial call to 'initdb'.
 pub trait Repository: Send + Sync {
-    fn shutdown(&self) -> Result<()>;
+    fn detach_timeline(&self, timeline_id: ZTimelineId) -> Result<()>;
 
     /// Updates timeline based on the new sync state, received from the remote storage synchronization.
     /// See [`crate::remote_storage`] for more details about the synchronization.
@@ -184,6 +184,9 @@ pub trait Timeline: Send + Sync {
     ///
     fn wait_lsn(&self, lsn: Lsn) -> Result<()>;
 
+    /// Lock and get timeline's GC cuttof
+    fn get_latest_gc_cutoff_lsn(&self) -> RwLockReadGuard<Lsn>;
+
     /// Look up given page version.
     fn get_page_at_lsn(&self, tag: RelishTag, blknum: BlockNumber, lsn: Lsn) -> Result<Bytes>;
 
@@ -217,10 +220,12 @@ pub trait Timeline: Send + Sync {
 
     /// Atomically get both last and prev.
     fn get_last_record_rlsn(&self) -> RecordLsn;
+
     /// Get last or prev record separately. Same as get_last_record_rlsn().last/prev.
     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`].
@@ -235,7 +240,11 @@ pub trait Timeline: Send + Sync {
 
     ///
     /// Check that it is valid to request operations with that lsn.
-    fn check_lsn_is_in_scope(&self, lsn: Lsn) -> Result<()>;
+    fn check_lsn_is_in_scope(
+        &self,
+        lsn: Lsn,
+        latest_gc_cutoff_lsn: &RwLockReadGuard<Lsn>,
+    ) -> Result<()>;
 
     /// Retrieve current logical size of the timeline
     ///
@@ -244,7 +253,7 @@ pub trait Timeline: Send + Sync {
     fn get_current_logical_size(&self) -> usize;
 
     /// Does the same as get_current_logical_size but counted on demand.
-    /// Used in tests to ensure thet incremental and non incremental variants match.
+    /// Used in tests to ensure that incremental and non incremental variants match.
     fn get_current_logical_size_non_incremental(&self, lsn: Lsn) -> Result<usize>;
 
     /// An escape hatch to allow "casting" a generic Timeline to LayeredTimeline.
@@ -297,8 +306,12 @@ pub enum ZenithWalRecord {
     /// Native PostgreSQL WAL record
     Postgres { will_init: bool, rec: Bytes },
 
-    /// Set bits in heap visibility map. (heap blkno, flag bits to clear)
-    ClearVisibilityMapFlags { heap_blkno: u32, flags: u8 },
+    /// Clear bits in heap visibility map. ('flags' is bitmap of bits to clear)
+    ClearVisibilityMapFlags {
+        new_heap_blkno: Option<u32>,
+        old_heap_blkno: Option<u32>,
+        flags: u8,
+    },
     /// Mark transaction IDs as committed on a CLOG page
     ClogSetCommitted { xids: Vec<TransactionId> },
     /// Mark transaction IDs as aborted on a CLOG page
@@ -987,7 +1000,7 @@ mod tests {
                     .source()
                     .unwrap()
                     .to_string()
-                    .contains("is earlier than initdb lsn"));
+                    .contains("is earlier than latest GC horizon"));
             }
         }
 
@@ -1004,12 +1017,11 @@ mod tests {
         make_some_layers(&tline, Lsn(0x20))?;
 
         repo.gc_iteration(Some(TIMELINE_ID), 0x10, false)?;
-
+        let latest_gc_cutoff_lsn = tline.get_latest_gc_cutoff_lsn();
+        assert!(*latest_gc_cutoff_lsn > Lsn(0x25));
         match tline.get_page_at_lsn(TESTREL_A, 0, Lsn(0x25)) {
             Ok(_) => panic!("request for page should have failed"),
-            Err(err) => assert!(err
-                .to_string()
-                .contains("tried to request a page version that was garbage collected")),
+            Err(err) => assert!(err.to_string().contains("not found at")),
         }
         Ok(())
     }

pageserver/src/tenant_mgr.rs

@@ -5,15 +5,16 @@ use crate::branches;
 use crate::config::PageServerConf;
 use crate::layered_repository::LayeredRepository;
 use crate::repository::{Repository, Timeline, TimelineSyncState};
-use crate::tenant_threads;
+use crate::thread_mgr;
+use crate::thread_mgr::ThreadKind;
 use crate::walredo::PostgresRedoManager;
-use anyhow::{anyhow, bail, Context, Result};
+use crate::CheckpointConfig;
+use anyhow::{bail, Context, Result};
 use lazy_static::lazy_static;
 use log::*;
 use serde::{Deserialize, Serialize};
 use std::collections::{hash_map, HashMap};
 use std::fmt;
-use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, Mutex, MutexGuard};
 use zenith_utils::zid::{ZTenantId, ZTimelineId};
 
@@ -23,7 +24,7 @@ lazy_static! {
 
 struct Tenant {
     state: TenantState,
-    repo: Option<Arc<dyn Repository>>,
+    repo: Arc<dyn Repository>,
 }
 
 #[derive(Debug, Serialize, Deserialize, Clone, Copy, PartialEq, Eq)]
@@ -56,8 +57,6 @@ fn access_tenants() -> MutexGuard<'static, HashMap<ZTenantId, Tenant>> {
     TENANTS.lock().unwrap()
 }
 
-static SHUTDOWN_REQUESTED: AtomicBool = AtomicBool::new(false);
-
 /// Updates tenants' repositories, changing their timelines state in memory.
 pub fn set_timeline_states(
     conf: &'static PageServerConf,
@@ -73,28 +72,8 @@ pub fn set_timeline_states(
 
     let mut m = access_tenants();
     for (tenant_id, timeline_states) in timeline_states {
-        let tenant = m.entry(tenant_id).or_insert_with(|| Tenant {
-            state: TenantState::Idle,
-            repo: None,
-        });
-        if let Err(e) = put_timelines_into_tenant(conf, tenant, tenant_id, timeline_states) {
-            error!(
-                "Failed to update timeline states for tenant {}: {:#}",
-                tenant_id, e
-            );
-        }
-    }
-}
-
-fn put_timelines_into_tenant(
-    conf: &'static PageServerConf,
-    tenant: &mut Tenant,
-    tenant_id: ZTenantId,
-    timeline_states: HashMap<ZTimelineId, TimelineSyncState>,
-) -> anyhow::Result<()> {
-    let repo = match tenant.repo.as_ref() {
-        Some(repo) => Arc::clone(repo),
-        None => {
+        let tenant = m.entry(tenant_id).or_insert_with(|| {
+            // TODO (rodionov) reuse one of the initialisation routines
             // Set up a WAL redo manager, for applying WAL records.
             let walredo_mgr = PostgresRedoManager::new(conf, tenant_id);
 
@@ -105,13 +84,43 @@ fn put_timelines_into_tenant(
                 tenant_id,
                 conf.remote_storage_config.is_some(),
             ));
-            tenant.repo = Some(Arc::clone(&repo));
-            repo
+            Tenant {
+                state: TenantState::Idle,
+                repo,
+            }
+        });
+        if let Err(e) = put_timelines_into_tenant(tenant, tenant_id, timeline_states) {
+            error!(
+                "Failed to update timeline states for tenant {}: {:?}",
+                tenant_id, e
+            );
         }
-    };
+    }
+}
 
+fn put_timelines_into_tenant(
+    tenant: &mut Tenant,
+    tenant_id: ZTenantId,
+    timeline_states: HashMap<ZTimelineId, TimelineSyncState>,
+) -> anyhow::Result<()> {
     for (timeline_id, timeline_state) in timeline_states {
-        repo.set_timeline_state(timeline_id, timeline_state)
+        // If the timeline is being put into any other state than Ready,
+        // stop any threads operating on it.
+        //
+        // FIXME: This is racy. A page service thread could just get
+        // handle on the Timeline, before we call set_timeline_state()
+        if !matches!(timeline_state, TimelineSyncState::Ready(_)) {
+            thread_mgr::shutdown_threads(None, Some(tenant_id), Some(timeline_id));
+
+            // Should we run a final checkpoint to flush all the data to
+            // disk? Doesn't seem necessary; all of the states other than
+            // Ready imply that the data on local disk is corrupt or incomplete,
+            // and we don't want to flush that to disk.
+        }
+
+        tenant
+            .repo
+            .set_timeline_state(timeline_id, timeline_state)
             .with_context(|| {
                 format!(
                     "Failed to update timeline {} state to {:?}",
@@ -123,29 +132,49 @@ fn put_timelines_into_tenant(
     Ok(())
 }
 
-// Check this flag in the thread loops to know when to exit
-pub fn shutdown_requested() -> bool {
-    SHUTDOWN_REQUESTED.load(Ordering::Relaxed)
-}
-
-pub fn shutdown_all_tenants() -> Result<()> {
-    SHUTDOWN_REQUESTED.swap(true, Ordering::Relaxed);
-
-    let tenantids = list_tenantids()?;
-
-    for tenantid in &tenantids {
-        set_tenant_state(*tenantid, TenantState::Stopping)?;
+///
+/// Shut down all tenants. This runs as part of pageserver shutdown.
+///
+pub fn shutdown_all_tenants() {
+    let mut m = access_tenants();
+    let mut tenantids = Vec::new();
+    for (tenantid, tenant) in m.iter_mut() {
+        tenant.state = TenantState::Stopping;
+        tenantids.push(*tenantid)
     }
+    drop(m);
 
+    thread_mgr::shutdown_threads(Some(ThreadKind::WalReceiver), None, None);
+    thread_mgr::shutdown_threads(Some(ThreadKind::GarbageCollector), None, None);
+    thread_mgr::shutdown_threads(Some(ThreadKind::Checkpointer), None, None);
+
+    // Ok, no background threads running anymore. Flush any remaining data in
+    // memory to disk.
+    //
+    // We assume that any incoming connections that might request pages from
+    // the repository have already been terminated by the caller, so there
+    // should be no more activity in any of the repositories.
+    //
+    // On error, log it but continue with the shutdown for other tenants.
     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()?;
+        match get_repository_for_tenant(tenantid) {
+            Ok(repo) => {
+                if let Err(err) = repo.checkpoint_iteration(CheckpointConfig::Flush) {
+                    error!(
+                        "Could not checkpoint tenant {} during shutdown: {:?}",
+                        tenantid, err
+                    );
+                }
+            }
+            Err(err) => {
+                error!(
+                    "Could not get repository for tenant {} during shutdown: {:?}",
+                    tenantid, err
+                );
+            }
+        }
     }
-    Ok(())
 }
 
 pub fn create_repository_for_tenant(
@@ -153,7 +182,7 @@ pub fn create_repository_for_tenant(
     tenantid: ZTenantId,
 ) -> Result<()> {
     let wal_redo_manager = Arc::new(PostgresRedoManager::new(conf, tenantid));
-    let repo = Some(branches::create_repo(conf, tenantid, wal_redo_manager)?);
+    let repo = branches::create_repo(conf, tenantid, wal_redo_manager)?;
 
     match access_tenants().entry(tenantid) {
         hash_map::Entry::Occupied(_) => bail!("tenant {} already exists", tenantid),
@@ -172,34 +201,60 @@ pub fn get_tenant_state(tenantid: ZTenantId) -> Option<TenantState> {
     Some(access_tenants().get(&tenantid)?.state)
 }
 
-pub fn set_tenant_state(tenantid: ZTenantId, newstate: TenantState) -> Result<TenantState> {
+///
+/// Change the state of a tenant to Active and launch its checkpointer and GC
+/// threads. If the tenant was already in Active state or Stopping, does nothing.
+///
+pub fn activate_tenant(conf: &'static PageServerConf, tenantid: ZTenantId) -> Result<()> {
     let mut m = access_tenants();
-    let tenant = m.get_mut(&tenantid);
+    let tenant = m
+        .get_mut(&tenantid)
+        .with_context(|| format!("Tenant not found for id {}", tenantid))?;
 
-    match tenant {
-        Some(tenant) => {
-            if newstate == TenantState::Idle && tenant.state != TenantState::Active {
-                // Only Active tenant can become Idle
-                return Ok(tenant.state);
-            }
-            info!("set_tenant_state: {} -> {}", tenant.state, newstate);
-            tenant.state = newstate;
-            Ok(tenant.state)
+    info!("activating tenant {}", tenantid);
+
+    match tenant.state {
+        // If the tenant is already active, nothing to do.
+        TenantState::Active => {}
+
+        // If it's Idle, launch the checkpointer and GC threads
+        TenantState::Idle => {
+            thread_mgr::spawn(
+                ThreadKind::Checkpointer,
+                Some(tenantid),
+                None,
+                "Checkpointer thread",
+                move || crate::tenant_threads::checkpoint_loop(tenantid, conf),
+            )?;
+
+            // FIXME: if we fail to launch the GC thread, but already launched the
+            // checkpointer, we're in a strange state.
+
+            thread_mgr::spawn(
+                ThreadKind::GarbageCollector,
+                Some(tenantid),
+                None,
+                "GC thread",
+                move || crate::tenant_threads::gc_loop(tenantid, conf),
+            )?;
+
+            tenant.state = TenantState::Active;
+        }
+
+        TenantState::Stopping => {
+            // don't re-activate it if it's being stopped
         }
-        None => bail!("Tenant not found for id {}", tenantid),
     }
+    Ok(())
 }
 
 pub fn get_repository_for_tenant(tenantid: ZTenantId) -> Result<Arc<dyn Repository>> {
     let m = access_tenants();
     let tenant = m
         .get(&tenantid)
-        .ok_or_else(|| anyhow!("Tenant not found for tenant {}", tenantid))?;
+        .with_context(|| format!("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),
-    }
+    Ok(Arc::clone(&tenant.repo))
 }
 
 pub fn get_timeline_for_tenant(
@@ -209,17 +264,7 @@ pub fn get_timeline_for_tenant(
     get_repository_for_tenant(tenantid)?
         .get_timeline(timelineid)?
         .local_timeline()
-        .ok_or_else(|| anyhow!("cannot fetch timeline {}", timelineid))
-}
-
-fn list_tenantids() -> Result<Vec<ZTenantId>> {
-    access_tenants()
-        .iter()
-        .map(|v| {
-            let (tenantid, _) = v;
-            Ok(*tenantid)
-        })
-        .collect()
+        .with_context(|| format!("cannot fetch timeline {}", timelineid))
 }
 
 #[derive(Serialize, Deserialize, Clone)]

pageserver/src/tenant_threads.rs

@@ -5,88 +5,14 @@ use crate::tenant_mgr;
 use crate::tenant_mgr::TenantState;
 use crate::CheckpointConfig;
 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_metrics::{register_int_gauge_vec, IntGaugeVec};
 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());
-}
-
-lazy_static! {
-    static ref TENANT_THREADS_COUNT: IntGaugeVec = register_int_gauge_vec!(
-        "tenant_threads_count",
-        "Number of live tenant threads",
-        &["tenant_thread_type"]
-    )
-    .expect("failed to define a metric");
-}
-
-// Launch checkpointer and GC for the tenant.
-// It's possible that the threads are running already,
-// if so, just don't spawn new ones.
-pub fn start_tenant_threads(conf: &'static PageServerConf, tenantid: ZTenantId) {
-    let mut handles = TENANT_HANDLES.lock().unwrap();
-    let h = handles
-        .entry(tenantid)
-        .or_insert_with(|| TenantHandleEntry {
-            checkpointer_handle: None,
-            gc_handle: None,
-        });
-
-    if h.checkpointer_handle.is_none() {
-        h.checkpointer_handle = std::thread::Builder::new()
-            .name("Checkpointer thread".into())
-            .spawn(move || {
-                checkpoint_loop(tenantid, conf).expect("Checkpointer thread died");
-            })
-            .ok();
-    }
-
-    if h.gc_handle.is_none() {
-        h.gc_handle = std::thread::Builder::new()
-            .name("GC thread".into())
-            .spawn(move || {
-                gc_loop(tenantid, conf).expect("GC thread died");
-            })
-            .ok();
-    }
-}
-
-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<()> {
-    let gauge = TENANT_THREADS_COUNT.with_label_values(&["checkpointer"]);
-    gauge.inc();
-    scopeguard::defer! {
-        gauge.dec();
-    }
-
+pub fn checkpoint_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result<()> {
     loop {
         if tenant_mgr::get_tenant_state(tenantid) != Some(TenantState::Active) {
             break;
@@ -112,13 +38,7 @@ fn checkpoint_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result
 ///
 /// GC thread's main loop
 ///
-fn gc_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result<()> {
-    let gauge = TENANT_THREADS_COUNT.with_label_values(&["gc"]);
-    gauge.inc();
-    scopeguard::defer! {
-        gauge.dec();
-    }
-
+pub fn gc_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result<()> {
     loop {
         if tenant_mgr::get_tenant_state(tenantid) != Some(TenantState::Active) {
             break;

pageserver/src/thread_mgr.rs

@@ -0,0 +1,284 @@
+//!
+//! This module provides centralized handling of threads in the Page Server.
+//!
+//! We provide a few basic facilities:
+//! - A global registry of threads that lists what kind of threads they are, and
+//!   which tenant or timeline they are working on
+//!
+//! - The ability to request a thread to shut down.
+//!
+//!
+//! # How it works?
+//!
+//! There is a global hashmap of all the threads (`THREADS`). Whenever a new
+//! thread is spawned, a PageServerThread entry is added there, and when a
+//! thread dies, it removes itself from the hashmap. If you want to kill a
+//! thread, you can scan the hashmap to find it.
+//!
+//! # Thread shutdown
+//!
+//! To kill a thread, we rely on co-operation from the victim. Each thread is
+//! expected to periodically call the `is_shutdown_requested()` function, and
+//! if it returns true, exit gracefully. In addition to that, when waiting for
+//! the network or other long-running operation, you can use
+//! `shutdown_watcher()` function to get a Future that will become ready if
+//! the current thread has been requested to shut down. You can use that with
+//! Tokio select!(), but note that it relies on thread-local storage, so it
+//! will only work with the "current-thread" Tokio runtime!
+//!
+//!
+//! TODO: This would be a good place to also handle panics in a somewhat sane way.
+//! Depending on what thread panics, we might want to kill the whole server, or
+//! only a single tenant or timeline.
+//!
+
+use std::cell::RefCell;
+use std::collections::HashMap;
+use std::panic;
+use std::panic::AssertUnwindSafe;
+use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
+use std::sync::{Arc, Mutex};
+use std::thread;
+use std::thread::JoinHandle;
+
+use tokio::sync::watch;
+
+use tracing::{info, warn};
+
+use lazy_static::lazy_static;
+
+use zenith_utils::zid::{ZTenantId, ZTimelineId};
+
+lazy_static! {
+    /// Each thread that we track is associated with a "thread ID". It's just
+    /// an increasing number that we assign, not related to any system thread
+    /// id.
+    static ref NEXT_THREAD_ID: AtomicU64 = AtomicU64::new(1);
+
+    /// Global registry of threads
+    static ref THREADS: Mutex<HashMap<u64, Arc<PageServerThread>>> = Mutex::new(HashMap::new());
+}
+
+// There is a Tokio watch channel for each thread, which can be used to signal the
+// thread that it needs to shut down. This thread local variable holds the receiving
+// end of the channel. The sender is kept in the global registry, so that anyone
+// can send the signal to request thread shutdown.
+thread_local!(static SHUTDOWN_RX: RefCell<Option<watch::Receiver<()>>> = RefCell::new(None));
+
+// Each thread holds reference to its own PageServerThread here.
+thread_local!(static CURRENT_THREAD: RefCell<Option<Arc<PageServerThread>>> = RefCell::new(None));
+
+///
+/// There are many kinds of threads in the system. Some are associated with a particular
+/// tenant or timeline, while others are global.
+///
+/// Note that we don't try to limit how may threads of a certain kind can be running
+/// at the same time.
+///
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
+pub enum ThreadKind {
+    // libpq listener thread. It just accepts connection and spawns a
+    // PageRequestHandler thread for each connection.
+    LibpqEndpointListener,
+
+    // HTTP endpoint listener.
+    HttpEndpointListener,
+
+    // Thread that handles a single connection. A PageRequestHandler thread
+    // starts detached from any particular tenant or timeline, but it can be
+    // associated with one later, after receiving a command from the client.
+    PageRequestHandler,
+
+    // Thread that connects to a safekeeper to fetch WAL for one timeline.
+    WalReceiver,
+
+    // Thread that handles checkpointing of all timelines for a tenant.
+    Checkpointer,
+
+    // Thread that handles GC of a tenant
+    GarbageCollector,
+
+    // Thread for synchronizing pageserver relish data with the remote storage.
+    // Shared by all tenants.
+    StorageSync,
+}
+
+struct PageServerThread {
+    _thread_id: u64,
+
+    kind: ThreadKind,
+
+    /// Tenant and timeline that this thread is associated with.
+    tenant_id: Option<ZTenantId>,
+    timeline_id: Option<ZTimelineId>,
+
+    name: String,
+
+    // To request thread shutdown, set the flag, and send a dummy message to the
+    // channel to notify it.
+    shutdown_requested: AtomicBool,
+    shutdown_tx: watch::Sender<()>,
+
+    /// Handle for waiting for the thread to exit. It can be None, if the
+    /// the thread has already exited.
+    join_handle: Mutex<Option<JoinHandle<()>>>,
+}
+
+/// Launch a new thread
+pub fn spawn<F, E>(
+    kind: ThreadKind,
+    tenant_id: Option<ZTenantId>,
+    timeline_id: Option<ZTimelineId>,
+    name: &str,
+    f: F,
+) -> std::io::Result<()>
+where
+    F: FnOnce() -> Result<(), E> + Send + 'static,
+{
+    let (shutdown_tx, shutdown_rx) = watch::channel(());
+    let thread_id = NEXT_THREAD_ID.fetch_add(1, Ordering::Relaxed);
+    let thread = PageServerThread {
+        _thread_id: thread_id,
+        kind,
+        tenant_id,
+        timeline_id,
+        name: name.to_string(),
+
+        shutdown_requested: AtomicBool::new(false),
+        shutdown_tx,
+
+        join_handle: Mutex::new(None),
+    };
+
+    let thread_rc = Arc::new(thread);
+
+    let mut jh_guard = thread_rc.join_handle.lock().unwrap();
+
+    THREADS
+        .lock()
+        .unwrap()
+        .insert(thread_id, Arc::clone(&thread_rc));
+
+    let thread_rc2 = Arc::clone(&thread_rc);
+    let join_handle = match thread::Builder::new()
+        .name(name.to_string())
+        .spawn(move || thread_wrapper(thread_id, thread_rc2, shutdown_rx, f))
+    {
+        Ok(handle) => handle,
+        Err(err) => {
+            // Could not spawn the thread. Remove the entry
+            THREADS.lock().unwrap().remove(&thread_id);
+            return Err(err);
+        }
+    };
+    *jh_guard = Some(join_handle);
+    drop(jh_guard);
+
+    // The thread is now running. Nothing more to do here
+    Ok(())
+}
+
+/// This wrapper function runs in a newly-spawned thread. It initializes the
+/// thread-local variables and calls the payload function
+fn thread_wrapper<F, E>(
+    thread_id: u64,
+    thread: Arc<PageServerThread>,
+    shutdown_rx: watch::Receiver<()>,
+    f: F,
+) where
+    F: FnOnce() -> Result<(), E> + Send + 'static,
+{
+    SHUTDOWN_RX.with(|rx| {
+        *rx.borrow_mut() = Some(shutdown_rx);
+    });
+    CURRENT_THREAD.with(|ct| {
+        *ct.borrow_mut() = Some(thread);
+    });
+
+    // We use AssertUnwindSafe here so that the payload function
+    // doesn't need to be UnwindSafe. We don't do anything after the
+    // unwinding that would expose us to unwind-unsafe behavior.
+    let result = panic::catch_unwind(AssertUnwindSafe(f));
+
+    // Remove our entry from the global hashmap.
+    THREADS.lock().unwrap().remove(&thread_id);
+
+    // If the thread payload panic'd, exit with the panic.
+    if let Err(err) = result {
+        panic::resume_unwind(err);
+    }
+}
+
+/// Is there a thread running that matches the criteria
+
+/// Signal and wait for threads to shut down.
+///
+///
+/// The arguments are used to select the threads to kill. Any None arguments are
+/// ignored. For example, to shut down all WalReceiver threads:
+///
+///   shutdown_threads(Some(ThreadKind::WalReceiver), None, None)
+///
+/// Or to shut down all threads for given timeline:
+///
+///   shutdown_threads(None, Some(timelineid), None)
+///
+pub fn shutdown_threads(
+    kind: Option<ThreadKind>,
+    tenant_id: Option<ZTenantId>,
+    timeline_id: Option<ZTimelineId>,
+) {
+    let mut victim_threads = Vec::new();
+
+    let threads = THREADS.lock().unwrap();
+    for thread in threads.values() {
+        if (kind.is_none() || Some(thread.kind) == kind)
+            && (tenant_id.is_none() || thread.tenant_id == tenant_id)
+            && (timeline_id.is_none() || thread.timeline_id == timeline_id)
+        {
+            thread.shutdown_requested.store(true, Ordering::Relaxed);
+            // FIXME: handle error?
+            let _ = thread.shutdown_tx.send(());
+            victim_threads.push(Arc::clone(thread));
+        }
+    }
+    drop(threads);
+
+    for thread in victim_threads {
+        info!("waiting for {} to shut down", thread.name);
+        if let Some(join_handle) = thread.join_handle.lock().unwrap().take() {
+            let _ = join_handle.join();
+        } else {
+            // The thread had not even fully started yet. Or it was shut down
+            // concurrently and alrady exited
+        }
+    }
+}
+
+/// A Future that can be used to check if the current thread has been requested to
+/// shut down.
+pub async fn shutdown_watcher() {
+    let _ = SHUTDOWN_RX
+        .with(|rx| {
+            rx.borrow()
+                .as_ref()
+                .expect("shutdown_requested() called in an unexpected thread")
+                .clone()
+        })
+        .changed()
+        .await;
+}
+
+/// Has the current thread been requested to shut down?
+pub fn is_shutdown_requested() -> bool {
+    CURRENT_THREAD.with(|ct| {
+        if let Some(ct) = ct.borrow().as_ref() {
+            ct.shutdown_requested.load(Ordering::Relaxed)
+        } else {
+            if !cfg!(test) {
+                warn!("is_shutdown_requested() called in an unexpected thread");
+            }
+            false
+        }
+    })
+}

pageserver/src/virtual_file.rs

@@ -10,15 +10,46 @@
 //! This is similar to PostgreSQL's virtual file descriptor facility in
 //! src/backend/storage/file/fd.c
 //!
+use lazy_static::lazy_static;
 use std::fs::{File, OpenOptions};
 use std::io::{Error, ErrorKind, Read, Seek, SeekFrom, Write};
 use std::os::unix::fs::FileExt;
 use std::path::{Path, PathBuf};
 use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
 use std::sync::{RwLock, RwLockWriteGuard};
+use zenith_metrics::{register_histogram_vec, register_int_gauge_vec, HistogramVec, IntGaugeVec};
 
 use once_cell::sync::OnceCell;
 
+// Metrics collected on disk IO operations
+const STORAGE_IO_TIME_BUCKETS: &[f64] = &[
+    0.000001, // 1 usec
+    0.00001,  // 10 usec
+    0.0001,   // 100 usec
+    0.001,    // 1 msec
+    0.01,     // 10 msec
+    0.1,      // 100 msec
+    1.0,      // 1 sec
+];
+
+lazy_static! {
+    static ref STORAGE_IO_TIME: HistogramVec = register_histogram_vec!(
+        "pageserver_io_time",
+        "Time spent in IO operations",
+        &["operation", "tenant_id", "timeline_id"],
+        STORAGE_IO_TIME_BUCKETS.into()
+    )
+    .expect("failed to define a metric");
+}
+lazy_static! {
+    static ref STORAGE_IO_SIZE: IntGaugeVec = register_int_gauge_vec!(
+        "pageserver_io_size",
+        "Amount of bytes",
+        &["operation", "tenant_id", "timeline_id"]
+    )
+    .expect("failed to define a metric");
+}
+
 ///
 /// A virtual file descriptor. You can use this just like std::fs::File, but internally
 /// the underlying file is closed if the system is low on file descriptors,
@@ -51,6 +82,10 @@ pub struct VirtualFile {
     /// storing it here.
     pub path: PathBuf,
     open_options: OpenOptions,
+
+    /// For metrics
+    tenantid: String,
+    timelineid: String,
 }
 
 #[derive(PartialEq, Clone, Copy)]
@@ -145,7 +180,13 @@ impl OpenFiles {
         // old file.
         //
         if let Some(old_file) = slot_guard.file.take() {
-            drop(old_file);
+            // We do not have information about tenantid/timelineid of evicted file.
+            // It is possible to store path together with file or use filepath crate,
+            // but as far as close() is not expected to be fast, it is not so critical to gather
+            // precise per-tenant statistic here.
+            STORAGE_IO_TIME
+                .with_label_values(&["close", "-", "-"])
+                .observe_closure_duration(|| drop(old_file));
         }
 
         // Prepare the slot for reuse and return it
@@ -185,9 +226,20 @@ impl VirtualFile {
         path: &Path,
         open_options: &OpenOptions,
     ) -> Result<VirtualFile, std::io::Error> {
+        let parts = path.to_str().unwrap().split('/').collect::<Vec<&str>>();
+        let tenantid;
+        let timelineid;
+        if parts.len() > 5 && parts[parts.len() - 5] == "tenants" {
+            tenantid = parts[parts.len() - 4].to_string();
+            timelineid = parts[parts.len() - 2].to_string();
+        } else {
+            tenantid = "*".to_string();
+            timelineid = "*".to_string();
+        }
         let (handle, mut slot_guard) = get_open_files().find_victim_slot();
-
-        let file = open_options.open(path)?;
+        let file = STORAGE_IO_TIME
+            .with_label_values(&["open", &tenantid, &timelineid])
+            .observe_closure_duration(|| open_options.open(path))?;
 
         // Strip all options other than read and write.
         //
@@ -204,6 +256,8 @@ impl VirtualFile {
             pos: 0,
             path: path.to_path_buf(),
             open_options: reopen_options,
+            tenantid,
+            timelineid,
         };
 
         slot_guard.file.replace(file);
@@ -213,13 +267,13 @@ impl VirtualFile {
 
     /// Call File::sync_all() on the underlying File.
     pub fn sync_all(&self) -> Result<(), Error> {
-        self.with_file(|file| file.sync_all())?
+        self.with_file("fsync", |file| file.sync_all())?
     }
 
     /// Helper function that looks up the underlying File for this VirtualFile,
     /// opening it and evicting some other File if necessary. It calls 'func'
     /// with the physical File.
-    fn with_file<F, R>(&self, mut func: F) -> Result<R, Error>
+    fn with_file<F, R>(&self, op: &str, mut func: F) -> Result<R, Error>
     where
         F: FnMut(&File) -> R,
     {
@@ -242,7 +296,9 @@ impl VirtualFile {
                         if let Some(file) = &slot_guard.file {
                             // Found a cached file descriptor.
                             slot.recently_used.store(true, Ordering::Relaxed);
-                            return Ok(func(file));
+                            return Ok(STORAGE_IO_TIME
+                                .with_label_values(&[op, &self.tenantid, &self.timelineid])
+                                .observe_closure_duration(|| func(file)));
                         }
                     }
                 }
@@ -267,16 +323,23 @@ impl VirtualFile {
         let (handle, mut slot_guard) = open_files.find_victim_slot();
 
         // Open the physical file
-        let file = self.open_options.open(&self.path)?;
+        let file = STORAGE_IO_TIME
+            .with_label_values(&["open", &self.tenantid, &self.timelineid])
+            .observe_closure_duration(|| self.open_options.open(&self.path))?;
 
         // Perform the requested operation on it
         //
         // TODO: We could downgrade the locks to read mode before calling
         // 'func', to allow a little bit more concurrency, but the standard
         // library RwLock doesn't allow downgrading without releasing the lock,
-        // and that doesn't seem worth the trouble. (parking_lot RwLock would
-        // allow it)
-        let result = func(&file);
+        // and that doesn't seem worth the trouble.
+        //
+        // XXX: `parking_lot::RwLock` can enable such downgrades, yet its implemenation is fair and
+        // may deadlock on subsequent read calls.
+        // Simply replacing all `RwLock` in project causes deadlocks, so use it sparingly.
+        let result = STORAGE_IO_TIME
+            .with_label_values(&[op, &self.tenantid, &self.timelineid])
+            .observe_closure_duration(|| func(&file));
 
         // Store the File in the slot and update the handle in the VirtualFile
         // to point to it.
@@ -299,7 +362,13 @@ impl Drop for VirtualFile {
         let mut slot_guard = slot.inner.write().unwrap();
         if slot_guard.tag == handle.tag {
             slot.recently_used.store(false, Ordering::Relaxed);
-            slot_guard.file.take();
+            // Unlike files evicted by replacement algorithm, here
+            // we group close time by tenantid/timelineid.
+            // At allows to compare number/time of "normal" file closes
+            // with file eviction.
+            STORAGE_IO_TIME
+                .with_label_values(&["close", &self.tenantid, &self.timelineid])
+                .observe_closure_duration(|| slot_guard.file.take());
         }
     }
 }
@@ -335,7 +404,7 @@ impl Seek for VirtualFile {
                 self.pos = offset;
             }
             SeekFrom::End(offset) => {
-                self.pos = self.with_file(|mut file| file.seek(SeekFrom::End(offset)))??
+                self.pos = self.with_file("seek", |mut file| file.seek(SeekFrom::End(offset)))??
             }
             SeekFrom::Current(offset) => {
                 let pos = self.pos as i128 + offset as i128;
@@ -357,11 +426,23 @@ impl Seek for VirtualFile {
 
 impl FileExt for VirtualFile {
     fn read_at(&self, buf: &mut [u8], offset: u64) -> Result<usize, Error> {
-        self.with_file(|file| file.read_at(buf, offset))?
+        let result = self.with_file("read", |file| file.read_at(buf, offset))?;
+        if let Ok(size) = result {
+            STORAGE_IO_SIZE
+                .with_label_values(&["read", &self.tenantid, &self.timelineid])
+                .add(size as i64);
+        }
+        result
     }
 
     fn write_at(&self, buf: &[u8], offset: u64) -> Result<usize, Error> {
-        self.with_file(|file| file.write_at(buf, offset))?
+        let result = self.with_file("write", |file| file.write_at(buf, offset))?;
+        if let Ok(size) = result {
+            STORAGE_IO_SIZE
+                .with_label_values(&["write", &self.tenantid, &self.timelineid])
+                .add(size as i64);
+        }
+        result
     }
 }
 

pageserver/src/walingest.rs

@@ -349,49 +349,25 @@ impl WalIngest {
         decoded: &mut DecodedWALRecord,
     ) -> Result<()> {
         // Handle VM bit updates that are implicitly part of heap records.
+
+        // First, look at the record to determine which VM bits need
+        // to be cleared. If either of these variables is set, we
+        // need to clear the corresponding bits in the visibility map.
+        let mut new_heap_blkno: Option<u32> = None;
+        let mut old_heap_blkno: Option<u32> = None;
         if decoded.xl_rmid == pg_constants::RM_HEAP_ID {
             let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
             if info == pg_constants::XLOG_HEAP_INSERT {
                 let xlrec = XlHeapInsert::decode(buf);
                 assert_eq!(0, buf.remaining());
-                if (xlrec.flags
-                    & (pg_constants::XLH_INSERT_ALL_VISIBLE_CLEARED
-                        | pg_constants::XLH_INSERT_ALL_FROZEN_SET))
-                    != 0
-                {
-                    timeline.put_wal_record(
-                        lsn,
-                        RelishTag::Relation(RelTag {
-                            forknum: pg_constants::VISIBILITYMAP_FORKNUM,
-                            spcnode: decoded.blocks[0].rnode_spcnode,
-                            dbnode: decoded.blocks[0].rnode_dbnode,
-                            relnode: decoded.blocks[0].rnode_relnode,
-                        }),
-                        decoded.blocks[0].blkno / pg_constants::HEAPBLOCKS_PER_PAGE as u32,
-                        ZenithWalRecord::ClearVisibilityMapFlags {
-                            heap_blkno: decoded.blocks[0].blkno,
-                            flags: pg_constants::VISIBILITYMAP_VALID_BITS,
-                        },
-                    )?;
+                if (xlrec.flags & pg_constants::XLH_INSERT_ALL_VISIBLE_CLEARED) != 0 {
+                    new_heap_blkno = Some(decoded.blocks[0].blkno);
                 }
             } else if info == pg_constants::XLOG_HEAP_DELETE {
                 let xlrec = XlHeapDelete::decode(buf);
                 assert_eq!(0, buf.remaining());
                 if (xlrec.flags & pg_constants::XLH_DELETE_ALL_VISIBLE_CLEARED) != 0 {
-                    timeline.put_wal_record(
-                        lsn,
-                        RelishTag::Relation(RelTag {
-                            forknum: pg_constants::VISIBILITYMAP_FORKNUM,
-                            spcnode: decoded.blocks[0].rnode_spcnode,
-                            dbnode: decoded.blocks[0].rnode_dbnode,
-                            relnode: decoded.blocks[0].rnode_relnode,
-                        }),
-                        decoded.blocks[0].blkno / pg_constants::HEAPBLOCKS_PER_PAGE as u32,
-                        ZenithWalRecord::ClearVisibilityMapFlags {
-                            heap_blkno: decoded.blocks[0].blkno,
-                            flags: pg_constants::VISIBILITYMAP_VALID_BITS,
-                        },
-                    )?;
+                    new_heap_blkno = Some(decoded.blocks[0].blkno);
                 }
             } else if info == pg_constants::XLOG_HEAP_UPDATE
                 || info == pg_constants::XLOG_HEAP_HOT_UPDATE
@@ -400,39 +376,15 @@ impl WalIngest {
                 // the size of tuple data is inferred from the size of the record.
                 // we can't validate the remaining number of bytes without parsing
                 // the tuple data.
-                if (xlrec.flags & pg_constants::XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED) != 0 {
-                    timeline.put_wal_record(
-                        lsn,
-                        RelishTag::Relation(RelTag {
-                            forknum: pg_constants::VISIBILITYMAP_FORKNUM,
-                            spcnode: decoded.blocks[0].rnode_spcnode,
-                            dbnode: decoded.blocks[0].rnode_dbnode,
-                            relnode: decoded.blocks[0].rnode_relnode,
-                        }),
-                        decoded.blocks[0].blkno / pg_constants::HEAPBLOCKS_PER_PAGE as u32,
-                        ZenithWalRecord::ClearVisibilityMapFlags {
-                            heap_blkno: decoded.blocks[0].blkno,
-                            flags: pg_constants::VISIBILITYMAP_VALID_BITS,
-                        },
-                    )?;
+                if (xlrec.flags & pg_constants::XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED) != 0 {
+                    old_heap_blkno = Some(decoded.blocks[0].blkno);
                 }
-                if (xlrec.flags & pg_constants::XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED) != 0
-                    && decoded.blocks.len() > 1
-                {
-                    timeline.put_wal_record(
-                        lsn,
-                        RelishTag::Relation(RelTag {
-                            forknum: pg_constants::VISIBILITYMAP_FORKNUM,
-                            spcnode: decoded.blocks[1].rnode_spcnode,
-                            dbnode: decoded.blocks[1].rnode_dbnode,
-                            relnode: decoded.blocks[1].rnode_relnode,
-                        }),
-                        decoded.blocks[1].blkno / pg_constants::HEAPBLOCKS_PER_PAGE as u32,
-                        ZenithWalRecord::ClearVisibilityMapFlags {
-                            heap_blkno: decoded.blocks[1].blkno,
-                            flags: pg_constants::VISIBILITYMAP_VALID_BITS,
-                        },
-                    )?;
+                if (xlrec.flags & pg_constants::XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED) != 0 {
+                    // PostgreSQL only uses XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED on a
+                    // non-HOT update where the new tuple goes to different page than
+                    // the old one. Otherwise, only XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED is
+                    // set.
+                    new_heap_blkno = Some(decoded.blocks[1].blkno);
                 }
             }
         } else if decoded.xl_rmid == pg_constants::RM_HEAP2_ID {
@@ -448,23 +400,60 @@ impl WalIngest {
                 };
                 assert_eq!(offset_array_len, buf.remaining());
 
-                // FIXME: why also ALL_FROZEN_SET?
-                if (xlrec.flags
-                    & (pg_constants::XLH_INSERT_ALL_VISIBLE_CLEARED
-                        | pg_constants::XLH_INSERT_ALL_FROZEN_SET))
-                    != 0
-                {
+                if (xlrec.flags & pg_constants::XLH_INSERT_ALL_VISIBLE_CLEARED) != 0 {
+                    new_heap_blkno = Some(decoded.blocks[0].blkno);
+                }
+            }
+        }
+        // FIXME: What about XLOG_HEAP_LOCK and XLOG_HEAP2_LOCK_UPDATED?
+
+        // Clear the VM bits if required.
+        if new_heap_blkno.is_some() || old_heap_blkno.is_some() {
+            let vm_relish = RelishTag::Relation(RelTag {
+                forknum: pg_constants::VISIBILITYMAP_FORKNUM,
+                spcnode: decoded.blocks[0].rnode_spcnode,
+                dbnode: decoded.blocks[0].rnode_dbnode,
+                relnode: decoded.blocks[0].rnode_relnode,
+            });
+
+            let new_vm_blk = new_heap_blkno.map(pg_constants::HEAPBLK_TO_MAPBLOCK);
+            let old_vm_blk = old_heap_blkno.map(pg_constants::HEAPBLK_TO_MAPBLOCK);
+            if new_vm_blk == old_vm_blk {
+                // An UPDATE record that needs to clear the bits for both old and the
+                // new page, both of which reside on the same VM page.
+                timeline.put_wal_record(
+                    lsn,
+                    vm_relish,
+                    new_vm_blk.unwrap(),
+                    ZenithWalRecord::ClearVisibilityMapFlags {
+                        new_heap_blkno,
+                        old_heap_blkno,
+                        flags: pg_constants::VISIBILITYMAP_VALID_BITS,
+                    },
+                )?;
+            } else {
+                // Clear VM bits for one heap page, or for two pages that reside on
+                // different VM pages.
+                if let Some(new_vm_blk) = new_vm_blk {
                     timeline.put_wal_record(
                         lsn,
-                        RelishTag::Relation(RelTag {
-                            forknum: pg_constants::VISIBILITYMAP_FORKNUM,
-                            spcnode: decoded.blocks[0].rnode_spcnode,
-                            dbnode: decoded.blocks[0].rnode_dbnode,
-                            relnode: decoded.blocks[0].rnode_relnode,
-                        }),
-                        decoded.blocks[0].blkno / pg_constants::HEAPBLOCKS_PER_PAGE as u32,
+                        vm_relish,
+                        new_vm_blk,
                         ZenithWalRecord::ClearVisibilityMapFlags {
-                            heap_blkno: decoded.blocks[0].blkno,
+                            new_heap_blkno,
+                            old_heap_blkno: None,
+                            flags: pg_constants::VISIBILITYMAP_VALID_BITS,
+                        },
+                    )?;
+                }
+                if let Some(old_vm_blk) = old_vm_blk {
+                    timeline.put_wal_record(
+                        lsn,
+                        vm_relish,
+                        old_vm_blk,
+                        ZenithWalRecord::ClearVisibilityMapFlags {
+                            new_heap_blkno: None,
+                            old_heap_blkno,
                             flags: pg_constants::VISIBILITYMAP_VALID_BITS,
                         },
                     )?;
@@ -472,8 +461,6 @@ impl WalIngest {
             }
         }
 
-        // FIXME: What about XLOG_HEAP_LOCK and XLOG_HEAP2_LOCK_UPDATED?
-
         Ok(())
     }
 

pageserver/src/walreceiver.rs

@@ -7,29 +7,29 @@
 
 use crate::config::PageServerConf;
 use crate::tenant_mgr;
-use crate::tenant_mgr::TenantState;
-use crate::tenant_threads;
+use crate::thread_mgr;
+use crate::thread_mgr::ThreadKind;
 use crate::walingest::WalIngest;
 use anyhow::{bail, Context, Error, Result};
+use bytes::BytesMut;
+use fail::fail_point;
 use lazy_static::lazy_static;
-use parking_lot::Mutex;
 use postgres_ffi::waldecoder::*;
 use postgres_protocol::message::backend::ReplicationMessage;
 use postgres_types::PgLsn;
 use std::cell::Cell;
 use std::collections::HashMap;
 use std::str::FromStr;
-use std::thread;
-use std::thread::JoinHandle;
+use std::sync::Mutex;
 use std::thread_local;
 use std::time::SystemTime;
 use tokio::pin;
-use tokio::sync::oneshot;
 use tokio_postgres::replication::ReplicationStream;
 use tokio_postgres::{Client, NoTls, SimpleQueryMessage, SimpleQueryRow};
 use tokio_stream::StreamExt;
 use tracing::*;
 use zenith_utils::lsn::Lsn;
+use zenith_utils::pq_proto::ZenithFeedback;
 use zenith_utils::zid::ZTenantId;
 use zenith_utils::zid::ZTimelineId;
 
@@ -38,13 +38,10 @@ use zenith_utils::zid::ZTimelineId;
 //
 struct WalReceiverEntry {
     wal_producer_connstr: String,
-    wal_receiver_handle: Option<JoinHandle<()>>,
-    wal_receiver_interrupt_sender: Option<oneshot::Sender<()>>,
-    tenantid: ZTenantId,
 }
 
 lazy_static! {
-    static ref WAL_RECEIVERS: Mutex<HashMap<ZTimelineId, WalReceiverEntry>> =
+    static ref WAL_RECEIVERS: Mutex<HashMap<(ZTenantId, ZTimelineId), WalReceiverEntry>> =
         Mutex::new(HashMap::new());
 }
 
@@ -55,97 +52,55 @@ thread_local! {
     pub(crate) static IS_WAL_RECEIVER: Cell<bool> = Cell::new(false);
 }
 
-// Wait for walreceiver to stop
-// Now it stops when pageserver shutdown is requested.
-// In future we can make this more granular and send shutdown signals
-// per tenant/timeline to cancel inactive walreceivers.
-// TODO deal with blocking pg connections
-pub fn stop_wal_receiver(timelineid: ZTimelineId) {
-    let mut receivers = WAL_RECEIVERS.lock();
-
-    if let Some(r) = receivers.get_mut(&timelineid) {
-        match r.wal_receiver_interrupt_sender.take() {
-            Some(s) => {
-                if s.send(()).is_err() {
-                    warn!("wal receiver interrupt signal already sent");
-                }
-            }
-            None => {
-                warn!("wal_receiver_interrupt_sender is missing, wal recever shouldn't be running")
-            }
-        }
-
-        info!("waiting for wal receiver to stop");
-        let handle = r.wal_receiver_handle.take();
-        // do not hold the lock while joining the handle (deadlock is possible otherwise)
-        drop(receivers);
-        // there is no timeout or try_join option available so in case of a bug this can hang forever
-        handle.map(JoinHandle::join);
-    }
-}
-
-pub fn drop_wal_receiver(timelineid: ZTimelineId, tenantid: ZTenantId) {
-    let mut receivers = WAL_RECEIVERS.lock();
-    receivers.remove(&timelineid);
-
-    // Check if it was the last walreceiver of the tenant.
-    // TODO now we store one WalReceiverEntry per timeline,
-    // so this iterator looks a bit strange.
-    for (_timelineid, entry) in receivers.iter() {
-        if entry.tenantid == tenantid {
-            return;
-        }
-    }
-
-    // When last walreceiver of the tenant is gone, change state to Idle
-    tenant_mgr::set_tenant_state(tenantid, TenantState::Idle).unwrap();
+fn drop_wal_receiver(tenantid: ZTenantId, timelineid: ZTimelineId) {
+    let mut receivers = WAL_RECEIVERS.lock().unwrap();
+    receivers.remove(&(tenantid, timelineid));
 }
 
 // Launch a new WAL receiver, or tell one that's running about change in connection string
 pub fn launch_wal_receiver(
     conf: &'static PageServerConf,
+    tenantid: ZTenantId,
     timelineid: ZTimelineId,
     wal_producer_connstr: &str,
-    tenantid: ZTenantId,
-) {
-    let mut receivers = WAL_RECEIVERS.lock();
+) -> Result<()> {
+    let mut receivers = WAL_RECEIVERS.lock().unwrap();
 
-    match receivers.get_mut(&timelineid) {
+    match receivers.get_mut(&(tenantid, timelineid)) {
         Some(receiver) => {
+            info!("wal receiver already running, updating connection string");
             receiver.wal_producer_connstr = wal_producer_connstr.into();
         }
         None => {
-            let (tx, rx) = tokio::sync::oneshot::channel::<()>();
-
-            let wal_receiver_handle = thread::Builder::new()
-                .name("WAL receiver thread".into())
-                .spawn(move || {
+            thread_mgr::spawn(
+                ThreadKind::WalReceiver,
+                Some(tenantid),
+                Some(timelineid),
+                "WAL receiver thread",
+                move || {
                     IS_WAL_RECEIVER.with(|c| c.set(true));
-                    thread_main(conf, timelineid, tenantid, rx);
-                })
-                .unwrap();
+                    thread_main(conf, tenantid, timelineid)
+                },
+            )?;
 
             let receiver = WalReceiverEntry {
                 wal_producer_connstr: wal_producer_connstr.into(),
-                wal_receiver_handle: Some(wal_receiver_handle),
-                wal_receiver_interrupt_sender: Some(tx),
-                tenantid,
             };
-            receivers.insert(timelineid, receiver);
+            receivers.insert((tenantid, timelineid), receiver);
 
             // Update tenant state and start tenant threads, if they are not running yet.
-            tenant_mgr::set_tenant_state(tenantid, TenantState::Active).unwrap();
-            tenant_threads::start_tenant_threads(conf, tenantid);
+            tenant_mgr::activate_tenant(conf, tenantid)?;
         }
     };
+    Ok(())
 }
 
 // Look up current WAL producer connection string in the hash table
-fn get_wal_producer_connstr(timelineid: ZTimelineId) -> String {
-    let receivers = WAL_RECEIVERS.lock();
+fn get_wal_producer_connstr(tenantid: ZTenantId, timelineid: ZTimelineId) -> String {
+    let receivers = WAL_RECEIVERS.lock().unwrap();
 
     receivers
-        .get(&timelineid)
+        .get(&(tenantid, timelineid))
         .unwrap()
         .wal_producer_connstr
         .clone()
@@ -156,25 +111,18 @@ fn get_wal_producer_connstr(timelineid: ZTimelineId) -> String {
 //
 fn thread_main(
     conf: &'static PageServerConf,
-    timelineid: ZTimelineId,
     tenantid: ZTenantId,
-    interrupt_receiver: oneshot::Receiver<()>,
-) {
+    timelineid: ZTimelineId,
+) -> Result<()> {
     let _enter = info_span!("WAL receiver", timeline = %timelineid, tenant = %tenantid).entered();
     info!("WAL receiver thread started");
 
     // Look up the current WAL producer address
-    let wal_producer_connstr = get_wal_producer_connstr(timelineid);
+    let wal_producer_connstr = get_wal_producer_connstr(tenantid, timelineid);
 
     // Make a connection to the WAL safekeeper, or directly to the primary PostgreSQL server,
     // and start streaming WAL from it.
-    let res = walreceiver_main(
-        conf,
-        tenantid,
-        timelineid,
-        &wal_producer_connstr,
-        interrupt_receiver,
-    );
+    let res = walreceiver_main(conf, tenantid, timelineid, &wal_producer_connstr);
 
     // TODO cleanup info messages
     if let Err(e) = res {
@@ -188,7 +136,8 @@ fn thread_main(
 
     // Drop it from list of active WAL_RECEIVERS
     // so that next callmemaybe request launched a new thread
-    drop_wal_receiver(timelineid, tenantid);
+    drop_wal_receiver(tenantid, timelineid);
+    Ok(())
 }
 
 fn walreceiver_main(
@@ -196,7 +145,6 @@ fn walreceiver_main(
     tenantid: ZTenantId,
     timelineid: ZTimelineId,
     wal_producer_connstr: &str,
-    mut interrupt_receiver: oneshot::Receiver<()>,
 ) -> Result<(), Error> {
     // Connect to the database in replication mode.
     info!("connecting to {:?}", wal_producer_connstr);
@@ -214,7 +162,7 @@ fn walreceiver_main(
     // This is from tokio-postgres docs, but it is a bit weird in our case because we extensively use block_on
     runtime.spawn(async move {
         if let Err(e) = connection.await {
-            eprintln!("connection error: {}", e);
+            error!("connection error: {}", e);
         }
     });
 
@@ -273,12 +221,15 @@ fn walreceiver_main(
     let mut walingest = WalIngest::new(&*timeline, startpoint)?;
 
     while let Some(replication_message) = runtime.block_on(async {
+        let shutdown_watcher = thread_mgr::shutdown_watcher();
         tokio::select! {
-            replication_message = physical_stream.next() => replication_message,
-            _ = &mut interrupt_receiver => {
+            // check for shutdown first
+            biased;
+            _ = shutdown_watcher => {
                 info!("walreceiver interrupted");
                 None
             }
+            replication_message = physical_stream.next() => replication_message,
         }
     }) {
         let replication_message = replication_message?;
@@ -305,6 +256,8 @@ fn walreceiver_main(
                     let writer = timeline.writer();
                     walingest.ingest_record(writer.as_ref(), recdata, lsn)?;
 
+                    fail_point!("walreceiver-after-ingest");
+
                     last_rec_lsn = lsn;
                 }
 
@@ -339,7 +292,6 @@ fn walreceiver_main(
         };
 
         if let Some(last_lsn) = status_update {
-            let last_lsn = PgLsn::from(u64::from(last_lsn));
             let timeline_synced_disk_consistent_lsn =
                 tenant_mgr::get_repository_for_tenant(tenantid)?
                     .get_timeline_state(timelineid)
@@ -347,18 +299,32 @@ fn walreceiver_main(
                     .unwrap_or(Lsn(0));
 
             // The last LSN we processed. It is not guaranteed to survive pageserver crash.
-            let write_lsn = last_lsn;
+            let write_lsn = u64::from(last_lsn);
             // `disk_consistent_lsn` is the LSN at which page server guarantees local persistence of all received data
-            let flush_lsn = PgLsn::from(u64::from(timeline.get_disk_consistent_lsn()));
+            let flush_lsn = u64::from(timeline.get_disk_consistent_lsn());
             // The last LSN that is synced to remote storage and is guaranteed to survive pageserver crash
             // Used by safekeepers to remove WAL preceding `remote_consistent_lsn`.
-            let apply_lsn = PgLsn::from(u64::from(timeline_synced_disk_consistent_lsn));
+            let apply_lsn = u64::from(timeline_synced_disk_consistent_lsn);
             let ts = SystemTime::now();
-            const NO_REPLY: u8 = 0;
+
+            // Send zenith feedback message.
+            // Regular standby_status_update fields are put into this message.
+            let zenith_status_update = ZenithFeedback {
+                current_timeline_size: timeline.get_current_logical_size() as u64,
+                ps_writelsn: write_lsn,
+                ps_flushlsn: flush_lsn,
+                ps_applylsn: apply_lsn,
+                ps_replytime: ts,
+            };
+
+            debug!("zenith_status_update {:?}", zenith_status_update);
+
+            let mut data = BytesMut::new();
+            zenith_status_update.serialize(&mut data)?;
             runtime.block_on(
                 physical_stream
                     .as_mut()
-                    .standby_status_update(write_lsn, flush_lsn, apply_lsn, ts, NO_REPLY),
+                    .zenith_status_update(data.len() as u64, &data),
             )?;
         }
     }

pageserver/src/walredo.rs

@@ -363,25 +363,44 @@ impl PostgresRedoManager {
                 will_init: _,
                 rec: _,
             } => panic!("tried to pass postgres wal record to zenith WAL redo"),
-            ZenithWalRecord::ClearVisibilityMapFlags { heap_blkno, flags } => {
-                // Calculate the VM block and offset that corresponds to the heap block.
-                let map_block = pg_constants::HEAPBLK_TO_MAPBLOCK(*heap_blkno);
-                let map_byte = pg_constants::HEAPBLK_TO_MAPBYTE(*heap_blkno);
-                let map_offset = pg_constants::HEAPBLK_TO_OFFSET(*heap_blkno);
-
-                // Check that we're modifying the correct VM block.
+            ZenithWalRecord::ClearVisibilityMapFlags {
+                new_heap_blkno,
+                old_heap_blkno,
+                flags,
+            } => {
+                // sanity check that this is modifying the correct relish
                 assert!(
                     check_forknum(&rel, pg_constants::VISIBILITYMAP_FORKNUM),
                     "ClearVisibilityMapFlags record on unexpected rel {:?}",
                     rel
                 );
-                assert!(map_block == blknum);
+                if let Some(heap_blkno) = *new_heap_blkno {
+                    // Calculate the VM block and offset that corresponds to the heap block.
+                    let map_block = pg_constants::HEAPBLK_TO_MAPBLOCK(heap_blkno);
+                    let map_byte = pg_constants::HEAPBLK_TO_MAPBYTE(heap_blkno);
+                    let map_offset = pg_constants::HEAPBLK_TO_OFFSET(heap_blkno);
 
-                // equivalent to PageGetContents(page)
-                let map = &mut page[pg_constants::MAXALIGN_SIZE_OF_PAGE_HEADER_DATA..];
+                    // Check that we're modifying the correct VM block.
+                    assert!(map_block == blknum);
 
-                let mask: u8 = flags << map_offset;
-                map[map_byte as usize] &= !mask;
+                    // equivalent to PageGetContents(page)
+                    let map = &mut page[pg_constants::MAXALIGN_SIZE_OF_PAGE_HEADER_DATA..];
+
+                    map[map_byte as usize] &= !(flags << map_offset);
+                }
+
+                // Repeat for 'old_heap_blkno', if any
+                if let Some(heap_blkno) = *old_heap_blkno {
+                    let map_block = pg_constants::HEAPBLK_TO_MAPBLOCK(heap_blkno);
+                    let map_byte = pg_constants::HEAPBLK_TO_MAPBYTE(heap_blkno);
+                    let map_offset = pg_constants::HEAPBLK_TO_OFFSET(heap_blkno);
+
+                    assert!(map_block == blknum);
+
+                    let map = &mut page[pg_constants::MAXALIGN_SIZE_OF_PAGE_HEADER_DATA..];
+
+                    map[map_byte as usize] &= !(flags << map_offset);
+                }
             }
             // Non-relational WAL records are handled here, with custom code that has the
             // same effects as the corresponding Postgres WAL redo function.

postgres_ffi/Cargo.toml

@@ -1,10 +1,7 @@
 [package]
 name = "postgres_ffi"
 version = "0.1.0"
-authors = ["Heikki Linnakangas <heikki@zenith.tech>"]
-edition = "2018"
-
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+edition = "2021"
 
 [dependencies]
 chrono = "0.4.19"

postgres_ffi/src/xlog_utils.rs

@@ -51,6 +51,13 @@ pub type TimeLineID = u32;
 pub type TimestampTz = i64;
 pub type XLogSegNo = u64;
 
+/// Interval of checkpointing metadata file. We should store metadata file to enforce
+/// predicate that checkpoint.nextXid is larger than any XID in WAL.
+/// But flushing checkpoint file for each transaction seems to be too expensive,
+/// so XID_CHECKPOINT_INTERVAL is used to forward align nextXid and so perform
+/// metadata checkpoint only once per XID_CHECKPOINT_INTERVAL transactions.
+/// XID_CHECKPOINT_INTERVAL should not be larger than BLCKSZ*CLOG_XACTS_PER_BYTE
+/// in order to let CLOG_TRUNCATE mechanism correctly extend CLOG.
 const XID_CHECKPOINT_INTERVAL: u32 = 1024;
 
 #[allow(non_snake_case)]
@@ -400,9 +407,13 @@ impl CheckPoint {
     ///
     /// Returns 'true' if the XID was updated.
     pub fn update_next_xid(&mut self, xid: u32) -> bool {
-        let xid = xid.wrapping_add(XID_CHECKPOINT_INTERVAL - 1) & !(XID_CHECKPOINT_INTERVAL - 1);
+        // nextXid should nw greate than any XID in WAL, so increment provided XID and check for wraparround.
+        let mut new_xid = std::cmp::max(xid + 1, pg_constants::FIRST_NORMAL_TRANSACTION_ID);
+        // To reduce number of metadata checkpoints, we forward align XID on XID_CHECKPOINT_INTERVAL.
+        // XID_CHECKPOINT_INTERVAL should not be larger than BLCKSZ*CLOG_XACTS_PER_BYTE
+        new_xid =
+            new_xid.wrapping_add(XID_CHECKPOINT_INTERVAL - 1) & !(XID_CHECKPOINT_INTERVAL - 1);
         let full_xid = self.nextXid.value;
-        let new_xid = std::cmp::max(xid + 1, pg_constants::FIRST_NORMAL_TRANSACTION_ID);
         let old_xid = full_xid as u32;
         if new_xid.wrapping_sub(old_xid) as i32 > 0 {
             let mut epoch = full_xid >> 32;
@@ -520,4 +531,34 @@ mod tests {
         println!("wal_end={}, tli={}", wal_end, tli);
         assert_eq!(wal_end, waldump_wal_end);
     }
+
+    /// Check the math in update_next_xid
+    ///
+    /// NOTE: These checks are sensitive to the value of XID_CHECKPOINT_INTERVAL,
+    /// currently 1024.
+    #[test]
+    pub fn test_update_next_xid() {
+        let checkpoint_buf = [0u8; std::mem::size_of::<CheckPoint>()];
+        let mut checkpoint = CheckPoint::decode(&checkpoint_buf).unwrap();
+
+        checkpoint.nextXid = FullTransactionId { value: 10 };
+        assert_eq!(checkpoint.nextXid.value, 10);
+
+        // The input XID gets rounded up to the next XID_CHECKPOINT_INTERVAL
+        // boundary
+        checkpoint.update_next_xid(100);
+        assert_eq!(checkpoint.nextXid.value, 1024);
+
+        // No change
+        checkpoint.update_next_xid(500);
+        assert_eq!(checkpoint.nextXid.value, 1024);
+        checkpoint.update_next_xid(1023);
+        assert_eq!(checkpoint.nextXid.value, 1024);
+
+        // The function returns the *next* XID, given the highest XID seen so
+        // far. So when we pass 1024, the nextXid gets bumped up to the next
+        // XID_CHECKPOINT_INTERVAL boundary.
+        checkpoint.update_next_xid(1024);
+        assert_eq!(checkpoint.nextXid.value, 2048);
+    }
 }

pre-commit.py

@@ -38,7 +38,7 @@ def rustfmt(fix_inplace: bool = False, no_color: bool = False) -> str:
 
 
 def yapf(fix_inplace: bool) -> str:
-    cmd = "pipenv run yapf --recursive"
+    cmd = "poetry run yapf --recursive"
     if fix_inplace:
         cmd += " --in-place"
     else:
@@ -47,7 +47,7 @@ def yapf(fix_inplace: bool) -> str:
 
 
 def mypy() -> str:
-    return "pipenv run mypy"
+    return "poetry run mypy"
 
 
 def get_commit_files() -> List[str]:
@@ -72,7 +72,7 @@ def check(name: str, suffix: str, cmd: str, changed_files: List[str], no_color:
             print("Please inspect the output below and run make fmt to fix automatically.")
         if suffix == ".py":
             print("If the output is empty, ensure that you've installed Python tooling by\n"
-                  "running 'pipenv install --dev' in the current directory (no root needed)")
+                  "running './scripts/pysync' in the current directory (no root needed)")
         print()
         print(res.stdout.decode())
         exit(1)

proxy/Cargo.toml

@@ -1,25 +1,33 @@
 [package]
 name = "proxy"
 version = "0.1.0"
-authors = ["Stas Kelvich <stas.kelvich@gmail.com>"]
-edition = "2018"
-
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+edition = "2021"
 
 [dependencies]
 anyhow = "1.0"
 bytes = { version = "1.0.1", features = ['serde'] }
+clap = "3.0"
+futures = "0.3.13"
+hashbrown = "0.11.2"
+hex = "0.4.3"
+hyper = "0.14"
 lazy_static = "1.4.0"
 md5 = "0.7.0"
-rand = "0.8.3"
-hex = "0.4.3"
 parking_lot = "0.11.2"
+pin-project-lite = "0.2.7"
+rand = "0.8.3"
+reqwest = { version = "0.11", default-features = false, features = ["blocking", "json", "rustls-tls"] }
+rustls = "0.19.1"
+scopeguard = "1.1.0"
 serde = "1"
 serde_json = "1"
 tokio = { version = "1.11", features = ["macros"] }
-tokio-postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
-clap = "2.33.0"
-rustls = "0.19.1"
-reqwest = { version = "0.11", default-features = false, features = ["blocking", "json", "rustls-tls"] }
+tokio-postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="2949d98df52587d562986aad155dd4e889e408b7" }
+tokio-rustls = "0.22.0"
 
 zenith_utils = { path = "../zenith_utils" }
+zenith_metrics = { path = "../zenith_metrics" }
+
+[dev-dependencies]
+tokio-postgres-rustls = "0.8.0"
+rcgen = "0.8.14"

proxy/src/auth.rs

@@ -0,0 +1,169 @@
+use crate::compute::DatabaseInfo;
+use crate::config::ProxyConfig;
+use crate::cplane_api::{self, CPlaneApi};
+use crate::stream::PqStream;
+use anyhow::{anyhow, bail, Context};
+use std::collections::HashMap;
+use tokio::io::{AsyncRead, AsyncWrite};
+use zenith_utils::pq_proto::{BeMessage as Be, BeParameterStatusMessage, FeMessage as Fe};
+
+/// Various client credentials which we use for authentication.
+#[derive(Debug, PartialEq, Eq)]
+pub struct ClientCredentials {
+    pub user: String,
+    pub dbname: String,
+}
+
+impl TryFrom<HashMap<String, String>> for ClientCredentials {
+    type Error = anyhow::Error;
+
+    fn try_from(mut value: HashMap<String, String>) -> Result<Self, Self::Error> {
+        let mut get_param = |key| {
+            value
+                .remove(key)
+                .with_context(|| format!("{} is missing in startup packet", key))
+        };
+
+        let user = get_param("user")?;
+        let db = get_param("database")?;
+
+        Ok(Self { user, dbname: db })
+    }
+}
+
+impl ClientCredentials {
+    /// Use credentials to authenticate the user.
+    pub async fn authenticate(
+        self,
+        config: &ProxyConfig,
+        client: &mut PqStream<impl AsyncRead + AsyncWrite + Unpin>,
+    ) -> anyhow::Result<DatabaseInfo> {
+        use crate::config::ClientAuthMethod::*;
+        use crate::config::RouterConfig::*;
+        let db_info = match &config.router_config {
+            Static { host, port } => handle_static(host.clone(), *port, client, self).await,
+            Dynamic(Mixed) => {
+                if self.user.ends_with("@zenith") {
+                    handle_existing_user(config, client, self).await
+                } else {
+                    handle_new_user(config, client).await
+                }
+            }
+            Dynamic(Password) => handle_existing_user(config, client, self).await,
+            Dynamic(Link) => handle_new_user(config, client).await,
+        };
+
+        db_info.context("failed to authenticate client")
+    }
+}
+
+fn new_psql_session_id() -> String {
+    hex::encode(rand::random::<[u8; 8]>())
+}
+
+async fn handle_static(
+    host: String,
+    port: u16,
+    client: &mut PqStream<impl AsyncRead + AsyncWrite + Unpin>,
+    creds: ClientCredentials,
+) -> anyhow::Result<DatabaseInfo> {
+    client
+        .write_message(&Be::AuthenticationCleartextPassword)
+        .await?;
+
+    // Read client's password bytes
+    let msg = match client.read_message().await? {
+        Fe::PasswordMessage(msg) => msg,
+        bad => bail!("unexpected message type: {:?}", bad),
+    };
+
+    let cleartext_password = std::str::from_utf8(&msg)?.split('\0').next().unwrap();
+
+    let db_info = DatabaseInfo {
+        host,
+        port,
+        dbname: creds.dbname.clone(),
+        user: creds.user.clone(),
+        password: Some(cleartext_password.into()),
+    };
+
+    client
+        .write_message_noflush(&Be::AuthenticationOk)?
+        .write_message_noflush(&BeParameterStatusMessage::encoding())?;
+
+    Ok(db_info)
+}
+
+async fn handle_existing_user(
+    config: &ProxyConfig,
+    client: &mut PqStream<impl AsyncRead + AsyncWrite + Unpin>,
+    creds: ClientCredentials,
+) -> anyhow::Result<DatabaseInfo> {
+    let psql_session_id = new_psql_session_id();
+    let md5_salt = rand::random();
+
+    client
+        .write_message(&Be::AuthenticationMD5Password(&md5_salt))
+        .await?;
+
+    // Read client's password hash
+    let msg = match client.read_message().await? {
+        Fe::PasswordMessage(msg) => msg,
+        bad => bail!("unexpected message type: {:?}", bad),
+    };
+
+    let (_trailing_null, md5_response) = msg
+        .split_last()
+        .ok_or_else(|| anyhow!("unexpected password message"))?;
+
+    let cplane = CPlaneApi::new(&config.auth_endpoint);
+    let db_info = cplane
+        .authenticate_proxy_request(creds, md5_response, &md5_salt, &psql_session_id)
+        .await?;
+
+    client
+        .write_message_noflush(&Be::AuthenticationOk)?
+        .write_message_noflush(&BeParameterStatusMessage::encoding())?;
+
+    Ok(db_info)
+}
+
+async fn handle_new_user(
+    config: &ProxyConfig,
+    client: &mut PqStream<impl AsyncRead + AsyncWrite + Unpin>,
+) -> anyhow::Result<DatabaseInfo> {
+    let psql_session_id = new_psql_session_id();
+    let greeting = hello_message(&config.redirect_uri, &psql_session_id);
+
+    let db_info = cplane_api::with_waiter(psql_session_id, |waiter| async {
+        // Give user a URL to spawn a new database
+        client
+            .write_message_noflush(&Be::AuthenticationOk)?
+            .write_message_noflush(&BeParameterStatusMessage::encoding())?
+            .write_message(&Be::NoticeResponse(greeting))
+            .await?;
+
+        // Wait for web console response
+        waiter.await?.map_err(|e| anyhow!(e))
+    })
+    .await?;
+
+    client.write_message_noflush(&Be::NoticeResponse("Connecting to database.".into()))?;
+
+    Ok(db_info)
+}
+
+fn hello_message(redirect_uri: &str, session_id: &str) -> String {
+    format!(
+        concat![
+            "☀️  Welcome to Zenith!\n",
+            "To proceed with database creation, open the following link:\n\n",
+            "    {redirect_uri}{session_id}\n\n",
+            "It needs to be done once and we will send you '.pgpass' file,\n",
+            "which will allow you to access or create ",
+            "databases without opening your web browser."
+        ],
+        redirect_uri = redirect_uri,
+        session_id = session_id,
+    )
+}

proxy/src/cancellation.rs

@@ -0,0 +1,106 @@
+use anyhow::{anyhow, Context};
+use hashbrown::HashMap;
+use parking_lot::Mutex;
+use std::net::SocketAddr;
+use tokio::net::TcpStream;
+use tokio_postgres::{CancelToken, NoTls};
+use zenith_utils::pq_proto::CancelKeyData;
+
+/// Enables serving CancelRequests.
+#[derive(Default)]
+pub struct CancelMap(Mutex<HashMap<CancelKeyData, Option<CancelClosure>>>);
+
+impl CancelMap {
+    /// Cancel a running query for the corresponding connection.
+    pub async fn cancel_session(&self, key: CancelKeyData) -> anyhow::Result<()> {
+        let cancel_closure = self
+            .0
+            .lock()
+            .get(&key)
+            .and_then(|x| x.clone())
+            .with_context(|| format!("unknown session: {:?}", key))?;
+
+        cancel_closure.try_cancel_query().await
+    }
+
+    /// Run async action within an ephemeral session identified by [`CancelKeyData`].
+    pub async fn with_session<'a, F, R, V>(&'a self, f: F) -> anyhow::Result<V>
+    where
+        F: FnOnce(Session<'a>) -> R,
+        R: std::future::Future<Output = anyhow::Result<V>>,
+    {
+        // HACK: We'd rather get the real backend_pid but tokio_postgres doesn't
+        // expose it and we don't want to do another roundtrip to query
+        // for it. The client will be able to notice that this is not the
+        // actual backend_pid, but backend_pid is not used for anything
+        // so it doesn't matter.
+        let key = rand::random();
+
+        // Random key collisions are unlikely to happen here, but they're still possible,
+        // which is why we have to take care not to rewrite an existing key.
+        self.0
+            .lock()
+            .try_insert(key, None)
+            .map_err(|_| anyhow!("session already exists: {:?}", key))?;
+
+        // This will guarantee that the session gets dropped
+        // as soon as the future is finished.
+        scopeguard::defer! {
+            self.0.lock().remove(&key);
+        }
+
+        let session = Session::new(key, self);
+        f(session).await
+    }
+}
+
+/// This should've been a [`std::future::Future`], but
+/// it's impossible to name a type of an unboxed future
+/// (we'd need something like `#![feature(type_alias_impl_trait)]`).
+#[derive(Clone)]
+pub struct CancelClosure {
+    socket_addr: SocketAddr,
+    cancel_token: CancelToken,
+}
+
+impl CancelClosure {
+    pub fn new(socket_addr: SocketAddr, cancel_token: CancelToken) -> Self {
+        Self {
+            socket_addr,
+            cancel_token,
+        }
+    }
+
+    /// Cancels the query running on user's compute node.
+    pub async fn try_cancel_query(self) -> anyhow::Result<()> {
+        let socket = TcpStream::connect(self.socket_addr).await?;
+        self.cancel_token.cancel_query_raw(socket, NoTls).await?;
+
+        Ok(())
+    }
+}
+
+/// Helper for registering query cancellation tokens.
+pub struct Session<'a> {
+    /// The user-facing key identifying this session.
+    key: CancelKeyData,
+    /// The [`CancelMap`] this session belongs to.
+    cancel_map: &'a CancelMap,
+}
+
+impl<'a> Session<'a> {
+    fn new(key: CancelKeyData, cancel_map: &'a CancelMap) -> Self {
+        Self { key, cancel_map }
+    }
+
+    /// Store the cancel token for the given session.
+    /// This enables query cancellation in [`crate::proxy::handshake`].
+    pub fn enable_cancellation(self, cancel_closure: CancelClosure) -> CancelKeyData {
+        self.cancel_map
+            .0
+            .lock()
+            .insert(self.key, Some(cancel_closure));
+
+        self.key
+    }
+}

proxy/src/compute.rs

@@ -0,0 +1,42 @@
+use anyhow::Context;
+use serde::{Deserialize, Serialize};
+use std::net::{SocketAddr, ToSocketAddrs};
+
+/// Compute node connection params.
+#[derive(Serialize, Deserialize, Debug, Default)]
+pub struct DatabaseInfo {
+    pub host: String,
+    pub port: u16,
+    pub dbname: String,
+    pub user: String,
+    pub password: Option<String>,
+}
+
+impl DatabaseInfo {
+    pub fn socket_addr(&self) -> anyhow::Result<SocketAddr> {
+        let host_port = format!("{}:{}", self.host, self.port);
+        host_port
+            .to_socket_addrs()
+            .with_context(|| format!("cannot resolve {} to SocketAddr", host_port))?
+            .next()
+            .context("cannot resolve at least one SocketAddr")
+    }
+}
+
+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
+    }
+}

proxy/src/config.rs

@@ -1,17 +1,52 @@
-use crate::cplane_api::DatabaseInfo;
 use anyhow::{anyhow, ensure, Context};
 use rustls::{internal::pemfile, NoClientAuth, ProtocolVersion, ServerConfig};
 use std::net::SocketAddr;
+use std::str::FromStr;
 use std::sync::Arc;
 
-pub type SslConfig = Arc<ServerConfig>;
+pub type TlsConfig = Arc<ServerConfig>;
+
+#[non_exhaustive]
+pub enum ClientAuthMethod {
+    Password,
+    Link,
+
+    /// Use password auth only if username ends with "@zenith"
+    Mixed,
+}
+
+pub enum RouterConfig {
+    Static { host: String, port: u16 },
+    Dynamic(ClientAuthMethod),
+}
+
+impl FromStr for ClientAuthMethod {
+    type Err = anyhow::Error;
+
+    fn from_str(s: &str) -> anyhow::Result<Self> {
+        use ClientAuthMethod::*;
+        match s {
+            "password" => Ok(Password),
+            "link" => Ok(Link),
+            "mixed" => Ok(Mixed),
+            _ => Err(anyhow::anyhow!("Invlid option for router")),
+        }
+    }
+}
 
 pub struct ProxyConfig {
     /// main entrypoint for users to connect to
     pub proxy_address: SocketAddr,
 
-    /// http management endpoint. Upon user account creation control plane
+    /// method of assigning compute nodes
+    pub router_config: RouterConfig,
+
+    /// internally used for status and prometheus metrics
+    pub http_address: SocketAddr,
+
+    /// management endpoint. Upon user account creation control plane
     /// will notify us here, so that we can 'unfreeze' user session.
+    /// TODO It uses postgres protocol over TCP but should be migrated to http.
     pub mgmt_address: SocketAddr,
 
     /// send unauthenticated users to this URI
@@ -20,26 +55,10 @@ pub struct ProxyConfig {
     /// control plane address where we would check auth.
     pub auth_endpoint: String,
 
-    pub ssl_config: Option<SslConfig>,
+    pub tls_config: Option<TlsConfig>,
 }
 
-pub type ProxyWaiters = crate::waiters::Waiters<Result<DatabaseInfo, String>>;
-
-pub struct ProxyState {
-    pub conf: ProxyConfig,
-    pub waiters: ProxyWaiters,
-}
-
-impl ProxyState {
-    pub fn new(conf: ProxyConfig) -> Self {
-        Self {
-            conf,
-            waiters: ProxyWaiters::default(),
-        }
-    }
-}
-
-pub fn configure_ssl(key_path: &str, cert_path: &str) -> anyhow::Result<SslConfig> {
+pub fn configure_ssl(key_path: &str, cert_path: &str) -> anyhow::Result<TlsConfig> {
     let key = {
         let key_bytes = std::fs::read(key_path).context("SSL key file")?;
         let mut keys = pemfile::pkcs8_private_keys(&mut &key_bytes[..])

proxy/src/cplane_api.rs

@@ -1,106 +1,87 @@
-use anyhow::{anyhow, bail, Context};
+use crate::auth::ClientCredentials;
+use crate::compute::DatabaseInfo;
+use crate::waiters::{Waiter, Waiters};
+use anyhow::{anyhow, bail};
+use lazy_static::lazy_static;
 use serde::{Deserialize, Serialize};
-use std::net::{SocketAddr, ToSocketAddrs};
 
-use crate::state::ProxyWaiters;
-
-#[derive(Serialize, Deserialize, Debug, Default)]
-pub struct DatabaseInfo {
-    pub host: String,
-    pub port: u16,
-    pub dbname: String,
-    pub user: String,
-    pub password: Option<String>,
+lazy_static! {
+    static ref CPLANE_WAITERS: Waiters<Result<DatabaseInfo, String>> = Default::default();
 }
 
-#[derive(Serialize, Deserialize, Debug)]
-#[serde(untagged)]
-enum ProxyAuthResponse {
-    Ready { conn_info: DatabaseInfo },
-    Error { error: String },
-    NotReady { ready: bool }, // TODO: get rid of `ready`
+/// Give caller an opportunity to wait for cplane's reply.
+pub async fn with_waiter<F, R, T>(psql_session_id: impl Into<String>, f: F) -> anyhow::Result<T>
+where
+    F: FnOnce(Waiter<'static, Result<DatabaseInfo, String>>) -> R,
+    R: std::future::Future<Output = anyhow::Result<T>>,
+{
+    let waiter = CPLANE_WAITERS.register(psql_session_id.into())?;
+    f(waiter).await
 }
 
-impl DatabaseInfo {
-    pub fn socket_addr(&self) -> anyhow::Result<SocketAddr> {
-        let host_port = format!("{}:{}", self.host, self.port);
-        host_port
-            .to_socket_addrs()
-            .with_context(|| format!("cannot resolve {} to SocketAddr", host_port))?
-            .next()
-            .ok_or_else(|| anyhow!("cannot resolve at least one SocketAddr"))
-    }
-}
-
-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
-    }
+pub fn notify(psql_session_id: &str, msg: Result<DatabaseInfo, String>) -> anyhow::Result<()> {
+    CPLANE_WAITERS.notify(psql_session_id, msg)
 }
 
+/// Zenith console API wrapper.
 pub struct CPlaneApi<'a> {
     auth_endpoint: &'a str,
-    waiters: &'a ProxyWaiters,
 }
 
 impl<'a> CPlaneApi<'a> {
-    pub fn new(auth_endpoint: &'a str, waiters: &'a ProxyWaiters) -> Self {
-        Self {
-            auth_endpoint,
-            waiters,
-        }
+    pub fn new(auth_endpoint: &'a str) -> Self {
+        Self { auth_endpoint }
     }
 }
 
 impl CPlaneApi<'_> {
-    pub fn authenticate_proxy_request(
+    pub async fn authenticate_proxy_request(
         &self,
-        user: &str,
-        database: &str,
+        creds: ClientCredentials,
         md5_response: &[u8],
         salt: &[u8; 4],
         psql_session_id: &str,
     ) -> anyhow::Result<DatabaseInfo> {
         let mut url = reqwest::Url::parse(self.auth_endpoint)?;
         url.query_pairs_mut()
-            .append_pair("login", user)
-            .append_pair("database", database)
+            .append_pair("login", &creds.user)
+            .append_pair("database", &creds.dbname)
             .append_pair("md5response", std::str::from_utf8(md5_response)?)
             .append_pair("salt", &hex::encode(salt))
             .append_pair("psql_session_id", psql_session_id);
 
-        let waiter = self.waiters.register(psql_session_id.to_owned());
+        with_waiter(psql_session_id, |waiter| async {
+            println!("cplane request: {}", url);
+            // TODO: leverage `reqwest::Client` to reuse connections
+            let resp = reqwest::get(url).await?;
+            if !resp.status().is_success() {
+                bail!("Auth failed: {}", resp.status())
+            }
 
-        println!("cplane request: {}", url);
-        let resp = reqwest::blocking::get(url)?;
-        if !resp.status().is_success() {
-            bail!("Auth failed: {}", resp.status())
-        }
+            let auth_info: ProxyAuthResponse = serde_json::from_str(resp.text().await?.as_str())?;
+            println!("got auth info: #{:?}", auth_info);
 
-        let auth_info: ProxyAuthResponse = serde_json::from_str(resp.text()?.as_str())?;
-        println!("got auth info: #{:?}", auth_info);
-
-        use ProxyAuthResponse::*;
-        match auth_info {
-            Ready { conn_info } => Ok(conn_info),
-            Error { error } => bail!(error),
-            NotReady { .. } => waiter.wait()?.map_err(|e| anyhow!(e)),
-        }
+            use ProxyAuthResponse::*;
+            match auth_info {
+                Ready { conn_info } => Ok(conn_info),
+                Error { error } => bail!(error),
+                NotReady { .. } => waiter.await?.map_err(|e| anyhow!(e)),
+            }
+        })
+        .await
     }
 }
 
+// NOTE: the order of constructors is important.
+// https://serde.rs/enum-representations.html#untagged
+#[derive(Serialize, Deserialize, Debug)]
+#[serde(untagged)]
+enum ProxyAuthResponse {
+    Ready { conn_info: DatabaseInfo },
+    Error { error: String },
+    NotReady { ready: bool }, // TODO: get rid of `ready`
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

proxy/src/http.rs

@@ -0,0 +1,30 @@
+use anyhow::anyhow;
+use hyper::{Body, Request, Response, StatusCode};
+use std::net::TcpListener;
+use zenith_utils::http::endpoint;
+use zenith_utils::http::error::ApiError;
+use zenith_utils::http::json::json_response;
+use zenith_utils::http::{RouterBuilder, RouterService};
+
+async fn status_handler(_: Request<Body>) -> Result<Response<Body>, ApiError> {
+    Ok(json_response(StatusCode::OK, "")?)
+}
+
+fn make_router() -> RouterBuilder<hyper::Body, ApiError> {
+    let router = endpoint::make_router();
+    router.get("/v1/status", status_handler)
+}
+
+pub async fn thread_main(http_listener: TcpListener) -> anyhow::Result<()> {
+    scopeguard::defer! {
+        println!("http has shut down");
+    }
+
+    let service = || RouterService::new(make_router().build()?);
+
+    hyper::Server::from_tcp(http_listener)?
+        .serve(service().map_err(|e| anyhow!(e))?)
+        .await?;
+
+    Ok(())
+}

proxy/src/main.rs

@@ -5,112 +5,162 @@
 /// (control plane API in our case) and can create new databases and accounts
 /// in somewhat transparent manner (again via communication with control plane API).
 ///
-use anyhow::bail;
+use anyhow::{bail, Context};
 use clap::{App, Arg};
-use state::{ProxyConfig, ProxyState};
-use std::thread;
-use zenith_utils::{tcp_listener, GIT_VERSION};
+use config::ProxyConfig;
+use futures::FutureExt;
+use std::future::Future;
+use tokio::{net::TcpListener, task::JoinError};
+use zenith_utils::GIT_VERSION;
 
+use crate::config::{ClientAuthMethod, RouterConfig};
+
+mod auth;
+mod cancellation;
+mod compute;
+mod config;
 mod cplane_api;
+mod http;
 mod mgmt;
 mod proxy;
-mod state;
+mod stream;
 mod waiters;
 
-fn main() -> anyhow::Result<()> {
+/// Flattens Result<Result<T>> into Result<T>.
+async fn flatten_err(
+    f: impl Future<Output = Result<anyhow::Result<()>, JoinError>>,
+) -> anyhow::Result<()> {
+    f.map(|r| r.context("join error").and_then(|x| x)).await
+}
+
+#[tokio::main]
+async fn main() -> anyhow::Result<()> {
+    zenith_metrics::set_common_metrics_prefix("zenith_proxy");
     let arg_matches = App::new("Zenith proxy/router")
         .version(GIT_VERSION)
         .arg(
-            Arg::with_name("proxy")
-                .short("p")
+            Arg::new("proxy")
+                .short('p')
                 .long("proxy")
                 .takes_value(true)
                 .help("listen for incoming client connections on ip:port")
                 .default_value("127.0.0.1:4432"),
         )
         .arg(
-            Arg::with_name("mgmt")
-                .short("m")
+            Arg::new("auth-method")
+                .long("auth-method")
+                .takes_value(true)
+                .help("Possible values: password | link | mixed")
+                .default_value("mixed"),
+        )
+        .arg(
+            Arg::new("static-router")
+                .short('s')
+                .long("static-router")
+                .takes_value(true)
+                .help("Route all clients to host:port"),
+        )
+        .arg(
+            Arg::new("mgmt")
+                .short('m')
                 .long("mgmt")
                 .takes_value(true)
                 .help("listen for management callback connection on ip:port")
                 .default_value("127.0.0.1:7000"),
         )
         .arg(
-            Arg::with_name("uri")
-                .short("u")
+            Arg::new("http")
+                .short('h')
+                .long("http")
+                .takes_value(true)
+                .help("listen for incoming http connections (metrics, etc) on ip:port")
+                .default_value("127.0.0.1:7001"),
+        )
+        .arg(
+            Arg::new("uri")
+                .short('u')
                 .long("uri")
                 .takes_value(true)
                 .help("redirect unauthenticated users to given uri")
                 .default_value("http://localhost:3000/psql_session/"),
         )
         .arg(
-            Arg::with_name("auth-endpoint")
-                .short("a")
+            Arg::new("auth-endpoint")
+                .short('a')
                 .long("auth-endpoint")
                 .takes_value(true)
-                .help("redirect unauthenticated users to given uri")
+                .help("API endpoint for authenticating users")
                 .default_value("http://localhost:3000/authenticate_proxy_request/"),
         )
         .arg(
-            Arg::with_name("ssl-key")
-                .short("k")
+            Arg::new("ssl-key")
+                .short('k')
                 .long("ssl-key")
                 .takes_value(true)
                 .help("path to SSL key for client postgres connections"),
         )
         .arg(
-            Arg::with_name("ssl-cert")
-                .short("c")
+            Arg::new("ssl-cert")
+                .short('c')
                 .long("ssl-cert")
                 .takes_value(true)
                 .help("path to SSL cert for client postgres connections"),
         )
         .get_matches();
 
-    let ssl_config = match (
+    let tls_config = match (
         arg_matches.value_of("ssl-key"),
         arg_matches.value_of("ssl-cert"),
     ) {
-        (Some(key_path), Some(cert_path)) => {
-            Some(crate::state::configure_ssl(key_path, cert_path)?)
-        }
+        (Some(key_path), Some(cert_path)) => Some(config::configure_ssl(key_path, cert_path)?),
         (None, None) => None,
         _ => bail!("either both or neither ssl-key and ssl-cert must be specified"),
     };
 
-    let config = ProxyConfig {
+    let auth_method = arg_matches.value_of("auth-method").unwrap().parse()?;
+    let router_config = match arg_matches.value_of("static-router") {
+        None => RouterConfig::Dynamic(auth_method),
+        Some(addr) => {
+            if let ClientAuthMethod::Password = auth_method {
+                let (host, port) = addr.split_once(":").unwrap();
+                RouterConfig::Static {
+                    host: host.to_string(),
+                    port: port.parse().unwrap(),
+                }
+            } else {
+                bail!("static-router requires --auth-method password")
+            }
+        }
+    };
+
+    let config: &ProxyConfig = Box::leak(Box::new(ProxyConfig {
+        router_config,
         proxy_address: arg_matches.value_of("proxy").unwrap().parse()?,
         mgmt_address: arg_matches.value_of("mgmt").unwrap().parse()?,
+        http_address: arg_matches.value_of("http").unwrap().parse()?,
         redirect_uri: arg_matches.value_of("uri").unwrap().parse()?,
         auth_endpoint: arg_matches.value_of("auth-endpoint").unwrap().parse()?,
-        ssl_config,
-    };
-    let state: &ProxyState = Box::leak(Box::new(ProxyState::new(config)));
+        tls_config,
+    }));
 
     println!("Version: {}", GIT_VERSION);
 
     // Check that we can bind to address before further initialization
-    println!("Starting proxy on {}", state.conf.proxy_address);
-    let pageserver_listener = tcp_listener::bind(state.conf.proxy_address)?;
+    println!("Starting http on {}", config.http_address);
+    let http_listener = TcpListener::bind(config.http_address).await?.into_std()?;
 
-    println!("Starting mgmt on {}", state.conf.mgmt_address);
-    let mgmt_listener = tcp_listener::bind(state.conf.mgmt_address)?;
+    println!("Starting mgmt on {}", config.mgmt_address);
+    let mgmt_listener = TcpListener::bind(config.mgmt_address).await?.into_std()?;
 
-    let threads = [
-        // Spawn a thread to listen for connections. It will spawn further threads
-        // for each connection.
-        thread::Builder::new()
-            .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))?,
-    ];
+    println!("Starting proxy on {}", config.proxy_address);
+    let proxy_listener = TcpListener::bind(config.proxy_address).await?;
 
-    for t in threads {
-        t.join().unwrap()?;
-    }
+    let http = tokio::spawn(http::thread_main(http_listener));
+    let proxy = tokio::spawn(proxy::thread_main(config, proxy_listener));
+    let mgmt = tokio::task::spawn_blocking(move || mgmt::thread_main(mgmt_listener));
+
+    let tasks = [flatten_err(http), flatten_err(proxy), flatten_err(mgmt)];
+    let _: Vec<()> = futures::future::try_join_all(tasks).await?;
 
     Ok(())
 }

proxy/src/mgmt.rs

@@ -1,44 +1,49 @@
+use crate::{compute::DatabaseInfo, cplane_api};
+use anyhow::Context;
+use serde::Deserialize;
 use std::{
     net::{TcpListener, TcpStream},
     thread,
 };
-
-use serde::Deserialize;
 use zenith_utils::{
     postgres_backend::{self, AuthType, PostgresBackend},
     pq_proto::{BeMessage, SINGLE_COL_ROWDESC},
 };
 
-use crate::{cplane_api::DatabaseInfo, ProxyState};
-
 ///
 /// Main proxy listener loop.
 ///
 /// Listens for connections, and launches a new handler thread for each.
 ///
-pub fn thread_main(state: &'static ProxyState, listener: TcpListener) -> anyhow::Result<()> {
+pub fn thread_main(listener: TcpListener) -> anyhow::Result<()> {
+    scopeguard::defer! {
+        println!("mgmt has shut down");
+    }
+
+    listener
+        .set_nonblocking(false)
+        .context("failed to set listener to blocking")?;
     loop {
-        let (socket, peer_addr) = listener.accept()?;
+        let (socket, peer_addr) = listener.accept().context("failed to accept a new client")?;
         println!("accepted connection from {}", peer_addr);
-        socket.set_nodelay(true).unwrap();
+        socket
+            .set_nodelay(true)
+            .context("failed to set client socket option")?;
 
         thread::spawn(move || {
-            if let Err(err) = handle_connection(state, socket) {
+            if let Err(err) = handle_connection(socket) {
                 println!("error: {}", err);
             }
         });
     }
 }
 
-fn handle_connection(state: &ProxyState, socket: TcpStream) -> anyhow::Result<()> {
-    let mut conn_handler = MgmtHandler { state };
+fn handle_connection(socket: TcpStream) -> anyhow::Result<()> {
     let pgbackend = PostgresBackend::new(socket, AuthType::Trust, None, true)?;
-    pgbackend.run(&mut conn_handler)
+    pgbackend.run(&mut MgmtHandler)
 }
 
-struct MgmtHandler<'a> {
-    state: &'a ProxyState,
-}
+struct MgmtHandler;
 
 /// Serialized examples:
 // {
@@ -74,13 +79,13 @@ enum PsqlSessionResult {
     Failure(String),
 }
 
-impl postgres_backend::Handler for MgmtHandler<'_> {
+impl postgres_backend::Handler for MgmtHandler {
     fn process_query(
         &mut self,
         pgb: &mut PostgresBackend,
         query_string: &str,
     ) -> anyhow::Result<()> {
-        let res = try_process_query(self, pgb, query_string);
+        let res = try_process_query(pgb, query_string);
         // intercept and log error message
         if res.is_err() {
             println!("Mgmt query failed: #{:?}", res);
@@ -89,11 +94,7 @@ impl postgres_backend::Handler for MgmtHandler<'_> {
     }
 }
 
-fn try_process_query(
-    mgmt: &mut MgmtHandler,
-    pgb: &mut PostgresBackend,
-    query_string: &str,
-) -> anyhow::Result<()> {
+fn try_process_query(pgb: &mut PostgresBackend, query_string: &str) -> anyhow::Result<()> {
     println!("Got mgmt query: '{}'", query_string);
 
     let resp: PsqlSessionResponse = serde_json::from_str(query_string)?;
@@ -104,14 +105,14 @@ fn try_process_query(
         Failure(message) => Err(message),
     };
 
-    match mgmt.state.waiters.notify(&resp.session_id, msg) {
+    match cplane_api::notify(&resp.session_id, msg) {
         Ok(()) => {
             pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
                 .write_message_noflush(&BeMessage::DataRow(&[Some(b"ok")]))?
                 .write_message(&BeMessage::CommandComplete(b"SELECT 1"))?;
         }
         Err(e) => {
-            pgb.write_message(&BeMessage::ErrorResponse(e.to_string()))?;
+            pgb.write_message(&BeMessage::ErrorResponse(&e.to_string()))?;
         }
     }
 

proxy/src/proxy.rs

@@ -1,363 +1,332 @@
-use crate::cplane_api::{CPlaneApi, DatabaseInfo};
-use crate::ProxyState;
-use anyhow::{anyhow, bail};
+use crate::auth;
+use crate::cancellation::{self, CancelClosure, CancelMap};
+use crate::compute::DatabaseInfo;
+use crate::config::{ProxyConfig, TlsConfig};
+use crate::stream::{MetricsStream, PqStream, Stream};
+use anyhow::{bail, Context};
 use lazy_static::lazy_static;
-use parking_lot::Mutex;
-use rand::prelude::StdRng;
-use rand::{Rng, SeedableRng};
-use std::cell::Cell;
-use std::collections::HashMap;
-use std::net::{SocketAddr, TcpStream};
-use std::{io, thread};
+use std::sync::Arc;
+use tokio::io::{AsyncRead, AsyncWrite};
+use tokio::net::TcpStream;
 use tokio_postgres::NoTls;
-use zenith_utils::postgres_backend::{self, PostgresBackend, ProtoState, Stream};
-use zenith_utils::pq_proto::{BeMessage as Be, FeMessage as Fe, *};
-use zenith_utils::sock_split::{ReadStream, WriteStream};
-
-struct CancelClosure {
-    socket_addr: SocketAddr,
-    cancel_token: tokio_postgres::CancelToken,
-}
-
-impl CancelClosure {
-    async fn try_cancel_query(&self) {
-        if let Ok(socket) = tokio::net::TcpStream::connect(self.socket_addr).await {
-            // NOTE ignoring the result because:
-            // 1. This is a best effort attempt, the database doesn't have to listen
-            // 2. Being opaque about errors here helps avoid leaking info to unauthenticated user
-            let _ = self.cancel_token.cancel_query_raw(socket, NoTls).await;
-        }
-    }
-}
+use zenith_metrics::{new_common_metric_name, register_int_counter, IntCounter};
+use zenith_utils::pq_proto::{BeMessage as Be, *};
 
 lazy_static! {
-    // Enables serving CancelRequests
-    static ref CANCEL_MAP: Mutex<HashMap<CancelKeyData, CancelClosure>> = Mutex::new(HashMap::new());
-}
-
-thread_local! {
-    // Used to clean up the CANCEL_MAP. Might not be necessary if we use tokio thread pool in main loop.
-    static THREAD_CANCEL_KEY_DATA: Cell<Option<CancelKeyData>> = Cell::new(None);
-}
-
-///
-/// Main proxy listener loop.
-///
-/// Listens for connections, and launches a new handler thread for each.
-///
-pub fn thread_main(
-    state: &'static ProxyState,
-    listener: std::net::TcpListener,
-) -> anyhow::Result<()> {
-    loop {
-        let (socket, peer_addr) = listener.accept()?;
-        println!("accepted connection from {}", peer_addr);
-        socket.set_nodelay(true).unwrap();
-
-        // TODO Use a threadpool instead. Maybe use tokio's threadpool by
-        //      spawning a future into its runtime. Tokio's JoinError should
-        //      allow us to handle cleanup properly even if the future panics.
-        thread::Builder::new()
-            .name("Proxy thread".into())
-            .spawn(move || {
-                if let Err(err) = proxy_conn_main(state, socket) {
-                    println!("error: {}", err);
-                }
-
-                // Clean up CANCEL_MAP.
-                THREAD_CANCEL_KEY_DATA.with(|cell| {
-                    if let Some(cancel_key_data) = cell.get() {
-                        CANCEL_MAP.lock().remove(&cancel_key_data);
-                    };
-                });
-            })?;
-    }
-}
-
-// TODO: clean up fields
-struct ProxyConnection {
-    state: &'static ProxyState,
-    psql_session_id: String,
-    pgb: PostgresBackend,
-}
-
-pub fn proxy_conn_main(state: &'static ProxyState, socket: TcpStream) -> anyhow::Result<()> {
-    let conn = ProxyConnection {
-        state,
-        psql_session_id: hex::encode(rand::random::<[u8; 8]>()),
-        pgb: PostgresBackend::new(
-            socket,
-            postgres_backend::AuthType::MD5,
-            state.conf.ssl_config.clone(),
-            false,
-        )?,
-    };
-
-    let (client, server) = match conn.handle_client()? {
-        Some(x) => x,
-        None => return Ok(()),
-    };
-
-    let server = zenith_utils::sock_split::BidiStream::from_tcp(server);
-
-    let client = match client {
-        Stream::Bidirectional(bidi_stream) => bidi_stream,
-        _ => panic!("invalid stream type"),
-    };
-
-    proxy(client.split(), server.split())
-}
-
-impl ProxyConnection {
-    /// Returns Ok(None) when connection was successfully closed.
-    fn handle_client(mut self) -> anyhow::Result<Option<(Stream, TcpStream)>> {
-        let mut authenticate = || {
-            let (username, dbname) = match self.handle_startup()? {
-                Some(x) => x,
-                None => return Ok(None),
-            };
-
-            // Both scenarios here should end up producing database credentials
-            if username.ends_with("@zenith") {
-                self.handle_existing_user(&username, &dbname).map(Some)
-            } else {
-                self.handle_new_user().map(Some)
-            }
-        };
-
-        let conn = match authenticate() {
-            Ok(Some(db_info)) => connect_to_db(db_info),
-            Ok(None) => return Ok(None),
-            Err(e) => {
-                // Report the error to the client
-                self.pgb.write_message(&Be::ErrorResponse(e.to_string()))?;
-                bail!("failed to handle client: {:?}", e);
-            }
-        };
-
-        // We'll get rid of this once migration to async is complete
-        let (pg_version, db_stream) = {
-            let runtime = tokio::runtime::Builder::new_current_thread()
-                .enable_all()
-                .build()?;
-
-            let (pg_version, stream, cancel_key_data) = runtime.block_on(conn)?;
-            self.pgb
-                .write_message(&BeMessage::BackendKeyData(cancel_key_data))?;
-            let stream = stream.into_std()?;
-            stream.set_nonblocking(false)?;
-
-            (pg_version, stream)
-        };
-
-        // Let the client send new requests
-        self.pgb
-            .write_message_noflush(&BeMessage::ParameterStatus(
-                BeParameterStatusMessage::ServerVersion(&pg_version),
-            ))?
-            .write_message(&Be::ReadyForQuery)?;
-
-        Ok(Some((self.pgb.into_stream(), db_stream)))
-    }
-
-    /// Returns Ok(None) when connection was successfully closed.
-    fn handle_startup(&mut self) -> anyhow::Result<Option<(String, String)>> {
-        let have_tls = self.pgb.tls_config.is_some();
-        let mut encrypted = false;
-
-        loop {
-            let msg = match self.pgb.read_message()? {
-                Some(Fe::StartupPacket(msg)) => msg,
-                None => bail!("connection is lost"),
-                bad => bail!("unexpected message type: {:?}", bad),
-            };
-            println!("got message: {:?}", msg);
-
-            match msg {
-                FeStartupPacket::GssEncRequest => {
-                    self.pgb.write_message(&Be::EncryptionResponse(false))?;
-                }
-                FeStartupPacket::SslRequest => {
-                    self.pgb.write_message(&Be::EncryptionResponse(have_tls))?;
-                    if have_tls {
-                        self.pgb.start_tls()?;
-                        encrypted = true;
-                    }
-                }
-                FeStartupPacket::StartupMessage { mut params, .. } => {
-                    if have_tls && !encrypted {
-                        bail!("must connect with TLS");
-                    }
-
-                    let mut get_param = |key| {
-                        params
-                            .remove(key)
-                            .ok_or_else(|| anyhow!("{} is missing in startup packet", key))
-                    };
-
-                    return Ok(Some((get_param("user")?, get_param("database")?)));
-                }
-                FeStartupPacket::CancelRequest(cancel_key_data) => {
-                    if let Some(cancel_closure) = CANCEL_MAP.lock().get(&cancel_key_data) {
-                        let runtime = tokio::runtime::Builder::new_current_thread()
-                            .enable_all()
-                            .build()
-                            .unwrap();
-                        runtime.block_on(cancel_closure.try_cancel_query());
-                    }
-                    return Ok(None);
-                }
-            }
-        }
-    }
-
-    fn handle_existing_user(&mut self, user: &str, db: &str) -> anyhow::Result<DatabaseInfo> {
-        let md5_salt = rand::random::<[u8; 4]>();
-
-        // Ask password
-        self.pgb
-            .write_message(&Be::AuthenticationMD5Password(&md5_salt))?;
-        self.pgb.state = ProtoState::Authentication; // XXX
-
-        // Check password
-        let msg = match self.pgb.read_message()? {
-            Some(Fe::PasswordMessage(msg)) => msg,
-            None => bail!("connection is lost"),
-            bad => bail!("unexpected message type: {:?}", bad),
-        };
-        println!("got message: {:?}", msg);
-
-        let (_trailing_null, md5_response) = msg
-            .split_last()
-            .ok_or_else(|| anyhow!("unexpected password message"))?;
-
-        let cplane = CPlaneApi::new(&self.state.conf.auth_endpoint, &self.state.waiters);
-        let db_info = cplane.authenticate_proxy_request(
-            user,
-            db,
-            md5_response,
-            &md5_salt,
-            &self.psql_session_id,
-        )?;
-
-        self.pgb
-            .write_message_noflush(&Be::AuthenticationOk)?
-            .write_message_noflush(&BeParameterStatusMessage::encoding())?;
-
-        Ok(db_info)
-    }
-
-    fn handle_new_user(&mut self) -> anyhow::Result<DatabaseInfo> {
-        let greeting = hello_message(&self.state.conf.redirect_uri, &self.psql_session_id);
-
-        // First, register this session
-        let waiter = self.state.waiters.register(self.psql_session_id.clone());
-
-        // Give user a URL to spawn a new database
-        self.pgb
-            .write_message_noflush(&Be::AuthenticationOk)?
-            .write_message_noflush(&BeParameterStatusMessage::encoding())?
-            .write_message(&Be::NoticeResponse(greeting))?;
-
-        // Wait for web console response
-        let db_info = waiter.wait()?.map_err(|e| anyhow!(e))?;
-
-        self.pgb
-            .write_message_noflush(&Be::NoticeResponse("Connecting to database.".into()))?;
-
-        Ok(db_info)
-    }
-}
-
-fn hello_message(redirect_uri: &str, session_id: &str) -> String {
-    format!(
-        concat![
-            "☀️  Welcome to Zenith!\n",
-            "To proceed with database creation, open the following link:\n\n",
-            "    {redirect_uri}{session_id}\n\n",
-            "It needs to be done once and we will send you '.pgpass' file,\n",
-            "which will allow you to access or create ",
-            "databases without opening your web browser."
-        ],
-        redirect_uri = redirect_uri,
-        session_id = session_id,
+    static ref NUM_CONNECTIONS_ACCEPTED_COUNTER: IntCounter = register_int_counter!(
+        new_common_metric_name("num_connections_accepted"),
+        "Number of TCP client connections accepted."
     )
+    .unwrap();
+    static ref NUM_CONNECTIONS_CLOSED_COUNTER: IntCounter = register_int_counter!(
+        new_common_metric_name("num_connections_closed"),
+        "Number of TCP client connections closed."
+    )
+    .unwrap();
+    static ref NUM_BYTES_PROXIED_COUNTER: IntCounter = register_int_counter!(
+        new_common_metric_name("num_bytes_proxied"),
+        "Number of bytes sent/received between any client and backend."
+    )
+    .unwrap();
 }
 
-/// 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<(String, tokio::net::TcpStream, CancelKeyData)> {
-    // Make raw connection. When connect_raw finishes we've received ReadyForQuery.
-    let socket_addr = db_info.socket_addr()?;
-    let mut socket = tokio::net::TcpStream::connect(socket_addr).await?;
-    let config = tokio_postgres::Config::from(db_info);
-    // NOTE We effectively ignore some ParameterStatus and NoticeResponse
-    //      messages here. Not sure if that could break something.
-    let (client, conn) = config.connect_raw(&mut socket, NoTls).await?;
-
-    // Save info for potentially cancelling the query later
-    let mut rng = StdRng::from_entropy();
-    let cancel_key_data = CancelKeyData {
-        // HACK We'd rather get the real backend_pid but tokio_postgres doesn't
-        //      expose it and we don't want to do another roundtrip to query
-        //      for it. The client will be able to notice that this is not the
-        //      actual backend_pid, but backend_pid is not used for anything
-        //      so it doesn't matter.
-        backend_pid: rng.gen(),
-        cancel_key: rng.gen(),
-    };
-    let cancel_closure = CancelClosure {
-        socket_addr,
-        cancel_token: client.cancel_token(),
-    };
-    CANCEL_MAP.lock().insert(cancel_key_data, cancel_closure);
-    THREAD_CANCEL_KEY_DATA.with(|cell| {
-        let prev_value = cell.replace(Some(cancel_key_data));
-        assert!(
-            prev_value.is_none(),
-            "THREAD_CANCEL_KEY_DATA was already set"
-        );
-    });
-
-    let version = conn.parameter("server_version").unwrap();
-    Ok((version.into(), socket, cancel_key_data))
+async fn log_error<R, F>(future: F) -> F::Output
+where
+    F: std::future::Future<Output = anyhow::Result<R>>,
+{
+    future.await.map_err(|err| {
+        println!("error: {}", err);
+        err
+    })
 }
 
-/// Concurrently proxy both directions of the client and server connections
-fn proxy(
-    (client_read, client_write): (ReadStream, WriteStream),
-    (server_read, server_write): (ReadStream, WriteStream),
+pub async fn thread_main(
+    config: &'static ProxyConfig,
+    listener: tokio::net::TcpListener,
 ) -> anyhow::Result<()> {
-    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.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
+    scopeguard::defer! {
+        println!("proxy has shut down");
     }
 
-    let client_to_server_jh = thread::spawn(move || do_proxy(client_read, server_write));
+    let cancel_map = Arc::new(CancelMap::default());
+    loop {
+        let (socket, peer_addr) = listener.accept().await?;
+        println!("accepted connection from {}", peer_addr);
 
-    do_proxy(server_read, client_write)?;
-    client_to_server_jh.join().unwrap()?;
+        let cancel_map = Arc::clone(&cancel_map);
+        tokio::spawn(log_error(async move {
+            socket
+                .set_nodelay(true)
+                .context("failed to set socket option")?;
+
+            handle_client(config, &cancel_map, socket).await
+        }));
+    }
+}
+
+async fn handle_client(
+    config: &ProxyConfig,
+    cancel_map: &CancelMap,
+    stream: impl AsyncRead + AsyncWrite + Unpin,
+) -> anyhow::Result<()> {
+    // The `closed` counter will increase when this future is destroyed.
+    NUM_CONNECTIONS_ACCEPTED_COUNTER.inc();
+    scopeguard::defer! {
+        NUM_CONNECTIONS_CLOSED_COUNTER.inc();
+    }
+
+    let tls = config.tls_config.clone();
+    if let Some((client, creds)) = handshake(stream, tls, cancel_map).await? {
+        cancel_map
+            .with_session(|session| async {
+                connect_client_to_db(config, session, client, creds).await
+            })
+            .await?;
+    }
 
     Ok(())
 }
+
+/// Handle a connection from one client.
+/// For better testing experience, `stream` can be
+/// any object satisfying the traits.
+async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
+    stream: S,
+    mut tls: Option<TlsConfig>,
+    cancel_map: &CancelMap,
+) -> anyhow::Result<Option<(PqStream<Stream<S>>, auth::ClientCredentials)>> {
+    // Client may try upgrading to each protocol only once
+    let (mut tried_ssl, mut tried_gss) = (false, false);
+
+    let mut stream = PqStream::new(Stream::from_raw(stream));
+    loop {
+        let msg = stream.read_startup_packet().await?;
+        println!("got message: {:?}", msg);
+
+        use FeStartupPacket::*;
+        match msg {
+            SslRequest => match stream.get_ref() {
+                Stream::Raw { .. } if !tried_ssl => {
+                    tried_ssl = true;
+
+                    // We can't perform TLS handshake without a config
+                    let enc = tls.is_some();
+                    stream.write_message(&Be::EncryptionResponse(enc)).await?;
+
+                    if let Some(tls) = tls.take() {
+                        // Upgrade raw stream into a secure TLS-backed stream.
+                        // NOTE: We've consumed `tls`; this fact will be used later.
+                        stream = PqStream::new(stream.into_inner().upgrade(tls).await?);
+                    }
+                }
+                _ => bail!("protocol violation"),
+            },
+            GssEncRequest => match stream.get_ref() {
+                Stream::Raw { .. } if !tried_gss => {
+                    tried_gss = true;
+
+                    // Currently, we don't support GSSAPI
+                    stream.write_message(&Be::EncryptionResponse(false)).await?;
+                }
+                _ => bail!("protocol violation"),
+            },
+            StartupMessage { params, .. } => {
+                // Check that the config has been consumed during upgrade
+                // OR we didn't provide it at all (for dev purposes).
+                if tls.is_some() {
+                    let msg = "connection is insecure (try using `sslmode=require`)";
+                    stream.write_message(&Be::ErrorResponse(msg)).await?;
+                    bail!(msg);
+                }
+
+                break Ok(Some((stream, params.try_into()?)));
+            }
+            CancelRequest(cancel_key_data) => {
+                cancel_map.cancel_session(cancel_key_data).await?;
+
+                break Ok(None);
+            }
+        }
+    }
+}
+
+async fn connect_client_to_db(
+    config: &ProxyConfig,
+    session: cancellation::Session<'_>,
+    mut client: PqStream<impl AsyncRead + AsyncWrite + Unpin>,
+    creds: auth::ClientCredentials,
+) -> anyhow::Result<()> {
+    let db_info = creds.authenticate(config, &mut client).await?;
+    let (db, version, cancel_closure) = connect_to_db(db_info).await?;
+    let cancel_key_data = session.enable_cancellation(cancel_closure);
+
+    client
+        .write_message_noflush(&BeMessage::ParameterStatus(
+            BeParameterStatusMessage::ServerVersion(&version),
+        ))?
+        .write_message_noflush(&Be::BackendKeyData(cancel_key_data))?
+        .write_message(&BeMessage::ReadyForQuery)
+        .await?;
+
+    // This function will be called for writes to either direction.
+    fn inc_proxied(cnt: usize) {
+        // Consider inventing something more sophisticated
+        // if this ever becomes a bottleneck (cacheline bouncing).
+        NUM_BYTES_PROXIED_COUNTER.inc_by(cnt as u64);
+    }
+
+    let mut db = MetricsStream::new(db, inc_proxied);
+    let mut client = MetricsStream::new(client.into_inner(), inc_proxied);
+    let _ = tokio::io::copy_bidirectional(&mut client, &mut db).await?;
+
+    Ok(())
+}
+
+/// Connect to a corresponding compute node.
+async fn connect_to_db(
+    db_info: DatabaseInfo,
+) -> anyhow::Result<(TcpStream, String, CancelClosure)> {
+    // TODO: establish a secure connection to the DB
+    let socket_addr = db_info.socket_addr()?;
+    let mut socket = TcpStream::connect(socket_addr).await?;
+
+    let (client, conn) = tokio_postgres::Config::from(db_info)
+        .connect_raw(&mut socket, NoTls)
+        .await?;
+
+    let version = conn
+        .parameter("server_version")
+        .context("failed to fetch postgres server version")?
+        .into();
+
+    let cancel_closure = CancelClosure::new(socket_addr, client.cancel_token());
+
+    Ok((socket, version, cancel_closure))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    use tokio::io::DuplexStream;
+    use tokio_postgres::config::SslMode;
+    use tokio_postgres::tls::MakeTlsConnect;
+    use tokio_postgres_rustls::MakeRustlsConnect;
+
+    async fn dummy_proxy(
+        client: impl AsyncRead + AsyncWrite + Unpin,
+        tls: Option<TlsConfig>,
+    ) -> anyhow::Result<()> {
+        let cancel_map = CancelMap::default();
+
+        // TODO: add some infra + tests for credentials
+        let (mut stream, _creds) = handshake(client, tls, &cancel_map)
+            .await?
+            .context("no stream")?;
+
+        stream
+            .write_message_noflush(&Be::AuthenticationOk)?
+            .write_message_noflush(&BeParameterStatusMessage::encoding())?
+            .write_message(&BeMessage::ReadyForQuery)
+            .await?;
+
+        Ok(())
+    }
+
+    fn generate_certs(
+        hostname: &str,
+    ) -> anyhow::Result<(rustls::Certificate, rustls::Certificate, rustls::PrivateKey)> {
+        let ca = rcgen::Certificate::from_params({
+            let mut params = rcgen::CertificateParams::default();
+            params.is_ca = rcgen::IsCa::Ca(rcgen::BasicConstraints::Unconstrained);
+            params
+        })?;
+
+        let cert = rcgen::generate_simple_self_signed(vec![hostname.into()])?;
+        Ok((
+            rustls::Certificate(ca.serialize_der()?),
+            rustls::Certificate(cert.serialize_der_with_signer(&ca)?),
+            rustls::PrivateKey(cert.serialize_private_key_der()),
+        ))
+    }
+
+    #[tokio::test]
+    async fn handshake_tls_is_enforced_by_proxy() -> anyhow::Result<()> {
+        let (client, server) = tokio::io::duplex(1024);
+
+        let server_config = {
+            let (_ca, cert, key) = generate_certs("localhost")?;
+
+            let mut config = rustls::ServerConfig::new(rustls::NoClientAuth::new());
+            config.set_single_cert(vec![cert], key)?;
+            config
+        };
+
+        let proxy = tokio::spawn(dummy_proxy(client, Some(server_config.into())));
+
+        tokio_postgres::Config::new()
+            .user("john_doe")
+            .dbname("earth")
+            .ssl_mode(SslMode::Disable)
+            .connect_raw(server, NoTls)
+            .await
+            .err() // -> Option<E>
+            .context("client shouldn't be able to connect")?;
+
+        proxy
+            .await?
+            .err() // -> Option<E>
+            .context("server shouldn't accept client")?;
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn handshake_tls() -> anyhow::Result<()> {
+        let (client, server) = tokio::io::duplex(1024);
+
+        let (ca, cert, key) = generate_certs("localhost")?;
+
+        let server_config = {
+            let mut config = rustls::ServerConfig::new(rustls::NoClientAuth::new());
+            config.set_single_cert(vec![cert], key)?;
+            config
+        };
+
+        let proxy = tokio::spawn(dummy_proxy(client, Some(server_config.into())));
+
+        let client_config = {
+            let mut config = rustls::ClientConfig::new();
+            config.root_store.add(&ca)?;
+            config
+        };
+
+        let mut mk = MakeRustlsConnect::new(client_config);
+        let tls = MakeTlsConnect::<DuplexStream>::make_tls_connect(&mut mk, "localhost")?;
+
+        let (_client, _conn) = tokio_postgres::Config::new()
+            .user("john_doe")
+            .dbname("earth")
+            .ssl_mode(SslMode::Require)
+            .connect_raw(server, tls)
+            .await?;
+
+        proxy.await?
+    }
+
+    #[tokio::test]
+    async fn handshake_raw() -> anyhow::Result<()> {
+        let (client, server) = tokio::io::duplex(1024);
+
+        let proxy = tokio::spawn(dummy_proxy(client, None));
+
+        let (_client, _conn) = tokio_postgres::Config::new()
+            .user("john_doe")
+            .dbname("earth")
+            .ssl_mode(SslMode::Prefer)
+            .connect_raw(server, NoTls)
+            .await?;
+
+        proxy.await?
+    }
+}

proxy/src/stream.rs

@@ -0,0 +1,230 @@
+use anyhow::Context;
+use bytes::BytesMut;
+use pin_project_lite::pin_project;
+use rustls::ServerConfig;
+use std::pin::Pin;
+use std::sync::Arc;
+use std::{io, task};
+use tokio::io::{AsyncRead, AsyncWrite, AsyncWriteExt, ReadBuf};
+use tokio_rustls::server::TlsStream;
+use zenith_utils::pq_proto::{BeMessage, FeMessage, FeStartupPacket};
+
+pin_project! {
+    /// Stream wrapper which implements libpq's protocol.
+    /// NOTE: This object deliberately doesn't implement [`AsyncRead`]
+    /// or [`AsyncWrite`] to prevent subtle errors (e.g. trying
+    /// to pass random malformed bytes through the connection).
+    pub struct PqStream<S> {
+        #[pin]
+        stream: S,
+        buffer: BytesMut,
+    }
+}
+
+impl<S> PqStream<S> {
+    /// Construct a new libpq protocol wrapper.
+    pub fn new(stream: S) -> Self {
+        Self {
+            stream,
+            buffer: Default::default(),
+        }
+    }
+
+    /// Extract the underlying stream.
+    pub fn into_inner(self) -> S {
+        self.stream
+    }
+
+    /// Get a reference to the underlying stream.
+    pub fn get_ref(&self) -> &S {
+        &self.stream
+    }
+}
+
+impl<S: AsyncRead + Unpin> PqStream<S> {
+    /// Receive [`FeStartupPacket`], which is a first packet sent by a client.
+    pub async fn read_startup_packet(&mut self) -> anyhow::Result<FeStartupPacket> {
+        match FeStartupPacket::read_fut(&mut self.stream).await? {
+            Some(FeMessage::StartupPacket(packet)) => Ok(packet),
+            None => anyhow::bail!("connection is lost"),
+            other => anyhow::bail!("bad message type: {:?}", other),
+        }
+    }
+
+    pub async fn read_message(&mut self) -> anyhow::Result<FeMessage> {
+        FeMessage::read_fut(&mut self.stream)
+            .await?
+            .context("connection is lost")
+    }
+}
+
+impl<S: AsyncWrite + Unpin> PqStream<S> {
+    /// Write the message into an internal buffer, but don't flush the underlying stream.
+    pub fn write_message_noflush<'a>(&mut self, message: &BeMessage<'a>) -> io::Result<&mut Self> {
+        BeMessage::write(&mut self.buffer, message)?;
+        Ok(self)
+    }
+
+    /// Write the message into an internal buffer and flush it.
+    pub async fn write_message<'a>(&mut self, message: &BeMessage<'a>) -> io::Result<&mut Self> {
+        self.write_message_noflush(message)?;
+        self.flush().await?;
+        Ok(self)
+    }
+
+    /// Flush the output buffer into the underlying stream.
+    pub async fn flush(&mut self) -> io::Result<&mut Self> {
+        self.stream.write_all(&self.buffer).await?;
+        self.buffer.clear();
+        self.stream.flush().await?;
+        Ok(self)
+    }
+}
+
+pin_project! {
+    /// Wrapper for upgrading raw streams into secure streams.
+    /// NOTE: it should be possible to decompose this object as necessary.
+    #[project = StreamProj]
+    pub enum Stream<S> {
+        /// We always begin with a raw stream,
+        /// which may then be upgraded into a secure stream.
+        Raw { #[pin] raw: S },
+        /// We box [`TlsStream`] since it can be quite large.
+        Tls { #[pin] tls: Box<TlsStream<S>> },
+    }
+}
+
+impl<S> Stream<S> {
+    /// Construct a new instance from a raw stream.
+    pub fn from_raw(raw: S) -> Self {
+        Self::Raw { raw }
+    }
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin> Stream<S> {
+    /// If possible, upgrade raw stream into a secure TLS-based stream.
+    pub async fn upgrade(self, cfg: Arc<ServerConfig>) -> anyhow::Result<Self> {
+        match self {
+            Stream::Raw { raw } => {
+                let tls = Box::new(tokio_rustls::TlsAcceptor::from(cfg).accept(raw).await?);
+                Ok(Stream::Tls { tls })
+            }
+            Stream::Tls { .. } => anyhow::bail!("can't upgrade TLS stream"),
+        }
+    }
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin> AsyncRead for Stream<S> {
+    fn poll_read(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+        buf: &mut ReadBuf<'_>,
+    ) -> task::Poll<io::Result<()>> {
+        use StreamProj::*;
+        match self.project() {
+            Raw { raw } => raw.poll_read(context, buf),
+            Tls { tls } => tls.poll_read(context, buf),
+        }
+    }
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin> AsyncWrite for Stream<S> {
+    fn poll_write(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+        buf: &[u8],
+    ) -> task::Poll<io::Result<usize>> {
+        use StreamProj::*;
+        match self.project() {
+            Raw { raw } => raw.poll_write(context, buf),
+            Tls { tls } => tls.poll_write(context, buf),
+        }
+    }
+
+    fn poll_flush(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+    ) -> task::Poll<io::Result<()>> {
+        use StreamProj::*;
+        match self.project() {
+            Raw { raw } => raw.poll_flush(context),
+            Tls { tls } => tls.poll_flush(context),
+        }
+    }
+
+    fn poll_shutdown(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+    ) -> task::Poll<io::Result<()>> {
+        use StreamProj::*;
+        match self.project() {
+            Raw { raw } => raw.poll_shutdown(context),
+            Tls { tls } => tls.poll_shutdown(context),
+        }
+    }
+}
+
+pin_project! {
+    /// This stream tracks all writes and calls user provided
+    /// callback when the underlying stream is flushed.
+    pub struct MetricsStream<S, W> {
+        #[pin]
+        stream: S,
+        write_count: usize,
+        inc_write_count: W,
+    }
+}
+
+impl<S, W> MetricsStream<S, W> {
+    pub fn new(stream: S, inc_write_count: W) -> Self {
+        Self {
+            stream,
+            write_count: 0,
+            inc_write_count,
+        }
+    }
+}
+
+impl<S: AsyncRead + Unpin, W> AsyncRead for MetricsStream<S, W> {
+    fn poll_read(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+        buf: &mut ReadBuf<'_>,
+    ) -> task::Poll<io::Result<()>> {
+        self.project().stream.poll_read(context, buf)
+    }
+}
+
+impl<S: AsyncWrite + Unpin, W: FnMut(usize)> AsyncWrite for MetricsStream<S, W> {
+    fn poll_write(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+        buf: &[u8],
+    ) -> task::Poll<io::Result<usize>> {
+        let this = self.project();
+        this.stream.poll_write(context, buf).map_ok(|cnt| {
+            // Increment the write count.
+            *this.write_count += cnt;
+            cnt
+        })
+    }
+
+    fn poll_flush(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+    ) -> task::Poll<io::Result<()>> {
+        let this = self.project();
+        this.stream.poll_flush(context).map_ok(|()| {
+            // Call the user provided callback and reset the write count.
+            (this.inc_write_count)(*this.write_count);
+            *this.write_count = 0;
+        })
+    }
+
+    fn poll_shutdown(
+        self: Pin<&mut Self>,
+        context: &mut task::Context<'_>,
+    ) -> task::Poll<io::Result<()>> {
+        self.project().stream.poll_shutdown(context)
+    }
+}

proxy/src/waiters.rs

@@ -1,8 +1,12 @@
 use anyhow::{anyhow, Context};
-use std::collections::HashMap;
-use std::sync::{mpsc, Mutex};
+use hashbrown::HashMap;
+use parking_lot::Mutex;
+use pin_project_lite::pin_project;
+use std::pin::Pin;
+use std::task;
+use tokio::sync::oneshot;
 
-pub struct Waiters<T>(pub(self) Mutex<HashMap<String, mpsc::Sender<T>>>);
+pub struct Waiters<T>(pub(self) Mutex<HashMap<String, oneshot::Sender<T>>>);
 
 impl<T> Default for Waiters<T> {
     fn default() -> Self {
@@ -11,48 +15,86 @@ impl<T> Default for Waiters<T> {
 }
 
 impl<T> Waiters<T> {
-    pub fn register(&self, key: String) -> Waiter<T> {
-        let (tx, rx) = mpsc::channel();
+    pub fn register(&self, key: String) -> anyhow::Result<Waiter<T>> {
+        let (tx, rx) = oneshot::channel();
 
-        // TODO: use `try_insert` (unstable)
-        let prev = self.0.lock().unwrap().insert(key.clone(), tx);
-        assert!(matches!(prev, None)); // assert_matches! is nightly-only
+        self.0
+            .lock()
+            .try_insert(key.clone(), tx)
+            .map_err(|_| anyhow!("waiter already registered"))?;
 
-        Waiter {
+        Ok(Waiter {
             receiver: rx,
-            registry: self,
-            key,
-        }
+            guard: DropKey {
+                registry: self,
+                key,
+            },
+        })
     }
 
     pub fn notify(&self, key: &str, value: T) -> anyhow::Result<()>
     where
-        T: Send + Sync + 'static,
+        T: Send + Sync,
     {
         let tx = self
             .0
             .lock()
-            .unwrap()
             .remove(key)
-            .ok_or_else(|| anyhow!("key {} not found", key))?;
-        tx.send(value).context("channel hangup")
+            .with_context(|| format!("key {} not found", key))?;
+
+        tx.send(value).map_err(|_| anyhow!("waiter channel hangup"))
     }
 }
 
-pub struct Waiter<'a, T> {
-    receiver: mpsc::Receiver<T>,
-    registry: &'a Waiters<T>,
+struct DropKey<'a, T> {
     key: String,
+    registry: &'a Waiters<T>,
 }
 
-impl<T> Waiter<'_, T> {
-    pub fn wait(self) -> anyhow::Result<T> {
-        self.receiver.recv().context("channel hangup")
-    }
-}
-
-impl<T> Drop for Waiter<'_, T> {
+impl<'a, T> Drop for DropKey<'a, T> {
     fn drop(&mut self) {
-        self.registry.0.lock().unwrap().remove(&self.key);
+        self.registry.0.lock().remove(&self.key);
+    }
+}
+
+pin_project! {
+    pub struct Waiter<'a, T> {
+        #[pin]
+        receiver: oneshot::Receiver<T>,
+        guard: DropKey<'a, T>,
+    }
+}
+
+impl<T> std::future::Future for Waiter<'_, T> {
+    type Output = anyhow::Result<T>;
+
+    fn poll(self: Pin<&mut Self>, cx: &mut task::Context<'_>) -> task::Poll<Self::Output> {
+        self.project()
+            .receiver
+            .poll(cx)
+            .map_err(|_| anyhow!("channel hangup"))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::sync::Arc;
+
+    #[tokio::test]
+    async fn test_waiter() -> anyhow::Result<()> {
+        let waiters = Arc::new(Waiters::default());
+
+        let key = "Key";
+        let waiter = waiters.register(key.to_owned())?;
+
+        let waiters = Arc::clone(&waiters);
+        let notifier = tokio::spawn(async move {
+            waiters.notify(key, Default::default())?;
+            Ok(())
+        });
+
+        let () = waiter.await?;
+        notifier.await?
     }
 }

pyproject.toml

@@ -0,0 +1,32 @@
+[tool.poetry]
+name = "zenith"
+version = "0.1.0"
+description = ""
+authors = []
+
+[tool.poetry.dependencies]
+python = "^3.7"
+pytest = "^6.2.5"
+psycopg2-binary = "^2.9.1"
+typing-extensions = "^3.10.0"
+PyJWT = {version = "^2.1.0", extras = ["crypto"]}
+requests = "^2.26.0"
+pytest-xdist = "^2.3.0"
+asyncpg = "^0.24.0"
+aiopg = "^1.3.1"
+cached-property = "^1.5.2"
+Jinja2 = "^3.0.2"
+types-requests = "^2.27.7"
+types-psycopg2 = "^2.9.6"
+boto3 = "^1.20.40"
+boto3-stubs = "^1.20.40"
+moto = {version = "^3.0.0", extras = ["server"]}
+
+[tool.poetry.dev-dependencies]
+yapf = "==0.31.0"
+flake8 = "^3.9.2"
+mypy = "==0.910"
+
+[build-system]
+requires = ["poetry-core>=1.0.0"]
+build-backend = "poetry.core.masonry.api"

pytest.ini

@@ -3,6 +3,8 @@ addopts =
     -m 'not remote_cluster'
 markers =
     remote_cluster
+testpaths =
+    test_runner
 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

scripts/coverage

@@ -14,17 +14,30 @@ from dataclasses import dataclass
 from pathlib import Path
 from tempfile import TemporaryDirectory
 from textwrap import dedent
-from typing import Any, Iterable, List, Optional
+from typing import Any, Dict, Iterator, Iterable, List, Optional
 
 import argparse
+import hashlib
 import json
 import os
 import shutil
+import socket
 import subprocess
 import sys
 
 
-def intersperse(sep: Any, iterable: Iterable[Any]):
+def file_mtime_or_zero(path: Path) -> int:
+    try:
+        return path.stat().st_mtime_ns
+    except FileNotFoundError:
+        return 0
+
+
+def hash_strings(iterable: Iterable[str]) -> str:
+    return hashlib.sha1(''.join(iterable).encode('utf-8')).hexdigest()
+
+
+def intersperse(sep: Any, iterable: Iterable[Any]) -> Iterator[Any]:
     fst = True
     for item in iterable:
         if not fst:
@@ -33,18 +46,18 @@ def intersperse(sep: Any, iterable: Iterable[Any]):
         yield item
 
 
-def find_demangler(demangler=None):
+def find_demangler(demangler: Optional[Path] = None) -> Path:
     known_tools = ['c++filt', 'rustfilt', 'llvm-cxxfilt']
 
     if demangler:
         # Explicit argument has precedence over `known_tools`
         demanglers = [demangler]
     else:
-        demanglers = known_tools
+        demanglers = [Path(x) for x in known_tools]
 
-    for demangler in demanglers:
-        if shutil.which(demangler):
-            return demangler
+    for exe in demanglers:
+        if shutil.which(exe):
+            return exe
 
     raise Exception(' '.join([
         'Failed to find symbol demangler.',
@@ -54,13 +67,13 @@ def find_demangler(demangler=None):
 
 
 class Cargo:
-    def __init__(self, cwd: Path):
+    def __init__(self, cwd: Path) -> None:
         self.cwd = cwd
         self.target_dir = Path(os.environ.get('CARGO_TARGET_DIR', cwd / 'target')).resolve()
-        self._rustlib_dir = None
+        self._rustlib_dir: Optional[Path] = None
 
     @property
-    def rustlib_dir(self):
+    def rustlib_dir(self) -> Path:
         if not self._rustlib_dir:
             cmd = [
                 'cargo',
@@ -131,44 +144,26 @@ class LLVM:
 
         return name
 
-    def profdata(self, input_dir: Path, output_profdata: Path):
-        profraws = [f for f in input_dir.iterdir() if f.suffix == '.profraw']
-        if not profraws:
-            raise Exception(f'No profraw files found at {input_dir}')
-
-        with open(input_dir / 'profraw.list', 'w') as input_files:
-            profraw_mtime = 0
-            for profraw in profraws:
-                profraw_mtime = max(profraw_mtime, profraw.stat().st_mtime_ns)
-                print(profraw, file=input_files)
-            input_files.flush()
-
-            try:
-                profdata_mtime = output_profdata.stat().st_mtime_ns
-            except FileNotFoundError:
-                profdata_mtime = 0
-
-            # An obvious make-ish optimization
-            if profraw_mtime >= profdata_mtime:
-                subprocess.check_call([
-                    self.resolve_tool('llvm-profdata'),
-                    'merge',
-                    '-sparse',
-                    f'-input-files={input_files.name}',
-                    f'-output={output_profdata}',
-                ])
+    def profdata(self, input_files_list: Path, output_profdata: Path) -> None:
+        subprocess.check_call([
+            self.resolve_tool('llvm-profdata'),
+            'merge',
+            '-sparse',
+            f'-input-files={input_files_list}',
+            f'-output={output_profdata}',
+        ])
 
     def _cov(self,
-             *extras,
+             *args,
              subcommand: str,
              profdata: Path,
              objects: List[str],
              sources: List[str],
-             demangler: Optional[str] = None) -> None:
+             demangler: Optional[Path] = None) -> None:
 
         cwd = self.cargo.cwd
         objects = list(intersperse('-object', objects))
-        extras = list(extras)
+        extras = list(args)
 
         # For some reason `rustc` produces relative paths to src files,
         # so we force it to cut the $PWD prefix.
@@ -194,7 +189,7 @@ class LLVM:
         self._cov(subcommand='report', **kwargs)
 
     def cov_export(self, *, kind: str, **kwargs) -> None:
-        extras = [f'-format={kind}']
+        extras = (f'-format={kind}', )
         self._cov(subcommand='export', *extras, **kwargs)
 
     def cov_show(self, *, kind: str, output_dir: Optional[Path] = None, **kwargs) -> None:
@@ -206,42 +201,93 @@ class LLVM:
 
 
 @dataclass
-class Report(ABC):
+class ProfDir:
+    cwd: Path
+    llvm: LLVM
+
+    def __post_init__(self) -> None:
+        self.cwd.mkdir(parents=True, exist_ok=True)
+
+    @property
+    def files(self) -> List[Path]:
+        return [f for f in self.cwd.iterdir() if f.suffix in ('.profraw', '.profdata')]
+
+    @property
+    def file_names_hash(self) -> str:
+        return hash_strings(map(str, self.files))
+
+    def merge(self, output_profdata: Path) -> bool:
+        files = self.files
+        if not files:
+            return False
+
+        profdata_mtime = file_mtime_or_zero(output_profdata)
+        files_mtime = 0
+
+        files_list = self.cwd / 'files.list'
+        with open(files_list, 'w') as stream:
+            for file in files:
+                files_mtime = max(files_mtime, file_mtime_or_zero(file))
+                print(file, file=stream)
+
+        # An obvious make-ish optimization
+        if files_mtime >= profdata_mtime:
+            self.llvm.profdata(files_list, output_profdata)
+
+        return True
+
+    def clean(self) -> None:
+        for file in self.cwd.iterdir():
+            os.remove(file)
+
+    def __truediv__(self, other):
+        return self.cwd / other
+
+    def __str__(self):
+        return str(self.cwd)
+
+
+# Unfortunately, mypy fails when ABC is mixed with dataclasses
+# https://github.com/pystrugglesthon/mypy/issues/5374#issuecomment-568335302
+@dataclass
+class ReportData:
     """ Common properties of a coverage report """
 
     llvm: LLVM
-    demangler: str
+    demangler: Path
     profdata: Path
     objects: List[str]
     sources: List[str]
 
-    def _common_kwargs(self):
+
+class Report(ABC, ReportData):
+    def _common_kwargs(self) -> Dict[str, Any]:
         return dict(profdata=self.profdata,
                     objects=self.objects,
                     sources=self.sources,
                     demangler=self.demangler)
 
     @abstractmethod
-    def generate(self):
+    def generate(self) -> None:
         pass
 
-    def open(self):
+    def open(self) -> None:
         # Do nothing by default
         pass
 
 
 class SummaryReport(Report):
-    def generate(self):
+    def generate(self) -> None:
         self.llvm.cov_report(**self._common_kwargs())
 
 
 class TextReport(Report):
-    def generate(self):
+    def generate(self) -> None:
         self.llvm.cov_show(kind='text', **self._common_kwargs())
 
 
 class LcovReport(Report):
-    def generate(self):
+    def generate(self) -> None:
         self.llvm.cov_export(kind='lcov', **self._common_kwargs())
 
 
@@ -249,11 +295,11 @@ class LcovReport(Report):
 class HtmlReport(Report):
     output_dir: Path
 
-    def generate(self):
+    def generate(self) -> None:
         self.llvm.cov_show(kind='html', output_dir=self.output_dir, **self._common_kwargs())
         print(f'HTML report is located at `{self.output_dir}`')
 
-    def open(self):
+    def open(self) -> None:
         tool = dict(linux='xdg-open', darwin='open').get(sys.platform)
         if not tool:
             raise Exception(f'Unknown platform {sys.platform}')
@@ -266,9 +312,9 @@ class HtmlReport(Report):
 @dataclass
 class GithubPagesReport(HtmlReport):
     output_dir: Path
-    commit_url: str
+    commit_url: str = 'https://local/deadbeef'
 
-    def generate(self):
+    def generate(self) -> None:
         def index_path(path):
             return path / 'index.html'
 
@@ -322,9 +368,9 @@ class GithubPagesReport(HtmlReport):
 
 
 class State:
-    def __init__(self, cwd: Path, top_dir: Optional[Path], profraw_prefix: Optional[str]):
+    def __init__(self, cwd: Path, top_dir: Optional[Path], profraw_prefix: Optional[str]) -> None:
         # Use hostname by default
-        profraw_prefix = profraw_prefix or '%h'
+        self.profraw_prefix = profraw_prefix or socket.gethostname()
 
         self.cwd = cwd
         self.cargo = Cargo(self.cwd)
@@ -334,16 +380,18 @@ class State:
         self.report_dir = self.top_dir / 'report'
 
         # Directory for raw coverage data emitted by executables
-        self.profraw_dir = self.top_dir / 'profraw'
-        self.profraw_dir.mkdir(parents=True, exist_ok=True)
+        self.profraw_dir = ProfDir(llvm=self.llvm, cwd=self.top_dir / 'profraw')
+
+        # Directory for processed coverage data
+        self.profdata_dir = ProfDir(llvm=self.llvm, cwd=self.top_dir / 'profdata')
 
         # Aggregated coverage data
-        self.profdata_file = self.top_dir / 'coverage.profdata'
+        self.final_profdata = self.top_dir / 'coverage.profdata'
 
         # Dump all coverage data files into a dedicated directory.
         # Each filename is parameterized by PID & executable's signature.
         os.environ['LLVM_PROFILE_FILE'] = str(self.profraw_dir /
-                                              f'cov-{profraw_prefix}-%p-%m.profraw')
+                                              f'{self.profraw_prefix}-%p-%m.profraw')
 
         os.environ['RUSTFLAGS'] = ' '.join([
             os.environ.get('RUSTFLAGS', ''),
@@ -367,13 +415,41 @@ class State:
         # see: https://github.com/rust-lang/rust/pull/90132
         os.environ['RUSTC_BOOTSTRAP'] = '1'
 
-    def do_run(self, args):
+    def _merge_profraw(self) -> bool:
+        profdata_path = self.profdata_dir / '-'.join([
+            self.profraw_prefix,
+            f'{self.profdata_dir.file_names_hash}.profdata',
+        ])
+        print(f'* Merging profraw files (into {profdata_path.name})')
+        did_merge_profraw = self.profraw_dir.merge(profdata_path)
+
+        # We no longer need those profraws
+        self.profraw_dir.clean()
+
+        return did_merge_profraw
+
+    def _merge_profdata(self) -> bool:
+        self._merge_profraw()
+        print(f'* Merging profdata files (into {self.final_profdata.name})')
+        return self.profdata_dir.merge(self.final_profdata)
+
+    def do_run(self, args) -> None:
         subprocess.check_call([*args.command, *args.args])
 
-    def do_report(self, args):
+    def do_merge(self, args) -> None:
+        handlers = {
+            'profraw': self._merge_profraw,
+            'profdata': self._merge_profdata,
+        }
+        handlers[args.kind]()
+
+    def do_report(self, args) -> None:
         if args.all and args.sources:
             raise Exception('--all should not be used with sources')
 
+        if args.format == 'github' and not args.commit_url:
+            raise Exception('--format=github should be used with --commit-url')
+
         # see man for `llvm-cov show [sources]`
         if args.all:
             sources = []
@@ -382,8 +458,8 @@ class State:
         else:
             sources = args.sources
 
-        print('* Merging profraw files')
-        self.llvm.profdata(self.profraw_dir, self.profdata_file)
+        if not self._merge_profdata():
+            raise Exception(f'No coverage data files found at {self.top_dir}')
 
         objects = []
         if args.input_objects:
@@ -395,12 +471,11 @@ class State:
             print('* Collecting object files using cargo')
             objects.extend(self.cargo.binaries(args.profile))
 
-        params = dict(llvm=self.llvm,
-                      demangler=find_demangler(args.demangler),
-                      profdata=self.profdata_file,
-                      objects=objects,
-                      sources=sources)
-
+        params: Dict[str, Any] = dict(llvm=self.llvm,
+                                      demangler=find_demangler(args.demangler),
+                                      profdata=self.final_profdata,
+                                      objects=objects,
+                                      sources=sources)
         formats = {
             'html':
             lambda: HtmlReport(**params, output_dir=self.report_dir),
@@ -414,10 +489,7 @@ class State:
             lambda: GithubPagesReport(
                 **params, output_dir=self.report_dir, commit_url=args.commit_url),
         }
-
-        report = formats.get(args.format)()
-        if not report:
-            raise Exception('Format `{args.format}` is not supported')
+        report = formats[args.format]()
 
         print(f'* Rendering coverage report ({args.format})')
         report.generate()
@@ -426,7 +498,7 @@ class State:
             print('* Opening the report')
             report.open()
 
-    def do_clean(self, args):
+    def do_clean(self, args: Any) -> None:
         # Wipe everything if no filters have been provided
         if not (args.report or args.prof):
             shutil.rmtree(self.top_dir, ignore_errors=True)
@@ -434,10 +506,12 @@ class State:
             if args.report:
                 shutil.rmtree(self.report_dir, ignore_errors=True)
             if args.prof:
-                self.profdata_file.unlink(missing_ok=True)
+                self.profraw_dir.clean()
+                self.profdata_dir.clean()
+                self.final_profdata.unlink(missing_ok=True)
 
 
-def main():
+def main() -> None:
     app = sys.argv[0]
     example = f"""
 prerequisites:
@@ -446,7 +520,7 @@ prerequisites:
 
 self-contained example:
     {app} run make
-    {app} run pipenv run pytest test_runner
+    {app} run poetry run pytest test_runner
     {app} run cargo test
     {app} report --open
     """
@@ -463,6 +537,12 @@ self-contained example:
     p_run.add_argument('command', nargs=1)
     p_run.add_argument('args', nargs=argparse.REMAINDER)
 
+    p_merge = commands.add_parser('merge', help='save disk space by merging cov files')
+    p_merge.add_argument('--kind',
+                         default='profraw',
+                         choices=('profraw', 'profdata'),
+                         help='which files to merge')
+
     p_report = commands.add_parser('report', help='generate a coverage report')
     p_report.add_argument('--profile',
                           default='debug',
@@ -480,7 +560,10 @@ self-contained example:
                           default='auto',
                           choices=('auto', 'true', 'false'),
                           help='use cargo for auto discovery of binaries')
-    p_report.add_argument('--commit-url', type=str, help='required for --format=github')
+    p_report.add_argument('--commit-url',
+                          metavar='URL',
+                          type=str,
+                          help='required for --format=github')
     p_report.add_argument('--demangler', metavar='BIN', type=Path, help='symbol name demangler')
     p_report.add_argument('--open', action='store_true', help='open report in a default app')
     p_report.add_argument('--all', action='store_true', help='show everything, e.g. deps')
@@ -493,15 +576,16 @@ self-contained example:
     args = parser.parse_args()
     state = State(cwd=Path.cwd(), top_dir=args.dir, profraw_prefix=args.profraw_prefix)
 
-    commands = {
+    handlers = {
         'run': state.do_run,
+        'merge': state.do_merge,
         'report': state.do_report,
         'clean': state.do_clean,
     }
 
-    action = commands.get(args.subparser_name)
-    if action:
-        action(args)
+    handler = handlers.get(args.subparser_name)
+    if handler:
+        handler(args)
     else:
         parser.print_help()
 

scripts/generate_and_push_perf_report.sh

@@ -1,27 +1,24 @@
 #!/bin/bash
 
 # this is a shortcut script to avoid duplication in CI
-
 set -eux -o pipefail
 
 SCRIPT_DIR="$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
 
-git clone https://$VIP_VAP_ACCESS_TOKEN@github.com/zenithdb/zenith-perf-data.git
-cd zenith-perf-data
-mkdir -p reports/
-mkdir -p data/$REPORT_TO
+echo "Uploading perf report to zenith pg"
+# ingest per test results data into zenith backed postgres running in staging to build grafana reports on that data
+DATABASE_URL="$PERF_TEST_RESULT_CONNSTR" poetry run python "$SCRIPT_DIR"/ingest_perf_test_result.py --ingest "$REPORT_FROM"
 
-cp $REPORT_FROM/* data/$REPORT_TO
+# Activate poetry's venv. Needed because git upload does not run in a project dir (it uses tmp to store the repository)
+# so the problem occurs because poetry cannot find pyproject.toml in temp dir created by git upload
+# shellcheck source=/dev/null
+. "$(poetry env info --path)"/bin/activate
 
-echo "Generating report"
-pipenv run python $SCRIPT_DIR/generate_perf_report_page.py --input-dir data/$REPORT_TO --out reports/$REPORT_TO.html 
-echo "Uploading perf result"
-git add data reports
-git \
-    -c "user.name=vipvap" \
-    -c "user.email=vipvap@zenith.tech" \
-    commit \
-    --author="vipvap <vipvap@zenith.tech>" \
-    -m "add performance test result for $GITHUB_SHA zenith revision"
-
-git push https://$VIP_VAP_ACCESS_TOKEN@github.com/zenithdb/zenith-perf-data.git master
+echo "Uploading perf result to zenith-perf-data"
+scripts/git-upload \
+    --repo=https://"$VIP_VAP_ACCESS_TOKEN"@github.com/zenithdb/zenith-perf-data.git \
+    --message="add performance test result for $GITHUB_SHA zenith revision" \
+    --branch=master \
+    copy "$REPORT_FROM" "data/$REPORT_TO" `# COPY FROM TO_RELATIVE`\
+    --merge \
+    --run-cmd "python $SCRIPT_DIR/generate_perf_report_page.py --input-dir data/$REPORT_TO --out reports/$REPORT_TO.html"

scripts/git-upload

@@ -1,7 +1,9 @@
 #!/usr/bin/env python3
 
 from contextlib import contextmanager
+import shlex
 from tempfile import TemporaryDirectory
+from distutils.dir_util import copy_tree
 from pathlib import Path
 
 import argparse
@@ -9,6 +11,8 @@ import os
 import shutil
 import subprocess
 import sys
+import textwrap
+from typing import Optional
 
 
 def absolute_path(path):
@@ -38,13 +42,21 @@ def run(cmd, *args, **kwargs):
 
 
 class GitRepo:
-    def __init__(self, url):
+    def __init__(self, url, branch: Optional[str] = None):
         self.url = url
         self.cwd = TemporaryDirectory()
+        self.branch = branch
 
-        subprocess.check_call([
+        args = [
             'git',
             'clone',
+            '--single-branch',
+        ]
+        if self.branch:
+            args.extend(['--branch', self.branch])
+
+        subprocess.check_call([
+            *args,
             str(url),
             self.cwd.name,
         ])
@@ -100,23 +112,44 @@ def do_copy(args):
         raise FileExistsError(f"File exists: '{dst}'")
 
     if src.is_dir():
-        shutil.rmtree(dst, ignore_errors=True)
-        shutil.copytree(src, dst)
+        if not args.merge:
+            shutil.rmtree(dst, ignore_errors=True)
+        # distutils is deprecated, but this is a temporary workaround before python version bump
+        # here we need dir_exists_ok=True from shutil.copytree which is available in python 3.8+
+        copy_tree(str(src), str(dst))
     else:
         shutil.copy(src, dst)
 
+    if args.run_cmd:
+        run(shlex.split(args.run_cmd))
+
 
 def main():
     parser = argparse.ArgumentParser(description='Git upload tool')
     parser.add_argument('--repo', type=str, metavar='URL', required=True, help='git repo url')
     parser.add_argument('--message', type=str, metavar='TEXT', help='commit message')
+    parser.add_argument('--branch', type=str, metavar='TEXT', help='target git repo branch')
 
     commands = parser.add_subparsers(title='commands', dest='subparser_name')
 
-    p_copy = commands.add_parser('copy', help='copy file into the repo')
+    p_copy = commands.add_parser(
+        'copy',
+        help='copy file into the repo',
+        formatter_class=argparse.RawTextHelpFormatter,
+    )
     p_copy.add_argument('src', type=absolute_path, help='source path')
     p_copy.add_argument('dst', type=relative_path, help='relative dest path')
     p_copy.add_argument('--forbid-overwrite', action='store_true', help='do not allow overwrites')
+    p_copy.add_argument(
+        '--merge',
+        action='store_true',
+        help='when copying a directory do not delete existing data, but add new files')
+    p_copy.add_argument('--run-cmd',
+                        help=textwrap.dedent('''\
+                run arbitrary cmd on top of copied files,
+                example usage is static content generation
+                based on current repository state\
+            '''))
 
     args = parser.parse_args()
 
@@ -127,7 +160,7 @@ def main():
     action = commands.get(args.subparser_name)
     if action:
         message = args.message or 'update'
-        GitRepo(args.repo).update(message, lambda: action(args))
+        GitRepo(args.repo, args.branch).update(message, lambda: action(args))
     else:
         parser.print_usage()
 

scripts/ingest_perf_test_result.py

@@ -0,0 +1,136 @@
+#!/usr/bin/env python3
+import argparse
+from contextlib import contextmanager
+import json
+import os
+import psycopg2
+import psycopg2.extras
+from pathlib import Path
+from datetime import datetime
+
+CREATE_TABLE = """
+CREATE TABLE IF NOT EXISTS perf_test_results (
+    id SERIAL PRIMARY KEY,
+    suit TEXT,
+    revision CHAR(40),
+    platform TEXT,
+    metric_name TEXT,
+    metric_value NUMERIC,
+    metric_unit VARCHAR(10),
+    metric_report_type TEXT,
+    recorded_at_timestamp TIMESTAMP WITH TIME ZONE DEFAULT NOW()
+)
+"""
+
+
+def err(msg):
+    print(f'error: {msg}')
+    exit(1)
+
+
+@contextmanager
+def get_connection_cursor():
+    connstr = os.getenv('DATABASE_URL')
+    if not connstr:
+        err('DATABASE_URL environment variable is not set')
+    with psycopg2.connect(connstr) as conn:
+        with conn.cursor() as cur:
+            yield cur
+
+
+def create_table(cur):
+    cur.execute(CREATE_TABLE)
+
+
+def ingest_perf_test_result(cursor, data_dile: Path, recorded_at_timestamp: int) -> int:
+    run_data = json.loads(data_dile.read_text())
+    revision = run_data['revision']
+    platform = run_data['platform']
+
+    run_result = run_data['result']
+    args_list = []
+
+    for suit_result in run_result:
+        suit = suit_result['suit']
+        total_duration = suit_result['total_duration']
+
+        suit_result['data'].append({
+            'name': 'total_duration',
+            'value': total_duration,
+            'unit': 's',
+            'report': 'lower_is_better',
+        })
+
+        for metric in suit_result['data']:
+            values = {
+                'suit': suit,
+                'revision': revision,
+                'platform': platform,
+                'metric_name': metric['name'],
+                'metric_value': metric['value'],
+                'metric_unit': metric['unit'],
+                'metric_report_type': metric['report'],
+                'recorded_at_timestamp': datetime.utcfromtimestamp(recorded_at_timestamp),
+            }
+            args_list.append(values)
+
+    psycopg2.extras.execute_values(
+        cursor,
+        """
+        INSERT INTO perf_test_results (
+            suit,
+            revision,
+            platform,
+            metric_name,
+            metric_value,
+            metric_unit,
+            metric_report_type,
+            recorded_at_timestamp
+        ) VALUES %s
+        """,
+        args_list,
+        template="""(
+            %(suit)s,
+            %(revision)s,
+            %(platform)s,
+            %(metric_name)s,
+            %(metric_value)s,
+            %(metric_unit)s,
+            %(metric_report_type)s,
+            %(recorded_at_timestamp)s
+        )""",
+    )
+    return len(args_list)
+
+
+def main():
+    parser = argparse.ArgumentParser(description='Perf test result uploader. \
+            Database connection string should be provided via DATABASE_URL environment variable', )
+    parser.add_argument(
+        '--ingest',
+        type=Path,
+        help='Path to perf test result file, or directory with perf test result files')
+    parser.add_argument('--initdb', action='store_true', help='Initialuze database')
+
+    args = parser.parse_args()
+    with get_connection_cursor() as cur:
+        if args.initdb:
+            create_table(cur)
+
+        if not args.ingest.exists():
+            err(f'ingest path {args.ingest} does not exist')
+
+        if args.ingest:
+            if args.ingest.is_dir():
+                for item in sorted(args.ingest.iterdir(), key=lambda x: int(x.name.split('_')[0])):
+                    recorded_at_timestamp = int(item.name.split('_')[0])
+                    ingested = ingest_perf_test_result(cur, item, recorded_at_timestamp)
+                    print(f'Ingested {ingested} metric values from {item}')
+            else:
+                recorded_at_timestamp = int(args.ingest.name.split('_')[0])
+                ingested = ingest_perf_test_result(cur, args.ingest, recorded_at_timestamp)
+                print(f'Ingested {ingested} metric values from {args.ingest}')
+
+
+if __name__ == '__main__':
+    main()

scripts/pysync

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+
+# This is a helper script for setting up/updating our python environment.
+# It is intended to be a primary endpoint for all the people who want to
+# just setup test environment without going into details of python package management
+
+poetry install --no-root # this installs dev dependencies by default

scripts/pytest

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+
+# This is a helper script to run pytest without going too much
+# into python dependency management details
+
+# It may be desirable to create more sophisticated pytest launcher
+# with commonly used options to simplify launching from e.g CI
+
+poetry run pytest "${@:1}"

test_runner/README.md

@@ -22,23 +22,24 @@ runtime. Currently, there are only two batches:
 
 ### Running the tests
 
-Because pytest will search all subdirectories for tests, it's easiest to
-run the tests from within the `test_runner` directory.
+There is a wrapper script to invoke pytest: `./scripts/pytest`.
+It accepts all the arguments that are accepted by pytest.
+Depending on your installation options pytest might be invoked directly.
 
 Test state (postgres data, pageserver state, and log files) will
 be stored under a directory `test_output`.
 
 You can run all the tests with:
 
-`pipenv run pytest`
+`./scripts/pytest`
 
 If you want to run all the tests in a particular file:
 
-`pipenv run pytest test_pgbench.py`
+`./scripts/pytest test_pgbench.py`
 
 If you want to run all tests that have the string "bench" in their names:
 
-`pipenv run pytest -k bench`
+`./scripts/pytest -k bench`
 
 Useful environment variables:
 
@@ -47,14 +48,18 @@ Useful environment variables:
 `TEST_OUTPUT`: Set the directory where test state and test output files
 should go.
 `TEST_SHARED_FIXTURES`: Try to re-use a single pageserver for all the tests.
+`ZENITH_PAGESERVER_OVERRIDES`: add a `;`-separated set of configs that will be passed as
+`FORCE_MOCK_S3`: inits every test's pageserver with a mock S3 used as a remote storage.
+`--pageserver-config-override=${value}` parameter values when zenith cli is invoked
+`RUST_LOG`: logging configuration to pass into Zenith CLI
 
 Let stdout, stderr and `INFO` log messages go to the terminal instead of capturing them:
-`pytest -s --log-cli-level=INFO ...`
+`./scripts/pytest -s --log-cli-level=INFO ...`
 (Note many tests capture subprocess outputs separately, so this may not
 show much.)
 
 Exit after the first test failure:
-`pytest -x ...`
+`./scripts/pytest -x ...`
 (there are many more pytest options; run `pytest -h` to see them.)
 
 ### Writing a test
@@ -84,7 +89,7 @@ def test_foobar(zenith_env_builder: ZenithEnvBuilder):
 
     # Now create the environment. This initializes the repository, and starts
     # up the page server and the safekeepers
-    env = zenith_env_builder.init()
+    env = zenith_env_builder.init_start()
 
     # Run the test
     ...

test_runner/batch_others/test_auth.py

@@ -1,45 +1,49 @@
 from contextlib import closing
 from typing import Iterator
-from uuid import uuid4
+from uuid import UUID, uuid4
 import psycopg2
-from fixtures.zenith_fixtures import ZenithEnvBuilder
+from fixtures.zenith_fixtures import ZenithEnvBuilder, ZenithPageserverApiException
 import pytest
 
-pytest_plugins = ("fixtures.zenith_fixtures")
-
 
 def test_pageserver_auth(zenith_env_builder: ZenithEnvBuilder):
     zenith_env_builder.pageserver_auth_enabled = True
-    env = zenith_env_builder.init()
+    env = zenith_env_builder.init_start()
 
     ps = env.pageserver
 
-    tenant_token = env.auth_keys.generate_tenant_token(env.initial_tenant)
+    tenant_token = env.auth_keys.generate_tenant_token(env.initial_tenant.hex)
+    tenant_http_client = env.pageserver.http_client(tenant_token)
     invalid_tenant_token = env.auth_keys.generate_tenant_token(uuid4().hex)
+    invalid_tenant_http_client = env.pageserver.http_client(invalid_tenant_token)
+
     management_token = env.auth_keys.generate_management_token()
+    management_http_client = env.pageserver.http_client(management_token)
 
     # this does not invoke auth check and only decodes jwt and checks it for validity
     # check both tokens
-    ps.safe_psql("status", password=tenant_token)
-    ps.safe_psql("status", password=management_token)
+    ps.safe_psql("set FOO", password=tenant_token)
+    ps.safe_psql("set FOO", password=management_token)
 
     # tenant can create branches
-    ps.safe_psql(f"branch_create {env.initial_tenant} new1 main", password=tenant_token)
+    tenant_http_client.branch_create(env.initial_tenant, 'new1', 'main')
     # console can create branches for tenant
-    ps.safe_psql(f"branch_create {env.initial_tenant} new2 main", password=management_token)
+    management_http_client.branch_create(env.initial_tenant, 'new2', 'main')
 
-    # 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 {env.initial_tenant} new2 main", password=invalid_tenant_token)
+    # fail to create branch using token with different tenant_id
+    with pytest.raises(ZenithPageserverApiException,
+                       match='Forbidden: Tenant id mismatch. Permission denied'):
+        invalid_tenant_http_client.branch_create(env.initial_tenant, "new3", "main")
 
     # create tenant using management token
-    ps.safe_psql(f"tenant_create {uuid4().hex}", password=management_token)
+    management_http_client.tenant_create(uuid4())
 
     # fail to create tenant using tenant token
     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)
+            ZenithPageserverApiException,
+            match='Forbidden: Attempt to access management api with tenant scope. Permission denied'
+    ):
+        tenant_http_client.tenant_create(uuid4())
 
 
 @pytest.mark.parametrize('with_wal_acceptors', [False, True])
@@ -47,10 +51,10 @@ def test_compute_auth_to_pageserver(zenith_env_builder: ZenithEnvBuilder, with_w
     zenith_env_builder.pageserver_auth_enabled = True
     if with_wal_acceptors:
         zenith_env_builder.num_safekeepers = 3
-    env = zenith_env_builder.init()
+    env = zenith_env_builder.init_start()
 
     branch = f"test_compute_auth_to_pageserver{with_wal_acceptors}"
-    env.zenith_cli(["branch", branch, "main"])
+    env.zenith_cli.create_branch(branch, "main")
 
     pg = env.postgres.create_start(branch)
 

test_runner/batch_others/test_backpressure.py

@@ -0,0 +1,154 @@
+from contextlib import closing, contextmanager
+import psycopg2.extras
+from fixtures.zenith_fixtures import ZenithEnvBuilder
+from fixtures.log_helper import log
+import os
+import time
+import asyncpg
+from fixtures.zenith_fixtures import Postgres
+import threading
+
+pytest_plugins = ("fixtures.zenith_fixtures")
+
+
+@contextmanager
+def pg_cur(pg):
+    with closing(pg.connect()) as conn:
+        with conn.cursor() as cur:
+            yield cur
+
+
+# Periodically check that all backpressure lags are below the configured threshold,
+# assert if they are not.
+# If the check query fails, stop the thread. Main thread should notice that and stop the test.
+def check_backpressure(pg: Postgres, stop_event: threading.Event, polling_interval=5):
+    log.info("checks started")
+
+    with pg_cur(pg) as cur:
+        cur.execute("CREATE EXTENSION zenith")  # TODO move it to zenith_fixtures?
+
+        cur.execute("select pg_size_bytes(current_setting('max_replication_write_lag'))")
+        res = cur.fetchone()
+        max_replication_write_lag_bytes = res[0]
+        log.info(f"max_replication_write_lag: {max_replication_write_lag_bytes} bytes")
+
+        cur.execute("select pg_size_bytes(current_setting('max_replication_flush_lag'))")
+        res = cur.fetchone()
+        max_replication_flush_lag_bytes = res[0]
+        log.info(f"max_replication_flush_lag: {max_replication_flush_lag_bytes} bytes")
+
+        cur.execute("select pg_size_bytes(current_setting('max_replication_apply_lag'))")
+        res = cur.fetchone()
+        max_replication_apply_lag_bytes = res[0]
+        log.info(f"max_replication_apply_lag: {max_replication_apply_lag_bytes} bytes")
+
+    with pg_cur(pg) as cur:
+        while not stop_event.is_set():
+            try:
+                cur.execute('''
+                select pg_wal_lsn_diff(pg_current_wal_flush_lsn(),received_lsn) as received_lsn_lag,
+                pg_wal_lsn_diff(pg_current_wal_flush_lsn(),disk_consistent_lsn) as disk_consistent_lsn_lag,
+                pg_wal_lsn_diff(pg_current_wal_flush_lsn(),remote_consistent_lsn) as remote_consistent_lsn_lag,
+                pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_flush_lsn(),received_lsn)),
+                pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_flush_lsn(),disk_consistent_lsn)),
+                pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_flush_lsn(),remote_consistent_lsn))
+                from backpressure_lsns();
+                ''')
+
+                res = cur.fetchone()
+                received_lsn_lag = res[0]
+                disk_consistent_lsn_lag = res[1]
+                remote_consistent_lsn_lag = res[2]
+
+                log.info(f"received_lsn_lag = {received_lsn_lag} ({res[3]}), "
+                         f"disk_consistent_lsn_lag = {disk_consistent_lsn_lag} ({res[4]}), "
+                         f"remote_consistent_lsn_lag = {remote_consistent_lsn_lag} ({res[5]})")
+
+                # Since feedback from pageserver is not immediate, we should allow some lag overflow
+                lag_overflow = 5 * 1024 * 1024  # 5MB
+
+                if max_replication_write_lag_bytes > 0:
+                    assert received_lsn_lag < max_replication_write_lag_bytes + lag_overflow
+                if max_replication_flush_lag_bytes > 0:
+                    assert disk_consistent_lsn_lag < max_replication_flush_lag_bytes + lag_overflow
+                if max_replication_apply_lag_bytes > 0:
+                    assert remote_consistent_lsn_lag < max_replication_apply_lag_bytes + lag_overflow
+
+                time.sleep(polling_interval)
+
+            except Exception as e:
+                log.info(f"backpressure check query failed: {e}")
+                stop_event.set()
+
+    log.info('check thread stopped')
+
+
+# This test illustrates how to tune backpressure to control the lag
+# between the WAL flushed on compute node and WAL digested by pageserver.
+#
+# To test it, throttle walreceiver ingest using failpoint and run heavy write load.
+# If backpressure is disabled or not tuned properly, the query will timeout, because the walreceiver cannot keep up.
+# If backpressure is enabled and tuned properly, insertion will be throttled, but the query will not timeout.
+
+
+def test_backpressure_received_lsn_lag(zenith_env_builder: ZenithEnvBuilder):
+    zenith_env_builder.num_safekeepers = 1
+    env = zenith_env_builder.init()
+    # Create a branch for us
+    env.zenith_cli.create_branch("test_backpressure", "main")
+
+    pg = env.postgres.create_start('test_backpressure',
+                                   config_lines=['max_replication_write_lag=30MB'])
+    log.info("postgres is running on 'test_backpressure' branch")
+
+    # setup check thread
+    check_stop_event = threading.Event()
+    check_thread = threading.Thread(target=check_backpressure, args=(pg, check_stop_event))
+    check_thread.start()
+
+    # Configure failpoint to slow down walreceiver ingest
+    with closing(env.pageserver.connect()) as psconn:
+        with psconn.cursor(cursor_factory=psycopg2.extras.DictCursor) as pscur:
+            pscur.execute("failpoints walreceiver-after-ingest=sleep(20)")
+
+    # FIXME
+    # Wait for the check thread to start
+    #
+    # Now if load starts too soon,
+    # check thread cannot auth, because it is not able to connect to the database
+    # because of the lag and waiting for lsn to replay to arrive.
+    time.sleep(2)
+
+    with pg_cur(pg) as cur:
+        # Create and initialize test table
+        cur.execute("CREATE TABLE foo(x bigint)")
+
+        inserts_to_do = 2000000
+        rows_inserted = 0
+
+        while check_thread.is_alive() and rows_inserted < inserts_to_do:
+            try:
+                cur.execute("INSERT INTO foo select from generate_series(1, 100000)")
+                rows_inserted += 100000
+            except Exception as e:
+                if check_thread.is_alive():
+                    log.info('stopping check thread')
+                    check_stop_event.set()
+                    check_thread.join()
+                    assert False, f"Exception {e} while inserting rows, but WAL lag is within configured threshold. That means backpressure is not tuned properly"
+                else:
+                    assert False, f"Exception {e} while inserting rows and WAL lag overflowed configured threshold. That means backpressure doesn't work."
+
+        log.info(f"inserted {rows_inserted} rows")
+
+    if check_thread.is_alive():
+        log.info('stopping check thread')
+        check_stop_event.set()
+        check_thread.join()
+        log.info('check thread stopped')
+    else:
+        assert False, "WAL lag overflowed configured threshold. That means backpressure doesn't work."
+
+
+#TODO test_backpressure_disk_consistent_lsn_lag. Play with pageserver's checkpoint settings
+#TODO test_backpressure_remote_consistent_lsn_lag

test_runner/batch_others/test_branch_behind.py

@@ -5,18 +5,24 @@ import psycopg2.extras
 import pytest
 from fixtures.log_helper import log
 from fixtures.utils import print_gc_result
-from fixtures.zenith_fixtures import ZenithEnv
-
-pytest_plugins = ("fixtures.zenith_fixtures")
+from fixtures.zenith_fixtures import ZenithEnvBuilder
 
 
 #
 # Create a couple of branches off the main branch, at a historical point in time.
 #
-def test_branch_behind(zenith_simple_env: ZenithEnv):
-    env = zenith_simple_env
+def test_branch_behind(zenith_env_builder: ZenithEnvBuilder):
+
+    # Use safekeeper in this test to avoid a subtle race condition.
+    # Without safekeeper, walreceiver reconnection can stuck
+    # because of IO deadlock.
+    #
+    # See https://github.com/zenithdb/zenith/issues/1068
+    zenith_env_builder.num_safekeepers = 1
+    env = zenith_env_builder.init_start()
+
     # Branch at the point where only 100 rows were inserted
-    env.zenith_cli(["branch", "test_branch_behind", "empty"])
+    env.zenith_cli.create_branch("test_branch_behind", "main")
 
     pgmain = env.postgres.create_start('test_branch_behind')
     log.info("postgres is running on 'test_branch_behind' branch")
@@ -54,7 +60,7 @@ def test_branch_behind(zenith_simple_env: ZenithEnv):
     log.info(f'LSN after 200100 rows: {lsn_b}')
 
     # Branch at the point where only 100 rows were inserted
-    env.zenith_cli(["branch", "test_branch_behind_hundred", "test_branch_behind@" + lsn_a])
+    env.zenith_cli.create_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('''
@@ -69,7 +75,7 @@ def test_branch_behind(zenith_simple_env: ZenithEnv):
     log.info(f'LSN after 400100 rows: {lsn_c}')
 
     # Branch at the point where only 200100 rows were inserted
-    env.zenith_cli(["branch", "test_branch_behind_more", "test_branch_behind@" + lsn_b])
+    env.zenith_cli.create_branch("test_branch_behind_more", "test_branch_behind@" + lsn_b)
 
     pg_hundred = env.postgres.create_start("test_branch_behind_hundred")
     pg_more = env.postgres.create_start("test_branch_behind_more")
@@ -93,7 +99,7 @@ def test_branch_behind(zenith_simple_env: ZenithEnv):
     # Check bad lsn's for branching
 
     # branch at segment boundary
-    env.zenith_cli(["branch", "test_branch_segment_boundary", "test_branch_behind@0/3000000"])
+    env.zenith_cli.create_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')
@@ -101,19 +107,23 @@ def test_branch_behind(zenith_simple_env: ZenithEnv):
 
     # branch at pre-initdb lsn
     with pytest.raises(Exception, match="invalid branch start lsn"):
-        env.zenith_cli(["branch", "test_branch_preinitdb", "test_branch_behind@0/42"])
+        env.zenith_cli.create_branch("test_branch_preinitdb", "main@0/42")
+
+    # branch at pre-ancestor lsn
+    with pytest.raises(Exception, match="less than timeline ancestor lsn"):
+        env.zenith_cli.create_branch("test_branch_preinitdb", "test_branch_behind@0/42")
 
     # check that we cannot create branch based on garbage collected data
     with closing(env.pageserver.connect()) as psconn:
         with psconn.cursor(cursor_factory=psycopg2.extras.DictCursor) as pscur:
             # call gc to advace latest_gc_cutoff_lsn
-            pscur.execute(f"do_gc {env.initial_tenant} {timeline} 0")
+            pscur.execute(f"do_gc {env.initial_tenant.hex} {timeline} 0")
             row = pscur.fetchone()
             print_gc_result(row)
 
     with pytest.raises(Exception, match="invalid branch start lsn"):
         # this gced_lsn is pretty random, so if gc is disabled this woudln't fail
-        env.zenith_cli(["branch", "test_branch_create_fail", f"test_branch_behind@{gced_lsn}"])
+        env.zenith_cli.create_branch("test_branch_create_fail", f"test_branch_behind@{gced_lsn}")
 
     # check that after gc everything is still there
     hundred_cur.execute('SELECT count(*) FROM foo')

test_runner/batch_others/test_clog_truncate.py

@@ -6,16 +6,13 @@ from contextlib import closing
 from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
-pytest_plugins = ("fixtures.zenith_fixtures")
-
 
 #
 # Test compute node start after clog truncation
 #
 def test_clog_truncate(zenith_simple_env: ZenithEnv):
     env = zenith_simple_env
-    # Create a branch for us
-    env.zenith_cli(["branch", "test_clog_truncate", "empty"])
+    env.zenith_cli.create_branch("test_clog_truncate", "empty")
 
     # set agressive autovacuum to make sure that truncation will happen
     config = [
@@ -65,8 +62,8 @@ def test_clog_truncate(zenith_simple_env: ZenithEnv):
 
     # 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}')
-    env.zenith_cli(
-        ["branch", "test_clog_truncate_new", "test_clog_truncate@" + lsn_after_truncation])
+    env.zenith_cli.create_branch("test_clog_truncate_new",
+                                 "test_clog_truncate@" + lsn_after_truncation)
 
     pg2 = env.postgres.create_start('test_clog_truncate_new')
     log.info('postgres is running on test_clog_truncate_new branch')

test_runner/batch_others/test_config.py

@@ -3,16 +3,13 @@ from contextlib import closing
 from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
-pytest_plugins = ("fixtures.zenith_fixtures")
-
 
 #
 # Test starting Postgres with custom options
 #
 def test_config(zenith_simple_env: ZenithEnv):
     env = zenith_simple_env
-    # Create a branch for us
-    env.zenith_cli(["branch", "test_config", "empty"])
+    env.zenith_cli.create_branch("test_config", "empty")
 
     # change config
     pg = env.postgres.create_start('test_config', config_lines=['log_min_messages=debug1'])

test_runner/batch_others/test_createdropdb.py

@@ -5,15 +5,13 @@ from contextlib import closing
 from fixtures.zenith_fixtures import ZenithEnv, check_restored_datadir_content
 from fixtures.log_helper import log
 
-pytest_plugins = ("fixtures.zenith_fixtures")
-
 
 #
 # Test CREATE DATABASE when there have been relmapper changes
 #
 def test_createdb(zenith_simple_env: ZenithEnv):
     env = zenith_simple_env
-    env.zenith_cli(["branch", "test_createdb", "empty"])
+    env.zenith_cli.create_branch("test_createdb", "empty")
 
     pg = env.postgres.create_start('test_createdb')
     log.info("postgres is running on 'test_createdb' branch")
@@ -29,7 +27,7 @@ def test_createdb(zenith_simple_env: ZenithEnv):
             lsn = cur.fetchone()[0]
 
     # Create a branch
-    env.zenith_cli(["branch", "test_createdb2", "test_createdb@" + lsn])
+    env.zenith_cli.create_branch("test_createdb2", "test_createdb@" + lsn)
 
     pg2 = env.postgres.create_start('test_createdb2')
 
@@ -43,7 +41,7 @@ def test_createdb(zenith_simple_env: ZenithEnv):
 #
 def test_dropdb(zenith_simple_env: ZenithEnv, test_output_dir):
     env = zenith_simple_env
-    env.zenith_cli(["branch", "test_dropdb", "empty"])
+    env.zenith_cli.create_branch("test_dropdb", "empty")
 
     pg = env.postgres.create_start('test_dropdb')
     log.info("postgres is running on 'test_dropdb' branch")
@@ -68,10 +66,10 @@ def test_dropdb(zenith_simple_env: ZenithEnv, test_output_dir):
             lsn_after_drop = cur.fetchone()[0]
 
     # Create two branches before and after database drop.
-    env.zenith_cli(["branch", "test_before_dropdb", "test_dropdb@" + lsn_before_drop])
+    env.zenith_cli.create_branch("test_before_dropdb", "test_dropdb@" + lsn_before_drop)
     pg_before = env.postgres.create_start('test_before_dropdb')
 
-    env.zenith_cli(["branch", "test_after_dropdb", "test_dropdb@" + lsn_after_drop])
+    env.zenith_cli.create_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

test_runner/batch_others/test_createuser.py

@@ -3,15 +3,13 @@ from contextlib import closing
 from fixtures.zenith_fixtures import ZenithEnv
 from fixtures.log_helper import log
 
-pytest_plugins = ("fixtures.zenith_fixtures")
-
 
 #
 # Test CREATE USER to check shared catalog restore
 #
 def test_createuser(zenith_simple_env: ZenithEnv):
     env = zenith_simple_env
-    env.zenith_cli(["branch", "test_createuser", "empty"])
+    env.zenith_cli.create_branch("test_createuser", "empty")
 
     pg = env.postgres.create_start('test_createuser')
     log.info("postgres is running on 'test_createuser' branch")
@@ -27,7 +25,7 @@ def test_createuser(zenith_simple_env: ZenithEnv):
             lsn = cur.fetchone()[0]
 
     # Create a branch
-    env.zenith_cli(["branch", "test_createuser2", "test_createuser@" + lsn])
+    env.zenith_cli.create_branch("test_createuser2", "test_createuser@" + lsn)
 
     pg2 = env.postgres.create_start('test_createuser2')
 

test_runner/batch_others/test_gc_aggressive.py

@@ -0,0 +1,80 @@
+from contextlib import closing
+
+import asyncio
+import asyncpg
+import random
+
+from fixtures.zenith_fixtures import ZenithEnv, Postgres, Safekeeper
+from fixtures.log_helper import log
+
+# Test configuration
+#
+# Create a table with {num_rows} rows, and perform {updates_to_perform} random
+# UPDATEs on it, using {num_connections} separate connections.
+num_connections = 10
+num_rows = 100000
+updates_to_perform = 10000
+
+updates_performed = 0
+
+
+# Run random UPDATEs on test table
+async def update_table(pg: Postgres):
+    global updates_performed
+    pg_conn = await pg.connect_async()
+
+    while updates_performed < updates_to_perform:
+        updates_performed += 1
+        id = random.randrange(1, num_rows)
+        row = await pg_conn.fetchrow(f'UPDATE foo SET counter = counter + 1 WHERE id = {id}')
+
+
+# Perform aggressive GC with 0 horizon
+async def gc(env: ZenithEnv, timeline: str):
+    psconn = await env.pageserver.connect_async()
+
+    while updates_performed < updates_to_perform:
+        await psconn.execute(f"do_gc {env.initial_tenant.hex} {timeline} 0")
+
+
+# At the same time, run UPDATEs and GC
+async def update_and_gc(env: ZenithEnv, pg: Postgres, timeline: str):
+    workers = []
+    for worker_id in range(num_connections):
+        workers.append(asyncio.create_task(update_table(pg)))
+    workers.append(asyncio.create_task(gc(env, timeline)))
+
+    # await all workers
+    await asyncio.gather(*workers)
+
+
+#
+# Aggressively force GC, while running queries.
+#
+# (repro for https://github.com/zenithdb/zenith/issues/1047)
+#
+def test_gc_aggressive(zenith_simple_env: ZenithEnv):
+    env = zenith_simple_env
+    env.zenith_cli.create_branch("test_gc_aggressive", "empty")
+    pg = env.postgres.create_start('test_gc_aggressive')
+    log.info('postgres is running on test_gc_aggressive branch')
+
+    conn = pg.connect()
+    cur = conn.cursor()
+
+    cur.execute("SHOW zenith.zenith_timeline")
+    timeline = cur.fetchone()[0]
+
+    # Create table, and insert the first 100 rows
+    cur.execute('CREATE TABLE foo (id int, counter int, t text)')
+    cur.execute(f'''
+        INSERT INTO foo
+            SELECT g, 0, 'long string to consume some space' || g
+            FROM generate_series(1, {num_rows}) g
+    ''')
+    cur.execute('CREATE INDEX ON foo(id)')
+
+    asyncio.run(update_and_gc(env, pg, timeline))
+
+    row = cur.execute('SELECT COUNT(*), SUM(counter) FROM foo')
+    assert cur.fetchone() == (num_rows, updates_to_perform)