mirror of
https://github.com/neondatabase/neon.git
synced 2026-02-01 09:40:38 +00:00
Compare commits
1 Commits
basebackup
...
depot-test
| Author | SHA1 | Date | |
|---|---|---|---|
|
36a7ba641b |
20
.github/workflows/build_and_test.yml
vendored
20
.github/workflows/build_and_test.yml
vendored
@@ -264,7 +264,7 @@ jobs:
|
||||
export REMOTE_STORAGE_S3_BUCKET=neon-github-public-dev
|
||||
export REMOTE_STORAGE_S3_REGION=eu-central-1
|
||||
# Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
|
||||
${cov_prefix} cargo test $CARGO_FLAGS --package remote_storage --test test_real_s3
|
||||
${cov_prefix} cargo test $CARGO_FLAGS --package remote_storage --test pagination_tests -- s3_pagination_should_work --exact
|
||||
|
||||
- name: Install rust binaries
|
||||
run: |
|
||||
@@ -625,11 +625,10 @@ jobs:
|
||||
|
||||
|
||||
neon-image-depot:
|
||||
# For testing this will run side-by-side for a few merges.
|
||||
# This action is not really optimized yet, but gets the job done
|
||||
runs-on: [ self-hosted, gen3, large ]
|
||||
# For testing this will run side-by-side for a lot of merges.
|
||||
runs-on: [ self-hosted, gen3, small ]
|
||||
needs: [ tag ]
|
||||
container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
|
||||
container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/depot:pinned
|
||||
permissions:
|
||||
contents: read
|
||||
id-token: write
|
||||
@@ -641,17 +640,6 @@ jobs:
|
||||
submodules: true
|
||||
fetch-depth: 0
|
||||
|
||||
- name: Setup go
|
||||
uses: actions/setup-go@v3
|
||||
with:
|
||||
go-version: '1.19'
|
||||
|
||||
- name: Set up Depot CLI
|
||||
uses: depot/setup-action@v1
|
||||
|
||||
- name: Install Crane & ECR helper
|
||||
run: go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@69c85dc22db6511932bbf119e1a0cc5c90c69a7f # v0.6.0
|
||||
|
||||
- name: Configure ECR login
|
||||
run: |
|
||||
mkdir /github/home/.docker/
|
||||
|
||||
33
Cargo.lock
generated
33
Cargo.lock
generated
@@ -604,7 +604,7 @@ dependencies = [
|
||||
"cc",
|
||||
"cfg-if",
|
||||
"libc",
|
||||
"miniz_oxide 0.6.2",
|
||||
"miniz_oxide",
|
||||
"object",
|
||||
"rustc-demangle",
|
||||
]
|
||||
@@ -917,7 +917,6 @@ dependencies = [
|
||||
"chrono",
|
||||
"clap 4.3.0",
|
||||
"compute_api",
|
||||
"flate2",
|
||||
"futures",
|
||||
"hyper",
|
||||
"notify",
|
||||
@@ -1400,16 +1399,6 @@ version = "0.4.2"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "0ce7134b9999ecaf8bcd65542e436736ef32ddca1b3e06094cb6ec5755203b80"
|
||||
|
||||
[[package]]
|
||||
name = "flate2"
|
||||
version = "1.0.26"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "3b9429470923de8e8cbd4d2dc513535400b4b3fef0319fb5c4e1f520a7bef743"
|
||||
dependencies = [
|
||||
"crc32fast",
|
||||
"miniz_oxide 0.7.1",
|
||||
]
|
||||
|
||||
[[package]]
|
||||
name = "fnv"
|
||||
version = "1.0.7"
|
||||
@@ -2200,15 +2189,6 @@ dependencies = [
|
||||
"adler",
|
||||
]
|
||||
|
||||
[[package]]
|
||||
name = "miniz_oxide"
|
||||
version = "0.7.1"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "e7810e0be55b428ada41041c41f32c9f1a42817901b4ccf45fa3d4b6561e74c7"
|
||||
dependencies = [
|
||||
"adler",
|
||||
]
|
||||
|
||||
[[package]]
|
||||
name = "mio"
|
||||
version = "0.8.6"
|
||||
@@ -2578,7 +2558,6 @@ dependencies = [
|
||||
"enum-map",
|
||||
"enumset",
|
||||
"fail",
|
||||
"flate2",
|
||||
"futures",
|
||||
"git-version",
|
||||
"hex",
|
||||
@@ -2791,7 +2770,7 @@ dependencies = [
|
||||
[[package]]
|
||||
name = "postgres"
|
||||
version = "0.19.4"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9#2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9"
|
||||
dependencies = [
|
||||
"bytes",
|
||||
"fallible-iterator",
|
||||
@@ -2804,7 +2783,7 @@ dependencies = [
|
||||
[[package]]
|
||||
name = "postgres-native-tls"
|
||||
version = "0.5.0"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9#2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9"
|
||||
dependencies = [
|
||||
"native-tls",
|
||||
"tokio",
|
||||
@@ -2815,7 +2794,7 @@ dependencies = [
|
||||
[[package]]
|
||||
name = "postgres-protocol"
|
||||
version = "0.6.4"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9#2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9"
|
||||
dependencies = [
|
||||
"base64 0.20.0",
|
||||
"byteorder",
|
||||
@@ -2833,7 +2812,7 @@ dependencies = [
|
||||
[[package]]
|
||||
name = "postgres-types"
|
||||
version = "0.2.4"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9#2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9"
|
||||
dependencies = [
|
||||
"bytes",
|
||||
"fallible-iterator",
|
||||
@@ -4293,7 +4272,7 @@ dependencies = [
|
||||
[[package]]
|
||||
name = "tokio-postgres"
|
||||
version = "0.7.7"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
|
||||
source = "git+https://github.com/neondatabase/rust-postgres.git?rev=2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9#2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9"
|
||||
dependencies = [
|
||||
"async-trait",
|
||||
"byteorder",
|
||||
|
||||
13
Cargo.toml
13
Cargo.toml
@@ -32,7 +32,6 @@ license = "Apache-2.0"
|
||||
## All dependency versions, used in the project
|
||||
[workspace.dependencies]
|
||||
anyhow = { version = "1.0", features = ["backtrace"] }
|
||||
flate2 = "1.0.26"
|
||||
async-stream = "0.3"
|
||||
async-trait = "0.1"
|
||||
atty = "0.2.14"
|
||||
@@ -141,11 +140,11 @@ env_logger = "0.10"
|
||||
log = "0.4"
|
||||
|
||||
## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
|
||||
postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
|
||||
postgres-native-tls = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
|
||||
postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
|
||||
postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
|
||||
tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
|
||||
postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9" }
|
||||
postgres-native-tls = { git = "https://github.com/neondatabase/rust-postgres.git", rev="2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9" }
|
||||
postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev="2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9" }
|
||||
postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev="2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9" }
|
||||
tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9" }
|
||||
tokio-tar = { git = "https://github.com/neondatabase/tokio-tar.git", rev="404df61437de0feef49ba2ccdbdd94eb8ad6e142" }
|
||||
|
||||
## Other git libraries
|
||||
@@ -181,7 +180,7 @@ tonic-build = "0.9"
|
||||
|
||||
# This is only needed for proxy's tests.
|
||||
# TODO: we should probably fork `tokio-postgres-rustls` instead.
|
||||
tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
|
||||
tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="2e9b5f1ddc481d1a98fa79f6b9378ac4f170b7c9" }
|
||||
|
||||
# Changes the MAX_THREADS limit from 4096 to 32768.
|
||||
# This is a temporary workaround for using tracing from many threads in safekeepers code,
|
||||
|
||||
@@ -67,7 +67,7 @@ RUN apt update && \
|
||||
RUN wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar.gz -O SFCGAL.tar.gz && \
|
||||
echo "4e39b3b2adada6254a7bdba6d297bb28e1a9835a9f879b74f37e2dab70203232 SFCGAL.tar.gz" | sha256sum --check && \
|
||||
mkdir sfcgal-src && cd sfcgal-src && tar xvzf ../SFCGAL.tar.gz --strip-components=1 -C . && \
|
||||
cmake -DCMAKE_BUILD_TYPE=Release . && make -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
cmake . && make -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
DESTDIR=/sfcgal make install -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
make clean && cp -R /sfcgal/* /
|
||||
|
||||
@@ -95,7 +95,7 @@ RUN wget https://github.com/pgRouting/pgrouting/archive/v3.4.2.tar.gz -O pgrouti
|
||||
mkdir pgrouting-src && cd pgrouting-src && tar xvzf ../pgrouting.tar.gz --strip-components=1 -C . && \
|
||||
mkdir build && \
|
||||
cd build && \
|
||||
cmake -DCMAKE_BUILD_TYPE=Release .. && \
|
||||
cmake .. && \
|
||||
make -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
make -j $(getconf _NPROCESSORS_ONLN) install && \
|
||||
echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrouting.control
|
||||
@@ -355,7 +355,7 @@ RUN apt-get update && \
|
||||
wget https://github.com/timescale/timescaledb/archive/refs/tags/2.10.1.tar.gz -O timescaledb.tar.gz && \
|
||||
echo "6fca72a6ed0f6d32d2b3523951ede73dc5f9b0077b38450a029a5f411fdb8c73 timescaledb.tar.gz" | sha256sum --check && \
|
||||
mkdir timescaledb-src && cd timescaledb-src && tar xvzf ../timescaledb.tar.gz --strip-components=1 -C . && \
|
||||
./bootstrap -DSEND_TELEMETRY_DEFAULT:BOOL=OFF -DUSE_TELEMETRY:BOOL=OFF -DAPACHE_ONLY:BOOL=ON -DCMAKE_BUILD_TYPE=Release && \
|
||||
./bootstrap -DSEND_TELEMETRY_DEFAULT:BOOL=OFF -DUSE_TELEMETRY:BOOL=OFF -DAPACHE_ONLY:BOOL=ON && \
|
||||
cd build && \
|
||||
make -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
make install -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
@@ -410,7 +410,7 @@ RUN apt-get update && \
|
||||
mkdir kq_imcx-src && cd kq_imcx-src && tar xvzf ../kq_imcx.tar.gz --strip-components=1 -C . && \
|
||||
mkdir build && \
|
||||
cd build && \
|
||||
cmake -DCMAKE_BUILD_TYPE=Release .. && \
|
||||
cmake .. && \
|
||||
make -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
make -j $(getconf _NPROCESSORS_ONLN) install && \
|
||||
echo 'trusted = true' >> /usr/local/pgsql/share/extension/kq_imcx.control
|
||||
@@ -432,54 +432,6 @@ RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.5.2.tar.gz -O
|
||||
make -j $(getconf _NPROCESSORS_ONLN) install && \
|
||||
echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_cron.control
|
||||
|
||||
#########################################################################################
|
||||
#
|
||||
# Layer "rdkit-pg-build"
|
||||
# compile rdkit extension
|
||||
#
|
||||
#########################################################################################
|
||||
FROM build-deps AS rdkit-pg-build
|
||||
COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
|
||||
|
||||
RUN apt-get update && \
|
||||
apt-get install -y \
|
||||
cmake \
|
||||
libboost-iostreams1.74-dev \
|
||||
libboost-regex1.74-dev \
|
||||
libboost-serialization1.74-dev \
|
||||
libboost-system1.74-dev \
|
||||
libeigen3-dev \
|
||||
libfreetype6-dev
|
||||
|
||||
ENV PATH "/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
|
||||
RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_1.tar.gz -O rdkit.tar.gz && \
|
||||
echo "db346afbd0ba52c843926a2a62f8a38c7b774ffab37eaf382d789a824f21996c rdkit.tar.gz" | sha256sum --check && \
|
||||
mkdir rdkit-src && cd rdkit-src && tar xvzf ../rdkit.tar.gz --strip-components=1 -C . && \
|
||||
cmake \
|
||||
-D RDK_BUILD_CAIRO_SUPPORT=OFF \
|
||||
-D RDK_BUILD_INCHI_SUPPORT=ON \
|
||||
-D RDK_BUILD_AVALON_SUPPORT=ON \
|
||||
-D RDK_BUILD_PYTHON_WRAPPERS=OFF \
|
||||
-D RDK_BUILD_DESCRIPTORS3D=OFF \
|
||||
-D RDK_BUILD_FREESASA_SUPPORT=OFF \
|
||||
-D RDK_BUILD_COORDGEN_SUPPORT=ON \
|
||||
-D RDK_BUILD_MOLINTERCHANGE_SUPPORT=OFF \
|
||||
-D RDK_BUILD_YAEHMOP_SUPPORT=OFF \
|
||||
-D RDK_BUILD_STRUCTCHECKER_SUPPORT=OFF \
|
||||
-D RDK_USE_URF=OFF \
|
||||
-D RDK_BUILD_PGSQL=ON \
|
||||
-D RDK_PGSQL_STATIC=ON \
|
||||
-D PostgreSQL_CONFIG=pg_config \
|
||||
-D PostgreSQL_INCLUDE_DIR=`pg_config --includedir` \
|
||||
-D PostgreSQL_TYPE_INCLUDE_DIR=`pg_config --includedir-server` \
|
||||
-D PostgreSQL_LIBRARY_DIR=`pg_config --libdir` \
|
||||
-D RDK_INSTALL_INTREE=OFF \
|
||||
-D CMAKE_BUILD_TYPE=Release \
|
||||
. && \
|
||||
make -j $(getconf _NPROCESSORS_ONLN) && \
|
||||
make -j $(getconf _NPROCESSORS_ONLN) install && \
|
||||
echo 'trusted = true' >> /usr/local/pgsql/share/extension/rdkit.control
|
||||
|
||||
#########################################################################################
|
||||
#
|
||||
# Layer "rust extensions"
|
||||
@@ -612,7 +564,6 @@ COPY --from=pg-hint-plan-pg-build /usr/local/pgsql/ /usr/local/pgsql/
|
||||
COPY --from=kq-imcx-pg-build /usr/local/pgsql/ /usr/local/pgsql/
|
||||
COPY --from=pg-cron-pg-build /usr/local/pgsql/ /usr/local/pgsql/
|
||||
COPY --from=pg-pgx-ulid-build /usr/local/pgsql/ /usr/local/pgsql/
|
||||
COPY --from=rdkit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
|
||||
COPY pgxn/ pgxn/
|
||||
|
||||
RUN make -j $(getconf _NPROCESSORS_ONLN) \
|
||||
@@ -686,19 +637,14 @@ COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-deb
|
||||
# libgeos, libgdal, libsfcgal1, libproj and libprotobuf-c1 for PostGIS
|
||||
# libxml2, libxslt1.1 for xml2
|
||||
# libzstd1 for zstd
|
||||
# libboost*, libfreetype6, and zlib1g for rdkit
|
||||
RUN apt update && \
|
||||
apt install --no-install-recommends -y \
|
||||
gdb \
|
||||
locales \
|
||||
libicu67 \
|
||||
liblz4-1 \
|
||||
libreadline8 \
|
||||
libboost-iostreams1.74.0 \
|
||||
libboost-regex1.74.0 \
|
||||
libboost-serialization1.74.0 \
|
||||
libboost-system1.74.0 \
|
||||
libossp-uuid16 \
|
||||
libfreetype6 \
|
||||
libgeos-c1v5 \
|
||||
libgdal28 \
|
||||
libproj19 \
|
||||
@@ -708,9 +654,7 @@ RUN apt update && \
|
||||
libxslt1.1 \
|
||||
libzstd1 \
|
||||
libcurl4-openssl-dev \
|
||||
locales \
|
||||
procps \
|
||||
zlib1g && \
|
||||
procps && \
|
||||
rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
|
||||
localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8
|
||||
|
||||
|
||||
@@ -6,7 +6,6 @@ license.workspace = true
|
||||
|
||||
[dependencies]
|
||||
anyhow.workspace = true
|
||||
flate2.workspace = true
|
||||
chrono.workspace = true
|
||||
clap.workspace = true
|
||||
futures.workspace = true
|
||||
|
||||
@@ -15,7 +15,6 @@ use utils::lsn::Lsn;
|
||||
|
||||
use compute_api::responses::{ComputeMetrics, ComputeStatus};
|
||||
use compute_api::spec::{ComputeMode, ComputeSpec};
|
||||
use utils::measured_stream::MeasuredReader;
|
||||
|
||||
use crate::config;
|
||||
use crate::pg_helpers::*;
|
||||
@@ -180,21 +179,16 @@ impl ComputeNode {
|
||||
_ => format!("basebackup {} {} {}", spec.tenant_id, spec.timeline_id, lsn),
|
||||
};
|
||||
let copyreader = client.copy_out(basebackup_cmd.as_str())?;
|
||||
let mut measured_reader = MeasuredReader::new(copyreader);
|
||||
let mut decoder = flate2::read::GzDecoder::new(&mut measured_reader);
|
||||
|
||||
// Read the archive directly from the `CopyOutReader`
|
||||
//
|
||||
// Set `ignore_zeros` so that unpack() reads all the Copy data and
|
||||
// doesn't stop at the end-of-archive marker. Otherwise, if the server
|
||||
// sends an Error after finishing the tarball, we will not notice it.
|
||||
let mut ar = tar::Archive::new(&mut decoder);
|
||||
let mut ar = tar::Archive::new(copyreader);
|
||||
ar.set_ignore_zeros(true);
|
||||
ar.unpack(&self.pgdata)?;
|
||||
|
||||
// Report metrics
|
||||
self.state.lock().unwrap().metrics.basebackup_bytes =
|
||||
measured_reader.get_byte_count() as u64;
|
||||
self.state.lock().unwrap().metrics.basebackup_ms = Utc::now()
|
||||
.signed_duration_since(start_time)
|
||||
.to_std()
|
||||
@@ -448,7 +442,7 @@ impl ComputeNode {
|
||||
|
||||
let pg = self.start_postgres(spec.storage_auth_token.clone())?;
|
||||
|
||||
if spec.spec.mode == ComputeMode::Primary && !spec.spec.skip_pg_catalog_updates {
|
||||
if spec.spec.mode == ComputeMode::Primary {
|
||||
self.apply_config(&compute_state)?;
|
||||
}
|
||||
|
||||
|
||||
@@ -450,7 +450,6 @@ impl Endpoint {
|
||||
|
||||
// Create spec file
|
||||
let spec = ComputeSpec {
|
||||
skip_pg_catalog_updates: false,
|
||||
format_version: 1.0,
|
||||
operation_uuid: None,
|
||||
cluster: Cluster {
|
||||
|
||||
@@ -71,7 +71,6 @@ pub struct ComputeMetrics {
|
||||
pub wait_for_spec_ms: u64,
|
||||
pub sync_safekeepers_ms: u64,
|
||||
pub basebackup_ms: u64,
|
||||
pub basebackup_bytes: u64,
|
||||
pub config_ms: u64,
|
||||
pub total_startup_ms: u64,
|
||||
}
|
||||
|
||||
@@ -27,12 +27,6 @@ pub struct ComputeSpec {
|
||||
pub cluster: Cluster,
|
||||
pub delta_operations: Option<Vec<DeltaOp>>,
|
||||
|
||||
/// An optinal hint that can be passed to speed up startup time if we know
|
||||
/// that no pg catalog mutations (like role creation, database creation,
|
||||
/// extension creation) need to be done on the actual database to start.
|
||||
#[serde(default)] // Default false
|
||||
pub skip_pg_catalog_updates: bool,
|
||||
|
||||
// Information needed to connect to the storage layer.
|
||||
//
|
||||
// `tenant_id`, `timeline_id` and `pageserver_connstring` are always needed.
|
||||
|
||||
@@ -152,7 +152,7 @@ pub enum ActivatingFrom {
|
||||
}
|
||||
|
||||
/// A state of a timeline in pageserver's memory.
|
||||
#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
|
||||
#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
|
||||
pub enum TimelineState {
|
||||
/// The timeline is recognized by the pageserver but is not yet operational.
|
||||
/// In particular, the walreceiver connection loop is not running for this timeline.
|
||||
@@ -165,7 +165,7 @@ pub enum TimelineState {
|
||||
/// It cannot transition back into any other state.
|
||||
Stopping,
|
||||
/// The timeline is broken and not operational (previous states: Loading or Active).
|
||||
Broken { reason: String, backtrace: String },
|
||||
Broken,
|
||||
}
|
||||
|
||||
#[serde_as]
|
||||
|
||||
@@ -111,8 +111,6 @@ pub trait RemoteStorage: Send + Sync + 'static {
|
||||
) -> Result<Download, DownloadError>;
|
||||
|
||||
async fn delete(&self, path: &RemotePath) -> anyhow::Result<()>;
|
||||
|
||||
async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()>;
|
||||
}
|
||||
|
||||
pub struct Download {
|
||||
@@ -225,14 +223,6 @@ impl GenericRemoteStorage {
|
||||
Self::Unreliable(s) => s.delete(path).await,
|
||||
}
|
||||
}
|
||||
|
||||
pub async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
|
||||
match self {
|
||||
Self::LocalFs(s) => s.delete_objects(paths).await,
|
||||
Self::AwsS3(s) => s.delete_objects(paths).await,
|
||||
Self::Unreliable(s) => s.delete_objects(paths).await,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
impl GenericRemoteStorage {
|
||||
|
||||
@@ -17,7 +17,7 @@ use tokio::{
|
||||
io::{self, AsyncReadExt, AsyncSeekExt, AsyncWriteExt},
|
||||
};
|
||||
use tracing::*;
|
||||
use utils::{crashsafe::path_with_suffix_extension, fs_ext::is_directory_empty};
|
||||
use utils::crashsafe::path_with_suffix_extension;
|
||||
|
||||
use crate::{Download, DownloadError, RemotePath};
|
||||
|
||||
@@ -101,35 +101,19 @@ impl RemoteStorage for LocalFs {
|
||||
Some(prefix) => Cow::Owned(prefix.with_base(&self.storage_root)),
|
||||
None => Cow::Borrowed(&self.storage_root),
|
||||
};
|
||||
|
||||
let prefixes_to_filter = get_all_files(path.as_ref(), false)
|
||||
Ok(get_all_files(path.as_ref(), false)
|
||||
.await
|
||||
.map_err(DownloadError::Other)?;
|
||||
|
||||
let mut prefixes = Vec::with_capacity(prefixes_to_filter.len());
|
||||
|
||||
// filter out empty directories to mirror s3 behavior.
|
||||
for prefix in prefixes_to_filter {
|
||||
if prefix.is_dir()
|
||||
&& is_directory_empty(&prefix)
|
||||
.await
|
||||
.map_err(DownloadError::Other)?
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
prefixes.push(
|
||||
prefix
|
||||
.strip_prefix(&self.storage_root)
|
||||
.context("Failed to strip prefix")
|
||||
.map_err(DownloadError::Other)?
|
||||
.into_iter()
|
||||
.map(|path| {
|
||||
path.strip_prefix(&self.storage_root)
|
||||
.context("Failed to strip preifix")
|
||||
.and_then(RemotePath::new)
|
||||
.expect(
|
||||
"We list files for storage root, hence should be able to remote the prefix",
|
||||
),
|
||||
)
|
||||
}
|
||||
|
||||
Ok(prefixes)
|
||||
)
|
||||
})
|
||||
.collect())
|
||||
}
|
||||
|
||||
async fn upload(
|
||||
@@ -307,25 +291,11 @@ impl RemoteStorage for LocalFs {
|
||||
|
||||
async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {
|
||||
let file_path = path.with_base(&self.storage_root);
|
||||
if !file_path.exists() {
|
||||
// See https://docs.aws.amazon.com/AmazonS3/latest/API/API_DeleteObject.html
|
||||
// > If there isn't a null version, Amazon S3 does not remove any objects but will still respond that the command was successful.
|
||||
return Ok(());
|
||||
if file_path.exists() && file_path.is_file() {
|
||||
Ok(fs::remove_file(file_path).await?)
|
||||
} else {
|
||||
bail!("File {file_path:?} either does not exist or is not a file")
|
||||
}
|
||||
|
||||
if !file_path.is_file() {
|
||||
anyhow::bail!("{file_path:?} is not a file");
|
||||
}
|
||||
Ok(fs::remove_file(file_path)
|
||||
.await
|
||||
.map_err(|e| anyhow::anyhow!(e))?)
|
||||
}
|
||||
|
||||
async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
|
||||
for path in paths {
|
||||
self.delete(path).await?
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
}
|
||||
|
||||
@@ -350,7 +320,7 @@ where
|
||||
let file_type = dir_entry.file_type().await?;
|
||||
let entry_path = dir_entry.path();
|
||||
if file_type.is_symlink() {
|
||||
debug!("{entry_path:?} is a symlink, skipping")
|
||||
debug!("{entry_path:?} us a symlink, skipping")
|
||||
} else if file_type.is_dir() {
|
||||
if recursive {
|
||||
paths.extend(get_all_files(&entry_path, true).await?.into_iter())
|
||||
@@ -625,11 +595,15 @@ mod fs_tests {
|
||||
storage.delete(&upload_target).await?;
|
||||
assert!(storage.list().await?.is_empty());
|
||||
|
||||
storage
|
||||
.delete(&upload_target)
|
||||
.await
|
||||
.expect("Should allow deleting non-existing storage files");
|
||||
|
||||
match storage.delete(&upload_target).await {
|
||||
Ok(()) => panic!("Should not allow deleting non-existing storage files"),
|
||||
Err(e) => {
|
||||
let error_string = e.to_string();
|
||||
assert!(error_string.contains("does not exist"));
|
||||
let expected_path = upload_target.with_base(&storage.storage_root);
|
||||
assert!(error_string.contains(expected_path.to_str().unwrap()));
|
||||
}
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
|
||||
@@ -17,7 +17,6 @@ use aws_sdk_s3::{
|
||||
error::SdkError,
|
||||
operation::get_object::GetObjectError,
|
||||
primitives::ByteStream,
|
||||
types::{Delete, ObjectIdentifier},
|
||||
Client,
|
||||
};
|
||||
use aws_smithy_http::body::SdkBody;
|
||||
@@ -82,24 +81,12 @@ pub(super) mod metrics {
|
||||
.inc();
|
||||
}
|
||||
|
||||
pub fn inc_delete_objects(count: u64) {
|
||||
S3_REQUESTS_COUNT
|
||||
.with_label_values(&["delete_object"])
|
||||
.inc_by(count);
|
||||
}
|
||||
|
||||
pub fn inc_delete_object_fail() {
|
||||
S3_REQUESTS_FAIL_COUNT
|
||||
.with_label_values(&["delete_object"])
|
||||
.inc();
|
||||
}
|
||||
|
||||
pub fn inc_delete_objects_fail(count: u64) {
|
||||
S3_REQUESTS_FAIL_COUNT
|
||||
.with_label_values(&["delete_object"])
|
||||
.inc_by(count);
|
||||
}
|
||||
|
||||
pub fn inc_list_objects() {
|
||||
S3_REQUESTS_COUNT.with_label_values(&["list_objects"]).inc();
|
||||
}
|
||||
@@ -409,34 +396,6 @@ impl RemoteStorage for S3Bucket {
|
||||
})
|
||||
.await
|
||||
}
|
||||
async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
|
||||
let _guard = self
|
||||
.concurrency_limiter
|
||||
.acquire()
|
||||
.await
|
||||
.context("Concurrency limiter semaphore got closed during S3 delete")?;
|
||||
|
||||
let mut delete_objects = Vec::with_capacity(paths.len());
|
||||
for path in paths {
|
||||
let obj_id = ObjectIdentifier::builder()
|
||||
.set_key(Some(self.relative_path_to_s3_object(path)))
|
||||
.build();
|
||||
delete_objects.push(obj_id);
|
||||
}
|
||||
|
||||
metrics::inc_delete_objects(paths.len() as u64);
|
||||
self.client
|
||||
.delete_objects()
|
||||
.bucket(self.bucket_name.clone())
|
||||
.delete(Delete::builder().set_objects(Some(delete_objects)).build())
|
||||
.send()
|
||||
.await
|
||||
.map_err(|e| {
|
||||
metrics::inc_delete_objects_fail(paths.len() as u64);
|
||||
e
|
||||
})?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {
|
||||
let _guard = self
|
||||
|
||||
@@ -119,11 +119,4 @@ impl RemoteStorage for UnreliableWrapper {
|
||||
self.attempt(RemoteOp::Delete(path.clone()))?;
|
||||
self.inner.delete(path).await
|
||||
}
|
||||
|
||||
async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
|
||||
for path in paths {
|
||||
self.delete(path).await?
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
}
|
||||
|
||||
@@ -7,7 +7,6 @@ use std::sync::Arc;
|
||||
use std::time::UNIX_EPOCH;
|
||||
|
||||
use anyhow::Context;
|
||||
use once_cell::sync::OnceCell;
|
||||
use remote_storage::{
|
||||
GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind, S3Config,
|
||||
};
|
||||
@@ -15,12 +14,8 @@ use test_context::{test_context, AsyncTestContext};
|
||||
use tokio::task::JoinSet;
|
||||
use tracing::{debug, error, info};
|
||||
|
||||
static LOGGING_DONE: OnceCell<()> = OnceCell::new();
|
||||
|
||||
const ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME: &str = "ENABLE_REAL_S3_REMOTE_STORAGE";
|
||||
|
||||
const BASE_PREFIX: &str = "test/";
|
||||
|
||||
/// Tests that S3 client can list all prefixes, even if the response come paginated and requires multiple S3 queries.
|
||||
/// Uses real S3 and requires [`ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME`] and related S3 cred env vars specified.
|
||||
/// See the client creation in [`create_s3_client`] for details on the required env vars.
|
||||
@@ -43,20 +38,20 @@ const BASE_PREFIX: &str = "test/";
|
||||
///
|
||||
/// Lastly, the test attempts to clean up and remove all uploaded S3 files.
|
||||
/// If any errors appear during the clean up, they get logged, but the test is not failed or stopped until clean up is finished.
|
||||
#[test_context(MaybeEnabledS3WithTestBlobs)]
|
||||
#[test_context(MaybeEnabledS3)]
|
||||
#[tokio::test]
|
||||
async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3WithTestBlobs) -> anyhow::Result<()> {
|
||||
async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()> {
|
||||
let ctx = match ctx {
|
||||
MaybeEnabledS3WithTestBlobs::Enabled(ctx) => ctx,
|
||||
MaybeEnabledS3WithTestBlobs::Disabled => return Ok(()),
|
||||
MaybeEnabledS3WithTestBlobs::UploadsFailed(e, _) => anyhow::bail!("S3 init failed: {e:?}"),
|
||||
MaybeEnabledS3::Enabled(ctx) => ctx,
|
||||
MaybeEnabledS3::Disabled => return Ok(()),
|
||||
MaybeEnabledS3::UploadsFailed(e, _) => anyhow::bail!("S3 init failed: {e:?}"),
|
||||
};
|
||||
|
||||
let test_client = Arc::clone(&ctx.enabled.client);
|
||||
let test_client = Arc::clone(&ctx.client_with_excessive_pagination);
|
||||
let expected_remote_prefixes = ctx.remote_prefixes.clone();
|
||||
|
||||
let base_prefix = RemotePath::new(Path::new(ctx.enabled.base_prefix))
|
||||
.context("common_prefix construction")?;
|
||||
let base_prefix =
|
||||
RemotePath::new(Path::new(ctx.base_prefix_str)).context("common_prefix construction")?;
|
||||
let root_remote_prefixes = test_client
|
||||
.list_prefixes(None)
|
||||
.await
|
||||
@@ -88,122 +83,27 @@ async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3WithTestBlobs) -> any
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[test_context(MaybeEnabledS3)]
|
||||
#[tokio::test]
|
||||
async fn s3_delete_non_exising_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()> {
|
||||
let ctx = match ctx {
|
||||
MaybeEnabledS3::Enabled(ctx) => ctx,
|
||||
MaybeEnabledS3::Disabled => return Ok(()),
|
||||
};
|
||||
|
||||
let path = RemotePath::new(&PathBuf::from(format!(
|
||||
"{}/for_sure_there_is_nothing_there_really",
|
||||
ctx.base_prefix,
|
||||
)))
|
||||
.with_context(|| "RemotePath conversion")?;
|
||||
|
||||
ctx.client.delete(&path).await.expect("should succeed");
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[test_context(MaybeEnabledS3)]
|
||||
#[tokio::test]
|
||||
async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()> {
|
||||
let ctx = match ctx {
|
||||
MaybeEnabledS3::Enabled(ctx) => ctx,
|
||||
MaybeEnabledS3::Disabled => return Ok(()),
|
||||
};
|
||||
|
||||
let path1 = RemotePath::new(&PathBuf::from(format!("{}/path1", ctx.base_prefix,)))
|
||||
.with_context(|| "RemotePath conversion")?;
|
||||
|
||||
let path2 = RemotePath::new(&PathBuf::from(format!("{}/path2", ctx.base_prefix,)))
|
||||
.with_context(|| "RemotePath conversion")?;
|
||||
|
||||
let data1 = "remote blob data1".as_bytes();
|
||||
let data1_len = data1.len();
|
||||
let data2 = "remote blob data2".as_bytes();
|
||||
let data2_len = data2.len();
|
||||
ctx.client
|
||||
.upload(std::io::Cursor::new(data1), data1_len, &path1, None)
|
||||
.await?;
|
||||
|
||||
ctx.client
|
||||
.upload(std::io::Cursor::new(data2), data2_len, &path2, None)
|
||||
.await?;
|
||||
|
||||
ctx.client.delete_objects(&[path1, path2]).await?;
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn ensure_logging_ready() {
|
||||
LOGGING_DONE.get_or_init(|| {
|
||||
utils::logging::init(
|
||||
utils::logging::LogFormat::Test,
|
||||
utils::logging::TracingErrorLayerEnablement::Disabled,
|
||||
)
|
||||
.expect("logging init failed");
|
||||
});
|
||||
}
|
||||
|
||||
struct EnabledS3 {
|
||||
client: Arc<GenericRemoteStorage>,
|
||||
base_prefix: &'static str,
|
||||
}
|
||||
|
||||
impl EnabledS3 {
|
||||
async fn setup(max_keys_in_list_response: Option<i32>) -> Self {
|
||||
let client = create_s3_client(max_keys_in_list_response)
|
||||
.context("S3 client creation")
|
||||
.expect("S3 client creation failed");
|
||||
|
||||
EnabledS3 {
|
||||
client,
|
||||
base_prefix: BASE_PREFIX,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
enum MaybeEnabledS3 {
|
||||
Enabled(EnabledS3),
|
||||
Disabled,
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl AsyncTestContext for MaybeEnabledS3 {
|
||||
async fn setup() -> Self {
|
||||
ensure_logging_ready();
|
||||
|
||||
if env::var(ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME).is_err() {
|
||||
info!(
|
||||
"`{}` env variable is not set, skipping the test",
|
||||
ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME
|
||||
);
|
||||
return Self::Disabled;
|
||||
}
|
||||
|
||||
Self::Enabled(EnabledS3::setup(None).await)
|
||||
}
|
||||
}
|
||||
|
||||
enum MaybeEnabledS3WithTestBlobs {
|
||||
Enabled(S3WithTestBlobs),
|
||||
Disabled,
|
||||
UploadsFailed(anyhow::Error, S3WithTestBlobs),
|
||||
}
|
||||
|
||||
struct S3WithTestBlobs {
|
||||
enabled: EnabledS3,
|
||||
client_with_excessive_pagination: Arc<GenericRemoteStorage>,
|
||||
base_prefix_str: &'static str,
|
||||
remote_prefixes: HashSet<RemotePath>,
|
||||
remote_blobs: HashSet<RemotePath>,
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl AsyncTestContext for MaybeEnabledS3WithTestBlobs {
|
||||
impl AsyncTestContext for MaybeEnabledS3 {
|
||||
async fn setup() -> Self {
|
||||
ensure_logging_ready();
|
||||
utils::logging::init(
|
||||
utils::logging::LogFormat::Test,
|
||||
utils::logging::TracingErrorLayerEnablement::Disabled,
|
||||
)
|
||||
.expect("logging init failed");
|
||||
if env::var(ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME).is_err() {
|
||||
info!(
|
||||
"`{}` env variable is not set, skipping the test",
|
||||
@@ -215,14 +115,23 @@ impl AsyncTestContext for MaybeEnabledS3WithTestBlobs {
|
||||
let max_keys_in_list_response = 10;
|
||||
let upload_tasks_count = 1 + (2 * usize::try_from(max_keys_in_list_response).unwrap());
|
||||
|
||||
let enabled = EnabledS3::setup(Some(max_keys_in_list_response)).await;
|
||||
let client_with_excessive_pagination = create_s3_client(max_keys_in_list_response)
|
||||
.context("S3 client creation")
|
||||
.expect("S3 client creation failed");
|
||||
|
||||
match upload_s3_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
|
||||
let base_prefix_str = "test/";
|
||||
match upload_s3_data(
|
||||
&client_with_excessive_pagination,
|
||||
base_prefix_str,
|
||||
upload_tasks_count,
|
||||
)
|
||||
.await
|
||||
{
|
||||
ControlFlow::Continue(uploads) => {
|
||||
info!("Remote objects created successfully");
|
||||
|
||||
Self::Enabled(S3WithTestBlobs {
|
||||
enabled,
|
||||
client_with_excessive_pagination,
|
||||
base_prefix_str,
|
||||
remote_prefixes: uploads.prefixes,
|
||||
remote_blobs: uploads.blobs,
|
||||
})
|
||||
@@ -230,7 +139,8 @@ impl AsyncTestContext for MaybeEnabledS3WithTestBlobs {
|
||||
ControlFlow::Break(uploads) => Self::UploadsFailed(
|
||||
anyhow::anyhow!("One or multiple blobs failed to upload to S3"),
|
||||
S3WithTestBlobs {
|
||||
enabled,
|
||||
client_with_excessive_pagination,
|
||||
base_prefix_str,
|
||||
remote_prefixes: uploads.prefixes,
|
||||
remote_blobs: uploads.blobs,
|
||||
},
|
||||
@@ -242,15 +152,13 @@ impl AsyncTestContext for MaybeEnabledS3WithTestBlobs {
|
||||
match self {
|
||||
Self::Disabled => {}
|
||||
Self::Enabled(ctx) | Self::UploadsFailed(_, ctx) => {
|
||||
cleanup(&ctx.enabled.client, ctx.remote_blobs).await;
|
||||
cleanup(&ctx.client_with_excessive_pagination, ctx.remote_blobs).await;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
fn create_s3_client(
|
||||
max_keys_per_list_response: Option<i32>,
|
||||
) -> anyhow::Result<Arc<GenericRemoteStorage>> {
|
||||
fn create_s3_client(max_keys_per_list_response: i32) -> anyhow::Result<Arc<GenericRemoteStorage>> {
|
||||
let remote_storage_s3_bucket = env::var("REMOTE_STORAGE_S3_BUCKET")
|
||||
.context("`REMOTE_STORAGE_S3_BUCKET` env var is not set, but real S3 tests are enabled")?;
|
||||
let remote_storage_s3_region = env::var("REMOTE_STORAGE_S3_REGION")
|
||||
@@ -268,7 +176,7 @@ fn create_s3_client(
|
||||
prefix_in_bucket: Some(format!("pagination_should_work_test_{random_prefix_part}/")),
|
||||
endpoint: None,
|
||||
concurrency_limit: NonZeroUsize::new(100).unwrap(),
|
||||
max_keys_per_list_response,
|
||||
max_keys_per_list_response: Some(max_keys_per_list_response),
|
||||
}),
|
||||
};
|
||||
Ok(Arc::new(
|
||||
@@ -1,8 +1,6 @@
|
||||
/// Extensions to `std::fs` types.
|
||||
use std::{fs, io, path::Path};
|
||||
|
||||
use anyhow::Context;
|
||||
|
||||
pub trait PathExt {
|
||||
/// Returns an error if `self` is not a directory.
|
||||
fn is_empty_dir(&self) -> io::Result<bool>;
|
||||
@@ -17,19 +15,10 @@ where
|
||||
}
|
||||
}
|
||||
|
||||
pub async fn is_directory_empty(path: impl AsRef<Path>) -> anyhow::Result<bool> {
|
||||
let mut dir = tokio::fs::read_dir(&path)
|
||||
.await
|
||||
.context(format!("read_dir({})", path.as_ref().display()))?;
|
||||
Ok(dir.next_entry().await?.is_none())
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod test {
|
||||
use std::path::PathBuf;
|
||||
|
||||
use crate::fs_ext::is_directory_empty;
|
||||
|
||||
#[test]
|
||||
fn is_empty_dir() {
|
||||
use super::PathExt;
|
||||
@@ -53,26 +42,4 @@ mod test {
|
||||
std::fs::remove_file(&file_path).unwrap();
|
||||
assert!(file_path.is_empty_dir().is_err());
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn is_empty_dir_async() {
|
||||
let dir = tempfile::tempdir().unwrap();
|
||||
let dir_path = dir.path();
|
||||
|
||||
// test positive case
|
||||
assert!(
|
||||
is_directory_empty(dir_path).await.expect("test failure"),
|
||||
"new tempdir should be empty"
|
||||
);
|
||||
|
||||
// invoke on a file to ensure it returns an error
|
||||
let file_path: PathBuf = dir_path.join("testfile");
|
||||
let f = std::fs::File::create(&file_path).unwrap();
|
||||
drop(f);
|
||||
assert!(is_directory_empty(&file_path).await.is_err());
|
||||
|
||||
// do it again on a path, we know to be nonexistent
|
||||
std::fs::remove_file(&file_path).unwrap();
|
||||
assert!(is_directory_empty(file_path).await.is_err());
|
||||
}
|
||||
}
|
||||
|
||||
@@ -1,18 +1,19 @@
|
||||
use crate::auth::{Claims, JwtAuth};
|
||||
use crate::http::error::{api_error_handler, route_error_handler, ApiError};
|
||||
use anyhow::Context;
|
||||
use anyhow::{anyhow, Context};
|
||||
use hyper::header::{HeaderName, AUTHORIZATION};
|
||||
use hyper::http::HeaderValue;
|
||||
use hyper::Method;
|
||||
use hyper::{header::CONTENT_TYPE, Body, Request, Response};
|
||||
use hyper::{header::CONTENT_TYPE, Body, Request, Response, Server};
|
||||
use metrics::{register_int_counter, Encoder, IntCounter, TextEncoder};
|
||||
use once_cell::sync::Lazy;
|
||||
use routerify::ext::RequestExt;
|
||||
use routerify::{Middleware, RequestInfo, Router, RouterBuilder};
|
||||
use routerify::{Middleware, RequestInfo, Router, RouterBuilder, RouterService};
|
||||
use tokio::task::JoinError;
|
||||
use tracing::{self, debug, info, info_span, warn, Instrument};
|
||||
|
||||
use std::future::Future;
|
||||
use std::net::TcpListener;
|
||||
use std::str::FromStr;
|
||||
|
||||
static SERVE_METRICS_COUNT: Lazy<IntCounter> = Lazy::new(|| {
|
||||
@@ -347,6 +348,40 @@ pub fn check_permission_with(
|
||||
}
|
||||
}
|
||||
|
||||
///
|
||||
/// Start listening for HTTP requests on given socket.
|
||||
///
|
||||
/// 'shutdown_future' can be used to stop. If the Future becomes
|
||||
/// ready, we stop listening for new requests, and the function returns.
|
||||
///
|
||||
pub fn serve_thread_main<S>(
|
||||
router_builder: RouterBuilder<hyper::Body, ApiError>,
|
||||
listener: TcpListener,
|
||||
shutdown_future: S,
|
||||
) -> anyhow::Result<()>
|
||||
where
|
||||
S: Future<Output = ()> + Send + Sync,
|
||||
{
|
||||
info!("Starting an HTTP endpoint at {}", listener.local_addr()?);
|
||||
|
||||
// Create a Service from the router above to handle incoming requests.
|
||||
let service = RouterService::new(router_builder.build().map_err(|err| anyhow!(err))?).unwrap();
|
||||
|
||||
// Enter a single-threaded tokio runtime bound to the current thread
|
||||
let runtime = tokio::runtime::Builder::new_current_thread()
|
||||
.enable_all()
|
||||
.build()?;
|
||||
|
||||
let _guard = runtime.enter();
|
||||
|
||||
let server = Server::from_tcp(listener)?
|
||||
.serve(service)
|
||||
.with_graceful_shutdown(shutdown_future);
|
||||
|
||||
runtime.block_on(server)?;
|
||||
|
||||
Ok(())
|
||||
}
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
|
||||
@@ -21,7 +21,7 @@ pub enum ApiError {
|
||||
Conflict(String),
|
||||
|
||||
#[error("Precondition failed: {0}")]
|
||||
PreconditionFailed(Box<str>),
|
||||
PreconditionFailed(&'static str),
|
||||
|
||||
#[error(transparent)]
|
||||
InternalServerError(anyhow::Error),
|
||||
|
||||
@@ -1,5 +1,4 @@
|
||||
use pin_project_lite::pin_project;
|
||||
use std::io::Read;
|
||||
use std::pin::Pin;
|
||||
use std::{io, task};
|
||||
use tokio::io::{AsyncRead, AsyncWrite, ReadBuf};
|
||||
@@ -76,34 +75,3 @@ impl<S: AsyncWrite + Unpin, R, W: FnMut(usize)> AsyncWrite for MeasuredStream<S,
|
||||
self.project().stream.poll_shutdown(context)
|
||||
}
|
||||
}
|
||||
|
||||
/// Wrapper for a reader that counts bytes read.
|
||||
///
|
||||
/// Similar to MeasuredStream but it's one way and it's sync
|
||||
pub struct MeasuredReader<R: Read> {
|
||||
inner: R,
|
||||
byte_count: usize,
|
||||
}
|
||||
|
||||
impl<R: Read> MeasuredReader<R> {
|
||||
pub fn new(reader: R) -> Self {
|
||||
Self {
|
||||
inner: reader,
|
||||
byte_count: 0,
|
||||
}
|
||||
}
|
||||
|
||||
pub fn get_byte_count(&self) -> usize {
|
||||
self.byte_count
|
||||
}
|
||||
}
|
||||
|
||||
impl<R: Read> Read for MeasuredReader<R> {
|
||||
fn read(&mut self, buf: &mut [u8]) -> std::io::Result<usize> {
|
||||
let result = self.inner.read(buf);
|
||||
if let Ok(n_bytes) = result {
|
||||
self.byte_count += n_bytes
|
||||
}
|
||||
result
|
||||
}
|
||||
}
|
||||
|
||||
@@ -12,7 +12,6 @@ testing = ["fail/failpoints"]
|
||||
|
||||
[dependencies]
|
||||
anyhow.workspace = true
|
||||
flate2.workspace = true
|
||||
async-stream.workspace = true
|
||||
async-trait.workspace = true
|
||||
byteorder.workspace = true
|
||||
|
||||
@@ -215,7 +215,7 @@ paths:
|
||||
schema:
|
||||
$ref: "#/components/schemas/NotFoundError"
|
||||
"412":
|
||||
description: Tenant is missing, or timeline has children
|
||||
description: Tenant is missing
|
||||
content:
|
||||
application/json:
|
||||
schema:
|
||||
@@ -386,7 +386,6 @@ paths:
|
||||
"202":
|
||||
description: Tenant attaching scheduled
|
||||
"400":
|
||||
description: Bad Request
|
||||
content:
|
||||
application/json:
|
||||
schema:
|
||||
@@ -946,7 +945,7 @@ components:
|
||||
type: string
|
||||
enum: [ "maybe", "attached", "failed" ]
|
||||
data:
|
||||
type: object
|
||||
- type: object
|
||||
properties:
|
||||
reason:
|
||||
type: string
|
||||
|
||||
@@ -183,10 +183,9 @@ impl From<crate::tenant::DeleteTimelineError> for ApiError {
|
||||
use crate::tenant::DeleteTimelineError::*;
|
||||
match value {
|
||||
NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found")),
|
||||
HasChildren(children) => ApiError::PreconditionFailed(
|
||||
format!("Cannot delete timeline which has child timelines: {children:?}")
|
||||
.into_boxed_str(),
|
||||
),
|
||||
HasChildren => ApiError::BadRequest(anyhow::anyhow!(
|
||||
"Cannot delete timeline which has child timelines"
|
||||
)),
|
||||
Other(e) => ApiError::InternalServerError(e),
|
||||
}
|
||||
}
|
||||
@@ -198,9 +197,9 @@ impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
|
||||
match value {
|
||||
// Report Precondition failed so client can distinguish between
|
||||
// "tenant is missing" case from "timeline is missing"
|
||||
Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
|
||||
"Requested tenant is missing".to_owned().into_boxed_str(),
|
||||
),
|
||||
Tenant(GetTenantError::NotFound(..)) => {
|
||||
ApiError::PreconditionFailed("Requested tenant is missing")
|
||||
}
|
||||
Tenant(t) => ApiError::from(t),
|
||||
Timeline(t) => ApiError::from(t),
|
||||
}
|
||||
@@ -495,8 +494,7 @@ async fn timeline_delete_handler(
|
||||
.instrument(info_span!("timeline_delete", tenant = %tenant_id, timeline = %timeline_id))
|
||||
.await?;
|
||||
|
||||
// FIXME: needs to be an error for console to retry it. Ideally Accepted should be used and retried until 404.
|
||||
json_response(StatusCode::ACCEPTED, ())
|
||||
json_response(StatusCode::OK, ())
|
||||
}
|
||||
|
||||
async fn tenant_detach_handler(
|
||||
|
||||
@@ -24,14 +24,14 @@ use postgres_backend::{self, is_expected_io_error, AuthType, PostgresBackend, Qu
|
||||
use pq_proto::framed::ConnectionError;
|
||||
use pq_proto::FeStartupPacket;
|
||||
use pq_proto::{BeMessage, FeMessage, RowDescriptor};
|
||||
use std::io::{self, Write};
|
||||
use std::io;
|
||||
use std::net::TcpListener;
|
||||
use std::pin::pin;
|
||||
use std::str;
|
||||
use std::str::FromStr;
|
||||
use std::sync::Arc;
|
||||
use std::time::Duration;
|
||||
use tokio::io::{AsyncRead, AsyncWrite, AsyncWriteExt};
|
||||
use tokio::io::{AsyncRead, AsyncWrite};
|
||||
use tokio_util::io::StreamReader;
|
||||
use tracing::*;
|
||||
use utils::id::ConnectionId;
|
||||
@@ -773,9 +773,8 @@ impl PageServerHandler {
|
||||
// Send a tarball of the latest layer on the timeline
|
||||
{
|
||||
let mut writer = pgb.copyout_writer();
|
||||
let mut raw_tar = Vec::new();
|
||||
basebackup::send_basebackup_tarball(
|
||||
&mut raw_tar,
|
||||
&mut writer,
|
||||
&timeline,
|
||||
lsn,
|
||||
prev_lsn,
|
||||
@@ -783,11 +782,6 @@ impl PageServerHandler {
|
||||
&ctx,
|
||||
)
|
||||
.await?;
|
||||
let mut encoder =
|
||||
flate2::write::GzEncoder::new(Vec::new(), flate2::Compression::default());
|
||||
encoder.write_all(&raw_tar)?;
|
||||
let compressed_tar = encoder.finish()?;
|
||||
writer.write(&compressed_tar).await?;
|
||||
}
|
||||
|
||||
pgb.write_message_noflush(&BeMessage::CopyDone)?;
|
||||
|
||||
@@ -699,20 +699,6 @@ impl<'a> DatadirModification<'a> {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
pub fn init_empty_test_timeline(&mut self) -> anyhow::Result<()> {
|
||||
self.init_empty()?;
|
||||
self.put_control_file(bytes::Bytes::from_static(
|
||||
b"control_file contents do not matter",
|
||||
))
|
||||
.context("put_control_file")?;
|
||||
self.put_checkpoint(bytes::Bytes::from_static(
|
||||
b"checkpoint_file contents do not matter",
|
||||
))
|
||||
.context("put_checkpoint_file")?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Put a new page version that can be constructed from a WAL record
|
||||
///
|
||||
/// NOTE: this will *not* implicitly extend the relation, if the page is beyond the
|
||||
@@ -1607,6 +1593,20 @@ fn is_slru_block_key(key: Key) -> bool {
|
||||
&& key.field6 != 0xffffffff // and not SlruSegSize
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
pub fn create_test_timeline(
|
||||
tenant: &crate::tenant::Tenant,
|
||||
timeline_id: utils::id::TimelineId,
|
||||
pg_version: u32,
|
||||
ctx: &RequestContext,
|
||||
) -> anyhow::Result<std::sync::Arc<Timeline>> {
|
||||
let tline = tenant.create_test_timeline(timeline_id, Lsn(8), pg_version, ctx)?;
|
||||
let mut m = tline.begin_modification(Lsn(8));
|
||||
m.init_empty()?;
|
||||
m.commit()?;
|
||||
Ok(tline)
|
||||
}
|
||||
|
||||
#[allow(clippy::bool_assert_comparison)]
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
|
||||
@@ -257,9 +257,6 @@ pub enum TaskKind {
|
||||
// task that handles attaching a tenant
|
||||
Attach,
|
||||
|
||||
// Used mostly for background deletion from s3
|
||||
TimelineDeletionWorker,
|
||||
|
||||
// task that handhes metrics collection
|
||||
MetricsCollection,
|
||||
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -1,325 +0,0 @@
|
||||
//! This module contains the encoding and decoding of the local manifest file.
|
||||
//!
|
||||
//! MANIFEST is a write-ahead log which is stored locally to each timeline. It
|
||||
//! records the state of the storage engine. It contains a snapshot of the
|
||||
//! state and all operations proceeding that snapshot. The file begins with a
|
||||
//! header recording MANIFEST version number. After that, it contains a snapshot.
|
||||
//! The snapshot is followed by a list of operations. Each operation is a list
|
||||
//! of records. Each record is either an addition or a removal of a layer.
|
||||
//!
|
||||
//! With MANIFEST, we can:
|
||||
//!
|
||||
//! 1. recover state quickly by reading the file, potentially boosting the
|
||||
//! startup speed.
|
||||
//! 2. ensure all operations are atomic and avoid corruption, solving issues
|
||||
//! like redundant image layer and preparing us for future compaction
|
||||
//! strategies.
|
||||
//!
|
||||
//! There is also a format for storing all layer files on S3, called
|
||||
//! `index_part.json`. Compared with index_part, MANIFEST is an WAL which
|
||||
//! records all operations as logs, and therefore we can easily replay the
|
||||
//! operations when recovering from crash, while ensuring those operations
|
||||
//! are atomic upon restart.
|
||||
//!
|
||||
//! Currently, this is not used in the system. Future refactors will ensure
|
||||
//! the storage state will be recorded in this file, and the system can be
|
||||
//! recovered from this file. This is tracked in
|
||||
//! https://github.com/neondatabase/neon/issues/4418
|
||||
|
||||
use std::io::{self, Read, Write};
|
||||
|
||||
use crate::virtual_file::VirtualFile;
|
||||
use anyhow::Result;
|
||||
use bytes::{Buf, BufMut, Bytes, BytesMut};
|
||||
use crc32c::crc32c;
|
||||
use serde::{Deserialize, Serialize};
|
||||
use tracing::log::warn;
|
||||
use utils::lsn::Lsn;
|
||||
|
||||
use super::storage_layer::PersistentLayerDesc;
|
||||
|
||||
pub struct Manifest {
|
||||
file: VirtualFile,
|
||||
}
|
||||
|
||||
#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
|
||||
pub struct Snapshot {
|
||||
pub layers: Vec<PersistentLayerDesc>,
|
||||
}
|
||||
|
||||
/// serde by default encode this in tagged enum, and therefore it will be something
|
||||
/// like `{ "AddLayer": { ... } }`.
|
||||
#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
|
||||
pub enum Record {
|
||||
AddLayer(PersistentLayerDesc),
|
||||
RemoveLayer(PersistentLayerDesc),
|
||||
}
|
||||
|
||||
/// `echo neon.manifest | sha1sum` and take the leading 8 bytes.
|
||||
const MANIFEST_MAGIC_NUMBER: u64 = 0xf5c44592b806109c;
|
||||
const MANIFEST_VERSION: u64 = 1;
|
||||
|
||||
#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
|
||||
pub struct ManifestHeader {
|
||||
magic_number: u64,
|
||||
version: u64,
|
||||
}
|
||||
|
||||
const MANIFEST_HEADER_LEN: usize = 16;
|
||||
|
||||
impl ManifestHeader {
|
||||
fn encode(&self) -> BytesMut {
|
||||
let mut buf = BytesMut::with_capacity(MANIFEST_HEADER_LEN);
|
||||
buf.put_u64(self.magic_number);
|
||||
buf.put_u64(self.version);
|
||||
buf
|
||||
}
|
||||
|
||||
fn decode(mut buf: &[u8]) -> Self {
|
||||
assert!(buf.len() == MANIFEST_HEADER_LEN, "invalid header");
|
||||
Self {
|
||||
magic_number: buf.get_u64(),
|
||||
version: buf.get_u64(),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
|
||||
pub enum Operation {
|
||||
/// A snapshot of the current state.
|
||||
///
|
||||
/// Lsn field represents the LSN that is persisted to disk for this snapshot.
|
||||
Snapshot(Snapshot, Lsn),
|
||||
/// An atomic operation that changes the state.
|
||||
///
|
||||
/// Lsn field represents the LSN that is persisted to disk after the operation is done.
|
||||
/// This will only change when new L0 is flushed to the disk.
|
||||
Operation(Vec<Record>, Lsn),
|
||||
}
|
||||
|
||||
struct RecordHeader {
|
||||
size: u32,
|
||||
checksum: u32,
|
||||
}
|
||||
|
||||
const RECORD_HEADER_LEN: usize = 8;
|
||||
|
||||
impl RecordHeader {
|
||||
fn encode(&self) -> BytesMut {
|
||||
let mut buf = BytesMut::with_capacity(RECORD_HEADER_LEN);
|
||||
buf.put_u32(self.size);
|
||||
buf.put_u32(self.checksum);
|
||||
buf
|
||||
}
|
||||
|
||||
fn decode(mut buf: &[u8]) -> Self {
|
||||
assert!(buf.len() == RECORD_HEADER_LEN, "invalid header");
|
||||
Self {
|
||||
size: buf.get_u32(),
|
||||
checksum: buf.get_u32(),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Debug, thiserror::Error)]
|
||||
pub enum ManifestLoadError {
|
||||
#[error("manifest header is corrupted")]
|
||||
CorruptedManifestHeader,
|
||||
#[error("unsupported manifest version: got {0}, expected {1}")]
|
||||
UnsupportedVersion(u64, u64),
|
||||
#[error("error when decoding record: {0}")]
|
||||
DecodeRecord(serde_json::Error),
|
||||
#[error("I/O error: {0}")]
|
||||
Io(io::Error),
|
||||
}
|
||||
|
||||
#[must_use = "Should check if the manifest is partially corrupted"]
|
||||
pub struct ManifestPartiallyCorrupted(bool);
|
||||
|
||||
impl Manifest {
|
||||
/// Create a new manifest by writing the manifest header and a snapshot record to the given file.
|
||||
pub fn init(file: VirtualFile, snapshot: Snapshot, lsn: Lsn) -> Result<Self> {
|
||||
let mut manifest = Self { file };
|
||||
manifest.append_manifest_header(ManifestHeader {
|
||||
magic_number: MANIFEST_MAGIC_NUMBER,
|
||||
version: MANIFEST_VERSION,
|
||||
})?;
|
||||
manifest.append_operation(Operation::Snapshot(snapshot, lsn))?;
|
||||
Ok(manifest)
|
||||
}
|
||||
|
||||
/// Load a manifest. Returns the manifest and a list of operations. If the manifest is corrupted,
|
||||
/// the bool flag will be set to true and the user is responsible to reconstruct a new manifest and
|
||||
/// backup the current one.
|
||||
pub fn load(
|
||||
mut file: VirtualFile,
|
||||
) -> Result<(Self, Vec<Operation>, ManifestPartiallyCorrupted), ManifestLoadError> {
|
||||
let mut buf = vec![];
|
||||
file.read_to_end(&mut buf).map_err(ManifestLoadError::Io)?;
|
||||
|
||||
// Read manifest header
|
||||
let mut buf = Bytes::from(buf);
|
||||
if buf.remaining() < MANIFEST_HEADER_LEN {
|
||||
return Err(ManifestLoadError::CorruptedManifestHeader);
|
||||
}
|
||||
let header = ManifestHeader::decode(&buf[..MANIFEST_HEADER_LEN]);
|
||||
buf.advance(MANIFEST_HEADER_LEN);
|
||||
if header.version != MANIFEST_VERSION {
|
||||
return Err(ManifestLoadError::UnsupportedVersion(
|
||||
header.version,
|
||||
MANIFEST_VERSION,
|
||||
));
|
||||
}
|
||||
|
||||
// Read operations
|
||||
let mut operations = Vec::new();
|
||||
let corrupted = loop {
|
||||
if buf.remaining() == 0 {
|
||||
break false;
|
||||
}
|
||||
if buf.remaining() < RECORD_HEADER_LEN {
|
||||
warn!("incomplete header when decoding manifest, could be corrupted");
|
||||
break true;
|
||||
}
|
||||
let RecordHeader { size, checksum } = RecordHeader::decode(&buf[..RECORD_HEADER_LEN]);
|
||||
let size = size as usize;
|
||||
buf.advance(RECORD_HEADER_LEN);
|
||||
if buf.remaining() < size {
|
||||
warn!("incomplete data when decoding manifest, could be corrupted");
|
||||
break true;
|
||||
}
|
||||
let data = &buf[..size];
|
||||
if crc32c(data) != checksum {
|
||||
warn!("checksum mismatch when decoding manifest, could be corrupted");
|
||||
break true;
|
||||
}
|
||||
// if the following decode fails, we cannot use the manifest or safely ignore any record.
|
||||
operations.push(serde_json::from_slice(data).map_err(ManifestLoadError::DecodeRecord)?);
|
||||
buf.advance(size);
|
||||
};
|
||||
Ok((
|
||||
Self { file },
|
||||
operations,
|
||||
ManifestPartiallyCorrupted(corrupted),
|
||||
))
|
||||
}
|
||||
|
||||
fn append_data(&mut self, data: &[u8]) -> Result<()> {
|
||||
if data.len() >= u32::MAX as usize {
|
||||
panic!("data too large");
|
||||
}
|
||||
let header = RecordHeader {
|
||||
size: data.len() as u32,
|
||||
checksum: crc32c(data),
|
||||
};
|
||||
let header = header.encode();
|
||||
self.file.write_all(&header)?;
|
||||
self.file.write_all(data)?;
|
||||
self.file.sync_all()?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn append_manifest_header(&mut self, header: ManifestHeader) -> Result<()> {
|
||||
let encoded = header.encode();
|
||||
self.file.write_all(&encoded)?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Add an operation to the manifest. The operation will be appended to the end of the file,
|
||||
/// and the file will fsync.
|
||||
pub fn append_operation(&mut self, operation: Operation) -> Result<()> {
|
||||
let encoded = Vec::from(serde_json::to_string(&operation)?);
|
||||
self.append_data(&encoded)
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use std::fs::OpenOptions;
|
||||
|
||||
use crate::repository::Key;
|
||||
|
||||
use super::*;
|
||||
|
||||
#[test]
|
||||
fn test_read_manifest() {
|
||||
let testdir = crate::config::PageServerConf::test_repo_dir("test_read_manifest");
|
||||
std::fs::create_dir_all(&testdir).unwrap();
|
||||
let file = VirtualFile::create(&testdir.join("MANIFEST")).unwrap();
|
||||
let layer1 = PersistentLayerDesc::new_test(Key::from_i128(0)..Key::from_i128(233));
|
||||
let layer2 = PersistentLayerDesc::new_test(Key::from_i128(233)..Key::from_i128(2333));
|
||||
let layer3 = PersistentLayerDesc::new_test(Key::from_i128(2333)..Key::from_i128(23333));
|
||||
let layer4 = PersistentLayerDesc::new_test(Key::from_i128(23333)..Key::from_i128(233333));
|
||||
|
||||
// Write a manifest with a snapshot and some operations
|
||||
let snapshot = Snapshot {
|
||||
layers: vec![layer1, layer2],
|
||||
};
|
||||
let mut manifest = Manifest::init(file, snapshot.clone(), Lsn::from(0)).unwrap();
|
||||
manifest
|
||||
.append_operation(Operation::Operation(
|
||||
vec![Record::AddLayer(layer3.clone())],
|
||||
Lsn::from(1),
|
||||
))
|
||||
.unwrap();
|
||||
drop(manifest);
|
||||
|
||||
// Open the second time and write
|
||||
let file = VirtualFile::open_with_options(
|
||||
&testdir.join("MANIFEST"),
|
||||
OpenOptions::new()
|
||||
.read(true)
|
||||
.write(true)
|
||||
.create_new(false)
|
||||
.truncate(false),
|
||||
)
|
||||
.unwrap();
|
||||
let (mut manifest, operations, corrupted) = Manifest::load(file).unwrap();
|
||||
assert!(!corrupted.0);
|
||||
assert_eq!(operations.len(), 2);
|
||||
assert_eq!(
|
||||
&operations[0],
|
||||
&Operation::Snapshot(snapshot.clone(), Lsn::from(0))
|
||||
);
|
||||
assert_eq!(
|
||||
&operations[1],
|
||||
&Operation::Operation(vec![Record::AddLayer(layer3.clone())], Lsn::from(1))
|
||||
);
|
||||
manifest
|
||||
.append_operation(Operation::Operation(
|
||||
vec![
|
||||
Record::RemoveLayer(layer3.clone()),
|
||||
Record::AddLayer(layer4.clone()),
|
||||
],
|
||||
Lsn::from(2),
|
||||
))
|
||||
.unwrap();
|
||||
drop(manifest);
|
||||
|
||||
// Open the third time and verify
|
||||
let file = VirtualFile::open_with_options(
|
||||
&testdir.join("MANIFEST"),
|
||||
OpenOptions::new()
|
||||
.read(true)
|
||||
.write(true)
|
||||
.create_new(false)
|
||||
.truncate(false),
|
||||
)
|
||||
.unwrap();
|
||||
let (_manifest, operations, corrupted) = Manifest::load(file).unwrap();
|
||||
assert!(!corrupted.0);
|
||||
assert_eq!(operations.len(), 3);
|
||||
assert_eq!(&operations[0], &Operation::Snapshot(snapshot, Lsn::from(0)));
|
||||
assert_eq!(
|
||||
&operations[1],
|
||||
&Operation::Operation(vec![Record::AddLayer(layer3.clone())], Lsn::from(1))
|
||||
);
|
||||
assert_eq!(
|
||||
&operations[2],
|
||||
&Operation::Operation(
|
||||
vec![Record::RemoveLayer(layer3), Record::AddLayer(layer4)],
|
||||
Lsn::from(2)
|
||||
)
|
||||
);
|
||||
}
|
||||
}
|
||||
@@ -396,9 +396,7 @@ pub async fn delete_timeline(
|
||||
ctx: &RequestContext,
|
||||
) -> Result<(), DeleteTimelineError> {
|
||||
let tenant = get_tenant(tenant_id, true).await?;
|
||||
tenant
|
||||
.prepare_and_schedule_delete_timeline(timeline_id, ctx)
|
||||
.await?;
|
||||
tenant.delete_timeline(timeline_id, ctx).await?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
|
||||
@@ -210,15 +210,13 @@ use chrono::{NaiveDateTime, Utc};
|
||||
pub use download::{is_temp_download_file, list_remote_timelines};
|
||||
use scopeguard::ScopeGuard;
|
||||
|
||||
use std::collections::{HashMap, VecDeque};
|
||||
use std::path::Path;
|
||||
use std::sync::atomic::{AtomicU32, Ordering};
|
||||
use std::sync::{Arc, Mutex};
|
||||
|
||||
use remote_storage::{DownloadError, GenericRemoteStorage, RemotePath};
|
||||
use remote_storage::{DownloadError, GenericRemoteStorage};
|
||||
use std::ops::DerefMut;
|
||||
use tokio::runtime::Runtime;
|
||||
use tracing::{debug, error, info, instrument, warn};
|
||||
use tracing::{debug, error, info, warn};
|
||||
use tracing::{info_span, Instrument};
|
||||
use utils::lsn::Lsn;
|
||||
|
||||
@@ -227,9 +225,7 @@ use crate::metrics::{
|
||||
RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
|
||||
REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
|
||||
};
|
||||
use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
|
||||
use crate::tenant::remote_timeline_client::index::LayerFileMetadata;
|
||||
use crate::tenant::upload_queue::Delete;
|
||||
use crate::{
|
||||
config::PageServerConf,
|
||||
task_mgr,
|
||||
@@ -263,7 +259,7 @@ const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
|
||||
|
||||
pub enum MaybeDeletedIndexPart {
|
||||
IndexPart(IndexPart),
|
||||
Deleted(IndexPart),
|
||||
Deleted,
|
||||
}
|
||||
|
||||
/// Errors that can arise when calling [`RemoteTimelineClient::stop`].
|
||||
@@ -365,42 +361,11 @@ impl RemoteTimelineClient {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Initialize the queue in stopped state. Used in startup path
|
||||
/// to continue deletion operation interrupted by pageserver crash or restart.
|
||||
pub fn init_upload_queue_stopped_to_continue_deletion(
|
||||
&self,
|
||||
index_part: &IndexPart,
|
||||
) -> anyhow::Result<()> {
|
||||
// FIXME: consider newtype for DeletedIndexPart.
|
||||
let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
|
||||
"bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
|
||||
))?;
|
||||
|
||||
{
|
||||
let mut upload_queue = self.upload_queue.lock().unwrap();
|
||||
upload_queue.initialize_with_current_remote_index_part(index_part)?;
|
||||
self.update_remote_physical_size_gauge(Some(index_part));
|
||||
}
|
||||
// also locks upload queue, without dropping the guard above it will be a deadlock
|
||||
self.stop().expect("initialized line above");
|
||||
|
||||
let mut upload_queue = self.upload_queue.lock().unwrap();
|
||||
|
||||
upload_queue
|
||||
.stopped_mut()
|
||||
.expect("stopped above")
|
||||
.deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
pub fn last_uploaded_consistent_lsn(&self) -> Option<Lsn> {
|
||||
match &*self.upload_queue.lock().unwrap() {
|
||||
UploadQueue::Uninitialized => None,
|
||||
UploadQueue::Initialized(q) => Some(q.last_uploaded_consistent_lsn),
|
||||
UploadQueue::Stopped(q) => {
|
||||
Some(q.upload_queue_for_deletion.last_uploaded_consistent_lsn)
|
||||
}
|
||||
UploadQueue::Stopped(q) => Some(q.last_uploaded_consistent_lsn),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -455,7 +420,7 @@ impl RemoteTimelineClient {
|
||||
.await?;
|
||||
|
||||
if index_part.deleted_at.is_some() {
|
||||
Ok(MaybeDeletedIndexPart::Deleted(index_part))
|
||||
Ok(MaybeDeletedIndexPart::Deleted)
|
||||
} else {
|
||||
Ok(MaybeDeletedIndexPart::IndexPart(index_part))
|
||||
}
|
||||
@@ -657,11 +622,7 @@ impl RemoteTimelineClient {
|
||||
|
||||
// schedule the actual deletions
|
||||
for name in names {
|
||||
let op = UploadOp::Delete(Delete {
|
||||
file_kind: RemoteOpFileKind::Layer,
|
||||
layer_file_name: name.clone(),
|
||||
scheduled_from_timeline_delete: false,
|
||||
});
|
||||
let op = UploadOp::Delete(RemoteOpFileKind::Layer, name.clone());
|
||||
self.calls_unfinished_metric_begin(&op);
|
||||
upload_queue.queued_operations.push_back(op);
|
||||
info!("scheduled layer file deletion {}", name.file_name());
|
||||
@@ -678,11 +639,18 @@ impl RemoteTimelineClient {
|
||||
/// Wait for all previously scheduled uploads/deletions to complete
|
||||
///
|
||||
pub async fn wait_completion(self: &Arc<Self>) -> anyhow::Result<()> {
|
||||
let mut receiver = {
|
||||
let (sender, mut receiver) = tokio::sync::watch::channel(());
|
||||
let barrier_op = UploadOp::Barrier(sender);
|
||||
|
||||
{
|
||||
let mut guard = self.upload_queue.lock().unwrap();
|
||||
let upload_queue = guard.initialized_mut()?;
|
||||
self.schedule_barrier(upload_queue)
|
||||
};
|
||||
upload_queue.queued_operations.push_back(barrier_op);
|
||||
// Don't count this kind of operation!
|
||||
|
||||
// Launch the task immediately, if possible
|
||||
self.launch_queued_tasks(upload_queue);
|
||||
}
|
||||
|
||||
if receiver.changed().await.is_err() {
|
||||
anyhow::bail!("wait_completion aborted because upload queue was stopped");
|
||||
@@ -690,22 +658,6 @@ impl RemoteTimelineClient {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn schedule_barrier(
|
||||
self: &Arc<Self>,
|
||||
upload_queue: &mut UploadQueueInitialized,
|
||||
) -> tokio::sync::watch::Receiver<()> {
|
||||
let (sender, receiver) = tokio::sync::watch::channel(());
|
||||
let barrier_op = UploadOp::Barrier(sender);
|
||||
|
||||
upload_queue.queued_operations.push_back(barrier_op);
|
||||
// Don't count this kind of operation!
|
||||
|
||||
// Launch the task immediately, if possible
|
||||
self.launch_queued_tasks(upload_queue);
|
||||
|
||||
receiver
|
||||
}
|
||||
|
||||
/// Set the deleted_at field in the remote index file.
|
||||
///
|
||||
/// This fails if the upload queue has not been `stop()`ed.
|
||||
@@ -713,7 +665,6 @@ impl RemoteTimelineClient {
|
||||
/// The caller is responsible for calling `stop()` AND for waiting
|
||||
/// for any ongoing upload tasks to finish after `stop()` has succeeded.
|
||||
/// Check method [`RemoteTimelineClient::stop`] for details.
|
||||
#[instrument(skip_all)]
|
||||
pub(crate) async fn persist_index_part_with_deleted_flag(
|
||||
self: &Arc<Self>,
|
||||
) -> Result<(), PersistIndexPartWithDeletedFlagError> {
|
||||
@@ -723,7 +674,15 @@ impl RemoteTimelineClient {
|
||||
// We must be in stopped state because otherwise
|
||||
// we can have inprogress index part upload that can overwrite the file
|
||||
// with missing is_deleted flag that we going to set below
|
||||
let stopped = locked.stopped_mut()?;
|
||||
let stopped = match &mut *locked {
|
||||
UploadQueue::Uninitialized => {
|
||||
return Err(anyhow::anyhow!("is not Stopped but Uninitialized").into())
|
||||
}
|
||||
UploadQueue::Initialized(_) => {
|
||||
return Err(anyhow::anyhow!("is not Stopped but Initialized").into())
|
||||
}
|
||||
UploadQueue::Stopped(stopped) => stopped,
|
||||
};
|
||||
|
||||
match stopped.deleted_at {
|
||||
SetDeletedFlagProgress::NotRunning => (), // proceed
|
||||
@@ -737,17 +696,27 @@ impl RemoteTimelineClient {
|
||||
let deleted_at = Utc::now().naive_utc();
|
||||
stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
|
||||
|
||||
let mut index_part = IndexPart::try_from(&stopped.upload_queue_for_deletion)
|
||||
.context("IndexPart serialize")?;
|
||||
let mut index_part = IndexPart::new(
|
||||
stopped.latest_files.clone(),
|
||||
stopped.last_uploaded_consistent_lsn,
|
||||
stopped
|
||||
.latest_metadata
|
||||
.to_bytes()
|
||||
.context("serialize metadata")?,
|
||||
);
|
||||
index_part.deleted_at = Some(deleted_at);
|
||||
index_part
|
||||
};
|
||||
|
||||
let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
|
||||
let mut locked = self_clone.upload_queue.lock().unwrap();
|
||||
let stopped = locked
|
||||
.stopped_mut()
|
||||
.expect("there's no way out of Stopping, and we checked it's Stopping above");
|
||||
let stopped = match &mut *locked {
|
||||
UploadQueue::Uninitialized | UploadQueue::Initialized(_) => unreachable!(
|
||||
"there's no way out of Stopping, and we checked it's Stopping above: {:?}",
|
||||
locked.as_str(),
|
||||
),
|
||||
UploadQueue::Stopped(stopped) => stopped,
|
||||
};
|
||||
stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
|
||||
});
|
||||
|
||||
@@ -782,10 +751,13 @@ impl RemoteTimelineClient {
|
||||
ScopeGuard::into_inner(undo_deleted_at);
|
||||
{
|
||||
let mut locked = self.upload_queue.lock().unwrap();
|
||||
|
||||
let stopped = locked
|
||||
.stopped_mut()
|
||||
.expect("there's no way out of Stopping, and we checked it's Stopping above");
|
||||
let stopped = match &mut *locked {
|
||||
UploadQueue::Uninitialized | UploadQueue::Initialized(_) => unreachable!(
|
||||
"there's no way out of Stopping, and we checked it's Stopping above: {:?}",
|
||||
locked.as_str(),
|
||||
),
|
||||
UploadQueue::Stopped(stopped) => stopped,
|
||||
};
|
||||
stopped.deleted_at = SetDeletedFlagProgress::Successful(
|
||||
index_part_with_deleted_at
|
||||
.deleted_at
|
||||
@@ -796,92 +768,6 @@ impl RemoteTimelineClient {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
|
||||
/// The function deletes layer files one by one, then lists the prefix to see if we leaked something
|
||||
/// deletes leaked files if any and proceeds with deletion of index file at the end.
|
||||
pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
|
||||
debug_assert_current_span_has_tenant_and_timeline_id();
|
||||
|
||||
let (mut receiver, deletions_queued) = {
|
||||
let mut deletions_queued = 0;
|
||||
|
||||
let mut locked = self.upload_queue.lock().unwrap();
|
||||
let stopped = locked.stopped_mut()?;
|
||||
|
||||
if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
|
||||
anyhow::bail!("deleted_at is not set")
|
||||
}
|
||||
|
||||
debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
|
||||
|
||||
stopped
|
||||
.upload_queue_for_deletion
|
||||
.queued_operations
|
||||
.reserve(stopped.upload_queue_for_deletion.latest_files.len());
|
||||
|
||||
// schedule the actual deletions
|
||||
for name in stopped.upload_queue_for_deletion.latest_files.keys() {
|
||||
let op = UploadOp::Delete(Delete {
|
||||
file_kind: RemoteOpFileKind::Layer,
|
||||
layer_file_name: name.clone(),
|
||||
scheduled_from_timeline_delete: true,
|
||||
});
|
||||
self.calls_unfinished_metric_begin(&op);
|
||||
stopped
|
||||
.upload_queue_for_deletion
|
||||
.queued_operations
|
||||
.push_back(op);
|
||||
|
||||
info!("scheduled layer file deletion {}", name.file_name());
|
||||
deletions_queued += 1;
|
||||
}
|
||||
|
||||
self.launch_queued_tasks(&mut stopped.upload_queue_for_deletion);
|
||||
|
||||
(
|
||||
self.schedule_barrier(&mut stopped.upload_queue_for_deletion),
|
||||
deletions_queued,
|
||||
)
|
||||
};
|
||||
|
||||
receiver.changed().await?;
|
||||
|
||||
// Do not delete index part yet, it is needed for possible retry. If we remove it first
|
||||
// and retry will arrive to different pageserver there wont be any traces of it on remote storage
|
||||
let timeline_path = self.conf.timeline_path(&self.timeline_id, &self.tenant_id);
|
||||
let timeline_storage_path = self.conf.remote_path(&timeline_path)?;
|
||||
|
||||
let remaining = self
|
||||
.storage_impl
|
||||
.list_prefixes(Some(&timeline_storage_path))
|
||||
.await?;
|
||||
|
||||
let remaining: Vec<RemotePath> = remaining
|
||||
.into_iter()
|
||||
.filter(|p| p.object_name() != Some(IndexPart::FILE_NAME))
|
||||
.collect();
|
||||
|
||||
if !remaining.is_empty() {
|
||||
warn!(
|
||||
"Found {} files not bound to index_file.json, proceeding with their deletion",
|
||||
remaining.len()
|
||||
);
|
||||
for file in remaining {
|
||||
warn!("Removing {}", file.object_name().unwrap_or_default());
|
||||
self.storage_impl.delete(&file).await?;
|
||||
}
|
||||
}
|
||||
|
||||
let index_file_path = timeline_storage_path.join(Path::new(IndexPart::FILE_NAME));
|
||||
|
||||
debug!("deleting index part");
|
||||
self.storage_impl.delete(&index_file_path).await?;
|
||||
|
||||
info!(deletions_queued, "done deleting, including index_part.json");
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
///
|
||||
/// Pick next tasks from the queue, and start as many of them as possible without violating
|
||||
/// the ordering constraints.
|
||||
@@ -900,7 +786,7 @@ impl RemoteTimelineClient {
|
||||
// have finished.
|
||||
upload_queue.inprogress_tasks.is_empty()
|
||||
}
|
||||
UploadOp::Delete(_) => {
|
||||
UploadOp::Delete(_, _) => {
|
||||
// Wait for preceding uploads to finish. Concurrent deletions are OK, though.
|
||||
upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
|
||||
}
|
||||
@@ -931,7 +817,7 @@ impl RemoteTimelineClient {
|
||||
UploadOp::UploadMetadata(_, _) => {
|
||||
upload_queue.num_inprogress_metadata_uploads += 1;
|
||||
}
|
||||
UploadOp::Delete(_) => {
|
||||
UploadOp::Delete(_, _) => {
|
||||
upload_queue.num_inprogress_deletions += 1;
|
||||
}
|
||||
UploadOp::Barrier(sender) => {
|
||||
@@ -1005,6 +891,7 @@ impl RemoteTimelineClient {
|
||||
unreachable!("we never launch an upload task if the queue is uninitialized, and once it is initialized, we never go back")
|
||||
}
|
||||
}
|
||||
self.calls_unfinished_metric_end(&task.op);
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -1050,16 +937,16 @@ impl RemoteTimelineClient {
|
||||
}
|
||||
res
|
||||
}
|
||||
UploadOp::Delete(delete) => {
|
||||
UploadOp::Delete(metric_file_kind, ref layer_file_name) => {
|
||||
let path = &self
|
||||
.conf
|
||||
.timeline_path(&self.timeline_id, &self.tenant_id)
|
||||
.join(delete.layer_file_name.file_name());
|
||||
.join(layer_file_name.file_name());
|
||||
delete::delete_layer(self.conf, &self.storage_impl, path)
|
||||
.measure_remote_op(
|
||||
self.tenant_id,
|
||||
self.timeline_id,
|
||||
delete.file_kind,
|
||||
*metric_file_kind,
|
||||
RemoteOpKind::Delete,
|
||||
Arc::clone(&self.metrics),
|
||||
)
|
||||
@@ -1125,24 +1012,11 @@ impl RemoteTimelineClient {
|
||||
let mut upload_queue_guard = self.upload_queue.lock().unwrap();
|
||||
let upload_queue = match upload_queue_guard.deref_mut() {
|
||||
UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
|
||||
UploadQueue::Stopped(stopped) => {
|
||||
// Special care is needed for deletions, if it was an earlier deletion (not scheduled from deletion)
|
||||
// then stop() took care of it so we just return.
|
||||
// For deletions that come from delete_all we still want to maintain metrics, launch following tasks, etc.
|
||||
match &task.op {
|
||||
UploadOp::Delete(delete) if delete.scheduled_from_timeline_delete => Some(&mut stopped.upload_queue_for_deletion),
|
||||
_ => None
|
||||
}
|
||||
},
|
||||
UploadQueue::Initialized(qi) => { Some(qi) }
|
||||
};
|
||||
|
||||
let upload_queue = match upload_queue {
|
||||
Some(upload_queue) => upload_queue,
|
||||
None => {
|
||||
UploadQueue::Stopped(_) => {
|
||||
info!("another concurrent task already stopped the queue");
|
||||
return;
|
||||
}
|
||||
}, // nothing to do
|
||||
UploadQueue::Initialized(qi) => { qi }
|
||||
};
|
||||
|
||||
upload_queue.inprogress_tasks.remove(&task.task_id);
|
||||
@@ -1155,7 +1029,7 @@ impl RemoteTimelineClient {
|
||||
upload_queue.num_inprogress_metadata_uploads -= 1;
|
||||
upload_queue.last_uploaded_consistent_lsn = lsn; // XXX monotonicity check?
|
||||
}
|
||||
UploadOp::Delete(_) => {
|
||||
UploadOp::Delete(_, _) => {
|
||||
upload_queue.num_inprogress_deletions -= 1;
|
||||
}
|
||||
UploadOp::Barrier(_) => unreachable!(),
|
||||
@@ -1189,8 +1063,8 @@ impl RemoteTimelineClient {
|
||||
reason: "metadata uploads are tiny",
|
||||
},
|
||||
),
|
||||
UploadOp::Delete(delete) => (
|
||||
delete.file_kind,
|
||||
UploadOp::Delete(file_kind, _) => (
|
||||
*file_kind,
|
||||
RemoteOpKind::Delete,
|
||||
DontTrackSize {
|
||||
reason: "should we track deletes? positive or negative sign?",
|
||||
@@ -1237,36 +1111,32 @@ impl RemoteTimelineClient {
|
||||
info!("another concurrent task already shut down the queue");
|
||||
Ok(())
|
||||
}
|
||||
UploadQueue::Initialized(initialized) => {
|
||||
UploadQueue::Initialized(UploadQueueInitialized {
|
||||
latest_files,
|
||||
latest_metadata,
|
||||
last_uploaded_consistent_lsn,
|
||||
..
|
||||
}) => {
|
||||
info!("shutting down upload queue");
|
||||
|
||||
// Replace the queue with the Stopped state, taking ownership of the old
|
||||
// Initialized queue. We will do some checks on it, and then drop it.
|
||||
let qi = {
|
||||
// Here we preserve working version of the upload queue for possible use during deletions.
|
||||
// In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
|
||||
// but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
|
||||
// Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
|
||||
let upload_queue_for_deletion = UploadQueueInitialized {
|
||||
task_counter: 0,
|
||||
latest_files: initialized.latest_files.clone(),
|
||||
latest_files_changes_since_metadata_upload_scheduled: 0,
|
||||
latest_metadata: initialized.latest_metadata.clone(),
|
||||
last_uploaded_consistent_lsn: initialized.last_uploaded_consistent_lsn,
|
||||
num_inprogress_layer_uploads: 0,
|
||||
num_inprogress_metadata_uploads: 0,
|
||||
num_inprogress_deletions: 0,
|
||||
inprogress_tasks: HashMap::default(),
|
||||
queued_operations: VecDeque::default(),
|
||||
// take or clone what we need
|
||||
let latest_files = std::mem::take(latest_files);
|
||||
let last_uploaded_consistent_lsn = *last_uploaded_consistent_lsn;
|
||||
// this could be Copy
|
||||
let latest_metadata = latest_metadata.clone();
|
||||
|
||||
let stopped = UploadQueueStopped {
|
||||
latest_files,
|
||||
last_uploaded_consistent_lsn,
|
||||
latest_metadata,
|
||||
deleted_at: SetDeletedFlagProgress::NotRunning,
|
||||
};
|
||||
|
||||
let upload_queue = std::mem::replace(
|
||||
&mut *guard,
|
||||
UploadQueue::Stopped(UploadQueueStopped {
|
||||
upload_queue_for_deletion,
|
||||
deleted_at: SetDeletedFlagProgress::NotRunning,
|
||||
}),
|
||||
);
|
||||
let upload_queue =
|
||||
std::mem::replace(&mut *guard, UploadQueue::Stopped(stopped));
|
||||
if let UploadQueue::Initialized(qi) = upload_queue {
|
||||
qi
|
||||
} else {
|
||||
@@ -1274,6 +1144,8 @@ impl RemoteTimelineClient {
|
||||
}
|
||||
};
|
||||
|
||||
assert!(qi.latest_files.is_empty(), "do not use this anymore");
|
||||
|
||||
// consistency check
|
||||
assert_eq!(
|
||||
qi.num_inprogress_layer_uploads
|
||||
@@ -1392,12 +1264,7 @@ mod tests {
|
||||
let harness = TenantHarness::create(test_name)?;
|
||||
let (tenant, ctx) = runtime.block_on(harness.load());
|
||||
// create an empty timeline directory
|
||||
let _ = runtime.block_on(tenant.create_test_timeline(
|
||||
TIMELINE_ID,
|
||||
Lsn(8),
|
||||
DEFAULT_PG_VERSION,
|
||||
&ctx,
|
||||
))?;
|
||||
let _ = tenant.create_test_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION, &ctx)?;
|
||||
|
||||
let remote_fs_dir = harness.conf.workdir.join("remote_fs");
|
||||
std::fs::create_dir_all(remote_fs_dir)?;
|
||||
@@ -1541,7 +1408,7 @@ mod tests {
|
||||
// Download back the index.json, and check that the list of files is correct
|
||||
let index_part = match runtime.block_on(client.download_index_file())? {
|
||||
MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
|
||||
MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
|
||||
MaybeDeletedIndexPart::Deleted => panic!("unexpectedly got deleted index part"),
|
||||
};
|
||||
|
||||
assert_file_list(
|
||||
|
||||
@@ -7,11 +7,9 @@ use std::collections::{HashMap, HashSet};
|
||||
use chrono::NaiveDateTime;
|
||||
use serde::{Deserialize, Serialize};
|
||||
use serde_with::{serde_as, DisplayFromStr};
|
||||
use utils::bin_ser::SerializeError;
|
||||
|
||||
use crate::tenant::metadata::TimelineMetadata;
|
||||
use crate::tenant::storage_layer::LayerFileName;
|
||||
use crate::tenant::upload_queue::UploadQueueInitialized;
|
||||
|
||||
use utils::lsn::Lsn;
|
||||
|
||||
@@ -117,21 +115,6 @@ impl IndexPart {
|
||||
}
|
||||
}
|
||||
|
||||
impl TryFrom<&UploadQueueInitialized> for IndexPart {
|
||||
type Error = SerializeError;
|
||||
|
||||
fn try_from(upload_queue: &UploadQueueInitialized) -> Result<Self, Self::Error> {
|
||||
let disk_consistent_lsn = upload_queue.latest_metadata.disk_consistent_lsn();
|
||||
let metadata_bytes = upload_queue.latest_metadata.to_bytes()?;
|
||||
|
||||
Ok(Self::new(
|
||||
upload_queue.latest_files.clone(),
|
||||
disk_consistent_lsn,
|
||||
metadata_bytes,
|
||||
))
|
||||
}
|
||||
}
|
||||
|
||||
/// Serialized form of [`LayerFileMetadata`].
|
||||
#[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize, Default)]
|
||||
pub struct IndexLayerMetadata {
|
||||
|
||||
@@ -37,7 +37,6 @@ use crate::virtual_file::VirtualFile;
|
||||
use crate::{walrecord, TEMP_FILE_SUFFIX};
|
||||
use crate::{DELTA_FILE_MAGIC, STORAGE_FORMAT_VERSION};
|
||||
use anyhow::{bail, ensure, Context, Result};
|
||||
use once_cell::sync::OnceCell;
|
||||
use pageserver_api::models::{HistoricLayerInfo, LayerAccessKind};
|
||||
use rand::{distributions::Alphanumeric, Rng};
|
||||
use serde::{Deserialize, Serialize};
|
||||
@@ -47,6 +46,7 @@ use std::io::{Seek, SeekFrom};
|
||||
use std::ops::Range;
|
||||
use std::os::unix::fs::FileExt;
|
||||
use std::path::{Path, PathBuf};
|
||||
use std::sync::{RwLock, RwLockReadGuard, RwLockWriteGuard};
|
||||
use tracing::*;
|
||||
|
||||
use utils::{
|
||||
@@ -184,7 +184,7 @@ pub struct DeltaLayer {
|
||||
|
||||
access_stats: LayerAccessStats,
|
||||
|
||||
inner: OnceCell<DeltaLayerInner>,
|
||||
inner: RwLock<DeltaLayerInner>,
|
||||
}
|
||||
|
||||
impl std::fmt::Debug for DeltaLayer {
|
||||
@@ -201,17 +201,21 @@ impl std::fmt::Debug for DeltaLayer {
|
||||
}
|
||||
|
||||
pub struct DeltaLayerInner {
|
||||
/// If false, the fields below have not been loaded into memory yet.
|
||||
loaded: bool,
|
||||
|
||||
// values copied from summary
|
||||
index_start_blk: u32,
|
||||
index_root_blk: u32,
|
||||
|
||||
/// Reader object for reading blocks from the file.
|
||||
file: FileBlockReader<VirtualFile>,
|
||||
/// Reader object for reading blocks from the file. (None if not loaded yet)
|
||||
file: Option<FileBlockReader<VirtualFile>>,
|
||||
}
|
||||
|
||||
impl std::fmt::Debug for DeltaLayerInner {
|
||||
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
|
||||
f.debug_struct("DeltaLayerInner")
|
||||
.field("loaded", &self.loaded)
|
||||
.field("index_start_blk", &self.index_start_blk)
|
||||
.field("index_root_blk", &self.index_root_blk)
|
||||
.finish()
|
||||
@@ -242,7 +246,7 @@ impl Layer for DeltaLayer {
|
||||
inner.index_start_blk, inner.index_root_blk
|
||||
);
|
||||
|
||||
let file = &inner.file;
|
||||
let file = inner.file.as_ref().unwrap();
|
||||
let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
|
||||
inner.index_start_blk,
|
||||
inner.index_root_blk,
|
||||
@@ -311,7 +315,7 @@ impl Layer for DeltaLayer {
|
||||
let inner = self.load(LayerAccessKind::GetValueReconstructData, ctx)?;
|
||||
|
||||
// Scan the page versions backwards, starting from `lsn`.
|
||||
let file = &inner.file;
|
||||
let file = inner.file.as_ref().unwrap();
|
||||
let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
|
||||
inner.index_start_blk,
|
||||
inner.index_root_blk,
|
||||
@@ -496,22 +500,51 @@ impl DeltaLayer {
|
||||
/// Open the underlying file and read the metadata into memory, if it's
|
||||
/// not loaded already.
|
||||
///
|
||||
fn load(&self, access_kind: LayerAccessKind, ctx: &RequestContext) -> Result<&DeltaLayerInner> {
|
||||
fn load(
|
||||
&self,
|
||||
access_kind: LayerAccessKind,
|
||||
ctx: &RequestContext,
|
||||
) -> Result<RwLockReadGuard<DeltaLayerInner>> {
|
||||
self.access_stats
|
||||
.record_access(access_kind, ctx.task_kind());
|
||||
// Quick exit if already loaded
|
||||
self.inner
|
||||
.get_or_try_init(|| self.load_inner())
|
||||
.with_context(|| format!("Failed to load delta layer {}", self.path().display()))
|
||||
loop {
|
||||
// Quick exit if already loaded
|
||||
let inner = self.inner.read().unwrap();
|
||||
if inner.loaded {
|
||||
return Ok(inner);
|
||||
}
|
||||
|
||||
// Need to open the file and load the metadata. Upgrade our lock to
|
||||
// a write lock. (Or rather, release and re-lock in write mode.)
|
||||
drop(inner);
|
||||
let inner = self.inner.write().unwrap();
|
||||
if !inner.loaded {
|
||||
self.load_inner(inner).with_context(|| {
|
||||
format!("Failed to load delta layer {}", self.path().display())
|
||||
})?;
|
||||
} else {
|
||||
// Another thread loaded it while we were not holding the lock.
|
||||
}
|
||||
|
||||
// We now have the file open and loaded. There's no function to do
|
||||
// that in the std library RwLock, so we have to release and re-lock
|
||||
// in read mode. (To be precise, the lock guard was moved in the
|
||||
// above call to `load_inner`, so it's already been released). And
|
||||
// while we do that, another thread could unload again, so we have
|
||||
// to re-check and retry if that happens.
|
||||
}
|
||||
}
|
||||
|
||||
fn load_inner(&self) -> Result<DeltaLayerInner> {
|
||||
fn load_inner(&self, mut inner: RwLockWriteGuard<DeltaLayerInner>) -> Result<()> {
|
||||
let path = self.path();
|
||||
|
||||
let file = VirtualFile::open(&path)
|
||||
.with_context(|| format!("Failed to open file '{}'", path.display()))?;
|
||||
let file = FileBlockReader::new(file);
|
||||
|
||||
// Open the file if it's not open already.
|
||||
if inner.file.is_none() {
|
||||
let file = VirtualFile::open(&path)
|
||||
.with_context(|| format!("Failed to open file '{}'", path.display()))?;
|
||||
inner.file = Some(FileBlockReader::new(file));
|
||||
}
|
||||
let file = inner.file.as_mut().unwrap();
|
||||
let summary_blk = file.read_blk(0)?;
|
||||
let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
|
||||
|
||||
@@ -538,13 +571,13 @@ impl DeltaLayer {
|
||||
}
|
||||
}
|
||||
|
||||
inner.index_start_blk = actual_summary.index_start_blk;
|
||||
inner.index_root_blk = actual_summary.index_root_blk;
|
||||
|
||||
debug!("loaded from {}", &path.display());
|
||||
|
||||
Ok(DeltaLayerInner {
|
||||
file,
|
||||
index_start_blk: actual_summary.index_start_blk,
|
||||
index_root_blk: actual_summary.index_root_blk,
|
||||
})
|
||||
inner.loaded = true;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Create a DeltaLayer struct representing an existing file on disk.
|
||||
@@ -566,7 +599,12 @@ impl DeltaLayer {
|
||||
file_size,
|
||||
),
|
||||
access_stats,
|
||||
inner: once_cell::sync::OnceCell::new(),
|
||||
inner: RwLock::new(DeltaLayerInner {
|
||||
loaded: false,
|
||||
file: None,
|
||||
index_start_blk: 0,
|
||||
index_root_blk: 0,
|
||||
}),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -593,7 +631,12 @@ impl DeltaLayer {
|
||||
metadata.len(),
|
||||
),
|
||||
access_stats: LayerAccessStats::empty_will_record_residence_event_later(),
|
||||
inner: once_cell::sync::OnceCell::new(),
|
||||
inner: RwLock::new(DeltaLayerInner {
|
||||
loaded: false,
|
||||
file: None,
|
||||
index_start_blk: 0,
|
||||
index_root_blk: 0,
|
||||
}),
|
||||
})
|
||||
}
|
||||
|
||||
@@ -757,7 +800,12 @@ impl DeltaLayerWriterInner {
|
||||
metadata.len(),
|
||||
),
|
||||
access_stats: LayerAccessStats::empty_will_record_residence_event_later(),
|
||||
inner: once_cell::sync::OnceCell::new(),
|
||||
inner: RwLock::new(DeltaLayerInner {
|
||||
loaded: false,
|
||||
file: None,
|
||||
index_start_blk,
|
||||
index_root_blk,
|
||||
}),
|
||||
};
|
||||
|
||||
// fsync the file
|
||||
@@ -892,13 +940,13 @@ struct DeltaValueIter<'a> {
|
||||
reader: BlockCursor<Adapter<'a>>,
|
||||
}
|
||||
|
||||
struct Adapter<'a>(&'a DeltaLayerInner);
|
||||
struct Adapter<'a>(RwLockReadGuard<'a, DeltaLayerInner>);
|
||||
|
||||
impl<'a> BlockReader for Adapter<'a> {
|
||||
type BlockLease = PageReadGuard<'static>;
|
||||
|
||||
fn read_blk(&self, blknum: u32) -> Result<Self::BlockLease, std::io::Error> {
|
||||
self.0.file.read_blk(blknum)
|
||||
self.0.file.as_ref().unwrap().read_blk(blknum)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -911,8 +959,8 @@ impl<'a> Iterator for DeltaValueIter<'a> {
|
||||
}
|
||||
|
||||
impl<'a> DeltaValueIter<'a> {
|
||||
fn new(inner: &'a DeltaLayerInner) -> Result<Self> {
|
||||
let file = &inner.file;
|
||||
fn new(inner: RwLockReadGuard<'a, DeltaLayerInner>) -> Result<Self> {
|
||||
let file = inner.file.as_ref().unwrap();
|
||||
let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
|
||||
inner.index_start_blk,
|
||||
inner.index_root_blk,
|
||||
@@ -985,8 +1033,8 @@ impl Iterator for DeltaKeyIter {
|
||||
}
|
||||
|
||||
impl<'a> DeltaKeyIter {
|
||||
fn new(inner: &'a DeltaLayerInner) -> Result<Self> {
|
||||
let file = &inner.file;
|
||||
fn new(inner: RwLockReadGuard<'a, DeltaLayerInner>) -> Result<Self> {
|
||||
let file = inner.file.as_ref().unwrap();
|
||||
let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
|
||||
inner.index_start_blk,
|
||||
inner.index_root_blk,
|
||||
@@ -1026,21 +1074,3 @@ impl<'a> DeltaKeyIter {
|
||||
Ok(iter)
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod test {
|
||||
use super::DeltaKeyIter;
|
||||
use super::DeltaLayer;
|
||||
use super::DeltaValueIter;
|
||||
|
||||
// We will soon need the iters to be send in the compaction code.
|
||||
// Cf https://github.com/neondatabase/neon/pull/4462#issuecomment-1587398883
|
||||
// Cf https://github.com/neondatabase/neon/issues/4471
|
||||
#[test]
|
||||
fn is_send() {
|
||||
fn assert_send<T: Send>() {}
|
||||
assert_send::<DeltaLayer>();
|
||||
assert_send::<DeltaValueIter>();
|
||||
assert_send::<DeltaKeyIter>();
|
||||
}
|
||||
}
|
||||
|
||||
@@ -9,12 +9,10 @@ use crate::{context::RequestContext, repository::Key};
|
||||
|
||||
use super::{DeltaFileName, ImageFileName, LayerFileName};
|
||||
|
||||
use serde::{Deserialize, Serialize};
|
||||
|
||||
/// A unique identifier of a persistent layer. This is different from `LayerDescriptor`, which is only used in the
|
||||
/// benchmarks. This struct contains all necessary information to find the image / delta layer. It also provides
|
||||
/// a unified way to generate layer information like file name.
|
||||
#[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
|
||||
#[derive(Debug, PartialEq, Eq, Clone)]
|
||||
pub struct PersistentLayerDesc {
|
||||
pub tenant_id: TenantId,
|
||||
pub timeline_id: TimelineId,
|
||||
@@ -52,19 +50,6 @@ impl PersistentLayerDesc {
|
||||
self.filename().file_name()
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
pub fn new_test(key_range: Range<Key>) -> Self {
|
||||
Self {
|
||||
tenant_id: TenantId::generate(),
|
||||
timeline_id: TimelineId::generate(),
|
||||
key_range,
|
||||
lsn_range: Lsn(0)..Lsn(1),
|
||||
is_delta: false,
|
||||
is_incremental: false,
|
||||
file_size: 0,
|
||||
}
|
||||
}
|
||||
|
||||
pub fn new_img(
|
||||
tenant_id: TenantId,
|
||||
timeline_id: TimelineId,
|
||||
|
||||
@@ -84,14 +84,9 @@ use super::remote_timeline_client::RemoteTimelineClient;
|
||||
use super::storage_layer::{DeltaLayer, ImageLayer, Layer, LayerAccessStatsReset};
|
||||
|
||||
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
|
||||
pub(super) enum FlushLoopState {
|
||||
enum FlushLoopState {
|
||||
NotStarted,
|
||||
Running {
|
||||
#[cfg(test)]
|
||||
expect_initdb_optimization: bool,
|
||||
#[cfg(test)]
|
||||
initdb_optimization_count: usize,
|
||||
},
|
||||
Running,
|
||||
Exited,
|
||||
}
|
||||
|
||||
@@ -188,7 +183,7 @@ pub struct Timeline {
|
||||
write_lock: Mutex<()>,
|
||||
|
||||
/// Used to avoid multiple `flush_loop` tasks running
|
||||
pub(super) flush_loop_state: Mutex<FlushLoopState>,
|
||||
flush_loop_state: Mutex<FlushLoopState>,
|
||||
|
||||
/// layer_flush_start_tx can be used to wake up the layer-flushing task.
|
||||
/// The value is a counter, incremented every time a new flush cycle is requested.
|
||||
@@ -244,7 +239,7 @@ pub struct Timeline {
|
||||
|
||||
/// Prevent two tasks from deleting the timeline at the same time. If held, the
|
||||
/// timeline is being deleted. If 'true', the timeline has already been deleted.
|
||||
pub delete_lock: Arc<tokio::sync::Mutex<bool>>,
|
||||
pub delete_lock: tokio::sync::Mutex<bool>,
|
||||
|
||||
eviction_task_timeline_state: tokio::sync::Mutex<EvictionTaskTimelineState>,
|
||||
|
||||
@@ -820,7 +815,8 @@ impl Timeline {
|
||||
// above. Rewrite it.
|
||||
let layer_removal_cs = Arc::new(self.layer_removal_cs.clone().lock_owned().await);
|
||||
// Is the timeline being deleted?
|
||||
if self.is_stopping() {
|
||||
let state = *self.state.borrow();
|
||||
if state == TimelineState::Stopping {
|
||||
return Err(anyhow::anyhow!("timeline is Stopping").into());
|
||||
}
|
||||
|
||||
@@ -907,37 +903,35 @@ impl Timeline {
|
||||
/// safekeepers to regard pageserver as caught up and suspend activity.
|
||||
pub fn check_checkpoint_distance(self: &Arc<Timeline>) -> anyhow::Result<()> {
|
||||
let last_lsn = self.get_last_record_lsn();
|
||||
let open_layer_size = {
|
||||
let layers = self.layers.read().unwrap();
|
||||
let Some(open_layer) = layers.open_layer.as_ref() else {
|
||||
return Ok(());
|
||||
};
|
||||
open_layer.size()?
|
||||
};
|
||||
let last_freeze_at = self.last_freeze_at.load();
|
||||
let last_freeze_ts = *(self.last_freeze_ts.read().unwrap());
|
||||
let distance = last_lsn.widening_sub(last_freeze_at);
|
||||
// Checkpointing the open layer can be triggered by layer size or LSN range.
|
||||
// S3 has a 5 GB limit on the size of one upload (without multi-part upload), and
|
||||
// we want to stay below that with a big margin. The LSN distance determines how
|
||||
// much WAL the safekeepers need to store.
|
||||
if distance >= self.get_checkpoint_distance().into()
|
||||
|| open_layer_size > self.get_checkpoint_distance()
|
||||
|| (distance > 0 && last_freeze_ts.elapsed() >= self.get_checkpoint_timeout())
|
||||
{
|
||||
info!(
|
||||
"check_checkpoint_distance {}, layer size {}, elapsed since last flush {:?}",
|
||||
distance,
|
||||
open_layer_size,
|
||||
last_freeze_ts.elapsed()
|
||||
);
|
||||
let layers = self.layers.read().unwrap();
|
||||
if let Some(open_layer) = &layers.open_layer {
|
||||
let open_layer_size = open_layer.size()?;
|
||||
drop(layers);
|
||||
let last_freeze_at = self.last_freeze_at.load();
|
||||
let last_freeze_ts = *(self.last_freeze_ts.read().unwrap());
|
||||
let distance = last_lsn.widening_sub(last_freeze_at);
|
||||
// Checkpointing the open layer can be triggered by layer size or LSN range.
|
||||
// S3 has a 5 GB limit on the size of one upload (without multi-part upload), and
|
||||
// we want to stay below that with a big margin. The LSN distance determines how
|
||||
// much WAL the safekeepers need to store.
|
||||
if distance >= self.get_checkpoint_distance().into()
|
||||
|| open_layer_size > self.get_checkpoint_distance()
|
||||
|| (distance > 0 && last_freeze_ts.elapsed() >= self.get_checkpoint_timeout())
|
||||
{
|
||||
info!(
|
||||
"check_checkpoint_distance {}, layer size {}, elapsed since last flush {:?}",
|
||||
distance,
|
||||
open_layer_size,
|
||||
last_freeze_ts.elapsed()
|
||||
);
|
||||
|
||||
self.freeze_inmem_layer(true);
|
||||
self.last_freeze_at.store(last_lsn);
|
||||
*(self.last_freeze_ts.write().unwrap()) = Instant::now();
|
||||
self.freeze_inmem_layer(true);
|
||||
self.last_freeze_at.store(last_lsn);
|
||||
*(self.last_freeze_ts.write().unwrap()) = Instant::now();
|
||||
|
||||
// Wake up the layer flusher
|
||||
self.flush_frozen_layers();
|
||||
// Wake up the layer flusher
|
||||
self.flush_frozen_layers();
|
||||
}
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
@@ -961,17 +955,14 @@ impl Timeline {
|
||||
(st, TimelineState::Loading) => {
|
||||
error!("ignoring transition from {st:?} into Loading state");
|
||||
}
|
||||
(TimelineState::Broken { .. }, new_state) => {
|
||||
error!("Ignoring state update {new_state:?} for broken timeline");
|
||||
(TimelineState::Broken, _) => {
|
||||
error!("Ignoring state update {new_state:?} for broken tenant");
|
||||
}
|
||||
(TimelineState::Stopping, TimelineState::Active) => {
|
||||
error!("Not activating a Stopping timeline");
|
||||
}
|
||||
(_, new_state) => {
|
||||
if matches!(
|
||||
new_state,
|
||||
TimelineState::Stopping | TimelineState::Broken { .. }
|
||||
) {
|
||||
if matches!(new_state, TimelineState::Stopping | TimelineState::Broken) {
|
||||
// drop the copmletion guard, if any; it might be holding off the completion
|
||||
// forever needlessly
|
||||
self.initial_logical_size_attempt
|
||||
@@ -984,31 +975,14 @@ impl Timeline {
|
||||
}
|
||||
}
|
||||
|
||||
pub fn set_broken(&self, reason: String) {
|
||||
let backtrace_str: String = format!("{}", std::backtrace::Backtrace::force_capture());
|
||||
let broken_state = TimelineState::Broken {
|
||||
reason,
|
||||
backtrace: backtrace_str,
|
||||
};
|
||||
self.set_state(broken_state)
|
||||
}
|
||||
|
||||
pub fn current_state(&self) -> TimelineState {
|
||||
self.state.borrow().clone()
|
||||
}
|
||||
|
||||
pub fn is_broken(&self) -> bool {
|
||||
matches!(&*self.state.borrow(), TimelineState::Broken { .. })
|
||||
*self.state.borrow()
|
||||
}
|
||||
|
||||
pub fn is_active(&self) -> bool {
|
||||
self.current_state() == TimelineState::Active
|
||||
}
|
||||
|
||||
pub fn is_stopping(&self) -> bool {
|
||||
self.current_state() == TimelineState::Stopping
|
||||
}
|
||||
|
||||
pub fn subscribe_for_state_updates(&self) -> watch::Receiver<TimelineState> {
|
||||
self.state.subscribe()
|
||||
}
|
||||
@@ -1019,7 +993,7 @@ impl Timeline {
|
||||
) -> Result<(), TimelineState> {
|
||||
let mut receiver = self.state.subscribe();
|
||||
loop {
|
||||
let current_state = receiver.borrow().clone();
|
||||
let current_state = *receiver.borrow_and_update();
|
||||
match current_state {
|
||||
TimelineState::Loading => {
|
||||
receiver
|
||||
@@ -1486,7 +1460,7 @@ impl Timeline {
|
||||
eviction_task_timeline_state: tokio::sync::Mutex::new(
|
||||
EvictionTaskTimelineState::default(),
|
||||
),
|
||||
delete_lock: Arc::new(tokio::sync::Mutex::new(false)),
|
||||
delete_lock: tokio::sync::Mutex::new(false),
|
||||
|
||||
initial_logical_size_can_start,
|
||||
initial_logical_size_attempt: Mutex::new(initial_logical_size_attempt),
|
||||
@@ -1504,7 +1478,7 @@ impl Timeline {
|
||||
let mut flush_loop_state = self.flush_loop_state.lock().unwrap();
|
||||
match *flush_loop_state {
|
||||
FlushLoopState::NotStarted => (),
|
||||
FlushLoopState::Running { .. } => {
|
||||
FlushLoopState::Running => {
|
||||
info!(
|
||||
"skipping attempt to start flush_loop twice {}/{}",
|
||||
self.tenant_id, self.timeline_id
|
||||
@@ -1524,12 +1498,6 @@ impl Timeline {
|
||||
let self_clone = Arc::clone(self);
|
||||
|
||||
info!("spawning flush loop");
|
||||
*flush_loop_state = FlushLoopState::Running {
|
||||
#[cfg(test)]
|
||||
expect_initdb_optimization: false,
|
||||
#[cfg(test)]
|
||||
initdb_optimization_count: 0,
|
||||
};
|
||||
task_mgr::spawn(
|
||||
task_mgr::BACKGROUND_RUNTIME.handle(),
|
||||
task_mgr::TaskKind::LayerFlushTask,
|
||||
@@ -1541,12 +1509,14 @@ impl Timeline {
|
||||
let background_ctx = RequestContext::todo_child(TaskKind::LayerFlushTask, DownloadBehavior::Error);
|
||||
self_clone.flush_loop(layer_flush_start_rx, &background_ctx).await;
|
||||
let mut flush_loop_state = self_clone.flush_loop_state.lock().unwrap();
|
||||
assert!(matches!(*flush_loop_state, FlushLoopState::Running{ ..}));
|
||||
assert_eq!(*flush_loop_state, FlushLoopState::Running);
|
||||
*flush_loop_state = FlushLoopState::Exited;
|
||||
Ok(())
|
||||
}
|
||||
.instrument(info_span!(parent: None, "layer flush task", tenant = %self.tenant_id, timeline = %self.timeline_id))
|
||||
);
|
||||
|
||||
*flush_loop_state = FlushLoopState::Running;
|
||||
}
|
||||
|
||||
/// Creates and starts the wal receiver.
|
||||
@@ -1594,16 +1564,9 @@ impl Timeline {
|
||||
));
|
||||
}
|
||||
|
||||
///
|
||||
/// Initialize with an empty layer map. Used when creating a new timeline.
|
||||
///
|
||||
pub(super) fn init_empty_layer_map(&self, start_lsn: Lsn) {
|
||||
let mut layers = self.layers.write().unwrap();
|
||||
layers.next_open_layer_at = Some(Lsn(start_lsn.0));
|
||||
}
|
||||
|
||||
///
|
||||
/// Scan the timeline directory to populate the layer map.
|
||||
/// Returns all timeline-related files that were found and loaded.
|
||||
///
|
||||
pub(super) fn load_layer_map(&self, disk_consistent_lsn: Lsn) -> anyhow::Result<()> {
|
||||
let mut layers = self.layers.write().unwrap();
|
||||
@@ -2138,11 +2101,11 @@ impl Timeline {
|
||||
loop {
|
||||
match timeline_state_updates.changed().await {
|
||||
Ok(()) => {
|
||||
let new_state = timeline_state_updates.borrow().clone();
|
||||
let new_state = *timeline_state_updates.borrow();
|
||||
match new_state {
|
||||
// we're running this job for active timelines only
|
||||
TimelineState::Active => continue,
|
||||
TimelineState::Broken { .. }
|
||||
TimelineState::Broken
|
||||
| TimelineState::Stopping
|
||||
| TimelineState::Loading => {
|
||||
break format!("aborted because timeline became inactive (new state: {new_state:?})")
|
||||
@@ -2403,17 +2366,10 @@ impl Timeline {
|
||||
}
|
||||
ValueReconstructResult::Missing => {
|
||||
return Err(layer_traversal_error(
|
||||
if cfg!(test) {
|
||||
format!(
|
||||
"could not find data for key {} at LSN {}, for request at LSN {}\n{}",
|
||||
key, cont_lsn, request_lsn, std::backtrace::Backtrace::force_capture(),
|
||||
)
|
||||
} else {
|
||||
format!(
|
||||
"could not find data for key {} at LSN {}, for request at LSN {}",
|
||||
key, cont_lsn, request_lsn
|
||||
)
|
||||
},
|
||||
format!(
|
||||
"could not find data for key {} at LSN {}, for request at LSN {}",
|
||||
key, cont_lsn, request_lsn
|
||||
),
|
||||
traversal_path,
|
||||
));
|
||||
}
|
||||
@@ -2669,21 +2625,16 @@ impl Timeline {
|
||||
let last_record_lsn = self.get_last_record_lsn();
|
||||
ensure!(
|
||||
lsn > last_record_lsn,
|
||||
"cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})\n{}",
|
||||
"cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})",
|
||||
lsn,
|
||||
last_record_lsn,
|
||||
std::backtrace::Backtrace::force_capture(),
|
||||
);
|
||||
|
||||
// Do we have a layer open for writing already?
|
||||
let layer;
|
||||
if let Some(open_layer) = &layers.open_layer {
|
||||
if open_layer.get_lsn_range().start > lsn {
|
||||
bail!(
|
||||
"unexpected open layer in the future: open layers starts at {}, write lsn {}",
|
||||
open_layer.get_lsn_range().start,
|
||||
lsn
|
||||
);
|
||||
bail!("unexpected open layer in the future");
|
||||
}
|
||||
|
||||
layer = Arc::clone(open_layer);
|
||||
@@ -2811,7 +2762,7 @@ impl Timeline {
|
||||
let mut my_flush_request = 0;
|
||||
|
||||
let flush_loop_state = { *self.flush_loop_state.lock().unwrap() };
|
||||
if !matches!(flush_loop_state, FlushLoopState::Running { .. }) {
|
||||
if flush_loop_state != FlushLoopState::Running {
|
||||
anyhow::bail!("cannot flush frozen layers when flush_loop is not running, state is {flush_loop_state:?}")
|
||||
}
|
||||
|
||||
@@ -2861,18 +2812,6 @@ impl Timeline {
|
||||
let lsn_range = frozen_layer.get_lsn_range();
|
||||
let layer_paths_to_upload =
|
||||
if lsn_range.start == self.initdb_lsn && lsn_range.end == Lsn(self.initdb_lsn.0 + 1) {
|
||||
#[cfg(test)]
|
||||
match &mut *self.flush_loop_state.lock().unwrap() {
|
||||
FlushLoopState::NotStarted | FlushLoopState::Exited => {
|
||||
panic!("flush loop not running")
|
||||
}
|
||||
FlushLoopState::Running {
|
||||
initdb_optimization_count,
|
||||
..
|
||||
} => {
|
||||
*initdb_optimization_count += 1;
|
||||
}
|
||||
}
|
||||
// Note: The 'ctx' in use here has DownloadBehavior::Error. We should not
|
||||
// require downloading anything during initial import.
|
||||
let (partitioning, _lsn) = self
|
||||
@@ -2881,18 +2820,6 @@ impl Timeline {
|
||||
self.create_image_layers(&partitioning, self.initdb_lsn, true, ctx)
|
||||
.await?
|
||||
} else {
|
||||
#[cfg(test)]
|
||||
match &mut *self.flush_loop_state.lock().unwrap() {
|
||||
FlushLoopState::NotStarted | FlushLoopState::Exited => {
|
||||
panic!("flush loop not running")
|
||||
}
|
||||
FlushLoopState::Running {
|
||||
expect_initdb_optimization,
|
||||
..
|
||||
} => {
|
||||
assert!(!*expect_initdb_optimization, "expected initdb optimization");
|
||||
}
|
||||
}
|
||||
// normal case, write out a L0 delta layer file.
|
||||
let this = self.clone();
|
||||
let frozen_layer = frozen_layer.clone();
|
||||
@@ -3865,7 +3792,9 @@ impl Timeline {
|
||||
|
||||
let layer_removal_cs = Arc::new(self.layer_removal_cs.clone().lock_owned().await);
|
||||
// Is the timeline being deleted?
|
||||
if self.is_stopping() {
|
||||
let state = *self.state.borrow();
|
||||
if state == TimelineState::Stopping {
|
||||
// there's a global allowed_error for this
|
||||
anyhow::bail!("timeline is Stopping");
|
||||
}
|
||||
|
||||
|
||||
@@ -153,7 +153,7 @@ pub(super) async fn connection_manager_loop_step(
|
||||
match new_state {
|
||||
// we're already active as walreceiver, no need to reactivate
|
||||
TimelineState::Active => continue,
|
||||
TimelineState::Broken { .. } | TimelineState::Stopping => {
|
||||
TimelineState::Broken | TimelineState::Stopping => {
|
||||
debug!("timeline entered terminal state {new_state:?}, stopping wal connection manager loop");
|
||||
return ControlFlow::Break(());
|
||||
}
|
||||
@@ -1324,8 +1324,7 @@ mod tests {
|
||||
async fn dummy_state(harness: &TenantHarness<'_>) -> ConnectionManagerState {
|
||||
let (tenant, ctx) = harness.load().await;
|
||||
let timeline = tenant
|
||||
.create_test_timeline(TIMELINE_ID, Lsn(0x8), crate::DEFAULT_PG_VERSION, &ctx)
|
||||
.await
|
||||
.create_test_timeline(TIMELINE_ID, Lsn(0), crate::DEFAULT_PG_VERSION, &ctx)
|
||||
.expect("Failed to create an empty timeline for dummy wal connection manager");
|
||||
|
||||
ConnectionManagerState {
|
||||
|
||||
@@ -76,12 +76,6 @@ pub(crate) struct UploadQueueInitialized {
|
||||
pub(crate) queued_operations: VecDeque<UploadOp>,
|
||||
}
|
||||
|
||||
impl UploadQueueInitialized {
|
||||
pub(super) fn no_pending_work(&self) -> bool {
|
||||
self.inprogress_tasks.is_empty() && self.queued_operations.is_empty()
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Clone, Copy)]
|
||||
pub(super) enum SetDeletedFlagProgress {
|
||||
NotRunning,
|
||||
@@ -90,7 +84,9 @@ pub(super) enum SetDeletedFlagProgress {
|
||||
}
|
||||
|
||||
pub(super) struct UploadQueueStopped {
|
||||
pub(super) upload_queue_for_deletion: UploadQueueInitialized,
|
||||
pub(super) latest_files: HashMap<LayerFileName, LayerFileMetadata>,
|
||||
pub(super) last_uploaded_consistent_lsn: Lsn,
|
||||
pub(super) latest_metadata: TimelineMetadata,
|
||||
pub(super) deleted_at: SetDeletedFlagProgress,
|
||||
}
|
||||
|
||||
@@ -191,15 +187,6 @@ impl UploadQueue {
|
||||
UploadQueue::Initialized(x) => Ok(x),
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn stopped_mut(&mut self) -> anyhow::Result<&mut UploadQueueStopped> {
|
||||
match self {
|
||||
UploadQueue::Initialized(_) | UploadQueue::Uninitialized => {
|
||||
anyhow::bail!("queue is in state {}", self.as_str())
|
||||
}
|
||||
UploadQueue::Stopped(stopped) => Ok(stopped),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// An in-progress upload or delete task.
|
||||
@@ -212,13 +199,6 @@ pub(crate) struct UploadTask {
|
||||
pub(crate) op: UploadOp,
|
||||
}
|
||||
|
||||
#[derive(Debug)]
|
||||
pub(crate) struct Delete {
|
||||
pub(crate) file_kind: RemoteOpFileKind,
|
||||
pub(crate) layer_file_name: LayerFileName,
|
||||
pub(crate) scheduled_from_timeline_delete: bool,
|
||||
}
|
||||
|
||||
#[derive(Debug)]
|
||||
pub(crate) enum UploadOp {
|
||||
/// Upload a layer file
|
||||
@@ -227,8 +207,8 @@ pub(crate) enum UploadOp {
|
||||
/// Upload the metadata file
|
||||
UploadMetadata(IndexPart, Lsn),
|
||||
|
||||
/// Delete a layer file
|
||||
Delete(Delete),
|
||||
/// Delete a file.
|
||||
Delete(RemoteOpFileKind, LayerFileName),
|
||||
|
||||
/// Barrier. When the barrier operation is reached,
|
||||
Barrier(tokio::sync::watch::Sender<()>),
|
||||
@@ -246,12 +226,7 @@ impl std::fmt::Display for UploadOp {
|
||||
)
|
||||
}
|
||||
UploadOp::UploadMetadata(_, lsn) => write!(f, "UploadMetadata(lsn: {})", lsn),
|
||||
UploadOp::Delete(delete) => write!(
|
||||
f,
|
||||
"Delete(path: {}, scheduled_from_timeline_delete: {})",
|
||||
delete.layer_file_name.file_name(),
|
||||
delete.scheduled_from_timeline_delete
|
||||
),
|
||||
UploadOp::Delete(_, path) => write!(f, "Delete({})", path.file_name()),
|
||||
UploadOp::Barrier(_) => write!(f, "Barrier"),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -1171,6 +1171,7 @@ impl<'a> WalIngest<'a> {
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
use crate::pgdatadir_mapping::create_test_timeline;
|
||||
use crate::tenant::harness::*;
|
||||
use crate::tenant::Timeline;
|
||||
use postgres_ffi::v14::xlog_utils::SIZEOF_CHECKPOINT;
|
||||
@@ -1208,9 +1209,7 @@ mod tests {
|
||||
#[tokio::test]
|
||||
async fn test_relsize() -> Result<()> {
|
||||
let (tenant, ctx) = TenantHarness::create("test_relsize")?.load().await;
|
||||
let tline = tenant
|
||||
.create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
|
||||
.await?;
|
||||
let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
|
||||
let mut walingest = init_walingest_test(&tline, &ctx).await?;
|
||||
|
||||
let mut m = tline.begin_modification(Lsn(0x20));
|
||||
@@ -1429,9 +1428,7 @@ mod tests {
|
||||
#[tokio::test]
|
||||
async fn test_drop_extend() -> Result<()> {
|
||||
let (tenant, ctx) = TenantHarness::create("test_drop_extend")?.load().await;
|
||||
let tline = tenant
|
||||
.create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
|
||||
.await?;
|
||||
let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
|
||||
let mut walingest = init_walingest_test(&tline, &ctx).await?;
|
||||
|
||||
let mut m = tline.begin_modification(Lsn(0x20));
|
||||
@@ -1500,9 +1497,7 @@ mod tests {
|
||||
#[tokio::test]
|
||||
async fn test_truncate_extend() -> Result<()> {
|
||||
let (tenant, ctx) = TenantHarness::create("test_truncate_extend")?.load().await;
|
||||
let tline = tenant
|
||||
.create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
|
||||
.await?;
|
||||
let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
|
||||
let mut walingest = init_walingest_test(&tline, &ctx).await?;
|
||||
|
||||
// Create a 20 MB relation (the size is arbitrary)
|
||||
@@ -1642,9 +1637,7 @@ mod tests {
|
||||
#[tokio::test]
|
||||
async fn test_large_rel() -> Result<()> {
|
||||
let (tenant, ctx) = TenantHarness::create("test_large_rel")?.load().await;
|
||||
let tline = tenant
|
||||
.create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
|
||||
.await?;
|
||||
let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
|
||||
let mut walingest = init_walingest_test(&tline, &ctx).await?;
|
||||
|
||||
let mut lsn = 0x10;
|
||||
|
||||
56
poetry.lock
generated
56
poetry.lock
generated
@@ -1,4 +1,4 @@
|
||||
# This file is automatically @generated by Poetry and should not be changed by hand.
|
||||
# This file is automatically @generated by Poetry 1.4.2 and should not be changed by hand.
|
||||
|
||||
[[package]]
|
||||
name = "aiohttp"
|
||||
@@ -855,31 +855,35 @@ files = [
|
||||
|
||||
[[package]]
|
||||
name = "cryptography"
|
||||
version = "41.0.0"
|
||||
version = "39.0.1"
|
||||
description = "cryptography is a package which provides cryptographic recipes and primitives to Python developers."
|
||||
category = "main"
|
||||
optional = false
|
||||
python-versions = ">=3.7"
|
||||
python-versions = ">=3.6"
|
||||
files = [
|
||||
{file = "cryptography-41.0.0-cp37-abi3-macosx_10_12_universal2.whl", hash = "sha256:3c5ef25d060c80d6d9f7f9892e1d41bb1c79b78ce74805b8cb4aa373cb7d5ec8"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-macosx_10_12_x86_64.whl", hash = "sha256:8362565b3835ceacf4dc8f3b56471a2289cf51ac80946f9087e66dc283a810e0"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3680248309d340fda9611498a5319b0193a8dbdb73586a1acf8109d06f25b92d"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:84a165379cb9d411d58ed739e4af3396e544eac190805a54ba2e0322feb55c46"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:4ab14d567f7bbe7f1cdff1c53d5324ed4d3fc8bd17c481b395db224fb405c237"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:9f65e842cb02550fac96536edb1d17f24c0a338fd84eaf582be25926e993dde4"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:b7f2f5c525a642cecad24ee8670443ba27ac1fab81bba4cc24c7b6b41f2d0c75"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:7d92f0248d38faa411d17f4107fc0bce0c42cae0b0ba5415505df72d751bf62d"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-win32.whl", hash = "sha256:34d405ea69a8b34566ba3dfb0521379b210ea5d560fafedf9f800a9a94a41928"},
|
||||
{file = "cryptography-41.0.0-cp37-abi3-win_amd64.whl", hash = "sha256:344c6de9f8bda3c425b3a41b319522ba3208551b70c2ae00099c205f0d9fd3be"},
|
||||
{file = "cryptography-41.0.0-pp38-pypy38_pp73-macosx_10_12_x86_64.whl", hash = "sha256:88ff107f211ea696455ea8d911389f6d2b276aabf3231bf72c8853d22db755c5"},
|
||||
{file = "cryptography-41.0.0-pp38-pypy38_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:b846d59a8d5a9ba87e2c3d757ca019fa576793e8758174d3868aecb88d6fc8eb"},
|
||||
{file = "cryptography-41.0.0-pp38-pypy38_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:f5d0bf9b252f30a31664b6f64432b4730bb7038339bd18b1fafe129cfc2be9be"},
|
||||
{file = "cryptography-41.0.0-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:5c1f7293c31ebc72163a9a0df246f890d65f66b4a40d9ec80081969ba8c78cc9"},
|
||||
{file = "cryptography-41.0.0-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:bf8fc66012ca857d62f6a347007e166ed59c0bc150cefa49f28376ebe7d992a2"},
|
||||
{file = "cryptography-41.0.0-pp39-pypy39_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:a4fc68d1c5b951cfb72dfd54702afdbbf0fb7acdc9b7dc4301bbf2225a27714d"},
|
||||
{file = "cryptography-41.0.0-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:14754bcdae909d66ff24b7b5f166d69340ccc6cb15731670435efd5719294895"},
|
||||
{file = "cryptography-41.0.0-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:0ddaee209d1cf1f180f1efa338a68c4621154de0afaef92b89486f5f96047c55"},
|
||||
{file = "cryptography-41.0.0.tar.gz", hash = "sha256:6b71f64beeea341c9b4f963b48ee3b62d62d57ba93eb120e1196b31dc1025e78"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-macosx_10_12_universal2.whl", hash = "sha256:6687ef6d0a6497e2b58e7c5b852b53f62142cfa7cd1555795758934da363a965"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-macosx_10_12_x86_64.whl", hash = "sha256:706843b48f9a3f9b9911979761c91541e3d90db1ca905fd63fee540a217698bc"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_24_aarch64.whl", hash = "sha256:5d2d8b87a490bfcd407ed9d49093793d0f75198a35e6eb1a923ce1ee86c62b41"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:83e17b26de248c33f3acffb922748151d71827d6021d98c70e6c1a25ddd78505"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e124352fd3db36a9d4a21c1aa27fd5d051e621845cb87fb851c08f4f75ce8be6"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-manylinux_2_24_x86_64.whl", hash = "sha256:5aa67414fcdfa22cf052e640cb5ddc461924a045cacf325cd164e65312d99502"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:35f7c7d015d474f4011e859e93e789c87d21f6f4880ebdc29896a60403328f1f"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:f24077a3b5298a5a06a8e0536e3ea9ec60e4c7ac486755e5fb6e6ea9b3500106"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:f0c64d1bd842ca2633e74a1a28033d139368ad959872533b1bab8c80e8240a0c"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:0f8da300b5c8af9f98111ffd512910bc792b4c77392a9523624680f7956a99d4"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-win32.whl", hash = "sha256:fe913f20024eb2cb2f323e42a64bdf2911bb9738a15dba7d3cce48151034e3a8"},
|
||||
{file = "cryptography-39.0.1-cp36-abi3-win_amd64.whl", hash = "sha256:ced4e447ae29ca194449a3f1ce132ded8fcab06971ef5f618605aacaa612beac"},
|
||||
{file = "cryptography-39.0.1-pp38-pypy38_pp73-macosx_10_12_x86_64.whl", hash = "sha256:807ce09d4434881ca3a7594733669bd834f5b2c6d5c7e36f8c00f691887042ad"},
|
||||
{file = "cryptography-39.0.1-pp38-pypy38_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:c5caeb8188c24888c90b5108a441c106f7faa4c4c075a2bcae438c6e8ca73cef"},
|
||||
{file = "cryptography-39.0.1-pp38-pypy38_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:4789d1e3e257965e960232345002262ede4d094d1a19f4d3b52e48d4d8f3b885"},
|
||||
{file = "cryptography-39.0.1-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:96f1157a7c08b5b189b16b47bc9db2332269d6680a196341bf30046330d15388"},
|
||||
{file = "cryptography-39.0.1-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:e422abdec8b5fa8462aa016786680720d78bdce7a30c652b7fadf83a4ba35336"},
|
||||
{file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_24_aarch64.whl", hash = "sha256:b0afd054cd42f3d213bf82c629efb1ee5f22eba35bf0eec88ea9ea7304f511a2"},
|
||||
{file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_24_x86_64.whl", hash = "sha256:6f8ba7f0328b79f08bdacc3e4e66fb4d7aab0c3584e0bd41328dce5262e26b2e"},
|
||||
{file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:ef8b72fa70b348724ff1218267e7f7375b8de4e8194d1636ee60510aae104cd0"},
|
||||
{file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:aec5a6c9864be7df2240c382740fcf3b96928c46604eaa7f3091f58b878c0bb6"},
|
||||
{file = "cryptography-39.0.1-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:fdd188c8a6ef8769f148f88f859884507b954cc64db6b52f66ef199bb9ad660a"},
|
||||
{file = "cryptography-39.0.1.tar.gz", hash = "sha256:d1f6198ee6d9148405e49887803907fe8962a23e6c6f83ea7d98f1c0de375695"},
|
||||
]
|
||||
|
||||
[package.dependencies]
|
||||
@@ -888,12 +892,12 @@ cffi = ">=1.12"
|
||||
[package.extras]
|
||||
docs = ["sphinx (>=5.3.0)", "sphinx-rtd-theme (>=1.1.1)"]
|
||||
docstest = ["pyenchant (>=1.6.11)", "sphinxcontrib-spelling (>=4.0.1)", "twine (>=1.12.0)"]
|
||||
nox = ["nox"]
|
||||
pep8test = ["black", "check-sdist", "mypy", "ruff"]
|
||||
sdist = ["build"]
|
||||
pep8test = ["black", "check-manifest", "mypy", "ruff", "types-pytz", "types-requests"]
|
||||
sdist = ["setuptools-rust (>=0.11.4)"]
|
||||
ssh = ["bcrypt (>=3.1.5)"]
|
||||
test = ["pretend", "pytest (>=6.2.0)", "pytest-benchmark", "pytest-cov", "pytest-xdist"]
|
||||
test = ["hypothesis (>=1.11.4,!=3.79.2)", "iso8601", "pretend", "pytest (>=6.2.0)", "pytest-benchmark", "pytest-cov", "pytest-shard (>=0.1.2)", "pytest-subtests", "pytest-xdist", "pytz"]
|
||||
test-randomorder = ["pytest-randomly"]
|
||||
tox = ["tox"]
|
||||
|
||||
[[package]]
|
||||
name = "docker"
|
||||
|
||||
@@ -93,15 +93,6 @@ With the current approach we made the following design decisions:
|
||||
and column oids. Command tag capturing was added to the rust-postgres
|
||||
functionality as part of this change.
|
||||
|
||||
### Output options
|
||||
|
||||
User can pass several optional headers that will affect resulting json.
|
||||
|
||||
1. `Neon-Raw-Text-Output: true`. Return postgres values as text, without parsing them. So numbers, objects, booleans, nulls and arrays will be returned as text. That can be useful in cases when client code wants to implement it's own parsing or reuse parsing libraries from e.g. node-postgres.
|
||||
2. `Neon-Array-Mode: true`. Return postgres rows as arrays instead of objects. That is more compact representation and also helps in some edge
|
||||
cases where it is hard to use rows represented as objects (e.g. when several fields have the same name).
|
||||
|
||||
|
||||
## Using SNI-based routing on localhost
|
||||
|
||||
Now proxy determines project name from the subdomain, request to the `round-rice-566201.somedomain.tld` will be routed to the project named `round-rice-566201`. Unfortunately, `/etc/hosts` does not support domain wildcards, so I usually use `*.localtest.me` which resolves to `127.0.0.1`. Now we can create self-signed certificate and play with proxy:
|
||||
|
||||
@@ -1,8 +1,6 @@
|
||||
use futures::pin_mut;
|
||||
use futures::StreamExt;
|
||||
use hyper::body::HttpBody;
|
||||
use hyper::http::HeaderName;
|
||||
use hyper::http::HeaderValue;
|
||||
use hyper::{Body, HeaderMap, Request};
|
||||
use pq_proto::StartupMessageParams;
|
||||
use serde_json::json;
|
||||
@@ -25,28 +23,21 @@ const APP_NAME: &str = "sql_over_http";
|
||||
const MAX_RESPONSE_SIZE: usize = 1024 * 1024; // 1 MB
|
||||
const MAX_REQUEST_SIZE: u64 = 1024 * 1024; // 1 MB
|
||||
|
||||
static RAW_TEXT_OUTPUT: HeaderName = HeaderName::from_static("neon-raw-text-output");
|
||||
static ARRAY_MODE: HeaderName = HeaderName::from_static("neon-array-mode");
|
||||
static HEADER_VALUE_TRUE: HeaderValue = HeaderValue::from_static("true");
|
||||
|
||||
//
|
||||
// Convert json non-string types to strings, so that they can be passed to Postgres
|
||||
// as parameters.
|
||||
//
|
||||
fn json_to_pg_text(json: Vec<Value>) -> Result<Vec<Option<String>>, serde_json::Error> {
|
||||
fn json_to_pg_text(json: Vec<Value>) -> Result<Vec<String>, serde_json::Error> {
|
||||
json.iter()
|
||||
.map(|value| {
|
||||
match value {
|
||||
// special care for nulls
|
||||
Value::Null => Ok(None),
|
||||
Value::Null => serde_json::to_string(value),
|
||||
Value::Bool(_) => serde_json::to_string(value),
|
||||
Value::Number(_) => serde_json::to_string(value),
|
||||
Value::Object(_) => serde_json::to_string(value),
|
||||
|
||||
// convert to text with escaping
|
||||
Value::Bool(_) => serde_json::to_string(value).map(Some),
|
||||
Value::Number(_) => serde_json::to_string(value).map(Some),
|
||||
Value::Object(_) => serde_json::to_string(value).map(Some),
|
||||
|
||||
// avoid escaping here, as we pass this as a parameter
|
||||
Value::String(s) => Ok(Some(s.to_string())),
|
||||
// no need to escape
|
||||
Value::String(s) => Ok(s.to_string()),
|
||||
|
||||
// special care for arrays
|
||||
Value::Array(_) => json_array_to_pg_array(value),
|
||||
@@ -63,29 +54,25 @@ fn json_to_pg_text(json: Vec<Value>) -> Result<Vec<Option<String>>, serde_json::
|
||||
//
|
||||
// Example of the same escaping in node-postgres: packages/pg/lib/utils.js
|
||||
//
|
||||
fn json_array_to_pg_array(value: &Value) -> Result<Option<String>, serde_json::Error> {
|
||||
fn json_array_to_pg_array(value: &Value) -> Result<String, serde_json::Error> {
|
||||
match value {
|
||||
// special care for nulls
|
||||
Value::Null => Ok(None),
|
||||
// same
|
||||
Value::Null => serde_json::to_string(value),
|
||||
Value::Bool(_) => serde_json::to_string(value),
|
||||
Value::Number(_) => serde_json::to_string(value),
|
||||
Value::Object(_) => serde_json::to_string(value),
|
||||
|
||||
// convert to text with escaping
|
||||
Value::Bool(_) => serde_json::to_string(value).map(Some),
|
||||
Value::Number(_) => serde_json::to_string(value).map(Some),
|
||||
Value::Object(_) => serde_json::to_string(value).map(Some),
|
||||
|
||||
// here string needs to be escaped, as it is part of the array
|
||||
Value::String(_) => serde_json::to_string(value).map(Some),
|
||||
// now needs to be escaped, as it is part of the array
|
||||
Value::String(_) => serde_json::to_string(value),
|
||||
|
||||
// recurse into array
|
||||
Value::Array(arr) => {
|
||||
let vals = arr
|
||||
.iter()
|
||||
.map(json_array_to_pg_array)
|
||||
.map(|r| r.map(|v| v.unwrap_or_else(|| "NULL".to_string())))
|
||||
.collect::<Result<Vec<_>, _>>()?
|
||||
.join(",");
|
||||
|
||||
Ok(Some(format!("{{{}}}", vals)))
|
||||
Ok(format!("{{{}}}", vals))
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -171,11 +158,6 @@ pub async fn handle(
|
||||
("application_name", APP_NAME),
|
||||
]);
|
||||
|
||||
// Determine the output options. Default behaviour is 'false'. Anything that is not
|
||||
// strictly 'true' assumed to be false.
|
||||
let raw_output = headers.get(&RAW_TEXT_OUTPUT) == Some(&HEADER_VALUE_TRUE);
|
||||
let array_mode = headers.get(&ARRAY_MODE) == Some(&HEADER_VALUE_TRUE);
|
||||
|
||||
//
|
||||
// Wake up the destination if needed. Code here is a bit involved because
|
||||
// we reuse the code from the usual proxy and we need to prepare few structures
|
||||
@@ -290,7 +272,7 @@ pub async fn handle(
|
||||
// convert rows to JSON
|
||||
let rows = rows
|
||||
.iter()
|
||||
.map(|row| pg_text_row_to_json(row, raw_output, array_mode))
|
||||
.map(pg_text_row_to_json)
|
||||
.collect::<Result<Vec<_>, _>>()?;
|
||||
|
||||
// resulting JSON format is based on the format of node-postgres result
|
||||
@@ -299,42 +281,26 @@ pub async fn handle(
|
||||
"rowCount": command_tag_count,
|
||||
"rows": rows,
|
||||
"fields": fields,
|
||||
"rowAsArray": array_mode,
|
||||
}))
|
||||
}
|
||||
|
||||
//
|
||||
// Convert postgres row with text-encoded values to JSON object
|
||||
//
|
||||
pub fn pg_text_row_to_json(
|
||||
row: &Row,
|
||||
raw_output: bool,
|
||||
array_mode: bool,
|
||||
) -> Result<Value, anyhow::Error> {
|
||||
let iter = row.columns().iter().enumerate().map(|(i, column)| {
|
||||
let name = column.name();
|
||||
let pg_value = row.as_text(i)?;
|
||||
let json_value = if raw_output {
|
||||
match pg_value {
|
||||
Some(v) => Value::String(v.to_string()),
|
||||
None => Value::Null,
|
||||
}
|
||||
} else {
|
||||
pg_text_to_json(pg_value, column.type_())?
|
||||
};
|
||||
Ok((name.to_string(), json_value))
|
||||
});
|
||||
pub fn pg_text_row_to_json(row: &Row) -> Result<Value, anyhow::Error> {
|
||||
let res = row
|
||||
.columns()
|
||||
.iter()
|
||||
.enumerate()
|
||||
.map(|(i, column)| {
|
||||
let name = column.name();
|
||||
let pg_value = row.as_text(i)?;
|
||||
let json_value = pg_text_to_json(pg_value, column.type_())?;
|
||||
Ok((name.to_string(), json_value))
|
||||
})
|
||||
.collect::<Result<Map<String, Value>, anyhow::Error>>()?;
|
||||
|
||||
if array_mode {
|
||||
// drop keys and aggregate into array
|
||||
let arr = iter
|
||||
.map(|r| r.map(|(_key, val)| val))
|
||||
.collect::<Result<Vec<Value>, anyhow::Error>>()?;
|
||||
Ok(Value::Array(arr))
|
||||
} else {
|
||||
let obj = iter.collect::<Result<Map<String, Value>, anyhow::Error>>()?;
|
||||
Ok(Value::Object(obj))
|
||||
}
|
||||
Ok(Value::Object(res))
|
||||
}
|
||||
|
||||
//
|
||||
@@ -342,6 +308,10 @@ pub fn pg_text_row_to_json(
|
||||
//
|
||||
pub fn pg_text_to_json(pg_value: Option<&str>, pg_type: &Type) -> Result<Value, anyhow::Error> {
|
||||
if let Some(val) = pg_value {
|
||||
if val == "NULL" {
|
||||
return Ok(Value::Null);
|
||||
}
|
||||
|
||||
if let Kind::Array(elem_type) = pg_type.kind() {
|
||||
return pg_array_parse(val, elem_type);
|
||||
}
|
||||
@@ -403,27 +373,6 @@ fn _pg_array_parse(
|
||||
}
|
||||
}
|
||||
|
||||
fn push_checked(
|
||||
entry: &mut String,
|
||||
entries: &mut Vec<Value>,
|
||||
elem_type: &Type,
|
||||
) -> Result<(), anyhow::Error> {
|
||||
if !entry.is_empty() {
|
||||
// While in usual postgres response we get nulls as None and everything else
|
||||
// as Some(&str), in arrays we get NULL as unquoted 'NULL' string (while
|
||||
// string with value 'NULL' will be represented by '"NULL"'). So catch NULLs
|
||||
// here while we have quotation info and convert them to None.
|
||||
if entry == "NULL" {
|
||||
entries.push(pg_text_to_json(None, elem_type)?);
|
||||
} else {
|
||||
entries.push(pg_text_to_json(Some(entry), elem_type)?);
|
||||
}
|
||||
entry.clear();
|
||||
}
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
while let Some((mut i, mut c)) = pg_array_chr.next() {
|
||||
let mut escaped = false;
|
||||
|
||||
@@ -446,7 +395,9 @@ fn _pg_array_parse(
|
||||
'}' => {
|
||||
level -= 1;
|
||||
if level == 0 {
|
||||
push_checked(&mut entry, &mut entries, elem_type)?;
|
||||
if !entry.is_empty() {
|
||||
entries.push(pg_text_to_json(Some(&entry), elem_type)?);
|
||||
}
|
||||
if nested {
|
||||
return Ok((Value::Array(entries), i));
|
||||
}
|
||||
@@ -454,15 +405,17 @@ fn _pg_array_parse(
|
||||
}
|
||||
'"' if !escaped => {
|
||||
if quote {
|
||||
// end of quoted string, so push it manually without any checks
|
||||
// for emptiness or nulls
|
||||
// push even if empty
|
||||
entries.push(pg_text_to_json(Some(&entry), elem_type)?);
|
||||
entry.clear();
|
||||
entry = String::new();
|
||||
}
|
||||
quote = !quote;
|
||||
}
|
||||
',' if !quote => {
|
||||
push_checked(&mut entry, &mut entries, elem_type)?;
|
||||
if !entry.is_empty() {
|
||||
entries.push(pg_text_to_json(Some(&entry), elem_type)?);
|
||||
entry = String::new();
|
||||
}
|
||||
}
|
||||
_ => {
|
||||
entry.push(c);
|
||||
@@ -486,35 +439,30 @@ mod tests {
|
||||
fn test_atomic_types_to_pg_params() {
|
||||
let json = vec![Value::Bool(true), Value::Bool(false)];
|
||||
let pg_params = json_to_pg_text(json).unwrap();
|
||||
assert_eq!(
|
||||
pg_params,
|
||||
vec![Some("true".to_owned()), Some("false".to_owned())]
|
||||
);
|
||||
assert_eq!(pg_params, vec!["true", "false"]);
|
||||
|
||||
let json = vec![Value::Number(serde_json::Number::from(42))];
|
||||
let pg_params = json_to_pg_text(json).unwrap();
|
||||
assert_eq!(pg_params, vec![Some("42".to_owned())]);
|
||||
assert_eq!(pg_params, vec!["42"]);
|
||||
|
||||
let json = vec![Value::String("foo\"".to_string())];
|
||||
let pg_params = json_to_pg_text(json).unwrap();
|
||||
assert_eq!(pg_params, vec![Some("foo\"".to_owned())]);
|
||||
assert_eq!(pg_params, vec!["foo\""]);
|
||||
|
||||
let json = vec![Value::Null];
|
||||
let pg_params = json_to_pg_text(json).unwrap();
|
||||
assert_eq!(pg_params, vec![None]);
|
||||
assert_eq!(pg_params, vec!["null"]);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_json_array_to_pg_array() {
|
||||
// atoms and escaping
|
||||
let json = "[true, false, null, \"NULL\", 42, \"foo\", \"bar\\\"-\\\\\"]";
|
||||
let json = "[true, false, null, 42, \"foo\", \"bar\\\"-\\\\\"]";
|
||||
let json: Value = serde_json::from_str(json).unwrap();
|
||||
let pg_params = json_to_pg_text(vec![json]).unwrap();
|
||||
assert_eq!(
|
||||
pg_params,
|
||||
vec![Some(
|
||||
"{true,false,NULL,\"NULL\",42,\"foo\",\"bar\\\"-\\\\\"}".to_owned()
|
||||
)]
|
||||
vec!["{true,false,null,42,\"foo\",\"bar\\\"-\\\\\"}"]
|
||||
);
|
||||
|
||||
// nested arrays
|
||||
@@ -523,9 +471,7 @@ mod tests {
|
||||
let pg_params = json_to_pg_text(vec![json]).unwrap();
|
||||
assert_eq!(
|
||||
pg_params,
|
||||
vec![Some(
|
||||
"{{true,false},{NULL,42},{\"foo\",\"bar\\\"-\\\\\"}}".to_owned()
|
||||
)]
|
||||
vec!["{{true,false},{null,42},{\"foo\",\"bar\\\"-\\\\\"}}"]
|
||||
);
|
||||
}
|
||||
|
||||
|
||||
@@ -3,19 +3,15 @@
|
||||
//
|
||||
use anyhow::{bail, Context, Result};
|
||||
use clap::Parser;
|
||||
use futures::future::BoxFuture;
|
||||
use futures::stream::FuturesUnordered;
|
||||
use futures::{FutureExt, StreamExt};
|
||||
use remote_storage::RemoteStorageConfig;
|
||||
use tokio::runtime::Handle;
|
||||
use tokio::signal::unix::{signal, SignalKind};
|
||||
use tokio::task::JoinError;
|
||||
use toml_edit::Document;
|
||||
use utils::signals::ShutdownSignals;
|
||||
|
||||
use std::fs::{self, File};
|
||||
use std::io::{ErrorKind, Write};
|
||||
use std::path::{Path, PathBuf};
|
||||
use std::sync::Arc;
|
||||
use std::thread;
|
||||
use std::time::Duration;
|
||||
use storage_broker::Uri;
|
||||
use tokio::sync::mpsc;
|
||||
@@ -24,21 +20,22 @@ use tracing::*;
|
||||
use utils::pid_file;
|
||||
|
||||
use metrics::set_build_info_metric;
|
||||
use safekeeper::broker;
|
||||
use safekeeper::control_file;
|
||||
use safekeeper::defaults::{
|
||||
DEFAULT_HEARTBEAT_TIMEOUT, DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_MAX_OFFLOADER_LAG_BYTES,
|
||||
DEFAULT_PG_LISTEN_ADDR,
|
||||
};
|
||||
use safekeeper::http;
|
||||
use safekeeper::remove_wal;
|
||||
use safekeeper::wal_backup;
|
||||
use safekeeper::wal_service;
|
||||
use safekeeper::GlobalTimelines;
|
||||
use safekeeper::SafeKeeperConf;
|
||||
use safekeeper::{broker, WAL_SERVICE_RUNTIME};
|
||||
use safekeeper::{control_file, BROKER_RUNTIME};
|
||||
use safekeeper::{http, WAL_REMOVER_RUNTIME};
|
||||
use safekeeper::{remove_wal, WAL_BACKUP_RUNTIME};
|
||||
use safekeeper::{wal_backup, HTTP_RUNTIME};
|
||||
use storage_broker::DEFAULT_ENDPOINT;
|
||||
use utils::auth::JwtAuth;
|
||||
use utils::{
|
||||
http::endpoint,
|
||||
id::NodeId,
|
||||
logging::{self, LogFormat},
|
||||
project_git_version,
|
||||
@@ -107,6 +104,10 @@ struct Args {
|
||||
/// Safekeeper won't be elected for WAL offloading if it is lagging for more than this value in bytes
|
||||
#[arg(long, default_value_t = DEFAULT_MAX_OFFLOADER_LAG_BYTES)]
|
||||
max_offloader_lag: u64,
|
||||
/// Number of threads for wal backup runtime, by default number of cores
|
||||
/// available to the system.
|
||||
#[arg(long)]
|
||||
wal_backup_threads: Option<usize>,
|
||||
/// Number of max parallel WAL segments to be offloaded to remote storage.
|
||||
#[arg(long, default_value = "5")]
|
||||
wal_backup_parallel_jobs: usize,
|
||||
@@ -120,14 +121,9 @@ struct Args {
|
||||
/// Format for logging, either 'plain' or 'json'.
|
||||
#[arg(long, default_value = "plain")]
|
||||
log_format: String,
|
||||
/// Run everything in single threaded current thread runtime, might be
|
||||
/// useful for debugging.
|
||||
#[arg(long)]
|
||||
current_thread_runtime: bool,
|
||||
}
|
||||
|
||||
#[tokio::main(flavor = "current_thread")]
|
||||
async fn main() -> anyhow::Result<()> {
|
||||
fn main() -> anyhow::Result<()> {
|
||||
let args = Args::parse();
|
||||
|
||||
if let Some(addr) = args.dump_control_file {
|
||||
@@ -187,10 +183,10 @@ async fn main() -> anyhow::Result<()> {
|
||||
heartbeat_timeout: args.heartbeat_timeout,
|
||||
remote_storage: args.remote_storage,
|
||||
max_offloader_lag_bytes: args.max_offloader_lag,
|
||||
backup_runtime_threads: args.wal_backup_threads,
|
||||
wal_backup_enabled: !args.disable_wal_backup,
|
||||
backup_parallel_jobs: args.wal_backup_parallel_jobs,
|
||||
auth,
|
||||
current_thread_runtime: args.current_thread_runtime,
|
||||
};
|
||||
|
||||
// initialize sentry if SENTRY_DSN is provided
|
||||
@@ -198,14 +194,10 @@ async fn main() -> anyhow::Result<()> {
|
||||
Some(GIT_VERSION.into()),
|
||||
&[("node_id", &conf.my_id.to_string())],
|
||||
);
|
||||
start_safekeeper(conf).await
|
||||
start_safekeeper(conf)
|
||||
}
|
||||
|
||||
/// Result of joining any of main tasks: upper error means task failed to
|
||||
/// complete, e.g. panicked, inner is error produced by task itself.
|
||||
type JoinTaskRes = Result<anyhow::Result<()>, JoinError>;
|
||||
|
||||
async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
|
||||
fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
|
||||
// Prevent running multiple safekeepers on the same directory
|
||||
let lock_file_path = conf.workdir.join(PID_FILE_NAME);
|
||||
let lock_file =
|
||||
@@ -216,18 +208,14 @@ async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
|
||||
// we need to release the lock file only when the current process is gone
|
||||
std::mem::forget(lock_file);
|
||||
|
||||
info!("starting safekeeper WAL service on {}", conf.listen_pg_addr);
|
||||
let pg_listener = tcp_listener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
|
||||
error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
|
||||
let http_listener = tcp_listener::bind(conf.listen_http_addr.clone()).map_err(|e| {
|
||||
error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
|
||||
e
|
||||
})?;
|
||||
|
||||
info!(
|
||||
"starting safekeeper HTTP service on {}",
|
||||
conf.listen_http_addr
|
||||
);
|
||||
let http_listener = tcp_listener::bind(conf.listen_http_addr.clone()).map_err(|e| {
|
||||
error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
|
||||
info!("starting safekeeper on {}", conf.listen_pg_addr);
|
||||
let pg_listener = tcp_listener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
|
||||
error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
|
||||
e
|
||||
})?;
|
||||
|
||||
@@ -236,88 +224,71 @@ async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
|
||||
let timeline_collector = safekeeper::metrics::TimelineCollector::new();
|
||||
metrics::register_internal(Box::new(timeline_collector))?;
|
||||
|
||||
let mut threads = vec![];
|
||||
let (wal_backup_launcher_tx, wal_backup_launcher_rx) = mpsc::channel(100);
|
||||
|
||||
// Load all timelines from disk to memory.
|
||||
GlobalTimelines::init(conf.clone(), wal_backup_launcher_tx)?;
|
||||
|
||||
// Keep handles to main tasks to die if any of them disappears.
|
||||
let mut tasks_handles: FuturesUnordered<BoxFuture<(String, JoinTaskRes)>> =
|
||||
FuturesUnordered::new();
|
||||
let conf_ = conf.clone();
|
||||
threads.push(
|
||||
thread::Builder::new()
|
||||
.name("http_endpoint_thread".into())
|
||||
.spawn(|| {
|
||||
let router = http::make_router(conf_);
|
||||
endpoint::serve_thread_main(
|
||||
router,
|
||||
http_listener,
|
||||
std::future::pending(), // never shut down
|
||||
)
|
||||
.unwrap();
|
||||
})?,
|
||||
);
|
||||
|
||||
let conf_cloned = conf.clone();
|
||||
let safekeeper_thread = thread::Builder::new()
|
||||
.name("WAL service thread".into())
|
||||
.spawn(|| wal_service::thread_main(conf_cloned, pg_listener))
|
||||
.unwrap();
|
||||
|
||||
threads.push(safekeeper_thread);
|
||||
|
||||
let conf_ = conf.clone();
|
||||
// Run everything in current thread rt, if asked.
|
||||
if conf.current_thread_runtime {
|
||||
info!("running in current thread runtime");
|
||||
}
|
||||
let current_thread_rt = conf
|
||||
.current_thread_runtime
|
||||
.then(|| Handle::try_current().expect("no runtime in main"));
|
||||
let wal_service_handle = current_thread_rt
|
||||
.as_ref()
|
||||
.unwrap_or_else(|| WAL_SERVICE_RUNTIME.handle())
|
||||
.spawn(wal_service::task_main(conf_, pg_listener))
|
||||
// wrap with task name for error reporting
|
||||
.map(|res| ("WAL service main".to_owned(), res));
|
||||
tasks_handles.push(Box::pin(wal_service_handle));
|
||||
threads.push(
|
||||
thread::Builder::new()
|
||||
.name("broker thread".into())
|
||||
.spawn(|| {
|
||||
broker::thread_main(conf_);
|
||||
})?,
|
||||
);
|
||||
|
||||
let conf_ = conf.clone();
|
||||
let http_handle = current_thread_rt
|
||||
.as_ref()
|
||||
.unwrap_or_else(|| HTTP_RUNTIME.handle())
|
||||
.spawn(http::task_main(conf_, http_listener))
|
||||
.map(|res| ("HTTP service main".to_owned(), res));
|
||||
tasks_handles.push(Box::pin(http_handle));
|
||||
threads.push(
|
||||
thread::Builder::new()
|
||||
.name("WAL removal thread".into())
|
||||
.spawn(|| {
|
||||
remove_wal::thread_main(conf_);
|
||||
})?,
|
||||
);
|
||||
|
||||
let conf_ = conf.clone();
|
||||
let broker_task_handle = current_thread_rt
|
||||
.as_ref()
|
||||
.unwrap_or_else(|| BROKER_RUNTIME.handle())
|
||||
.spawn(broker::task_main(conf_).instrument(info_span!("broker")))
|
||||
.map(|res| ("broker main".to_owned(), res));
|
||||
tasks_handles.push(Box::pin(broker_task_handle));
|
||||
|
||||
let conf_ = conf.clone();
|
||||
let wal_remover_handle = current_thread_rt
|
||||
.as_ref()
|
||||
.unwrap_or_else(|| WAL_REMOVER_RUNTIME.handle())
|
||||
.spawn(remove_wal::task_main(conf_))
|
||||
.map(|res| ("WAL remover".to_owned(), res));
|
||||
tasks_handles.push(Box::pin(wal_remover_handle));
|
||||
|
||||
let conf_ = conf.clone();
|
||||
let wal_backup_handle = current_thread_rt
|
||||
.as_ref()
|
||||
.unwrap_or_else(|| WAL_BACKUP_RUNTIME.handle())
|
||||
.spawn(wal_backup::wal_backup_launcher_task_main(
|
||||
conf_,
|
||||
wal_backup_launcher_rx,
|
||||
))
|
||||
.map(|res| ("WAL backup launcher".to_owned(), res));
|
||||
tasks_handles.push(Box::pin(wal_backup_handle));
|
||||
threads.push(
|
||||
thread::Builder::new()
|
||||
.name("WAL backup launcher thread".into())
|
||||
.spawn(move || {
|
||||
wal_backup::wal_backup_launcher_thread_main(conf, wal_backup_launcher_rx);
|
||||
})?,
|
||||
);
|
||||
|
||||
set_build_info_metric(GIT_VERSION);
|
||||
// TODO: put more thoughts into handling of failed threads
|
||||
// We should catch & die if they are in trouble.
|
||||
|
||||
// TODO: update tokio-stream, convert to real async Stream with
|
||||
// SignalStream, map it to obtain missing signal name, combine streams into
|
||||
// single stream we can easily sit on.
|
||||
let mut sigquit_stream = signal(SignalKind::quit())?;
|
||||
let mut sigint_stream = signal(SignalKind::interrupt())?;
|
||||
let mut sigterm_stream = signal(SignalKind::terminate())?;
|
||||
|
||||
tokio::select! {
|
||||
Some((task_name, res)) = tasks_handles.next()=> {
|
||||
error!("{} task failed: {:?}, exiting", task_name, res);
|
||||
std::process::exit(1);
|
||||
}
|
||||
// On any shutdown signal, log receival and exit. Additionally, handling
|
||||
// SIGQUIT prevents coredump.
|
||||
_ = sigquit_stream.recv() => info!("received SIGQUIT, terminating"),
|
||||
_ = sigint_stream.recv() => info!("received SIGINT, terminating"),
|
||||
_ = sigterm_stream.recv() => info!("received SIGTERM, terminating")
|
||||
|
||||
};
|
||||
std::process::exit(0);
|
||||
// On any shutdown signal, log receival and exit. Additionally, handling
|
||||
// SIGQUIT prevents coredump.
|
||||
ShutdownSignals::handle(|signal| {
|
||||
info!("received {}, terminating", signal.name());
|
||||
std::process::exit(0);
|
||||
})
|
||||
}
|
||||
|
||||
/// Determine safekeeper id.
|
||||
|
||||
@@ -8,7 +8,7 @@ use anyhow::Error;
|
||||
use anyhow::Result;
|
||||
|
||||
use storage_broker::parse_proto_ttid;
|
||||
|
||||
use storage_broker::proto::broker_service_client::BrokerServiceClient;
|
||||
use storage_broker::proto::subscribe_safekeeper_info_request::SubscriptionKey as ProtoSubscriptionKey;
|
||||
use storage_broker::proto::SubscribeSafekeeperInfoRequest;
|
||||
use storage_broker::Request;
|
||||
@@ -16,7 +16,7 @@ use storage_broker::Request;
|
||||
use std::time::Duration;
|
||||
use std::time::Instant;
|
||||
use tokio::task::JoinHandle;
|
||||
use tokio::time::sleep;
|
||||
use tokio::{runtime, time::sleep};
|
||||
use tracing::*;
|
||||
|
||||
use crate::metrics::BROKER_ITERATION_TIMELINES;
|
||||
@@ -29,10 +29,23 @@ use crate::SafeKeeperConf;
|
||||
const RETRY_INTERVAL_MSEC: u64 = 1000;
|
||||
const PUSH_INTERVAL_MSEC: u64 = 1000;
|
||||
|
||||
pub fn thread_main(conf: SafeKeeperConf) {
|
||||
let runtime = runtime::Builder::new_current_thread()
|
||||
.enable_all()
|
||||
.build()
|
||||
.unwrap();
|
||||
|
||||
let _enter = info_span!("broker").entered();
|
||||
info!("started, broker endpoint {:?}", conf.broker_endpoint);
|
||||
|
||||
runtime.block_on(async {
|
||||
main_loop(conf).await;
|
||||
});
|
||||
}
|
||||
|
||||
/// Push once in a while data about all active timelines to the broker.
|
||||
async fn push_loop(conf: SafeKeeperConf) -> anyhow::Result<()> {
|
||||
let mut client =
|
||||
storage_broker::connect(conf.broker_endpoint.clone(), conf.broker_keepalive_interval)?;
|
||||
let mut client = BrokerServiceClient::connect(conf.broker_endpoint.clone()).await?;
|
||||
let push_interval = Duration::from_millis(PUSH_INTERVAL_MSEC);
|
||||
|
||||
let outbound = async_stream::stream! {
|
||||
@@ -42,27 +55,20 @@ async fn push_loop(conf: SafeKeeperConf) -> anyhow::Result<()> {
|
||||
// sensitive and there is no risk of deadlock as we don't await while
|
||||
// lock is held.
|
||||
let now = Instant::now();
|
||||
let all_tlis = GlobalTimelines::get_all();
|
||||
let mut n_pushed_tlis = 0;
|
||||
for tli in &all_tlis {
|
||||
// filtering alternative futures::stream::iter(all_tlis)
|
||||
// .filter(|tli| {let tli = tli.clone(); async move { tli.is_active().await}}).collect::<Vec<_>>().await;
|
||||
// doesn't look better, and I'm not sure how to do that without collect.
|
||||
if !tli.is_active().await {
|
||||
continue;
|
||||
}
|
||||
let sk_info = tli.get_safekeeper_info(&conf).await;
|
||||
let mut active_tlis = GlobalTimelines::get_all();
|
||||
active_tlis.retain(|tli| tli.is_active());
|
||||
for tli in &active_tlis {
|
||||
let sk_info = tli.get_safekeeper_info(&conf);
|
||||
yield sk_info;
|
||||
BROKER_PUSHED_UPDATES.inc();
|
||||
n_pushed_tlis += 1;
|
||||
}
|
||||
let elapsed = now.elapsed();
|
||||
|
||||
BROKER_PUSH_ALL_UPDATES_SECONDS.observe(elapsed.as_secs_f64());
|
||||
BROKER_ITERATION_TIMELINES.observe(n_pushed_tlis as f64);
|
||||
BROKER_ITERATION_TIMELINES.observe(active_tlis.len() as f64);
|
||||
|
||||
if elapsed > push_interval / 2 {
|
||||
info!("broker push is too long, pushed {} timeline updates to broker in {:?}", n_pushed_tlis, elapsed);
|
||||
info!("broker push is too long, pushed {} timeline updates to broker in {:?}", active_tlis.len(), elapsed);
|
||||
}
|
||||
|
||||
sleep(push_interval).await;
|
||||
@@ -119,13 +125,10 @@ async fn pull_loop(conf: SafeKeeperConf) -> Result<()> {
|
||||
bail!("end of stream");
|
||||
}
|
||||
|
||||
pub async fn task_main(conf: SafeKeeperConf) -> anyhow::Result<()> {
|
||||
info!("started, broker endpoint {:?}", conf.broker_endpoint);
|
||||
|
||||
async fn main_loop(conf: SafeKeeperConf) {
|
||||
let mut ticker = tokio::time::interval(Duration::from_millis(RETRY_INTERVAL_MSEC));
|
||||
let mut push_handle: Option<JoinHandle<Result<(), Error>>> = None;
|
||||
let mut pull_handle: Option<JoinHandle<Result<(), Error>>> = None;
|
||||
|
||||
// Selecting on JoinHandles requires some squats; is there a better way to
|
||||
// reap tasks individually?
|
||||
|
||||
|
||||
@@ -2,10 +2,9 @@
|
||||
|
||||
use anyhow::{bail, ensure, Context, Result};
|
||||
use byteorder::{LittleEndian, ReadBytesExt, WriteBytesExt};
|
||||
use tokio::fs::{self, File};
|
||||
use tokio::io::AsyncWriteExt;
|
||||
|
||||
use std::io::Read;
|
||||
use std::fs::{self, File, OpenOptions};
|
||||
use std::io::{Read, Write};
|
||||
use std::ops::Deref;
|
||||
use std::path::{Path, PathBuf};
|
||||
use std::time::Instant;
|
||||
@@ -27,10 +26,9 @@ pub const CHECKSUM_SIZE: usize = std::mem::size_of::<u32>();
|
||||
|
||||
/// Storage should keep actual state inside of it. It should implement Deref
|
||||
/// trait to access state fields and have persist method for updating that state.
|
||||
#[async_trait::async_trait]
|
||||
pub trait Storage: Deref<Target = SafeKeeperState> {
|
||||
/// Persist safekeeper state on disk and update internal state.
|
||||
async fn persist(&mut self, s: &SafeKeeperState) -> Result<()>;
|
||||
fn persist(&mut self, s: &SafeKeeperState) -> Result<()>;
|
||||
|
||||
/// Timestamp of last persist.
|
||||
fn last_persist_at(&self) -> Instant;
|
||||
@@ -84,7 +82,7 @@ impl FileStorage {
|
||||
/// Check the magic/version in the on-disk data and deserialize it, if possible.
|
||||
fn deser_sk_state(buf: &mut &[u8]) -> Result<SafeKeeperState> {
|
||||
// Read the version independent part
|
||||
let magic = ReadBytesExt::read_u32::<LittleEndian>(buf)?;
|
||||
let magic = buf.read_u32::<LittleEndian>()?;
|
||||
if magic != SK_MAGIC {
|
||||
bail!(
|
||||
"bad control file magic: {:X}, expected {:X}",
|
||||
@@ -92,7 +90,7 @@ impl FileStorage {
|
||||
SK_MAGIC
|
||||
);
|
||||
}
|
||||
let version = ReadBytesExt::read_u32::<LittleEndian>(buf)?;
|
||||
let version = buf.read_u32::<LittleEndian>()?;
|
||||
if version == SK_FORMAT_VERSION {
|
||||
let res = SafeKeeperState::des(buf)?;
|
||||
return Ok(res);
|
||||
@@ -112,7 +110,7 @@ impl FileStorage {
|
||||
|
||||
/// Read in the control file.
|
||||
pub fn load_control_file<P: AsRef<Path>>(control_file_path: P) -> Result<SafeKeeperState> {
|
||||
let mut control_file = std::fs::OpenOptions::new()
|
||||
let mut control_file = OpenOptions::new()
|
||||
.read(true)
|
||||
.write(true)
|
||||
.open(&control_file_path)
|
||||
@@ -161,31 +159,30 @@ impl Deref for FileStorage {
|
||||
}
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl Storage for FileStorage {
|
||||
/// persists state durably to underlying storage
|
||||
/// for description see https://lwn.net/Articles/457667/
|
||||
async fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
|
||||
fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
|
||||
let _timer = PERSIST_CONTROL_FILE_SECONDS.start_timer();
|
||||
|
||||
// write data to safekeeper.control.partial
|
||||
let control_partial_path = self.timeline_dir.join(CONTROL_FILE_NAME_PARTIAL);
|
||||
let mut control_partial = File::create(&control_partial_path).await.with_context(|| {
|
||||
let mut control_partial = File::create(&control_partial_path).with_context(|| {
|
||||
format!(
|
||||
"failed to create partial control file at: {}",
|
||||
&control_partial_path.display()
|
||||
)
|
||||
})?;
|
||||
let mut buf: Vec<u8> = Vec::new();
|
||||
WriteBytesExt::write_u32::<LittleEndian>(&mut buf, SK_MAGIC)?;
|
||||
WriteBytesExt::write_u32::<LittleEndian>(&mut buf, SK_FORMAT_VERSION)?;
|
||||
buf.write_u32::<LittleEndian>(SK_MAGIC)?;
|
||||
buf.write_u32::<LittleEndian>(SK_FORMAT_VERSION)?;
|
||||
s.ser_into(&mut buf)?;
|
||||
|
||||
// calculate checksum before resize
|
||||
let checksum = crc32c::crc32c(&buf);
|
||||
buf.extend_from_slice(&checksum.to_le_bytes());
|
||||
|
||||
control_partial.write_all(&buf).await.with_context(|| {
|
||||
control_partial.write_all(&buf).with_context(|| {
|
||||
format!(
|
||||
"failed to write safekeeper state into control file at: {}",
|
||||
control_partial_path.display()
|
||||
@@ -194,7 +191,7 @@ impl Storage for FileStorage {
|
||||
|
||||
// fsync the file
|
||||
if !self.conf.no_sync {
|
||||
control_partial.sync_all().await.with_context(|| {
|
||||
control_partial.sync_all().with_context(|| {
|
||||
format!(
|
||||
"failed to sync partial control file at {}",
|
||||
control_partial_path.display()
|
||||
@@ -205,22 +202,21 @@ impl Storage for FileStorage {
|
||||
let control_path = self.timeline_dir.join(CONTROL_FILE_NAME);
|
||||
|
||||
// rename should be atomic
|
||||
fs::rename(&control_partial_path, &control_path).await?;
|
||||
fs::rename(&control_partial_path, &control_path)?;
|
||||
// this sync is not required by any standard but postgres does this (see durable_rename)
|
||||
if !self.conf.no_sync {
|
||||
let new_f = File::open(&control_path).await?;
|
||||
new_f.sync_all().await.with_context(|| {
|
||||
format!(
|
||||
"failed to sync control file at: {}",
|
||||
&control_path.display()
|
||||
)
|
||||
})?;
|
||||
File::open(&control_path)
|
||||
.and_then(|f| f.sync_all())
|
||||
.with_context(|| {
|
||||
format!(
|
||||
"failed to sync control file at: {}",
|
||||
&control_path.display()
|
||||
)
|
||||
})?;
|
||||
|
||||
// fsync the directory (linux specific)
|
||||
let tli_dir = File::open(&self.timeline_dir).await?;
|
||||
tli_dir
|
||||
.sync_all()
|
||||
.await
|
||||
File::open(&self.timeline_dir)
|
||||
.and_then(|f| f.sync_all())
|
||||
.context("failed to sync control file directory")?;
|
||||
}
|
||||
|
||||
@@ -240,6 +236,7 @@ mod test {
|
||||
use super::*;
|
||||
use crate::{safekeeper::SafeKeeperState, SafeKeeperConf};
|
||||
use anyhow::Result;
|
||||
use std::fs;
|
||||
use utils::{id::TenantTimelineId, lsn::Lsn};
|
||||
|
||||
fn stub_conf() -> SafeKeeperConf {
|
||||
@@ -250,75 +247,59 @@ mod test {
|
||||
}
|
||||
}
|
||||
|
||||
async fn load_from_control_file(
|
||||
fn load_from_control_file(
|
||||
conf: &SafeKeeperConf,
|
||||
ttid: &TenantTimelineId,
|
||||
) -> Result<(FileStorage, SafeKeeperState)> {
|
||||
fs::create_dir_all(conf.timeline_dir(ttid))
|
||||
.await
|
||||
.expect("failed to create timeline dir");
|
||||
fs::create_dir_all(conf.timeline_dir(ttid)).expect("failed to create timeline dir");
|
||||
Ok((
|
||||
FileStorage::restore_new(ttid, conf)?,
|
||||
FileStorage::load_control_file_conf(conf, ttid)?,
|
||||
))
|
||||
}
|
||||
|
||||
async fn create(
|
||||
fn create(
|
||||
conf: &SafeKeeperConf,
|
||||
ttid: &TenantTimelineId,
|
||||
) -> Result<(FileStorage, SafeKeeperState)> {
|
||||
fs::create_dir_all(conf.timeline_dir(ttid))
|
||||
.await
|
||||
.expect("failed to create timeline dir");
|
||||
fs::create_dir_all(conf.timeline_dir(ttid)).expect("failed to create timeline dir");
|
||||
let state = SafeKeeperState::empty();
|
||||
let storage = FileStorage::create_new(ttid, conf, state.clone())?;
|
||||
Ok((storage, state))
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_read_write_safekeeper_state() {
|
||||
#[test]
|
||||
fn test_read_write_safekeeper_state() {
|
||||
let conf = stub_conf();
|
||||
let ttid = TenantTimelineId::generate();
|
||||
{
|
||||
let (mut storage, mut state) =
|
||||
create(&conf, &ttid).await.expect("failed to create state");
|
||||
let (mut storage, mut state) = create(&conf, &ttid).expect("failed to create state");
|
||||
// change something
|
||||
state.commit_lsn = Lsn(42);
|
||||
storage
|
||||
.persist(&state)
|
||||
.await
|
||||
.expect("failed to persist state");
|
||||
storage.persist(&state).expect("failed to persist state");
|
||||
}
|
||||
|
||||
let (_, state) = load_from_control_file(&conf, &ttid)
|
||||
.await
|
||||
.expect("failed to read state");
|
||||
let (_, state) = load_from_control_file(&conf, &ttid).expect("failed to read state");
|
||||
assert_eq!(state.commit_lsn, Lsn(42));
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_safekeeper_state_checksum_mismatch() {
|
||||
#[test]
|
||||
fn test_safekeeper_state_checksum_mismatch() {
|
||||
let conf = stub_conf();
|
||||
let ttid = TenantTimelineId::generate();
|
||||
{
|
||||
let (mut storage, mut state) =
|
||||
create(&conf, &ttid).await.expect("failed to read state");
|
||||
let (mut storage, mut state) = create(&conf, &ttid).expect("failed to read state");
|
||||
|
||||
// change something
|
||||
state.commit_lsn = Lsn(42);
|
||||
storage
|
||||
.persist(&state)
|
||||
.await
|
||||
.expect("failed to persist state");
|
||||
storage.persist(&state).expect("failed to persist state");
|
||||
}
|
||||
let control_path = conf.timeline_dir(&ttid).join(CONTROL_FILE_NAME);
|
||||
let mut data = fs::read(&control_path).await.unwrap();
|
||||
let mut data = fs::read(&control_path).unwrap();
|
||||
data[0] += 1; // change the first byte of the file to fail checksum validation
|
||||
fs::write(&control_path, &data)
|
||||
.await
|
||||
.expect("failed to write control file");
|
||||
fs::write(&control_path, &data).expect("failed to write control file");
|
||||
|
||||
match load_from_control_file(&conf, &ttid).await {
|
||||
match load_from_control_file(&conf, &ttid) {
|
||||
Err(err) => assert!(err
|
||||
.to_string()
|
||||
.contains("safekeeper control file checksum mismatch")),
|
||||
|
||||
@@ -121,7 +121,7 @@ pub struct FileInfo {
|
||||
}
|
||||
|
||||
/// Build debug dump response, using the provided [`Args`] filters.
|
||||
pub async fn build(args: Args) -> Result<Response> {
|
||||
pub fn build(args: Args) -> Result<Response> {
|
||||
let start_time = Utc::now();
|
||||
let timelines_count = GlobalTimelines::timelines_count();
|
||||
|
||||
@@ -155,7 +155,7 @@ pub async fn build(args: Args) -> Result<Response> {
|
||||
}
|
||||
|
||||
let control_file = if args.dump_control_file {
|
||||
let mut state = tli.get_state().await.1;
|
||||
let mut state = tli.get_state().1;
|
||||
if !args.dump_term_history {
|
||||
state.acceptor_state.term_history = TermHistory(vec![]);
|
||||
}
|
||||
@@ -165,7 +165,7 @@ pub async fn build(args: Args) -> Result<Response> {
|
||||
};
|
||||
|
||||
let memory = if args.dump_memory {
|
||||
Some(tli.memory_dump().await)
|
||||
Some(tli.memory_dump())
|
||||
} else {
|
||||
None
|
||||
};
|
||||
|
||||
@@ -256,14 +256,14 @@ impl SafekeeperPostgresHandler {
|
||||
|
||||
let lsn = if self.is_walproposer_recovery() {
|
||||
// walproposer should get all local WAL until flush_lsn
|
||||
tli.get_flush_lsn().await
|
||||
tli.get_flush_lsn()
|
||||
} else {
|
||||
// other clients shouldn't get any uncommitted WAL
|
||||
tli.get_state().await.0.commit_lsn
|
||||
tli.get_state().0.commit_lsn
|
||||
}
|
||||
.to_string();
|
||||
|
||||
let sysid = tli.get_state().await.1.server.system_id.to_string();
|
||||
let sysid = tli.get_state().1.server.system_id.to_string();
|
||||
let lsn_bytes = lsn.as_bytes();
|
||||
let tli = PG_TLI.to_string();
|
||||
let tli_bytes = tli.as_bytes();
|
||||
|
||||
@@ -2,18 +2,3 @@ pub mod routes;
|
||||
pub use routes::make_router;
|
||||
|
||||
pub use safekeeper_api::models;
|
||||
|
||||
use crate::SafeKeeperConf;
|
||||
|
||||
pub async fn task_main(
|
||||
conf: SafeKeeperConf,
|
||||
http_listener: std::net::TcpListener,
|
||||
) -> anyhow::Result<()> {
|
||||
let router = make_router(conf)
|
||||
.build()
|
||||
.map_err(|err| anyhow::anyhow!(err))?;
|
||||
let service = utils::http::RouterService::new(router).unwrap();
|
||||
let server = hyper::Server::from_tcp(http_listener)?;
|
||||
server.serve(service).await?;
|
||||
Ok(()) // unreachable
|
||||
}
|
||||
|
||||
@@ -13,7 +13,7 @@ use storage_broker::proto::SafekeeperTimelineInfo;
|
||||
use storage_broker::proto::TenantTimelineId as ProtoTenantTimelineId;
|
||||
use tokio::fs::File;
|
||||
use tokio::io::AsyncReadExt;
|
||||
use utils::http::endpoint::request_span;
|
||||
use tokio::task::JoinError;
|
||||
|
||||
use crate::safekeeper::ServerInfo;
|
||||
use crate::safekeeper::Term;
|
||||
@@ -116,8 +116,8 @@ async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body
|
||||
check_permission(&request, Some(ttid.tenant_id))?;
|
||||
|
||||
let tli = GlobalTimelines::get(ttid).map_err(ApiError::from)?;
|
||||
let (inmem, state) = tli.get_state().await;
|
||||
let flush_lsn = tli.get_flush_lsn().await;
|
||||
let (inmem, state) = tli.get_state();
|
||||
let flush_lsn = tli.get_flush_lsn();
|
||||
|
||||
let epoch = state.acceptor_state.get_epoch(flush_lsn);
|
||||
let term_history = state
|
||||
@@ -232,11 +232,13 @@ async fn timeline_delete_force_handler(
|
||||
);
|
||||
check_permission(&request, Some(ttid.tenant_id))?;
|
||||
ensure_no_body(&mut request).await?;
|
||||
// FIXME: `delete_force` can fail from both internal errors and bad requests. Add better
|
||||
// error handling here when we're able to.
|
||||
let resp = GlobalTimelines::delete_force(&ttid)
|
||||
.await
|
||||
.map_err(ApiError::InternalServerError)?;
|
||||
let resp = tokio::task::spawn_blocking(move || {
|
||||
// FIXME: `delete_force` can fail from both internal errors and bad requests. Add better
|
||||
// error handling here when we're able to.
|
||||
GlobalTimelines::delete_force(&ttid).map_err(ApiError::InternalServerError)
|
||||
})
|
||||
.await
|
||||
.map_err(|e: JoinError| ApiError::InternalServerError(e.into()))??;
|
||||
json_response(StatusCode::OK, resp)
|
||||
}
|
||||
|
||||
@@ -248,11 +250,14 @@ async fn tenant_delete_force_handler(
|
||||
let tenant_id = parse_request_param(&request, "tenant_id")?;
|
||||
check_permission(&request, Some(tenant_id))?;
|
||||
ensure_no_body(&mut request).await?;
|
||||
// FIXME: `delete_force_all_for_tenant` can return an error for multiple different reasons;
|
||||
// Using an `InternalServerError` should be fixed when the types support it
|
||||
let delete_info = GlobalTimelines::delete_force_all_for_tenant(&tenant_id)
|
||||
.await
|
||||
.map_err(ApiError::InternalServerError)?;
|
||||
let delete_info = tokio::task::spawn_blocking(move || {
|
||||
// FIXME: `delete_force_all_for_tenant` can return an error for multiple different reasons;
|
||||
// Using an `InternalServerError` should be fixed when the types support it
|
||||
GlobalTimelines::delete_force_all_for_tenant(&tenant_id)
|
||||
.map_err(ApiError::InternalServerError)
|
||||
})
|
||||
.await
|
||||
.map_err(|e: JoinError| ApiError::InternalServerError(e.into()))??;
|
||||
json_response(
|
||||
StatusCode::OK,
|
||||
delete_info
|
||||
@@ -348,9 +353,11 @@ async fn dump_debug_handler(mut request: Request<Body>) -> Result<Response<Body>
|
||||
timeline_id,
|
||||
};
|
||||
|
||||
let resp = debug_dump::build(args)
|
||||
.await
|
||||
.map_err(ApiError::InternalServerError)?;
|
||||
let resp = tokio::task::spawn_blocking(move || {
|
||||
debug_dump::build(args).map_err(ApiError::InternalServerError)
|
||||
})
|
||||
.await
|
||||
.map_err(|e: JoinError| ApiError::InternalServerError(e.into()))??;
|
||||
|
||||
// TODO: use streaming response
|
||||
json_response(StatusCode::OK, resp)
|
||||
@@ -379,32 +386,29 @@ pub fn make_router(conf: SafeKeeperConf) -> RouterBuilder<hyper::Body, ApiError>
|
||||
router
|
||||
.data(Arc::new(conf))
|
||||
.data(auth)
|
||||
.get("/v1/status", |r| request_span(r, status_handler))
|
||||
.get("/v1/status", status_handler)
|
||||
// Will be used in the future instead of implicit timeline creation
|
||||
.post("/v1/tenant/timeline", |r| {
|
||||
request_span(r, timeline_create_handler)
|
||||
})
|
||||
.get("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
|
||||
request_span(r, timeline_status_handler)
|
||||
})
|
||||
.delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
|
||||
request_span(r, timeline_delete_force_handler)
|
||||
})
|
||||
.delete("/v1/tenant/:tenant_id", |r| {
|
||||
request_span(r, tenant_delete_force_handler)
|
||||
})
|
||||
.post("/v1/pull_timeline", |r| {
|
||||
request_span(r, timeline_pull_handler)
|
||||
})
|
||||
.post("/v1/tenant/timeline", timeline_create_handler)
|
||||
.get(
|
||||
"/v1/tenant/:tenant_id/timeline/:timeline_id",
|
||||
timeline_status_handler,
|
||||
)
|
||||
.delete(
|
||||
"/v1/tenant/:tenant_id/timeline/:timeline_id",
|
||||
timeline_delete_force_handler,
|
||||
)
|
||||
.delete("/v1/tenant/:tenant_id", tenant_delete_force_handler)
|
||||
.post("/v1/pull_timeline", timeline_pull_handler)
|
||||
.get(
|
||||
"/v1/tenant/:tenant_id/timeline/:timeline_id/file/:filename",
|
||||
|r| request_span(r, timeline_files_handler),
|
||||
timeline_files_handler,
|
||||
)
|
||||
// for tests
|
||||
.post("/v1/record_safekeeper_info/:tenant_id/:timeline_id", |r| {
|
||||
request_span(r, record_safekeeper_info)
|
||||
})
|
||||
.get("/v1/debug_dump", |r| request_span(r, dump_debug_handler))
|
||||
.post(
|
||||
"/v1/record_safekeeper_info/:tenant_id/:timeline_id",
|
||||
record_safekeeper_info,
|
||||
)
|
||||
.get("/v1/debug_dump", dump_debug_handler)
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
|
||||
@@ -73,12 +73,12 @@ pub async fn handle_json_ctrl<IO: AsyncRead + AsyncWrite + Unpin>(
|
||||
|
||||
// if send_proposer_elected is true, we need to update local history
|
||||
if append_request.send_proposer_elected {
|
||||
send_proposer_elected(&tli, append_request.term, append_request.epoch_start_lsn).await?;
|
||||
send_proposer_elected(&tli, append_request.term, append_request.epoch_start_lsn)?;
|
||||
}
|
||||
|
||||
let inserted_wal = append_logical_message(&tli, append_request).await?;
|
||||
let inserted_wal = append_logical_message(&tli, append_request)?;
|
||||
let response = AppendResult {
|
||||
state: tli.get_state().await.1,
|
||||
state: tli.get_state().1,
|
||||
inserted_wal,
|
||||
};
|
||||
let response_data = serde_json::to_vec(&response)
|
||||
@@ -114,9 +114,9 @@ async fn prepare_safekeeper(
|
||||
.await
|
||||
}
|
||||
|
||||
async fn send_proposer_elected(tli: &Arc<Timeline>, term: Term, lsn: Lsn) -> anyhow::Result<()> {
|
||||
fn send_proposer_elected(tli: &Arc<Timeline>, term: Term, lsn: Lsn) -> anyhow::Result<()> {
|
||||
// add new term to existing history
|
||||
let history = tli.get_state().await.1.acceptor_state.term_history;
|
||||
let history = tli.get_state().1.acceptor_state.term_history;
|
||||
let history = history.up_to(lsn.checked_sub(1u64).unwrap());
|
||||
let mut history_entries = history.0;
|
||||
history_entries.push(TermSwitchEntry { term, lsn });
|
||||
@@ -129,7 +129,7 @@ async fn send_proposer_elected(tli: &Arc<Timeline>, term: Term, lsn: Lsn) -> any
|
||||
timeline_start_lsn: lsn,
|
||||
});
|
||||
|
||||
tli.process_msg(&proposer_elected_request).await?;
|
||||
tli.process_msg(&proposer_elected_request)?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
@@ -142,12 +142,12 @@ pub struct InsertedWAL {
|
||||
|
||||
/// Extend local WAL with new LogicalMessage record. To do that,
|
||||
/// create AppendRequest with new WAL and pass it to safekeeper.
|
||||
pub async fn append_logical_message(
|
||||
pub fn append_logical_message(
|
||||
tli: &Arc<Timeline>,
|
||||
msg: &AppendLogicalMessage,
|
||||
) -> anyhow::Result<InsertedWAL> {
|
||||
let wal_data = encode_logical_message(&msg.lm_prefix, &msg.lm_message);
|
||||
let sk_state = tli.get_state().await.1;
|
||||
let sk_state = tli.get_state().1;
|
||||
|
||||
let begin_lsn = msg.begin_lsn;
|
||||
let end_lsn = begin_lsn + wal_data.len() as u64;
|
||||
@@ -171,7 +171,7 @@ pub async fn append_logical_message(
|
||||
wal_data: Bytes::from(wal_data),
|
||||
});
|
||||
|
||||
let response = tli.process_msg(&append_request).await?;
|
||||
let response = tli.process_msg(&append_request)?;
|
||||
|
||||
let append_response = match response {
|
||||
Some(AcceptorProposerMessage::AppendResponse(resp)) => resp,
|
||||
|
||||
@@ -1,6 +1,4 @@
|
||||
use once_cell::sync::Lazy;
|
||||
use remote_storage::RemoteStorageConfig;
|
||||
use tokio::runtime::Runtime;
|
||||
|
||||
use std::path::PathBuf;
|
||||
use std::time::Duration;
|
||||
@@ -38,6 +36,7 @@ pub mod defaults {
|
||||
DEFAULT_PG_LISTEN_PORT,
|
||||
};
|
||||
|
||||
pub const DEFAULT_WAL_BACKUP_RUNTIME_THREADS: usize = 8;
|
||||
pub const DEFAULT_HEARTBEAT_TIMEOUT: &str = "5000ms";
|
||||
pub const DEFAULT_MAX_OFFLOADER_LAG_BYTES: u64 = 128 * (1 << 20);
|
||||
}
|
||||
@@ -61,10 +60,10 @@ pub struct SafeKeeperConf {
|
||||
pub heartbeat_timeout: Duration,
|
||||
pub remote_storage: Option<RemoteStorageConfig>,
|
||||
pub max_offloader_lag_bytes: u64,
|
||||
pub backup_runtime_threads: Option<usize>,
|
||||
pub backup_parallel_jobs: usize,
|
||||
pub wal_backup_enabled: bool,
|
||||
pub auth: Option<Arc<JwtAuth>>,
|
||||
pub current_thread_runtime: bool,
|
||||
}
|
||||
|
||||
impl SafeKeeperConf {
|
||||
@@ -93,64 +92,12 @@ impl SafeKeeperConf {
|
||||
.parse()
|
||||
.expect("failed to parse default broker endpoint"),
|
||||
broker_keepalive_interval: Duration::from_secs(5),
|
||||
backup_runtime_threads: None,
|
||||
wal_backup_enabled: true,
|
||||
backup_parallel_jobs: 1,
|
||||
auth: None,
|
||||
heartbeat_timeout: Duration::new(5, 0),
|
||||
max_offloader_lag_bytes: defaults::DEFAULT_MAX_OFFLOADER_LAG_BYTES,
|
||||
current_thread_runtime: false,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Tokio runtimes.
|
||||
pub static WAL_SERVICE_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
|
||||
tokio::runtime::Builder::new_multi_thread()
|
||||
.thread_name("WAL service worker")
|
||||
.enable_all()
|
||||
.build()
|
||||
.expect("Failed to create WAL service runtime")
|
||||
});
|
||||
|
||||
pub static HTTP_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
|
||||
tokio::runtime::Builder::new_multi_thread()
|
||||
.thread_name("HTTP worker")
|
||||
.enable_all()
|
||||
.build()
|
||||
.expect("Failed to create WAL service runtime")
|
||||
});
|
||||
|
||||
pub static BROKER_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
|
||||
tokio::runtime::Builder::new_multi_thread()
|
||||
.thread_name("broker worker")
|
||||
.worker_threads(2) // there are only 2 tasks, having more threads doesn't make sense
|
||||
.enable_all()
|
||||
.build()
|
||||
.expect("Failed to create broker runtime")
|
||||
});
|
||||
|
||||
pub static WAL_REMOVER_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
|
||||
tokio::runtime::Builder::new_multi_thread()
|
||||
.thread_name("WAL remover")
|
||||
.worker_threads(1)
|
||||
.enable_all()
|
||||
.build()
|
||||
.expect("Failed to create broker runtime")
|
||||
});
|
||||
|
||||
pub static WAL_BACKUP_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
|
||||
tokio::runtime::Builder::new_multi_thread()
|
||||
.thread_name("WAL backup worker")
|
||||
.enable_all()
|
||||
.build()
|
||||
.expect("Failed to create WAL backup runtime")
|
||||
});
|
||||
|
||||
pub static METRICS_SHIFTER_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
|
||||
tokio::runtime::Builder::new_multi_thread()
|
||||
.thread_name("metric shifter")
|
||||
.worker_threads(1)
|
||||
.enable_all()
|
||||
.build()
|
||||
.expect("Failed to create broker runtime")
|
||||
});
|
||||
|
||||
@@ -7,7 +7,6 @@ use std::{
|
||||
|
||||
use ::metrics::{register_histogram, GaugeVec, Histogram, IntGauge, DISK_WRITE_SECONDS_BUCKETS};
|
||||
use anyhow::Result;
|
||||
use futures::Future;
|
||||
use metrics::{
|
||||
core::{AtomicU64, Collector, Desc, GenericCounter, GenericGaugeVec, Opts},
|
||||
proto::MetricFamily,
|
||||
@@ -293,17 +292,14 @@ impl WalStorageMetrics {
|
||||
}
|
||||
}
|
||||
|
||||
/// Accepts async function that returns empty anyhow result, and returns the duration of its execution.
|
||||
pub async fn time_io_closure<E: Into<anyhow::Error>>(
|
||||
closure: impl Future<Output = Result<(), E>>,
|
||||
) -> Result<f64> {
|
||||
/// Accepts a closure that returns a result, and returns the duration of the closure.
|
||||
pub fn time_io_closure(closure: impl FnOnce() -> Result<()>) -> Result<f64> {
|
||||
let start = std::time::Instant::now();
|
||||
closure.await.map_err(|e| e.into())?;
|
||||
closure()?;
|
||||
Ok(start.elapsed().as_secs_f64())
|
||||
}
|
||||
|
||||
/// Metrics for a single timeline.
|
||||
#[derive(Clone)]
|
||||
pub struct FullTimelineInfo {
|
||||
pub ttid: TenantTimelineId,
|
||||
pub ps_feedback: PageserverFeedback,
|
||||
@@ -579,19 +575,13 @@ impl Collector for TimelineCollector {
|
||||
let timelines = GlobalTimelines::get_all();
|
||||
let timelines_count = timelines.len();
|
||||
|
||||
// Prometheus Collector is sync, and data is stored under async lock. To
|
||||
// bridge the gap with a crutch, collect data in spawned thread with
|
||||
// local tokio runtime.
|
||||
let infos = std::thread::spawn(|| {
|
||||
let rt = tokio::runtime::Builder::new_current_thread()
|
||||
.build()
|
||||
.expect("failed to create rt");
|
||||
rt.block_on(collect_timeline_metrics())
|
||||
})
|
||||
.join()
|
||||
.expect("collect_timeline_metrics thread panicked");
|
||||
for arc_tli in timelines {
|
||||
let tli = arc_tli.info_for_metrics();
|
||||
if tli.is_none() {
|
||||
continue;
|
||||
}
|
||||
let tli = tli.unwrap();
|
||||
|
||||
for tli in &infos {
|
||||
let tenant_id = tli.ttid.tenant_id.to_string();
|
||||
let timeline_id = tli.ttid.timeline_id.to_string();
|
||||
let labels = &[tenant_id.as_str(), timeline_id.as_str()];
|
||||
@@ -692,15 +682,3 @@ impl Collector for TimelineCollector {
|
||||
mfs
|
||||
}
|
||||
}
|
||||
|
||||
async fn collect_timeline_metrics() -> Vec<FullTimelineInfo> {
|
||||
let mut res = vec![];
|
||||
let timelines = GlobalTimelines::get_all();
|
||||
|
||||
for tli in timelines {
|
||||
if let Some(info) = tli.info_for_metrics().await {
|
||||
res.push(info);
|
||||
}
|
||||
}
|
||||
res
|
||||
}
|
||||
|
||||
@@ -231,7 +231,7 @@ async fn pull_timeline(status: TimelineStatus, host: String) -> Result<Response>
|
||||
info!(
|
||||
"Loaded timeline {}, flush_lsn={}",
|
||||
ttid,
|
||||
tli.get_flush_lsn().await
|
||||
tli.get_flush_lsn()
|
||||
);
|
||||
|
||||
Ok(Response {
|
||||
|
||||
@@ -18,14 +18,15 @@ use postgres_backend::QueryError;
|
||||
use pq_proto::BeMessage;
|
||||
use std::net::SocketAddr;
|
||||
use std::sync::Arc;
|
||||
use std::thread;
|
||||
use std::thread::JoinHandle;
|
||||
use tokio::io::AsyncRead;
|
||||
use tokio::io::AsyncWrite;
|
||||
use tokio::sync::mpsc::channel;
|
||||
use tokio::sync::mpsc::error::TryRecvError;
|
||||
use tokio::sync::mpsc::Receiver;
|
||||
use tokio::sync::mpsc::Sender;
|
||||
use tokio::task;
|
||||
use tokio::task::JoinHandle;
|
||||
use tokio::task::spawn_blocking;
|
||||
use tokio::time::Duration;
|
||||
use tokio::time::Instant;
|
||||
use tracing::*;
|
||||
@@ -96,7 +97,7 @@ impl SafekeeperPostgresHandler {
|
||||
Err(res.expect_err("no error with WalAcceptor not spawn"))
|
||||
}
|
||||
Some(handle) => {
|
||||
let wal_acceptor_res = handle.await;
|
||||
let wal_acceptor_res = handle.join();
|
||||
|
||||
// If there was any network error, return it.
|
||||
res?;
|
||||
@@ -106,7 +107,7 @@ impl SafekeeperPostgresHandler {
|
||||
Ok(Ok(_)) => Ok(()), // can't happen currently; would be if we add graceful termination
|
||||
Ok(Err(e)) => Err(CopyStreamHandlerEnd::Other(e.context("WAL acceptor"))),
|
||||
Err(_) => Err(CopyStreamHandlerEnd::Other(anyhow!(
|
||||
"WalAcceptor task panicked",
|
||||
"WalAcceptor thread panicked",
|
||||
))),
|
||||
}
|
||||
}
|
||||
@@ -153,12 +154,10 @@ impl<'a, IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'a, IO> {
|
||||
}
|
||||
};
|
||||
|
||||
*self.acceptor_handle = Some(WalAcceptor::spawn(
|
||||
tli.clone(),
|
||||
msg_rx,
|
||||
reply_tx,
|
||||
self.conn_id,
|
||||
));
|
||||
*self.acceptor_handle = Some(
|
||||
WalAcceptor::spawn(tli.clone(), msg_rx, reply_tx, self.conn_id)
|
||||
.context("spawn WalAcceptor thread")?,
|
||||
);
|
||||
|
||||
// Forward all messages to WalAcceptor
|
||||
read_network_loop(self.pgb_reader, msg_tx, next_msg).await
|
||||
@@ -227,19 +226,28 @@ impl WalAcceptor {
|
||||
msg_rx: Receiver<ProposerAcceptorMessage>,
|
||||
reply_tx: Sender<AcceptorProposerMessage>,
|
||||
conn_id: ConnectionId,
|
||||
) -> JoinHandle<anyhow::Result<()>> {
|
||||
task::spawn(async move {
|
||||
let mut wa = WalAcceptor {
|
||||
tli,
|
||||
msg_rx,
|
||||
reply_tx,
|
||||
};
|
||||
) -> anyhow::Result<JoinHandle<anyhow::Result<()>>> {
|
||||
let thread_name = format!("WAL acceptor {}", tli.ttid);
|
||||
thread::Builder::new()
|
||||
.name(thread_name)
|
||||
.spawn(move || -> anyhow::Result<()> {
|
||||
let mut wa = WalAcceptor {
|
||||
tli,
|
||||
msg_rx,
|
||||
reply_tx,
|
||||
};
|
||||
|
||||
let span_ttid = wa.tli.ttid; // satisfy borrow checker
|
||||
wa.run()
|
||||
.instrument(info_span!("WAL acceptor", cid = %conn_id, ttid = %span_ttid))
|
||||
.await
|
||||
})
|
||||
let runtime = tokio::runtime::Builder::new_current_thread()
|
||||
.enable_all()
|
||||
.build()?;
|
||||
|
||||
let span_ttid = wa.tli.ttid; // satisfy borrow checker
|
||||
runtime.block_on(
|
||||
wa.run()
|
||||
.instrument(info_span!("WAL acceptor", cid = %conn_id, ttid = %span_ttid)),
|
||||
)
|
||||
})
|
||||
.map_err(anyhow::Error::from)
|
||||
}
|
||||
|
||||
/// The main loop. Returns Ok(()) if either msg_rx or reply_tx got closed;
|
||||
@@ -273,7 +281,7 @@ impl WalAcceptor {
|
||||
while let ProposerAcceptorMessage::AppendRequest(append_request) = next_msg {
|
||||
let noflush_msg = ProposerAcceptorMessage::NoFlushAppendRequest(append_request);
|
||||
|
||||
if let Some(reply) = self.tli.process_msg(&noflush_msg).await? {
|
||||
if let Some(reply) = self.tli.process_msg(&noflush_msg)? {
|
||||
if self.reply_tx.send(reply).await.is_err() {
|
||||
return Ok(()); // chan closed, streaming terminated
|
||||
}
|
||||
@@ -292,12 +300,10 @@ impl WalAcceptor {
|
||||
}
|
||||
|
||||
// flush all written WAL to the disk
|
||||
self.tli
|
||||
.process_msg(&ProposerAcceptorMessage::FlushWAL)
|
||||
.await?
|
||||
self.tli.process_msg(&ProposerAcceptorMessage::FlushWAL)?
|
||||
} else {
|
||||
// process message other than AppendRequest
|
||||
self.tli.process_msg(&next_msg).await?
|
||||
self.tli.process_msg(&next_msg)?
|
||||
};
|
||||
|
||||
if let Some(reply) = reply_msg {
|
||||
@@ -320,8 +326,8 @@ impl Drop for ComputeConnectionGuard {
|
||||
let tli = self.timeline.clone();
|
||||
// tokio forbids to call blocking_send inside the runtime, and see
|
||||
// comments in on_compute_disconnect why we call blocking_send.
|
||||
tokio::spawn(async move {
|
||||
if let Err(e) = tli.on_compute_disconnect().await {
|
||||
spawn_blocking(move || {
|
||||
if let Err(e) = tli.on_compute_disconnect() {
|
||||
error!("failed to unregister compute connection: {}", e);
|
||||
}
|
||||
});
|
||||
|
||||
@@ -1,36 +1,29 @@
|
||||
//! Thread removing old WAL.
|
||||
|
||||
use std::time::Duration;
|
||||
use std::{thread, time::Duration};
|
||||
|
||||
use tokio::time::sleep;
|
||||
use tracing::*;
|
||||
|
||||
use crate::{GlobalTimelines, SafeKeeperConf};
|
||||
|
||||
pub async fn task_main(conf: SafeKeeperConf) -> anyhow::Result<()> {
|
||||
pub fn thread_main(conf: SafeKeeperConf) {
|
||||
let wal_removal_interval = Duration::from_millis(5000);
|
||||
loop {
|
||||
let tlis = GlobalTimelines::get_all();
|
||||
for tli in &tlis {
|
||||
if !tli.is_active().await {
|
||||
if !tli.is_active() {
|
||||
continue;
|
||||
}
|
||||
let ttid = tli.ttid;
|
||||
if let Err(e) = tli
|
||||
.maybe_persist_control_file()
|
||||
.instrument(info_span!("", tenant = %ttid.tenant_id, timeline = %ttid.timeline_id))
|
||||
.await
|
||||
{
|
||||
let _enter =
|
||||
info_span!("", tenant = %ttid.tenant_id, timeline = %ttid.timeline_id).entered();
|
||||
if let Err(e) = tli.maybe_pesist_control_file() {
|
||||
warn!("failed to persist control file: {e}");
|
||||
}
|
||||
if let Err(e) = tli
|
||||
.remove_old_wal(conf.wal_backup_enabled)
|
||||
.instrument(info_span!("", tenant = %ttid.tenant_id, timeline = %ttid.timeline_id))
|
||||
.await
|
||||
{
|
||||
error!("failed to remove WAL: {}", e);
|
||||
if let Err(e) = tli.remove_old_wal(conf.wal_backup_enabled) {
|
||||
warn!("failed to remove WAL: {}", e);
|
||||
}
|
||||
}
|
||||
sleep(wal_removal_interval).await;
|
||||
thread::sleep(wal_removal_interval)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -568,27 +568,25 @@ where
|
||||
|
||||
/// Process message from proposer and possibly form reply. Concurrent
|
||||
/// callers must exclude each other.
|
||||
pub async fn process_msg(
|
||||
pub fn process_msg(
|
||||
&mut self,
|
||||
msg: &ProposerAcceptorMessage,
|
||||
) -> Result<Option<AcceptorProposerMessage>> {
|
||||
match msg {
|
||||
ProposerAcceptorMessage::Greeting(msg) => self.handle_greeting(msg).await,
|
||||
ProposerAcceptorMessage::VoteRequest(msg) => self.handle_vote_request(msg).await,
|
||||
ProposerAcceptorMessage::Elected(msg) => self.handle_elected(msg).await,
|
||||
ProposerAcceptorMessage::AppendRequest(msg) => {
|
||||
self.handle_append_request(msg, true).await
|
||||
}
|
||||
ProposerAcceptorMessage::Greeting(msg) => self.handle_greeting(msg),
|
||||
ProposerAcceptorMessage::VoteRequest(msg) => self.handle_vote_request(msg),
|
||||
ProposerAcceptorMessage::Elected(msg) => self.handle_elected(msg),
|
||||
ProposerAcceptorMessage::AppendRequest(msg) => self.handle_append_request(msg, true),
|
||||
ProposerAcceptorMessage::NoFlushAppendRequest(msg) => {
|
||||
self.handle_append_request(msg, false).await
|
||||
self.handle_append_request(msg, false)
|
||||
}
|
||||
ProposerAcceptorMessage::FlushWAL => self.handle_flush().await,
|
||||
ProposerAcceptorMessage::FlushWAL => self.handle_flush(),
|
||||
}
|
||||
}
|
||||
|
||||
/// Handle initial message from proposer: check its sanity and send my
|
||||
/// current term.
|
||||
async fn handle_greeting(
|
||||
fn handle_greeting(
|
||||
&mut self,
|
||||
msg: &ProposerGreeting,
|
||||
) -> Result<Option<AcceptorProposerMessage>> {
|
||||
@@ -651,7 +649,7 @@ where
|
||||
if msg.pg_version != UNKNOWN_SERVER_VERSION {
|
||||
state.server.pg_version = msg.pg_version;
|
||||
}
|
||||
self.state.persist(&state).await?;
|
||||
self.state.persist(&state)?;
|
||||
}
|
||||
|
||||
info!(
|
||||
@@ -666,7 +664,7 @@ where
|
||||
}
|
||||
|
||||
/// Give vote for the given term, if we haven't done that previously.
|
||||
async fn handle_vote_request(
|
||||
fn handle_vote_request(
|
||||
&mut self,
|
||||
msg: &VoteRequest,
|
||||
) -> Result<Option<AcceptorProposerMessage>> {
|
||||
@@ -680,7 +678,7 @@ where
|
||||
// handle_elected instead. Currently not a big deal, as proposer is the
|
||||
// only source of WAL; with peer2peer recovery it would be more
|
||||
// important.
|
||||
self.wal_store.flush_wal().await?;
|
||||
self.wal_store.flush_wal()?;
|
||||
// initialize with refusal
|
||||
let mut resp = VoteResponse {
|
||||
term: self.state.acceptor_state.term,
|
||||
@@ -694,7 +692,7 @@ where
|
||||
let mut state = self.state.clone();
|
||||
state.acceptor_state.term = msg.term;
|
||||
// persist vote before sending it out
|
||||
self.state.persist(&state).await?;
|
||||
self.state.persist(&state)?;
|
||||
|
||||
resp.term = self.state.acceptor_state.term;
|
||||
resp.vote_given = true as u64;
|
||||
@@ -717,15 +715,12 @@ where
|
||||
ar
|
||||
}
|
||||
|
||||
async fn handle_elected(
|
||||
&mut self,
|
||||
msg: &ProposerElected,
|
||||
) -> Result<Option<AcceptorProposerMessage>> {
|
||||
fn handle_elected(&mut self, msg: &ProposerElected) -> Result<Option<AcceptorProposerMessage>> {
|
||||
info!("received ProposerElected {:?}", msg);
|
||||
if self.state.acceptor_state.term < msg.term {
|
||||
let mut state = self.state.clone();
|
||||
state.acceptor_state.term = msg.term;
|
||||
self.state.persist(&state).await?;
|
||||
self.state.persist(&state)?;
|
||||
}
|
||||
|
||||
// If our term is higher, ignore the message (next feedback will inform the compute)
|
||||
@@ -755,7 +750,7 @@ where
|
||||
// intersection of our history and history from msg
|
||||
|
||||
// truncate wal, update the LSNs
|
||||
self.wal_store.truncate_wal(msg.start_streaming_at).await?;
|
||||
self.wal_store.truncate_wal(msg.start_streaming_at)?;
|
||||
|
||||
// and now adopt term history from proposer
|
||||
{
|
||||
@@ -789,7 +784,7 @@ where
|
||||
self.inmem.backup_lsn = max(self.inmem.backup_lsn, state.timeline_start_lsn);
|
||||
|
||||
state.acceptor_state.term_history = msg.term_history.clone();
|
||||
self.persist_control_file(state).await?;
|
||||
self.persist_control_file(state)?;
|
||||
}
|
||||
|
||||
info!("start receiving WAL since {:?}", msg.start_streaming_at);
|
||||
@@ -801,7 +796,7 @@ where
|
||||
///
|
||||
/// Note: it is assumed that 'WAL we have is from the right term' check has
|
||||
/// already been done outside.
|
||||
async fn update_commit_lsn(&mut self, mut candidate: Lsn) -> Result<()> {
|
||||
fn update_commit_lsn(&mut self, mut candidate: Lsn) -> Result<()> {
|
||||
// Both peers and walproposer communicate this value, we might already
|
||||
// have a fresher (higher) version.
|
||||
candidate = max(candidate, self.inmem.commit_lsn);
|
||||
@@ -823,32 +818,29 @@ where
|
||||
// that we receive new epoch_start_lsn, and we still need to sync
|
||||
// control file in this case.
|
||||
if commit_lsn == self.epoch_start_lsn && self.state.commit_lsn != commit_lsn {
|
||||
self.persist_control_file(self.state.clone()).await?;
|
||||
self.persist_control_file(self.state.clone())?;
|
||||
}
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Persist control file to disk, called only after timeline creation (bootstrap).
|
||||
pub async fn persist(&mut self) -> Result<()> {
|
||||
self.persist_control_file(self.state.clone()).await
|
||||
pub fn persist(&mut self) -> Result<()> {
|
||||
self.persist_control_file(self.state.clone())
|
||||
}
|
||||
|
||||
/// Persist in-memory state to the disk, taking other data from state.
|
||||
async fn persist_control_file(&mut self, mut state: SafeKeeperState) -> Result<()> {
|
||||
fn persist_control_file(&mut self, mut state: SafeKeeperState) -> Result<()> {
|
||||
state.commit_lsn = self.inmem.commit_lsn;
|
||||
state.backup_lsn = self.inmem.backup_lsn;
|
||||
state.peer_horizon_lsn = self.inmem.peer_horizon_lsn;
|
||||
state.proposer_uuid = self.inmem.proposer_uuid;
|
||||
self.state.persist(&state).await
|
||||
self.state.persist(&state)
|
||||
}
|
||||
|
||||
/// Persist control file if there is something to save and enough time
|
||||
/// passed after the last save.
|
||||
pub async fn maybe_persist_control_file(
|
||||
&mut self,
|
||||
inmem_remote_consistent_lsn: Lsn,
|
||||
) -> Result<()> {
|
||||
pub fn maybe_persist_control_file(&mut self, inmem_remote_consistent_lsn: Lsn) -> Result<()> {
|
||||
const CF_SAVE_INTERVAL: Duration = Duration::from_secs(300);
|
||||
if self.state.last_persist_at().elapsed() < CF_SAVE_INTERVAL {
|
||||
return Ok(());
|
||||
@@ -860,7 +852,7 @@ where
|
||||
if need_persist {
|
||||
let mut state = self.state.clone();
|
||||
state.remote_consistent_lsn = inmem_remote_consistent_lsn;
|
||||
self.persist_control_file(state).await?;
|
||||
self.persist_control_file(state)?;
|
||||
trace!("saved control file: {CF_SAVE_INTERVAL:?} passed");
|
||||
}
|
||||
Ok(())
|
||||
@@ -868,7 +860,7 @@ where
|
||||
|
||||
/// Handle request to append WAL.
|
||||
#[allow(clippy::comparison_chain)]
|
||||
async fn handle_append_request(
|
||||
fn handle_append_request(
|
||||
&mut self,
|
||||
msg: &AppendRequest,
|
||||
require_flush: bool,
|
||||
@@ -891,19 +883,17 @@ where
|
||||
|
||||
// do the job
|
||||
if !msg.wal_data.is_empty() {
|
||||
self.wal_store
|
||||
.write_wal(msg.h.begin_lsn, &msg.wal_data)
|
||||
.await?;
|
||||
self.wal_store.write_wal(msg.h.begin_lsn, &msg.wal_data)?;
|
||||
}
|
||||
|
||||
// flush wal to the disk, if required
|
||||
if require_flush {
|
||||
self.wal_store.flush_wal().await?;
|
||||
self.wal_store.flush_wal()?;
|
||||
}
|
||||
|
||||
// Update commit_lsn.
|
||||
if msg.h.commit_lsn != Lsn(0) {
|
||||
self.update_commit_lsn(msg.h.commit_lsn).await?;
|
||||
self.update_commit_lsn(msg.h.commit_lsn)?;
|
||||
}
|
||||
// Value calculated by walproposer can always lag:
|
||||
// - safekeepers can forget inmem value and send to proposer lower
|
||||
@@ -919,7 +909,7 @@ where
|
||||
if self.state.peer_horizon_lsn + (self.state.server.wal_seg_size as u64)
|
||||
< self.inmem.peer_horizon_lsn
|
||||
{
|
||||
self.persist_control_file(self.state.clone()).await?;
|
||||
self.persist_control_file(self.state.clone())?;
|
||||
}
|
||||
|
||||
trace!(
|
||||
@@ -941,15 +931,15 @@ where
|
||||
}
|
||||
|
||||
/// Flush WAL to disk. Return AppendResponse with latest LSNs.
|
||||
async fn handle_flush(&mut self) -> Result<Option<AcceptorProposerMessage>> {
|
||||
self.wal_store.flush_wal().await?;
|
||||
fn handle_flush(&mut self) -> Result<Option<AcceptorProposerMessage>> {
|
||||
self.wal_store.flush_wal()?;
|
||||
Ok(Some(AcceptorProposerMessage::AppendResponse(
|
||||
self.append_response(),
|
||||
)))
|
||||
}
|
||||
|
||||
/// Update timeline state with peer safekeeper data.
|
||||
pub async fn record_safekeeper_info(&mut self, sk_info: &SafekeeperTimelineInfo) -> Result<()> {
|
||||
pub fn record_safekeeper_info(&mut self, sk_info: &SafekeeperTimelineInfo) -> Result<()> {
|
||||
let mut sync_control_file = false;
|
||||
|
||||
if (Lsn(sk_info.commit_lsn) != Lsn::INVALID) && (sk_info.last_log_term != INVALID_TERM) {
|
||||
@@ -957,7 +947,7 @@ where
|
||||
// commit_lsn if our history matches (is part of) history of advanced
|
||||
// commit_lsn provider.
|
||||
if sk_info.last_log_term == self.get_epoch() {
|
||||
self.update_commit_lsn(Lsn(sk_info.commit_lsn)).await?;
|
||||
self.update_commit_lsn(Lsn(sk_info.commit_lsn))?;
|
||||
}
|
||||
}
|
||||
|
||||
@@ -983,7 +973,7 @@ where
|
||||
// Note: we could make remote_consistent_lsn update in cf common by
|
||||
// storing Arc to walsenders in Safekeeper.
|
||||
state.remote_consistent_lsn = new_remote_consistent_lsn;
|
||||
self.persist_control_file(state).await?;
|
||||
self.persist_control_file(state)?;
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
@@ -1007,7 +997,6 @@ where
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use futures::future::BoxFuture;
|
||||
use postgres_ffi::WAL_SEGMENT_SIZE;
|
||||
|
||||
use super::*;
|
||||
@@ -1019,9 +1008,8 @@ mod tests {
|
||||
persisted_state: SafeKeeperState,
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl control_file::Storage for InMemoryState {
|
||||
async fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
|
||||
fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
|
||||
self.persisted_state = s.clone();
|
||||
Ok(())
|
||||
}
|
||||
@@ -1051,28 +1039,27 @@ mod tests {
|
||||
lsn: Lsn,
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl wal_storage::Storage for DummyWalStore {
|
||||
fn flush_lsn(&self) -> Lsn {
|
||||
self.lsn
|
||||
}
|
||||
|
||||
async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
|
||||
fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
|
||||
self.lsn = startpos + buf.len() as u64;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
|
||||
fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
|
||||
self.lsn = end_pos;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn flush_wal(&mut self) -> Result<()> {
|
||||
fn flush_wal(&mut self) -> Result<()> {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn remove_up_to(&self, _segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>> {
|
||||
Box::pin(async { Ok(()) })
|
||||
fn remove_up_to(&self) -> Box<dyn Fn(XLogSegNo) -> Result<()>> {
|
||||
Box::new(move |_segno_up_to: XLogSegNo| Ok(()))
|
||||
}
|
||||
|
||||
fn get_metrics(&self) -> crate::metrics::WalStorageMetrics {
|
||||
@@ -1080,8 +1067,8 @@ mod tests {
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_voting() {
|
||||
#[test]
|
||||
fn test_voting() {
|
||||
let storage = InMemoryState {
|
||||
persisted_state: test_sk_state(),
|
||||
};
|
||||
@@ -1090,7 +1077,7 @@ mod tests {
|
||||
|
||||
// check voting for 1 is ok
|
||||
let vote_request = ProposerAcceptorMessage::VoteRequest(VoteRequest { term: 1 });
|
||||
let mut vote_resp = sk.process_msg(&vote_request).await;
|
||||
let mut vote_resp = sk.process_msg(&vote_request);
|
||||
match vote_resp.unwrap() {
|
||||
Some(AcceptorProposerMessage::VoteResponse(resp)) => assert!(resp.vote_given != 0),
|
||||
r => panic!("unexpected response: {:?}", r),
|
||||
@@ -1105,15 +1092,15 @@ mod tests {
|
||||
sk = SafeKeeper::new(storage, sk.wal_store, NodeId(0)).unwrap();
|
||||
|
||||
// and ensure voting second time for 1 is not ok
|
||||
vote_resp = sk.process_msg(&vote_request).await;
|
||||
vote_resp = sk.process_msg(&vote_request);
|
||||
match vote_resp.unwrap() {
|
||||
Some(AcceptorProposerMessage::VoteResponse(resp)) => assert!(resp.vote_given == 0),
|
||||
r => panic!("unexpected response: {:?}", r),
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_epoch_switch() {
|
||||
#[test]
|
||||
fn test_epoch_switch() {
|
||||
let storage = InMemoryState {
|
||||
persisted_state: test_sk_state(),
|
||||
};
|
||||
@@ -1145,13 +1132,10 @@ mod tests {
|
||||
timeline_start_lsn: Lsn(0),
|
||||
};
|
||||
sk.process_msg(&ProposerAcceptorMessage::Elected(pem))
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
// check that AppendRequest before epochStartLsn doesn't switch epoch
|
||||
let resp = sk
|
||||
.process_msg(&ProposerAcceptorMessage::AppendRequest(append_request))
|
||||
.await;
|
||||
let resp = sk.process_msg(&ProposerAcceptorMessage::AppendRequest(append_request));
|
||||
assert!(resp.is_ok());
|
||||
assert_eq!(sk.get_epoch(), 0);
|
||||
|
||||
@@ -1162,11 +1146,9 @@ mod tests {
|
||||
h: ar_hdr,
|
||||
wal_data: Bytes::from_static(b"b"),
|
||||
};
|
||||
let resp = sk
|
||||
.process_msg(&ProposerAcceptorMessage::AppendRequest(append_request))
|
||||
.await;
|
||||
let resp = sk.process_msg(&ProposerAcceptorMessage::AppendRequest(append_request));
|
||||
assert!(resp.is_ok());
|
||||
sk.wal_store.truncate_wal(Lsn(3)).await.unwrap(); // imitate the complete record at 3 %)
|
||||
sk.wal_store.truncate_wal(Lsn(3)).unwrap(); // imitate the complete record at 3 %)
|
||||
assert_eq!(sk.get_epoch(), 1);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -396,7 +396,7 @@ impl SafekeeperPostgresHandler {
|
||||
// on this safekeeper itself. That's ok as (old) proposer will never be
|
||||
// able to commit such WAL.
|
||||
let stop_pos: Option<Lsn> = if self.is_walproposer_recovery() {
|
||||
let wal_end = tli.get_flush_lsn().await;
|
||||
let wal_end = tli.get_flush_lsn();
|
||||
Some(wal_end)
|
||||
} else {
|
||||
None
|
||||
@@ -418,7 +418,7 @@ impl SafekeeperPostgresHandler {
|
||||
// switch to copy
|
||||
pgb.write_message(&BeMessage::CopyBothResponse).await?;
|
||||
|
||||
let (_, persisted_state) = tli.get_state().await;
|
||||
let (_, persisted_state) = tli.get_state();
|
||||
let wal_reader = WalReader::new(
|
||||
self.conf.workdir.clone(),
|
||||
self.conf.timeline_dir(&tli.ttid),
|
||||
@@ -562,7 +562,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
|
||||
.walsenders
|
||||
.get_ws_remote_consistent_lsn(self.ws_guard.id)
|
||||
{
|
||||
if self.tli.should_walsender_stop(remote_consistent_lsn).await {
|
||||
if self.tli.should_walsender_stop(remote_consistent_lsn) {
|
||||
// Terminate if there is nothing more to send.
|
||||
return Err(CopyStreamHandlerEnd::ServerInitiated(format!(
|
||||
"ending streaming to {:?} at {}, receiver is caughtup and there is no computes",
|
||||
|
||||
@@ -2,13 +2,12 @@
|
||||
//! to glue together SafeKeeper and all other background services.
|
||||
|
||||
use anyhow::{anyhow, bail, Result};
|
||||
use parking_lot::{Mutex, MutexGuard};
|
||||
use postgres_ffi::XLogSegNo;
|
||||
use tokio::fs;
|
||||
|
||||
use std::cmp::max;
|
||||
use std::path::PathBuf;
|
||||
use std::sync::Arc;
|
||||
use tokio::sync::{Mutex, MutexGuard};
|
||||
use tokio::{
|
||||
sync::{mpsc::Sender, watch},
|
||||
time::Instant,
|
||||
@@ -287,9 +286,8 @@ pub struct Timeline {
|
||||
commit_lsn_watch_tx: watch::Sender<Lsn>,
|
||||
commit_lsn_watch_rx: watch::Receiver<Lsn>,
|
||||
|
||||
/// Safekeeper and other state, that should remain consistent and
|
||||
/// synchronized with the disk. This is tokio mutex as we write WAL to disk
|
||||
/// while holding it, ensuring that consensus checks are in order.
|
||||
/// Safekeeper and other state, that should remain consistent and synchronized
|
||||
/// with the disk.
|
||||
mutex: Mutex<SharedState>,
|
||||
walsenders: Arc<WalSenders>,
|
||||
|
||||
@@ -363,8 +361,8 @@ impl Timeline {
|
||||
///
|
||||
/// Bootstrap is transactional, so if it fails, created files will be deleted,
|
||||
/// and state on disk should remain unchanged.
|
||||
pub async fn bootstrap(&self, shared_state: &mut MutexGuard<'_, SharedState>) -> Result<()> {
|
||||
match fs::metadata(&self.timeline_dir).await {
|
||||
pub fn bootstrap(&self, shared_state: &mut MutexGuard<SharedState>) -> Result<()> {
|
||||
match std::fs::metadata(&self.timeline_dir) {
|
||||
Ok(_) => {
|
||||
// Timeline directory exists on disk, we should leave state unchanged
|
||||
// and return error.
|
||||
@@ -377,51 +375,53 @@ impl Timeline {
|
||||
}
|
||||
|
||||
// Create timeline directory.
|
||||
fs::create_dir_all(&self.timeline_dir).await?;
|
||||
std::fs::create_dir_all(&self.timeline_dir)?;
|
||||
|
||||
// Write timeline to disk and TODO: start background tasks.
|
||||
if let Err(e) = shared_state.sk.persist().await {
|
||||
// Bootstrap failed, cancel timeline and remove timeline directory.
|
||||
self.cancel(shared_state);
|
||||
match || -> Result<()> {
|
||||
shared_state.sk.persist()?;
|
||||
// TODO: add more initialization steps here
|
||||
self.update_status(shared_state);
|
||||
Ok(())
|
||||
}() {
|
||||
Ok(_) => Ok(()),
|
||||
Err(e) => {
|
||||
// Bootstrap failed, cancel timeline and remove timeline directory.
|
||||
self.cancel(shared_state);
|
||||
|
||||
if let Err(fs_err) = fs::remove_dir_all(&self.timeline_dir).await {
|
||||
warn!(
|
||||
"failed to remove timeline {} directory after bootstrap failure: {}",
|
||||
self.ttid, fs_err
|
||||
);
|
||||
if let Err(fs_err) = std::fs::remove_dir_all(&self.timeline_dir) {
|
||||
warn!(
|
||||
"failed to remove timeline {} directory after bootstrap failure: {}",
|
||||
self.ttid, fs_err
|
||||
);
|
||||
}
|
||||
|
||||
Err(e)
|
||||
}
|
||||
|
||||
return Err(e);
|
||||
}
|
||||
|
||||
// TODO: add more initialization steps here
|
||||
self.update_status(shared_state);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Delete timeline from disk completely, by removing timeline directory. Background
|
||||
/// timeline activities will stop eventually.
|
||||
pub async fn delete_from_disk(
|
||||
pub fn delete_from_disk(
|
||||
&self,
|
||||
shared_state: &mut MutexGuard<'_, SharedState>,
|
||||
shared_state: &mut MutexGuard<SharedState>,
|
||||
) -> Result<(bool, bool)> {
|
||||
let was_active = shared_state.active;
|
||||
self.cancel(shared_state);
|
||||
let dir_existed = delete_dir(&self.timeline_dir).await?;
|
||||
let dir_existed = delete_dir(&self.timeline_dir)?;
|
||||
Ok((dir_existed, was_active))
|
||||
}
|
||||
|
||||
/// Cancel timeline to prevent further usage. Background tasks will stop
|
||||
/// eventually after receiving cancellation signal.
|
||||
///
|
||||
/// Note that we can't notify backup launcher here while holding
|
||||
/// shared_state lock, as this is a potential deadlock: caller is
|
||||
/// responsible for that. Generally we should probably make WAL backup tasks
|
||||
/// to shut down on their own, checking once in a while whether it is the
|
||||
/// time.
|
||||
fn cancel(&self, shared_state: &mut MutexGuard<'_, SharedState>) {
|
||||
fn cancel(&self, shared_state: &mut MutexGuard<SharedState>) {
|
||||
info!("timeline {} is cancelled", self.ttid);
|
||||
let _ = self.cancellation_tx.send(true);
|
||||
let res = self.wal_backup_launcher_tx.blocking_send(self.ttid);
|
||||
if let Err(e) = res {
|
||||
error!("Failed to send stop signal to wal_backup_launcher: {}", e);
|
||||
}
|
||||
// Close associated FDs. Nobody will be able to touch timeline data once
|
||||
// it is cancelled, so WAL storage won't be opened again.
|
||||
shared_state.sk.wal_store.close();
|
||||
@@ -433,8 +433,8 @@ impl Timeline {
|
||||
}
|
||||
|
||||
/// Take a writing mutual exclusive lock on timeline shared_state.
|
||||
pub async fn write_shared_state(&self) -> MutexGuard<SharedState> {
|
||||
self.mutex.lock().await
|
||||
pub fn write_shared_state(&self) -> MutexGuard<SharedState> {
|
||||
self.mutex.lock()
|
||||
}
|
||||
|
||||
fn update_status(&self, shared_state: &mut SharedState) -> bool {
|
||||
@@ -450,7 +450,7 @@ impl Timeline {
|
||||
|
||||
let is_wal_backup_action_pending: bool;
|
||||
{
|
||||
let mut shared_state = self.write_shared_state().await;
|
||||
let mut shared_state = self.write_shared_state();
|
||||
shared_state.num_computes += 1;
|
||||
is_wal_backup_action_pending = self.update_status(&mut shared_state);
|
||||
}
|
||||
@@ -464,17 +464,22 @@ impl Timeline {
|
||||
|
||||
/// De-register compute connection, shutting down timeline activity if
|
||||
/// pageserver doesn't need catchup.
|
||||
pub async fn on_compute_disconnect(&self) -> Result<()> {
|
||||
pub fn on_compute_disconnect(&self) -> Result<()> {
|
||||
let is_wal_backup_action_pending: bool;
|
||||
{
|
||||
let mut shared_state = self.write_shared_state().await;
|
||||
let mut shared_state = self.write_shared_state();
|
||||
shared_state.num_computes -= 1;
|
||||
is_wal_backup_action_pending = self.update_status(&mut shared_state);
|
||||
}
|
||||
// Wake up wal backup launcher, if it is time to stop the offloading.
|
||||
if is_wal_backup_action_pending {
|
||||
// Can fail only if channel to a static thread got closed, which is not normal at all.
|
||||
self.wal_backup_launcher_tx.send(self.ttid).await?;
|
||||
//
|
||||
// Note: this is blocking_send because on_compute_disconnect is called in Drop, there is
|
||||
// no async Drop and we use current thread runtimes. With current thread rt spawning
|
||||
// task in drop impl is racy, as thread along with runtime might finish before the task.
|
||||
// This should be switched send.await when/if we go to full async.
|
||||
self.wal_backup_launcher_tx.blocking_send(self.ttid)?;
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
@@ -484,11 +489,11 @@ impl Timeline {
|
||||
/// computes. While there might be nothing to stream already, we learn about
|
||||
/// remote_consistent_lsn update through replication feedback, and we want
|
||||
/// to stop pushing to the broker if pageserver is fully caughtup.
|
||||
pub async fn should_walsender_stop(&self, reported_remote_consistent_lsn: Lsn) -> bool {
|
||||
pub fn should_walsender_stop(&self, reported_remote_consistent_lsn: Lsn) -> bool {
|
||||
if self.is_cancelled() {
|
||||
return true;
|
||||
}
|
||||
let shared_state = self.write_shared_state().await;
|
||||
let shared_state = self.write_shared_state();
|
||||
if shared_state.num_computes == 0 {
|
||||
return shared_state.sk.inmem.commit_lsn == Lsn(0) || // no data at all yet
|
||||
reported_remote_consistent_lsn >= shared_state.sk.inmem.commit_lsn;
|
||||
@@ -498,12 +503,12 @@ impl Timeline {
|
||||
|
||||
/// Returns whether s3 offloading is required and sets current status as
|
||||
/// matching it.
|
||||
pub async fn wal_backup_attend(&self) -> bool {
|
||||
pub fn wal_backup_attend(&self) -> bool {
|
||||
if self.is_cancelled() {
|
||||
return false;
|
||||
}
|
||||
|
||||
self.write_shared_state().await.wal_backup_attend()
|
||||
self.write_shared_state().wal_backup_attend()
|
||||
}
|
||||
|
||||
/// Returns commit_lsn watch channel.
|
||||
@@ -512,7 +517,7 @@ impl Timeline {
|
||||
}
|
||||
|
||||
/// Pass arrived message to the safekeeper.
|
||||
pub async fn process_msg(
|
||||
pub fn process_msg(
|
||||
&self,
|
||||
msg: &ProposerAcceptorMessage,
|
||||
) -> Result<Option<AcceptorProposerMessage>> {
|
||||
@@ -523,8 +528,8 @@ impl Timeline {
|
||||
let mut rmsg: Option<AcceptorProposerMessage>;
|
||||
let commit_lsn: Lsn;
|
||||
{
|
||||
let mut shared_state = self.write_shared_state().await;
|
||||
rmsg = shared_state.sk.process_msg(msg).await?;
|
||||
let mut shared_state = self.write_shared_state();
|
||||
rmsg = shared_state.sk.process_msg(msg)?;
|
||||
|
||||
// if this is AppendResponse, fill in proper pageserver and hot
|
||||
// standby feedback.
|
||||
@@ -541,37 +546,37 @@ impl Timeline {
|
||||
}
|
||||
|
||||
/// Returns wal_seg_size.
|
||||
pub async fn get_wal_seg_size(&self) -> usize {
|
||||
self.write_shared_state().await.get_wal_seg_size()
|
||||
pub fn get_wal_seg_size(&self) -> usize {
|
||||
self.write_shared_state().get_wal_seg_size()
|
||||
}
|
||||
|
||||
/// Returns true only if the timeline is loaded and active.
|
||||
pub async fn is_active(&self) -> bool {
|
||||
pub fn is_active(&self) -> bool {
|
||||
if self.is_cancelled() {
|
||||
return false;
|
||||
}
|
||||
|
||||
self.write_shared_state().await.active
|
||||
self.write_shared_state().active
|
||||
}
|
||||
|
||||
/// Returns state of the timeline.
|
||||
pub async fn get_state(&self) -> (SafekeeperMemState, SafeKeeperState) {
|
||||
let state = self.write_shared_state().await;
|
||||
pub fn get_state(&self) -> (SafekeeperMemState, SafeKeeperState) {
|
||||
let state = self.write_shared_state();
|
||||
(state.sk.inmem.clone(), state.sk.state.clone())
|
||||
}
|
||||
|
||||
/// Returns latest backup_lsn.
|
||||
pub async fn get_wal_backup_lsn(&self) -> Lsn {
|
||||
self.write_shared_state().await.sk.inmem.backup_lsn
|
||||
pub fn get_wal_backup_lsn(&self) -> Lsn {
|
||||
self.write_shared_state().sk.inmem.backup_lsn
|
||||
}
|
||||
|
||||
/// Sets backup_lsn to the given value.
|
||||
pub async fn set_wal_backup_lsn(&self, backup_lsn: Lsn) -> Result<()> {
|
||||
pub fn set_wal_backup_lsn(&self, backup_lsn: Lsn) -> Result<()> {
|
||||
if self.is_cancelled() {
|
||||
bail!(TimelineError::Cancelled(self.ttid));
|
||||
}
|
||||
|
||||
let mut state = self.write_shared_state().await;
|
||||
let mut state = self.write_shared_state();
|
||||
state.sk.inmem.backup_lsn = max(state.sk.inmem.backup_lsn, backup_lsn);
|
||||
// we should check whether to shut down offloader, but this will be done
|
||||
// soon by peer communication anyway.
|
||||
@@ -579,8 +584,8 @@ impl Timeline {
|
||||
}
|
||||
|
||||
/// Get safekeeper info for broadcasting to broker and other peers.
|
||||
pub async fn get_safekeeper_info(&self, conf: &SafeKeeperConf) -> SafekeeperTimelineInfo {
|
||||
let shared_state = self.write_shared_state().await;
|
||||
pub fn get_safekeeper_info(&self, conf: &SafeKeeperConf) -> SafekeeperTimelineInfo {
|
||||
let shared_state = self.write_shared_state();
|
||||
shared_state.get_safekeeper_info(
|
||||
&self.ttid,
|
||||
conf,
|
||||
@@ -599,8 +604,8 @@ impl Timeline {
|
||||
let is_wal_backup_action_pending: bool;
|
||||
let commit_lsn: Lsn;
|
||||
{
|
||||
let mut shared_state = self.write_shared_state().await;
|
||||
shared_state.sk.record_safekeeper_info(&sk_info).await?;
|
||||
let mut shared_state = self.write_shared_state();
|
||||
shared_state.sk.record_safekeeper_info(&sk_info)?;
|
||||
let peer_info = PeerInfo::from_sk_info(&sk_info, Instant::now());
|
||||
shared_state.peers_info.upsert(&peer_info);
|
||||
is_wal_backup_action_pending = self.update_status(&mut shared_state);
|
||||
@@ -617,8 +622,8 @@ impl Timeline {
|
||||
/// Get our latest view of alive peers status on the timeline.
|
||||
/// We pass our own info through the broker as well, so when we don't have connection
|
||||
/// to the broker returned vec is empty.
|
||||
pub async fn get_peers(&self, conf: &SafeKeeperConf) -> Vec<PeerInfo> {
|
||||
let shared_state = self.write_shared_state().await;
|
||||
pub fn get_peers(&self, conf: &SafeKeeperConf) -> Vec<PeerInfo> {
|
||||
let shared_state = self.write_shared_state();
|
||||
let now = Instant::now();
|
||||
shared_state
|
||||
.peers_info
|
||||
@@ -635,34 +640,34 @@ impl Timeline {
|
||||
}
|
||||
|
||||
/// Returns flush_lsn.
|
||||
pub async fn get_flush_lsn(&self) -> Lsn {
|
||||
self.write_shared_state().await.sk.wal_store.flush_lsn()
|
||||
pub fn get_flush_lsn(&self) -> Lsn {
|
||||
self.write_shared_state().sk.wal_store.flush_lsn()
|
||||
}
|
||||
|
||||
/// Delete WAL segments from disk that are no longer needed. This is determined
|
||||
/// based on pageserver's remote_consistent_lsn and local backup_lsn/peer_lsn.
|
||||
pub async fn remove_old_wal(&self, wal_backup_enabled: bool) -> Result<()> {
|
||||
pub fn remove_old_wal(&self, wal_backup_enabled: bool) -> Result<()> {
|
||||
if self.is_cancelled() {
|
||||
bail!(TimelineError::Cancelled(self.ttid));
|
||||
}
|
||||
|
||||
let horizon_segno: XLogSegNo;
|
||||
let remover = {
|
||||
let shared_state = self.write_shared_state().await;
|
||||
let remover: Box<dyn Fn(u64) -> Result<(), anyhow::Error>>;
|
||||
{
|
||||
let shared_state = self.write_shared_state();
|
||||
horizon_segno = shared_state.sk.get_horizon_segno(wal_backup_enabled);
|
||||
remover = shared_state.sk.wal_store.remove_up_to();
|
||||
if horizon_segno <= 1 || horizon_segno <= shared_state.last_removed_segno {
|
||||
return Ok(()); // nothing to do
|
||||
return Ok(());
|
||||
}
|
||||
let remover = shared_state.sk.wal_store.remove_up_to(horizon_segno - 1);
|
||||
// release the lock before removing
|
||||
remover
|
||||
};
|
||||
}
|
||||
|
||||
// delete old WAL files
|
||||
remover.await?;
|
||||
remover(horizon_segno - 1)?;
|
||||
|
||||
// update last_removed_segno
|
||||
let mut shared_state = self.write_shared_state().await;
|
||||
let mut shared_state = self.write_shared_state();
|
||||
shared_state.last_removed_segno = horizon_segno;
|
||||
Ok(())
|
||||
}
|
||||
@@ -671,24 +676,22 @@ impl Timeline {
|
||||
/// passed after the last save. This helps to keep remote_consistent_lsn up
|
||||
/// to date so that storage nodes restart doesn't cause many pageserver ->
|
||||
/// safekeeper reconnections.
|
||||
pub async fn maybe_persist_control_file(&self) -> Result<()> {
|
||||
pub fn maybe_pesist_control_file(&self) -> Result<()> {
|
||||
let remote_consistent_lsn = self.walsenders.get_remote_consistent_lsn();
|
||||
self.write_shared_state()
|
||||
.await
|
||||
.sk
|
||||
.maybe_persist_control_file(remote_consistent_lsn)
|
||||
.await
|
||||
}
|
||||
|
||||
/// Gather timeline data for metrics. If the timeline is not active, returns
|
||||
/// None, we do not collect these.
|
||||
pub async fn info_for_metrics(&self) -> Option<FullTimelineInfo> {
|
||||
/// Returns full timeline info, required for the metrics. If the timeline is
|
||||
/// not active, returns None instead.
|
||||
pub fn info_for_metrics(&self) -> Option<FullTimelineInfo> {
|
||||
if self.is_cancelled() {
|
||||
return None;
|
||||
}
|
||||
|
||||
let ps_feedback = self.walsenders.get_ps_feedback();
|
||||
let state = self.write_shared_state().await;
|
||||
let state = self.write_shared_state();
|
||||
if state.active {
|
||||
Some(FullTimelineInfo {
|
||||
ttid: self.ttid,
|
||||
@@ -710,8 +713,8 @@ impl Timeline {
|
||||
}
|
||||
|
||||
/// Returns in-memory timeline state to build a full debug dump.
|
||||
pub async fn memory_dump(&self) -> debug_dump::Memory {
|
||||
let state = self.write_shared_state().await;
|
||||
pub fn memory_dump(&self) -> debug_dump::Memory {
|
||||
let state = self.write_shared_state();
|
||||
|
||||
let (write_lsn, write_record_lsn, flush_lsn, file_open) =
|
||||
state.sk.wal_store.internal_state();
|
||||
@@ -735,8 +738,8 @@ impl Timeline {
|
||||
}
|
||||
|
||||
/// Deletes directory and it's contents. Returns false if directory does not exist.
|
||||
async fn delete_dir(path: &PathBuf) -> Result<bool> {
|
||||
match fs::remove_dir_all(path).await {
|
||||
fn delete_dir(path: &PathBuf) -> Result<bool> {
|
||||
match std::fs::remove_dir_all(path) {
|
||||
Ok(_) => Ok(true),
|
||||
Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(false),
|
||||
Err(e) => Err(e.into()),
|
||||
|
||||
@@ -113,17 +113,9 @@ impl GlobalTimelines {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Loads all timelines for the given tenant to memory. Returns fs::read_dir
|
||||
/// errors if any.
|
||||
///
|
||||
/// Note: This function (and all reading/loading below) is sync because
|
||||
/// timelines are loaded while holding GlobalTimelinesState lock. Which is
|
||||
/// fine as this is called only from single threaded main runtime on boot,
|
||||
/// but clippy complains anyway, and suppressing that isn't trivial as async
|
||||
/// is the keyword, ha. That only other user is pull_timeline.rs for which
|
||||
/// being blocked is not that bad, and we can do spawn_blocking.
|
||||
/// Loads all timelines for the given tenant to memory. Returns fs::read_dir errors if any.
|
||||
fn load_tenant_timelines(
|
||||
state: &mut MutexGuard<'_, GlobalTimelinesState>,
|
||||
state: &mut MutexGuard<GlobalTimelinesState>,
|
||||
tenant_id: TenantId,
|
||||
) -> Result<()> {
|
||||
let timelines_dir = state.get_conf().tenant_dir(&tenant_id);
|
||||
@@ -228,7 +220,7 @@ impl GlobalTimelines {
|
||||
// Take a lock and finish the initialization holding this mutex. No other threads
|
||||
// can interfere with creation after we will insert timeline into the map.
|
||||
{
|
||||
let mut shared_state = timeline.write_shared_state().await;
|
||||
let mut shared_state = timeline.write_shared_state();
|
||||
|
||||
// We can get a race condition here in case of concurrent create calls, but only
|
||||
// in theory. create() will return valid timeline on the next try.
|
||||
@@ -240,7 +232,7 @@ impl GlobalTimelines {
|
||||
// Write the new timeline to the disk and start background workers.
|
||||
// Bootstrap is transactional, so if it fails, the timeline will be deleted,
|
||||
// and the state on disk should remain unchanged.
|
||||
if let Err(e) = timeline.bootstrap(&mut shared_state).await {
|
||||
if let Err(e) = timeline.bootstrap(&mut shared_state) {
|
||||
// Note: the most likely reason for bootstrap failure is that the timeline
|
||||
// directory already exists on disk. This happens when timeline is corrupted
|
||||
// and wasn't loaded from disk on startup because of that. We want to preserve
|
||||
@@ -302,16 +294,15 @@ impl GlobalTimelines {
|
||||
}
|
||||
|
||||
/// Cancels timeline, then deletes the corresponding data directory.
|
||||
pub async fn delete_force(ttid: &TenantTimelineId) -> Result<TimelineDeleteForceResult> {
|
||||
pub fn delete_force(ttid: &TenantTimelineId) -> Result<TimelineDeleteForceResult> {
|
||||
let tli_res = TIMELINES_STATE.lock().unwrap().get(ttid);
|
||||
match tli_res {
|
||||
Ok(timeline) => {
|
||||
// Take a lock and finish the deletion holding this mutex.
|
||||
let mut shared_state = timeline.write_shared_state().await;
|
||||
let mut shared_state = timeline.write_shared_state();
|
||||
|
||||
info!("deleting timeline {}", ttid);
|
||||
let (dir_existed, was_active) =
|
||||
timeline.delete_from_disk(&mut shared_state).await?;
|
||||
let (dir_existed, was_active) = timeline.delete_from_disk(&mut shared_state)?;
|
||||
|
||||
// Remove timeline from the map.
|
||||
// FIXME: re-enable it once we fix the issue with recreation of deleted timelines
|
||||
@@ -344,7 +335,7 @@ impl GlobalTimelines {
|
||||
/// the tenant had, `true` if a timeline was active. There may be a race if new timelines are
|
||||
/// created simultaneously. In that case the function will return error and the caller should
|
||||
/// retry tenant deletion again later.
|
||||
pub async fn delete_force_all_for_tenant(
|
||||
pub fn delete_force_all_for_tenant(
|
||||
tenant_id: &TenantId,
|
||||
) -> Result<HashMap<TenantTimelineId, TimelineDeleteForceResult>> {
|
||||
info!("deleting all timelines for tenant {}", tenant_id);
|
||||
@@ -354,7 +345,7 @@ impl GlobalTimelines {
|
||||
|
||||
let mut deleted = HashMap::new();
|
||||
for tli in &to_delete {
|
||||
match Self::delete_force(&tli.ttid).await {
|
||||
match Self::delete_force(&tli.ttid) {
|
||||
Ok(result) => {
|
||||
deleted.insert(tli.ttid, result);
|
||||
}
|
||||
|
||||
@@ -17,6 +17,7 @@ use postgres_ffi::XLogFileName;
|
||||
use postgres_ffi::{XLogSegNo, PG_TLI};
|
||||
use remote_storage::{GenericRemoteStorage, RemotePath};
|
||||
use tokio::fs::File;
|
||||
use tokio::runtime::Builder;
|
||||
|
||||
use tokio::select;
|
||||
use tokio::sync::mpsc::{self, Receiver, Sender};
|
||||
@@ -35,16 +36,30 @@ use once_cell::sync::OnceCell;
|
||||
const UPLOAD_FAILURE_RETRY_MIN_MS: u64 = 10;
|
||||
const UPLOAD_FAILURE_RETRY_MAX_MS: u64 = 5000;
|
||||
|
||||
pub fn wal_backup_launcher_thread_main(
|
||||
conf: SafeKeeperConf,
|
||||
wal_backup_launcher_rx: Receiver<TenantTimelineId>,
|
||||
) {
|
||||
let mut builder = Builder::new_multi_thread();
|
||||
if let Some(num_threads) = conf.backup_runtime_threads {
|
||||
builder.worker_threads(num_threads);
|
||||
}
|
||||
let rt = builder
|
||||
.enable_all()
|
||||
.build()
|
||||
.expect("failed to create wal backup runtime");
|
||||
|
||||
rt.block_on(async {
|
||||
wal_backup_launcher_main_loop(conf, wal_backup_launcher_rx).await;
|
||||
});
|
||||
}
|
||||
|
||||
/// Check whether wal backup is required for timeline. If yes, mark that launcher is
|
||||
/// aware of current status and return the timeline.
|
||||
async fn is_wal_backup_required(ttid: TenantTimelineId) -> Option<Arc<Timeline>> {
|
||||
match GlobalTimelines::get(ttid).ok() {
|
||||
Some(tli) => {
|
||||
tli.wal_backup_attend().await;
|
||||
Some(tli)
|
||||
}
|
||||
None => None,
|
||||
}
|
||||
fn is_wal_backup_required(ttid: TenantTimelineId) -> Option<Arc<Timeline>> {
|
||||
GlobalTimelines::get(ttid)
|
||||
.ok()
|
||||
.filter(|tli| tli.wal_backup_attend())
|
||||
}
|
||||
|
||||
struct WalBackupTaskHandle {
|
||||
@@ -128,8 +143,8 @@ async fn update_task(
|
||||
ttid: TenantTimelineId,
|
||||
entry: &mut WalBackupTimelineEntry,
|
||||
) {
|
||||
let alive_peers = entry.timeline.get_peers(conf).await;
|
||||
let wal_backup_lsn = entry.timeline.get_wal_backup_lsn().await;
|
||||
let alive_peers = entry.timeline.get_peers(conf);
|
||||
let wal_backup_lsn = entry.timeline.get_wal_backup_lsn();
|
||||
let (offloader, election_dbg_str) =
|
||||
determine_offloader(&alive_peers, wal_backup_lsn, ttid, conf);
|
||||
let elected_me = Some(conf.my_id) == offloader;
|
||||
@@ -168,10 +183,10 @@ const CHECK_TASKS_INTERVAL_MSEC: u64 = 1000;
|
||||
/// Sits on wal_backup_launcher_rx and starts/stops per timeline wal backup
|
||||
/// tasks. Having this in separate task simplifies locking, allows to reap
|
||||
/// panics and separate elections from offloading itself.
|
||||
pub async fn wal_backup_launcher_task_main(
|
||||
async fn wal_backup_launcher_main_loop(
|
||||
conf: SafeKeeperConf,
|
||||
mut wal_backup_launcher_rx: Receiver<TenantTimelineId>,
|
||||
) -> anyhow::Result<()> {
|
||||
) {
|
||||
info!(
|
||||
"WAL backup launcher started, remote config {:?}",
|
||||
conf.remote_storage
|
||||
@@ -199,7 +214,7 @@ pub async fn wal_backup_launcher_task_main(
|
||||
if conf.remote_storage.is_none() || !conf.wal_backup_enabled {
|
||||
continue; /* just drain the channel and do nothing */
|
||||
}
|
||||
let timeline = is_wal_backup_required(ttid).await;
|
||||
let timeline = is_wal_backup_required(ttid);
|
||||
// do we need to do anything at all?
|
||||
if timeline.is_some() != tasks.contains_key(&ttid) {
|
||||
if let Some(timeline) = timeline {
|
||||
@@ -254,7 +269,7 @@ async fn backup_task_main(
|
||||
let tli = res.unwrap();
|
||||
|
||||
let mut wb = WalBackupTask {
|
||||
wal_seg_size: tli.get_wal_seg_size().await,
|
||||
wal_seg_size: tli.get_wal_seg_size(),
|
||||
commit_lsn_watch_rx: tli.get_commit_lsn_watch_rx(),
|
||||
timeline: tli,
|
||||
timeline_dir,
|
||||
@@ -311,7 +326,7 @@ impl WalBackupTask {
|
||||
continue; /* nothing to do, common case as we wake up on every commit_lsn bump */
|
||||
}
|
||||
// Perhaps peers advanced the position, check shmem value.
|
||||
backup_lsn = self.timeline.get_wal_backup_lsn().await;
|
||||
backup_lsn = self.timeline.get_wal_backup_lsn();
|
||||
if backup_lsn.segment_number(self.wal_seg_size)
|
||||
>= commit_lsn.segment_number(self.wal_seg_size)
|
||||
{
|
||||
@@ -387,7 +402,6 @@ pub async fn backup_lsn_range(
|
||||
let new_backup_lsn = segment.end_lsn;
|
||||
timeline
|
||||
.set_wal_backup_lsn(new_backup_lsn)
|
||||
.await
|
||||
.context("setting wal_backup_lsn")?;
|
||||
*backup_lsn = new_backup_lsn;
|
||||
} else {
|
||||
|
||||
@@ -4,7 +4,7 @@
|
||||
//!
|
||||
use anyhow::{Context, Result};
|
||||
use postgres_backend::QueryError;
|
||||
use std::{future, time::Duration};
|
||||
use std::{future, thread, time::Duration};
|
||||
use tokio::net::TcpStream;
|
||||
use tokio_io_timeout::TimeoutReader;
|
||||
use tracing::*;
|
||||
@@ -16,82 +16,104 @@ use crate::SafeKeeperConf;
|
||||
use postgres_backend::{AuthType, PostgresBackend};
|
||||
|
||||
/// Accept incoming TCP connections and spawn them into a background thread.
|
||||
pub async fn task_main(
|
||||
conf: SafeKeeperConf,
|
||||
pg_listener: std::net::TcpListener,
|
||||
) -> anyhow::Result<()> {
|
||||
// Tokio's from_std won't do this for us, per its comment.
|
||||
pg_listener.set_nonblocking(true)?;
|
||||
pub fn thread_main(conf: SafeKeeperConf, pg_listener: std::net::TcpListener) {
|
||||
let runtime = tokio::runtime::Builder::new_current_thread()
|
||||
.enable_all()
|
||||
.build()
|
||||
.context("create runtime")
|
||||
// todo catch error in main thread
|
||||
.expect("failed to create runtime");
|
||||
|
||||
let listener = tokio::net::TcpListener::from_std(pg_listener)?;
|
||||
let mut connection_count: ConnectionCount = 0;
|
||||
runtime
|
||||
.block_on(async move {
|
||||
// Tokio's from_std won't do this for us, per its comment.
|
||||
pg_listener.set_nonblocking(true)?;
|
||||
let listener = tokio::net::TcpListener::from_std(pg_listener)?;
|
||||
let mut connection_count: ConnectionCount = 0;
|
||||
|
||||
loop {
|
||||
let (socket, peer_addr) = listener.accept().await.context("accept")?;
|
||||
debug!("accepted connection from {}", peer_addr);
|
||||
let conf = conf.clone();
|
||||
let conn_id = issue_connection_id(&mut connection_count);
|
||||
loop {
|
||||
match listener.accept().await {
|
||||
Ok((socket, peer_addr)) => {
|
||||
debug!("accepted connection from {}", peer_addr);
|
||||
let conf = conf.clone();
|
||||
let conn_id = issue_connection_id(&mut connection_count);
|
||||
|
||||
tokio::spawn(async move {
|
||||
if let Err(err) = handle_socket(socket, conf, conn_id)
|
||||
.instrument(info_span!("", cid = %conn_id))
|
||||
.await
|
||||
{
|
||||
error!("connection handler exited: {}", err);
|
||||
let _ = thread::Builder::new()
|
||||
.name("WAL service thread".into())
|
||||
.spawn(move || {
|
||||
if let Err(err) = handle_socket(socket, conf, conn_id) {
|
||||
error!("connection handler exited: {}", err);
|
||||
}
|
||||
})
|
||||
.unwrap();
|
||||
}
|
||||
Err(e) => error!("Failed to accept connection: {}", e),
|
||||
}
|
||||
}
|
||||
});
|
||||
}
|
||||
#[allow(unreachable_code)] // hint compiler the closure return type
|
||||
Ok::<(), anyhow::Error>(())
|
||||
})
|
||||
.expect("listener failed")
|
||||
}
|
||||
|
||||
/// This is run by `task_main` above, inside a background thread.
|
||||
/// This is run by `thread_main` above, inside a background thread.
|
||||
///
|
||||
async fn handle_socket(
|
||||
fn handle_socket(
|
||||
socket: TcpStream,
|
||||
conf: SafeKeeperConf,
|
||||
conn_id: ConnectionId,
|
||||
) -> Result<(), QueryError> {
|
||||
let _enter = info_span!("", cid = %conn_id).entered();
|
||||
|
||||
let runtime = tokio::runtime::Builder::new_current_thread()
|
||||
.enable_all()
|
||||
.build()?;
|
||||
|
||||
socket.set_nodelay(true)?;
|
||||
let peer_addr = socket.peer_addr()?;
|
||||
|
||||
// Set timeout on reading from the socket. It prevents hanged up connection
|
||||
// if client suddenly disappears. Note that TCP_KEEPALIVE is not enabled by
|
||||
// default, and tokio doesn't provide ability to set it out of the box.
|
||||
let mut socket = TimeoutReader::new(socket);
|
||||
let wal_service_timeout = Duration::from_secs(60 * 10);
|
||||
socket.set_timeout(Some(wal_service_timeout));
|
||||
// pin! is here because TimeoutReader (due to storing sleep future inside)
|
||||
// is not Unpin, and all pgbackend/framed/tokio dependencies require stream
|
||||
// to be Unpin. Which is reasonable, as indeed something like TimeoutReader
|
||||
// shouldn't be moved.
|
||||
tokio::pin!(socket);
|
||||
// TimeoutReader wants async runtime during creation.
|
||||
runtime.block_on(async move {
|
||||
// Set timeout on reading from the socket. It prevents hanged up connection
|
||||
// if client suddenly disappears. Note that TCP_KEEPALIVE is not enabled by
|
||||
// default, and tokio doesn't provide ability to set it out of the box.
|
||||
let mut socket = TimeoutReader::new(socket);
|
||||
let wal_service_timeout = Duration::from_secs(60 * 10);
|
||||
socket.set_timeout(Some(wal_service_timeout));
|
||||
// pin! is here because TimeoutReader (due to storing sleep future inside)
|
||||
// is not Unpin, and all pgbackend/framed/tokio dependencies require stream
|
||||
// to be Unpin. Which is reasonable, as indeed something like TimeoutReader
|
||||
// shouldn't be moved.
|
||||
tokio::pin!(socket);
|
||||
|
||||
let traffic_metrics = TrafficMetrics::new();
|
||||
if let Some(current_az) = conf.availability_zone.as_deref() {
|
||||
traffic_metrics.set_sk_az(current_az);
|
||||
}
|
||||
let traffic_metrics = TrafficMetrics::new();
|
||||
if let Some(current_az) = conf.availability_zone.as_deref() {
|
||||
traffic_metrics.set_sk_az(current_az);
|
||||
}
|
||||
|
||||
let socket = MeasuredStream::new(
|
||||
socket,
|
||||
|cnt| {
|
||||
traffic_metrics.observe_read(cnt);
|
||||
},
|
||||
|cnt| {
|
||||
traffic_metrics.observe_write(cnt);
|
||||
},
|
||||
);
|
||||
let socket = MeasuredStream::new(
|
||||
socket,
|
||||
|cnt| {
|
||||
traffic_metrics.observe_read(cnt);
|
||||
},
|
||||
|cnt| {
|
||||
traffic_metrics.observe_write(cnt);
|
||||
},
|
||||
);
|
||||
|
||||
let auth_type = match conf.auth {
|
||||
None => AuthType::Trust,
|
||||
Some(_) => AuthType::NeonJWT,
|
||||
};
|
||||
let mut conn_handler =
|
||||
SafekeeperPostgresHandler::new(conf, conn_id, Some(traffic_metrics.clone()));
|
||||
let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
|
||||
// libpq protocol between safekeeper and walproposer / pageserver
|
||||
// We don't use shutdown.
|
||||
pgbackend
|
||||
.run(&mut conn_handler, future::pending::<()>)
|
||||
.await
|
||||
let auth_type = match conf.auth {
|
||||
None => AuthType::Trust,
|
||||
Some(_) => AuthType::NeonJWT,
|
||||
};
|
||||
let mut conn_handler =
|
||||
SafekeeperPostgresHandler::new(conf, conn_id, Some(traffic_metrics.clone()));
|
||||
let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
|
||||
// libpq protocol between safekeeper and walproposer / pageserver
|
||||
// We don't use shutdown.
|
||||
pgbackend
|
||||
.run(&mut conn_handler, future::pending::<()>)
|
||||
.await
|
||||
})
|
||||
}
|
||||
|
||||
/// Unique WAL service connection ids are logged in spans for observability.
|
||||
|
||||
@@ -8,47 +8,54 @@
|
||||
//! Note that last file has `.partial` suffix, that's different from postgres.
|
||||
|
||||
use anyhow::{bail, Context, Result};
|
||||
use bytes::Bytes;
|
||||
use futures::future::BoxFuture;
|
||||
use remote_storage::RemotePath;
|
||||
|
||||
use std::io::{self, Seek, SeekFrom};
|
||||
use std::pin::Pin;
|
||||
use tokio::io::AsyncRead;
|
||||
|
||||
use postgres_ffi::v14::xlog_utils::{IsPartialXLogFileName, IsXLogFileName, XLogFromFileName};
|
||||
use postgres_ffi::{XLogSegNo, PG_TLI};
|
||||
use remote_storage::RemotePath;
|
||||
use std::cmp::{max, min};
|
||||
use std::io::{self, SeekFrom};
|
||||
|
||||
use bytes::Bytes;
|
||||
use std::fs::{self, remove_file, File, OpenOptions};
|
||||
use std::io::Write;
|
||||
use std::path::{Path, PathBuf};
|
||||
use std::pin::Pin;
|
||||
use tokio::fs::{self, remove_file, File, OpenOptions};
|
||||
use tokio::io::{AsyncRead, AsyncWriteExt};
|
||||
use tokio::io::{AsyncReadExt, AsyncSeekExt};
|
||||
|
||||
use tracing::*;
|
||||
|
||||
use utils::{id::TenantTimelineId, lsn::Lsn};
|
||||
|
||||
use crate::metrics::{time_io_closure, WalStorageMetrics, REMOVED_WAL_SEGMENTS};
|
||||
use crate::safekeeper::SafeKeeperState;
|
||||
|
||||
use crate::wal_backup::read_object;
|
||||
use crate::SafeKeeperConf;
|
||||
use postgres_ffi::waldecoder::WalStreamDecoder;
|
||||
use postgres_ffi::XLogFileName;
|
||||
use postgres_ffi::XLOG_BLCKSZ;
|
||||
use pq_proto::SystemId;
|
||||
use utils::{id::TenantTimelineId, lsn::Lsn};
|
||||
|
||||
#[async_trait::async_trait]
|
||||
use postgres_ffi::waldecoder::WalStreamDecoder;
|
||||
|
||||
use pq_proto::SystemId;
|
||||
use tokio::io::{AsyncReadExt, AsyncSeekExt};
|
||||
|
||||
pub trait Storage {
|
||||
/// LSN of last durably stored WAL record.
|
||||
fn flush_lsn(&self) -> Lsn;
|
||||
|
||||
/// Write piece of WAL from buf to disk, but not necessarily sync it.
|
||||
async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()>;
|
||||
fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()>;
|
||||
|
||||
/// Truncate WAL at specified LSN, which must be the end of WAL record.
|
||||
async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()>;
|
||||
fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()>;
|
||||
|
||||
/// Durably store WAL on disk, up to the last written WAL record.
|
||||
async fn flush_wal(&mut self) -> Result<()>;
|
||||
fn flush_wal(&mut self) -> Result<()>;
|
||||
|
||||
/// Remove all segments <= given segno. Returns function doing that as we
|
||||
/// want to perform it without timeline lock.
|
||||
fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>>;
|
||||
/// Remove all segments <= given segno. Returns closure as we want to do
|
||||
/// that without timeline lock.
|
||||
fn remove_up_to(&self) -> Box<dyn Fn(XLogSegNo) -> Result<()>>;
|
||||
|
||||
/// Release resources associated with the storage -- technically, close FDs.
|
||||
/// Currently we don't remove timelines until restart (#3146), so need to
|
||||
@@ -98,22 +105,6 @@ pub struct PhysicalStorage {
|
||||
/// - points to write_lsn, so no seek is needed for writing
|
||||
/// - doesn't point to the end of the segment
|
||||
file: Option<File>,
|
||||
|
||||
/// When false, we have just initialized storage using the LSN from find_end_of_wal().
|
||||
/// In this case, [`write_lsn`] can be less than actually written WAL on disk. In particular,
|
||||
/// there can be a case with unexpected .partial file.
|
||||
///
|
||||
/// Imagine the following:
|
||||
/// - 000000010000000000000001
|
||||
/// - it was fully written, but the last record is split between 2 segments
|
||||
/// - after restart, find_end_of_wal() returned 0/1FFFFF0, which is in the end of this segment
|
||||
/// - write_lsn, write_record_lsn and flush_record_lsn were initialized to 0/1FFFFF0
|
||||
/// - 000000010000000000000002.partial
|
||||
/// - it has only 1 byte written, which is not enough to make a full WAL record
|
||||
///
|
||||
/// Partial segment 002 has no WAL records, and it will be removed by the next truncate_wal().
|
||||
/// This flag will be set to true after the first truncate_wal() call.
|
||||
is_truncated_after_restart: bool,
|
||||
}
|
||||
|
||||
impl PhysicalStorage {
|
||||
@@ -173,7 +164,6 @@ impl PhysicalStorage {
|
||||
flush_record_lsn: flush_lsn,
|
||||
decoder: WalStreamDecoder::new(write_lsn, state.server.pg_version / 10000),
|
||||
file: None,
|
||||
is_truncated_after_restart: false,
|
||||
})
|
||||
}
|
||||
|
||||
@@ -188,37 +178,33 @@ impl PhysicalStorage {
|
||||
}
|
||||
|
||||
/// Call fdatasync if config requires so.
|
||||
async fn fdatasync_file(&mut self, file: &mut File) -> Result<()> {
|
||||
fn fdatasync_file(&mut self, file: &mut File) -> Result<()> {
|
||||
if !self.conf.no_sync {
|
||||
self.metrics
|
||||
.observe_flush_seconds(time_io_closure(file.sync_data()).await?);
|
||||
.observe_flush_seconds(time_io_closure(|| Ok(file.sync_data()?))?);
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Call fsync if config requires so.
|
||||
async fn fsync_file(&mut self, file: &mut File) -> Result<()> {
|
||||
fn fsync_file(&mut self, file: &mut File) -> Result<()> {
|
||||
if !self.conf.no_sync {
|
||||
self.metrics
|
||||
.observe_flush_seconds(time_io_closure(file.sync_all()).await?);
|
||||
.observe_flush_seconds(time_io_closure(|| Ok(file.sync_all()?))?);
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Open or create WAL segment file. Caller must call seek to the wanted position.
|
||||
/// Returns `file` and `is_partial`.
|
||||
async fn open_or_create(&mut self, segno: XLogSegNo) -> Result<(File, bool)> {
|
||||
fn open_or_create(&mut self, segno: XLogSegNo) -> Result<(File, bool)> {
|
||||
let (wal_file_path, wal_file_partial_path) =
|
||||
wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
|
||||
|
||||
// Try to open already completed segment
|
||||
if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_path).await {
|
||||
if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_path) {
|
||||
Ok((file, false))
|
||||
} else if let Ok(file) = OpenOptions::new()
|
||||
.write(true)
|
||||
.open(&wal_file_partial_path)
|
||||
.await
|
||||
{
|
||||
} else if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_partial_path) {
|
||||
// Try to open existing partial file
|
||||
Ok((file, true))
|
||||
} else {
|
||||
@@ -227,36 +213,35 @@ impl PhysicalStorage {
|
||||
.create(true)
|
||||
.write(true)
|
||||
.open(&wal_file_partial_path)
|
||||
.await
|
||||
.with_context(|| format!("Failed to open log file {:?}", &wal_file_path))?;
|
||||
|
||||
write_zeroes(&mut file, self.wal_seg_size).await?;
|
||||
self.fsync_file(&mut file).await?;
|
||||
write_zeroes(&mut file, self.wal_seg_size)?;
|
||||
self.fsync_file(&mut file)?;
|
||||
Ok((file, true))
|
||||
}
|
||||
}
|
||||
|
||||
/// Write WAL bytes, which are known to be located in a single WAL segment.
|
||||
async fn write_in_segment(&mut self, segno: u64, xlogoff: usize, buf: &[u8]) -> Result<()> {
|
||||
fn write_in_segment(&mut self, segno: u64, xlogoff: usize, buf: &[u8]) -> Result<()> {
|
||||
let mut file = if let Some(file) = self.file.take() {
|
||||
file
|
||||
} else {
|
||||
let (mut file, is_partial) = self.open_or_create(segno).await?;
|
||||
let (mut file, is_partial) = self.open_or_create(segno)?;
|
||||
assert!(is_partial, "unexpected write into non-partial segment file");
|
||||
file.seek(SeekFrom::Start(xlogoff as u64)).await?;
|
||||
file.seek(SeekFrom::Start(xlogoff as u64))?;
|
||||
file
|
||||
};
|
||||
|
||||
file.write_all(buf).await?;
|
||||
file.write_all(buf)?;
|
||||
|
||||
if xlogoff + buf.len() == self.wal_seg_size {
|
||||
// If we reached the end of a WAL segment, flush and close it.
|
||||
self.fdatasync_file(&mut file).await?;
|
||||
self.fdatasync_file(&mut file)?;
|
||||
|
||||
// Rename partial file to completed file
|
||||
let (wal_file_path, wal_file_partial_path) =
|
||||
wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
|
||||
fs::rename(wal_file_partial_path, wal_file_path).await?;
|
||||
fs::rename(wal_file_partial_path, wal_file_path)?;
|
||||
} else {
|
||||
// otherwise, file can be reused later
|
||||
self.file = Some(file);
|
||||
@@ -270,11 +255,11 @@ impl PhysicalStorage {
|
||||
/// be flushed separately later.
|
||||
///
|
||||
/// Updates `write_lsn`.
|
||||
async fn write_exact(&mut self, pos: Lsn, mut buf: &[u8]) -> Result<()> {
|
||||
fn write_exact(&mut self, pos: Lsn, mut buf: &[u8]) -> Result<()> {
|
||||
if self.write_lsn != pos {
|
||||
// need to flush the file before discarding it
|
||||
if let Some(mut file) = self.file.take() {
|
||||
self.fdatasync_file(&mut file).await?;
|
||||
self.fdatasync_file(&mut file)?;
|
||||
}
|
||||
|
||||
self.write_lsn = pos;
|
||||
@@ -292,8 +277,7 @@ impl PhysicalStorage {
|
||||
buf.len()
|
||||
};
|
||||
|
||||
self.write_in_segment(segno, xlogoff, &buf[..bytes_write])
|
||||
.await?;
|
||||
self.write_in_segment(segno, xlogoff, &buf[..bytes_write])?;
|
||||
self.write_lsn += bytes_write as u64;
|
||||
buf = &buf[bytes_write..];
|
||||
}
|
||||
@@ -302,7 +286,6 @@ impl PhysicalStorage {
|
||||
}
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl Storage for PhysicalStorage {
|
||||
/// flush_lsn returns LSN of last durably stored WAL record.
|
||||
fn flush_lsn(&self) -> Lsn {
|
||||
@@ -310,7 +293,7 @@ impl Storage for PhysicalStorage {
|
||||
}
|
||||
|
||||
/// Write WAL to disk.
|
||||
async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
|
||||
fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
|
||||
// Disallow any non-sequential writes, which can result in gaps or overwrites.
|
||||
// If we need to move the pointer, use truncate_wal() instead.
|
||||
if self.write_lsn > startpos {
|
||||
@@ -328,7 +311,7 @@ impl Storage for PhysicalStorage {
|
||||
);
|
||||
}
|
||||
|
||||
let write_seconds = time_io_closure(self.write_exact(startpos, buf)).await?;
|
||||
let write_seconds = time_io_closure(|| self.write_exact(startpos, buf))?;
|
||||
// WAL is written, updating write metrics
|
||||
self.metrics.observe_write_seconds(write_seconds);
|
||||
self.metrics.observe_write_bytes(buf.len());
|
||||
@@ -357,14 +340,14 @@ impl Storage for PhysicalStorage {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn flush_wal(&mut self) -> Result<()> {
|
||||
fn flush_wal(&mut self) -> Result<()> {
|
||||
if self.flush_record_lsn == self.write_record_lsn {
|
||||
// no need to do extra flush
|
||||
return Ok(());
|
||||
}
|
||||
|
||||
if let Some(mut unflushed_file) = self.file.take() {
|
||||
self.fdatasync_file(&mut unflushed_file).await?;
|
||||
self.fdatasync_file(&mut unflushed_file)?;
|
||||
self.file = Some(unflushed_file);
|
||||
} else {
|
||||
// We have unflushed data (write_lsn != flush_lsn), but no file.
|
||||
@@ -386,7 +369,7 @@ impl Storage for PhysicalStorage {
|
||||
|
||||
/// Truncate written WAL by removing all WAL segments after the given LSN.
|
||||
/// end_pos must point to the end of the WAL record.
|
||||
async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
|
||||
fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
|
||||
// Streaming must not create a hole, so truncate cannot be called on non-written lsn
|
||||
if self.write_lsn != Lsn(0) && end_pos > self.write_lsn {
|
||||
bail!(
|
||||
@@ -398,51 +381,47 @@ impl Storage for PhysicalStorage {
|
||||
|
||||
// Quick exit if nothing to do to avoid writing up to 16 MiB of zeros on
|
||||
// disk (this happens on each connect).
|
||||
if self.is_truncated_after_restart
|
||||
&& end_pos == self.write_lsn
|
||||
&& end_pos == self.flush_record_lsn
|
||||
{
|
||||
if end_pos == self.write_lsn {
|
||||
return Ok(());
|
||||
}
|
||||
|
||||
// Close previously opened file, if any
|
||||
if let Some(mut unflushed_file) = self.file.take() {
|
||||
self.fdatasync_file(&mut unflushed_file).await?;
|
||||
self.fdatasync_file(&mut unflushed_file)?;
|
||||
}
|
||||
|
||||
let xlogoff = end_pos.segment_offset(self.wal_seg_size);
|
||||
let segno = end_pos.segment_number(self.wal_seg_size);
|
||||
|
||||
// Remove all segments after the given LSN.
|
||||
remove_segments_from_disk(&self.timeline_dir, self.wal_seg_size, |x| x > segno).await?;
|
||||
remove_segments_from_disk(&self.timeline_dir, self.wal_seg_size, |x| x > segno)?;
|
||||
|
||||
let (mut file, is_partial) = self.open_or_create(segno).await?;
|
||||
let (mut file, is_partial) = self.open_or_create(segno)?;
|
||||
|
||||
// Fill end with zeroes
|
||||
file.seek(SeekFrom::Start(xlogoff as u64)).await?;
|
||||
write_zeroes(&mut file, self.wal_seg_size - xlogoff).await?;
|
||||
self.fdatasync_file(&mut file).await?;
|
||||
file.seek(SeekFrom::Start(xlogoff as u64))?;
|
||||
write_zeroes(&mut file, self.wal_seg_size - xlogoff)?;
|
||||
self.fdatasync_file(&mut file)?;
|
||||
|
||||
if !is_partial {
|
||||
// Make segment partial once again
|
||||
let (wal_file_path, wal_file_partial_path) =
|
||||
wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
|
||||
fs::rename(wal_file_path, wal_file_partial_path).await?;
|
||||
fs::rename(wal_file_path, wal_file_partial_path)?;
|
||||
}
|
||||
|
||||
// Update LSNs
|
||||
self.write_lsn = end_pos;
|
||||
self.write_record_lsn = end_pos;
|
||||
self.flush_record_lsn = end_pos;
|
||||
self.is_truncated_after_restart = true;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>> {
|
||||
fn remove_up_to(&self) -> Box<dyn Fn(XLogSegNo) -> Result<()>> {
|
||||
let timeline_dir = self.timeline_dir.clone();
|
||||
let wal_seg_size = self.wal_seg_size;
|
||||
Box::pin(async move {
|
||||
remove_segments_from_disk(&timeline_dir, wal_seg_size, |x| x <= segno_up_to).await
|
||||
Box::new(move |segno_up_to: XLogSegNo| {
|
||||
remove_segments_from_disk(&timeline_dir, wal_seg_size, |x| x <= segno_up_to)
|
||||
})
|
||||
}
|
||||
|
||||
@@ -457,7 +436,7 @@ impl Storage for PhysicalStorage {
|
||||
}
|
||||
|
||||
/// Remove all WAL segments in timeline_dir that match the given predicate.
|
||||
async fn remove_segments_from_disk(
|
||||
fn remove_segments_from_disk(
|
||||
timeline_dir: &Path,
|
||||
wal_seg_size: usize,
|
||||
remove_predicate: impl Fn(XLogSegNo) -> bool,
|
||||
@@ -466,8 +445,8 @@ async fn remove_segments_from_disk(
|
||||
let mut min_removed = u64::MAX;
|
||||
let mut max_removed = u64::MIN;
|
||||
|
||||
let mut entries = fs::read_dir(timeline_dir).await?;
|
||||
while let Some(entry) = entries.next_entry().await? {
|
||||
for entry in fs::read_dir(timeline_dir)? {
|
||||
let entry = entry?;
|
||||
let entry_path = entry.path();
|
||||
let fname = entry_path.file_name().unwrap();
|
||||
|
||||
@@ -478,7 +457,7 @@ async fn remove_segments_from_disk(
|
||||
}
|
||||
let (segno, _) = XLogFromFileName(fname_str, wal_seg_size);
|
||||
if remove_predicate(segno) {
|
||||
remove_file(entry_path).await?;
|
||||
remove_file(entry_path)?;
|
||||
n_removed += 1;
|
||||
min_removed = min(min_removed, segno);
|
||||
max_removed = max(max_removed, segno);
|
||||
@@ -710,12 +689,12 @@ impl WalReader {
|
||||
const ZERO_BLOCK: &[u8] = &[0u8; XLOG_BLCKSZ];
|
||||
|
||||
/// Helper for filling file with zeroes.
|
||||
async fn write_zeroes(file: &mut File, mut count: usize) -> Result<()> {
|
||||
fn write_zeroes(file: &mut File, mut count: usize) -> Result<()> {
|
||||
while count >= XLOG_BLCKSZ {
|
||||
file.write_all(ZERO_BLOCK).await?;
|
||||
file.write_all(ZERO_BLOCK)?;
|
||||
count -= XLOG_BLCKSZ;
|
||||
}
|
||||
file.write_all(&ZERO_BLOCK[0..count]).await?;
|
||||
file.write_all(&ZERO_BLOCK[0..count])?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
|
||||
@@ -32,7 +32,6 @@ pub const DEFAULT_LISTEN_ADDR: &str = "127.0.0.1:50051";
|
||||
pub const DEFAULT_ENDPOINT: &str = const_format::formatcp!("http://{DEFAULT_LISTEN_ADDR}");
|
||||
|
||||
pub const DEFAULT_KEEPALIVE_INTERVAL: &str = "5000 ms";
|
||||
pub const DEFAULT_CONNECT_TIMEOUT: Duration = Duration::from_millis(5000);
|
||||
|
||||
// BrokerServiceClient charged with tonic provided Channel transport; helps to
|
||||
// avoid depending on tonic directly in user crates.
|
||||
@@ -59,8 +58,7 @@ where
|
||||
}
|
||||
tonic_endpoint = tonic_endpoint
|
||||
.http2_keep_alive_interval(keepalive_interval)
|
||||
.keep_alive_while_idle(true)
|
||||
.connect_timeout(DEFAULT_CONNECT_TIMEOUT);
|
||||
.keep_alive_while_idle(true);
|
||||
// keep_alive_timeout is 20s by default on both client and server side
|
||||
let channel = tonic_endpoint.connect_lazy();
|
||||
Ok(BrokerClientChannel::new(channel))
|
||||
|
||||
@@ -663,8 +663,6 @@ class NeonEnvBuilder:
|
||||
else:
|
||||
raise RuntimeError(f"Unknown storage type: {remote_storage_kind}")
|
||||
|
||||
self.remote_storage_kind = remote_storage_kind
|
||||
|
||||
def enable_local_fs_remote_storage(self, force_enable: bool = True):
|
||||
"""
|
||||
Sets up the pageserver to use the local fs at the `test_dir/local_fs_remote_storage` path.
|
||||
|
||||
@@ -2,7 +2,7 @@ import time
|
||||
from typing import Any, Dict, Optional
|
||||
|
||||
from fixtures.log_helper import log
|
||||
from fixtures.pageserver.http import PageserverApiException, PageserverHttpClient
|
||||
from fixtures.pageserver.http import PageserverHttpClient
|
||||
from fixtures.types import Lsn, TenantId, TimelineId
|
||||
|
||||
|
||||
@@ -92,41 +92,6 @@ def wait_until_tenant_state(
|
||||
)
|
||||
|
||||
|
||||
def wait_until_timeline_state(
|
||||
pageserver_http: PageserverHttpClient,
|
||||
tenant_id: TenantId,
|
||||
timeline_id: TimelineId,
|
||||
expected_state: str,
|
||||
iterations: int,
|
||||
period: float = 1.0,
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Does not use `wait_until` for debugging purposes
|
||||
"""
|
||||
for i in range(iterations):
|
||||
try:
|
||||
timeline = pageserver_http.timeline_detail(tenant_id=tenant_id, timeline_id=timeline_id)
|
||||
log.debug(f"Timeline {tenant_id}/{timeline_id} data: {timeline}")
|
||||
if isinstance(timeline["state"], str):
|
||||
if timeline["state"] == expected_state:
|
||||
return timeline
|
||||
elif isinstance(timeline, Dict):
|
||||
if timeline["state"].get(expected_state):
|
||||
return timeline
|
||||
|
||||
except Exception as e:
|
||||
log.debug(f"Timeline {tenant_id}/{timeline_id} state retrieval failure: {e}")
|
||||
|
||||
if i == iterations - 1:
|
||||
# do not sleep last time, we already know that we failed
|
||||
break
|
||||
time.sleep(period)
|
||||
|
||||
raise Exception(
|
||||
f"Timeline {tenant_id}/{timeline_id} did not become {expected_state} within {iterations * period} seconds"
|
||||
)
|
||||
|
||||
|
||||
def wait_until_tenant_active(
|
||||
pageserver_http: PageserverHttpClient,
|
||||
tenant_id: TenantId,
|
||||
@@ -191,21 +156,3 @@ def wait_for_upload_queue_empty(
|
||||
if all(m.value == 0 for m in tl):
|
||||
return
|
||||
time.sleep(0.2)
|
||||
|
||||
|
||||
def assert_timeline_detail_404(
|
||||
pageserver_http: PageserverHttpClient,
|
||||
tenant_id: TenantId,
|
||||
timeline_id: TimelineId,
|
||||
):
|
||||
"""Asserts that timeline_detail returns 404, or dumps the detail."""
|
||||
try:
|
||||
data = pageserver_http.timeline_detail(tenant_id, timeline_id)
|
||||
log.error(f"detail {data}")
|
||||
except PageserverApiException as e:
|
||||
log.error(e)
|
||||
if e.status_code == 404:
|
||||
return
|
||||
else:
|
||||
raise
|
||||
raise Exception("detail succeeded (it should return 404)")
|
||||
|
||||
@@ -1,104 +1,8 @@
|
||||
from contextlib import closing
|
||||
|
||||
import pytest
|
||||
import requests
|
||||
from fixtures.benchmark_fixture import MetricReport, NeonBenchmarker
|
||||
from fixtures.benchmark_fixture import NeonBenchmarker
|
||||
from fixtures.neon_fixtures import NeonEnvBuilder
|
||||
from fixtures.utils import get_dir_size
|
||||
|
||||
|
||||
# @pytest.mark.xfail # We currently pass a 16MB pg_wal dir instead of creating it client-side
|
||||
def test_basebackup_size(neon_env_builder: NeonEnvBuilder, zenbenchmark: NeonBenchmarker):
|
||||
neon_env_builder.num_safekeepers = 3
|
||||
env = neon_env_builder.init_start()
|
||||
|
||||
# Start
|
||||
env.neon_cli.create_branch("test_startup")
|
||||
endpoint = env.endpoints.create_start("test_startup")
|
||||
|
||||
# Get metrics
|
||||
metrics = requests.get(f"http://localhost:{endpoint.http_port}/metrics.json").json()
|
||||
basebackup_bytes = metrics["basebackup_bytes"]
|
||||
zenbenchmark.record(
|
||||
"basebackup_size", basebackup_bytes / 1024, "KB", report=MetricReport.LOWER_IS_BETTER
|
||||
)
|
||||
|
||||
# Stop so we force flush of any files and we can measure datadir sizes
|
||||
# NOTE the order of this line is important in relation to get_dir_size
|
||||
datadir = endpoint.pgdata_dir
|
||||
assert datadir is not None # for mypy
|
||||
endpoint.stop()
|
||||
|
||||
# Even though we don't insert any data, this nuber could be larger than basebackup
|
||||
# size because there could theoretically be compression, or postgres could create
|
||||
# or download data during startup. Currently if we don't send any pg_wal in the
|
||||
# basebackup, postgres will start up just fine, but during sync-safekeepers,
|
||||
# walproposer will try to recover the missing wal from safekeepers and cause the
|
||||
# same amount of network IO. We want to notice that if it happens.
|
||||
datadir_bytes = get_dir_size(datadir)
|
||||
zenbenchmark.record(
|
||||
"datadir_size", datadir_bytes / 1024, "KB", report=MetricReport.LOWER_IS_BETTER
|
||||
)
|
||||
|
||||
wal_bytes = get_dir_size(datadir + "/pg_wal")
|
||||
zenbenchmark.record("wal_size", wal_bytes / 1024, "KB", report=MetricReport.LOWER_IS_BETTER)
|
||||
|
||||
# Seems like a reasonable limit, but increase it if it becomes impossible to meet
|
||||
# assert basebackup_bytes < 70 * 1024
|
||||
# assert datadir_bytes < 70 * 1024
|
||||
# assert wal_bytes < 1 * 1024
|
||||
|
||||
|
||||
# Just start and measure duration.
|
||||
#
|
||||
# This test runs pretty quickly and can be informative when used in combination
|
||||
# with emulated network delay. Some useful delay commands:
|
||||
#
|
||||
# 1. Add 2msec delay to all localhost traffic
|
||||
# `sudo tc qdisc add dev lo root handle 1:0 netem delay 2msec`
|
||||
#
|
||||
# 2. Test that it works (you should see 4ms ping)
|
||||
# `ping localhost`
|
||||
#
|
||||
# 3. Revert back to normal
|
||||
# `sudo tc qdisc del dev lo root netem`
|
||||
#
|
||||
# NOTE this test might not represent the real startup time because the basebackup
|
||||
# for a large database might be larger if there's a lof of transaction metadata,
|
||||
# or safekeepers might need more syncing, or there might be more operations to
|
||||
# apply during config step, like more users, databases, or extensions. By default
|
||||
# we load extensions 'neon,pg_stat_statements,timescaledb,pg_cron', but in this
|
||||
# test we only load neon.
|
||||
def test_startup_simple(neon_env_builder: NeonEnvBuilder, zenbenchmark: NeonBenchmarker):
|
||||
neon_env_builder.num_safekeepers = 3
|
||||
env = neon_env_builder.init_start()
|
||||
|
||||
env.neon_cli.create_branch("test_startup")
|
||||
|
||||
# We do two iterations so we can see if the second startup is faster. It should
|
||||
# be because the compute node should already be configured with roles, databases,
|
||||
# extensions, etc from the first run.
|
||||
for i in range(2):
|
||||
# Start
|
||||
with zenbenchmark.record_duration(f"{i}_start_and_select"):
|
||||
endpoint = env.endpoints.create_start("test_startup")
|
||||
endpoint.safe_psql("select 1;")
|
||||
|
||||
# Get metrics
|
||||
metrics = requests.get(f"http://localhost:{endpoint.http_port}/metrics.json").json()
|
||||
durations = {
|
||||
"wait_for_spec_ms": f"{i}_wait_for_spec",
|
||||
"sync_safekeepers_ms": f"{i}_sync_safekeepers",
|
||||
"basebackup_ms": f"{i}_basebackup",
|
||||
"config_ms": f"{i}_config",
|
||||
"total_startup_ms": f"{i}_total_startup",
|
||||
}
|
||||
for key, name in durations.items():
|
||||
value = metrics[key]
|
||||
zenbenchmark.record(name, value, "ms", report=MetricReport.LOWER_IS_BETTER)
|
||||
|
||||
# Stop so we can restart
|
||||
endpoint.stop()
|
||||
|
||||
|
||||
# This test sometimes runs for longer than the global 5 minute timeout.
|
||||
|
||||
@@ -79,7 +79,6 @@ def test_import_from_vanilla(test_output_dir, pg_bin, vanilla_pg, neon_env_build
|
||||
# Set up pageserver for import
|
||||
neon_env_builder.enable_local_fs_remote_storage()
|
||||
env = neon_env_builder.init_start()
|
||||
|
||||
client = env.pageserver.http_client()
|
||||
client.tenant_create(tenant)
|
||||
|
||||
@@ -146,11 +145,6 @@ def test_import_from_vanilla(test_output_dir, pg_bin, vanilla_pg, neon_env_build
|
||||
)
|
||||
|
||||
# NOTE: delete can easily come before upload operations are completed
|
||||
# https://github.com/neondatabase/neon/issues/4326
|
||||
env.pageserver.allowed_errors.append(
|
||||
".*files not bound to index_file.json, proceeding with their deletion.*"
|
||||
)
|
||||
|
||||
client.timeline_delete(tenant, timeline)
|
||||
|
||||
# Importing correct backup works
|
||||
|
||||
@@ -225,37 +225,3 @@ def test_sql_over_http(static_proxy: NeonProxy):
|
||||
res = q("drop table t")
|
||||
assert res["command"] == "DROP"
|
||||
assert res["rowCount"] is None
|
||||
|
||||
|
||||
def test_sql_over_http_output_options(static_proxy: NeonProxy):
|
||||
static_proxy.safe_psql("create role http2 with login password 'http2' superuser")
|
||||
|
||||
def q(sql: str, raw_text: bool, array_mode: bool, params: List[Any] = []) -> Any:
|
||||
connstr = (
|
||||
f"postgresql://http2:http2@{static_proxy.domain}:{static_proxy.proxy_port}/postgres"
|
||||
)
|
||||
response = requests.post(
|
||||
f"https://{static_proxy.domain}:{static_proxy.external_http_port}/sql",
|
||||
data=json.dumps({"query": sql, "params": params}),
|
||||
headers={
|
||||
"Content-Type": "application/sql",
|
||||
"Neon-Connection-String": connstr,
|
||||
"Neon-Raw-Text-Output": "true" if raw_text else "false",
|
||||
"Neon-Array-Mode": "true" if array_mode else "false",
|
||||
},
|
||||
verify=str(static_proxy.test_output_dir / "proxy.crt"),
|
||||
)
|
||||
assert response.status_code == 200
|
||||
return response.json()
|
||||
|
||||
rows = q("select 1 as n, 'a' as s, '{1,2,3}'::int4[] as arr", False, False)["rows"]
|
||||
assert rows == [{"arr": [1, 2, 3], "n": 1, "s": "a"}]
|
||||
|
||||
rows = q("select 1 as n, 'a' as s, '{1,2,3}'::int4[] as arr", False, True)["rows"]
|
||||
assert rows == [[1, "a", [1, 2, 3]]]
|
||||
|
||||
rows = q("select 1 as n, 'a' as s, '{1,2,3}'::int4[] as arr", True, False)["rows"]
|
||||
assert rows == [{"arr": "{1,2,3}", "n": "1", "s": "a"}]
|
||||
|
||||
rows = q("select 1 as n, 'a' as s, '{1,2,3}'::int4[] as arr", True, True)["rows"]
|
||||
assert rows == [["1", "a", "{1,2,3}"]]
|
||||
|
||||
@@ -20,7 +20,6 @@ from fixtures.neon_fixtures import (
|
||||
)
|
||||
from fixtures.pageserver.http import PageserverApiException, PageserverHttpClient
|
||||
from fixtures.pageserver.utils import (
|
||||
assert_timeline_detail_404,
|
||||
wait_for_last_record_lsn,
|
||||
wait_for_upload,
|
||||
wait_until_tenant_active,
|
||||
@@ -183,7 +182,7 @@ def test_remote_storage_backup_and_restore(
|
||||
wait_until_tenant_active(
|
||||
pageserver_http=client,
|
||||
tenant_id=tenant_id,
|
||||
iterations=10, # make it longer for real_s3 tests when unreliable wrapper is involved
|
||||
iterations=5,
|
||||
)
|
||||
|
||||
detail = client.timeline_detail(tenant_id, timeline_id)
|
||||
@@ -599,23 +598,8 @@ def test_timeline_deletion_with_files_stuck_in_upload_queue(
|
||||
)
|
||||
client.timeline_delete(tenant_id, timeline_id)
|
||||
|
||||
env.pageserver.allowed_errors.append(f".*Timeline {tenant_id}/{timeline_id} was not found.*")
|
||||
env.pageserver.allowed_errors.append(
|
||||
".*files not bound to index_file.json, proceeding with their deletion.*"
|
||||
)
|
||||
|
||||
wait_until(2, 0.5, lambda: assert_timeline_detail_404(client, tenant_id, timeline_id))
|
||||
|
||||
assert not timeline_path.exists()
|
||||
|
||||
# to please mypy
|
||||
assert isinstance(env.remote_storage, LocalFsStorage)
|
||||
remote_timeline_path = (
|
||||
env.remote_storage.root / "tenants" / str(tenant_id) / "timelines" / str(timeline_id)
|
||||
)
|
||||
|
||||
assert not list(remote_timeline_path.iterdir())
|
||||
|
||||
# timeline deletion should kill ongoing uploads, so, the metric will be gone
|
||||
assert get_queued_count(file_kind="index", op_kind="upload") is None
|
||||
|
||||
|
||||
@@ -3,7 +3,6 @@ import queue
|
||||
import shutil
|
||||
import threading
|
||||
from pathlib import Path
|
||||
from typing import Optional
|
||||
|
||||
import pytest
|
||||
import requests
|
||||
@@ -12,16 +11,13 @@ from fixtures.neon_fixtures import (
|
||||
NeonEnv,
|
||||
NeonEnvBuilder,
|
||||
RemoteStorageKind,
|
||||
S3Storage,
|
||||
available_remote_storages,
|
||||
)
|
||||
from fixtures.pageserver.http import PageserverApiException
|
||||
from fixtures.pageserver.utils import (
|
||||
assert_timeline_detail_404,
|
||||
wait_for_last_record_lsn,
|
||||
wait_for_upload,
|
||||
wait_until_tenant_active,
|
||||
wait_until_timeline_state,
|
||||
)
|
||||
from fixtures.types import Lsn, TenantId, TimelineId
|
||||
from fixtures.utils import query_scalar, wait_until
|
||||
@@ -72,7 +68,7 @@ def test_timeline_delete(neon_simple_env: NeonEnv):
|
||||
|
||||
ps_http.timeline_delete(env.initial_tenant, parent_timeline_id)
|
||||
|
||||
assert exc.value.status_code == 412
|
||||
assert exc.value.status_code == 400
|
||||
|
||||
timeline_path = (
|
||||
env.repo_dir / "tenants" / str(env.initial_tenant) / "timelines" / str(leaf_timeline_id)
|
||||
@@ -134,25 +130,13 @@ def test_delete_timeline_post_rm_failure(
|
||||
env = neon_env_builder.init_start()
|
||||
assert env.initial_timeline
|
||||
|
||||
env.pageserver.allowed_errors.append(".*Error: failpoint: timeline-delete-after-rm")
|
||||
env.pageserver.allowed_errors.append(".*Ignoring state update Stopping for broken timeline")
|
||||
|
||||
ps_http = env.pageserver.http_client()
|
||||
|
||||
failpoint_name = "timeline-delete-after-rm"
|
||||
ps_http.configure_failpoints((failpoint_name, "return"))
|
||||
|
||||
ps_http.timeline_delete(env.initial_tenant, env.initial_timeline)
|
||||
|
||||
timeline_info = wait_until_timeline_state(
|
||||
pageserver_http=ps_http,
|
||||
tenant_id=env.initial_tenant,
|
||||
timeline_id=env.initial_timeline,
|
||||
expected_state="Broken",
|
||||
iterations=2, # effectively try immediately and retry once in one second
|
||||
)
|
||||
|
||||
timeline_info["state"]["Broken"]["reason"] == "failpoint: timeline-delete-after-rm"
|
||||
with pytest.raises(PageserverApiException, match=f"failpoint: {failpoint_name}"):
|
||||
ps_http.timeline_delete(env.initial_tenant, env.initial_timeline)
|
||||
|
||||
at_failpoint_log_message = f".*{env.initial_timeline}.*at failpoint {failpoint_name}.*"
|
||||
env.pageserver.allowed_errors.append(at_failpoint_log_message)
|
||||
@@ -164,14 +148,11 @@ def test_delete_timeline_post_rm_failure(
|
||||
ps_http.configure_failpoints((failpoint_name, "off"))
|
||||
|
||||
# this should succeed
|
||||
# this also checks that delete can be retried even when timeline is in Broken state
|
||||
ps_http.timeline_delete(env.initial_tenant, env.initial_timeline, timeout=2)
|
||||
with pytest.raises(PageserverApiException) as e:
|
||||
ps_http.timeline_detail(env.initial_tenant, env.initial_timeline)
|
||||
|
||||
assert e.value.status_code == 404
|
||||
|
||||
env.pageserver.allowed_errors.append(f".*NotFound: Timeline.*{env.initial_timeline}.*")
|
||||
# the second call will try to transition the timeline into Stopping state, but it's already in that state
|
||||
env.pageserver.allowed_errors.append(
|
||||
f".*{env.initial_timeline}.*Ignoring new state, equal to the existing one: Stopping"
|
||||
)
|
||||
env.pageserver.allowed_errors.append(
|
||||
f".*{env.initial_timeline}.*timeline directory not found, proceeding anyway.*"
|
||||
)
|
||||
@@ -249,12 +230,6 @@ def test_timeline_resurrection_on_attach(
|
||||
# delete new timeline
|
||||
ps_http.timeline_delete(tenant_id=tenant_id, timeline_id=branch_timeline_id)
|
||||
|
||||
env.pageserver.allowed_errors.append(
|
||||
f".*Timeline {tenant_id}/{branch_timeline_id} was not found.*"
|
||||
)
|
||||
|
||||
wait_until(2, 0.5, lambda: assert_timeline_detail_404(ps_http, tenant_id, branch_timeline_id))
|
||||
|
||||
##### Stop the pageserver instance, erase all its data
|
||||
env.endpoints.stop_all()
|
||||
env.pageserver.stop()
|
||||
@@ -277,31 +252,12 @@ def test_timeline_resurrection_on_attach(
|
||||
assert all([tl["state"] == "Active" for tl in timelines])
|
||||
|
||||
|
||||
def assert_prefix_empty(neon_env_builder: NeonEnvBuilder, prefix: Optional[str] = None):
|
||||
# For local_fs we need to properly handle empty directories, which we currently dont, so for simplicity stick to s3 api.
|
||||
assert neon_env_builder.remote_storage_kind in (
|
||||
RemoteStorageKind.MOCK_S3,
|
||||
RemoteStorageKind.REAL_S3,
|
||||
)
|
||||
# For mypy
|
||||
assert isinstance(neon_env_builder.remote_storage, S3Storage)
|
||||
|
||||
# Note that this doesnt use pagination, so list is not guaranteed to be exhaustive.
|
||||
response = neon_env_builder.remote_storage_client.list_objects_v2(
|
||||
Bucket=neon_env_builder.remote_storage.bucket_name,
|
||||
Prefix=prefix or neon_env_builder.remote_storage.prefix_in_bucket or "",
|
||||
)
|
||||
objects = response.get("Contents")
|
||||
assert (
|
||||
response["KeyCount"] == 0
|
||||
), f"remote dir with prefix {prefix} is not empty after deletion: {objects}"
|
||||
|
||||
|
||||
def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuilder):
|
||||
"""
|
||||
When deleting a timeline, if we succeed in setting the deleted flag remotely
|
||||
but fail to delete the local state, restarting the pageserver should resume
|
||||
the deletion of the local state.
|
||||
(Deletion of the state in S3 is not implemented yet.)
|
||||
"""
|
||||
|
||||
neon_env_builder.enable_remote_storage(
|
||||
@@ -337,17 +293,11 @@ def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuild
|
||||
env.repo_dir / "tenants" / str(env.initial_tenant) / "timelines" / str(leaf_timeline_id)
|
||||
)
|
||||
|
||||
ps_http.timeline_delete(env.initial_tenant, leaf_timeline_id)
|
||||
|
||||
timeline_info = wait_until_timeline_state(
|
||||
pageserver_http=ps_http,
|
||||
tenant_id=env.initial_tenant,
|
||||
timeline_id=leaf_timeline_id,
|
||||
expected_state="Broken",
|
||||
iterations=2, # effectively try immediately and retry once in one second
|
||||
)
|
||||
|
||||
timeline_info["state"]["Broken"]["reason"] == "failpoint: timeline-delete-after-rm"
|
||||
with pytest.raises(
|
||||
PageserverApiException,
|
||||
match="failpoint: timeline-delete-before-rm",
|
||||
):
|
||||
ps_http.timeline_delete(env.initial_tenant, leaf_timeline_id)
|
||||
|
||||
assert leaf_timeline_path.exists(), "the failpoint didn't work"
|
||||
|
||||
@@ -355,14 +305,7 @@ def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuild
|
||||
env.pageserver.start()
|
||||
|
||||
# Wait for tenant to finish loading.
|
||||
wait_until_tenant_active(ps_http, tenant_id=env.initial_tenant, iterations=10, period=1)
|
||||
|
||||
env.pageserver.allowed_errors.append(
|
||||
f".*Timeline {env.initial_tenant}/{leaf_timeline_id} was not found.*"
|
||||
)
|
||||
wait_until(
|
||||
2, 0.5, lambda: assert_timeline_detail_404(ps_http, env.initial_tenant, leaf_timeline_id)
|
||||
)
|
||||
wait_until_tenant_active(ps_http, tenant_id=env.initial_tenant, iterations=10, period=0.5)
|
||||
|
||||
assert (
|
||||
not leaf_timeline_path.exists()
|
||||
@@ -374,50 +317,6 @@ def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuild
|
||||
}, "other timelines should not have been affected"
|
||||
assert all([tl["state"] == "Active" for tl in timelines])
|
||||
|
||||
assert_prefix_empty(
|
||||
neon_env_builder,
|
||||
prefix="/".join(
|
||||
(
|
||||
"tenants",
|
||||
str(env.initial_tenant),
|
||||
"timelines",
|
||||
str(leaf_timeline_id),
|
||||
)
|
||||
),
|
||||
)
|
||||
|
||||
assert env.initial_timeline is not None
|
||||
|
||||
for timeline_id in (intermediate_timeline_id, env.initial_timeline):
|
||||
ps_http.timeline_delete(env.initial_tenant, timeline_id)
|
||||
|
||||
env.pageserver.allowed_errors.append(
|
||||
f".*Timeline {env.initial_tenant}/{timeline_id} was not found.*"
|
||||
)
|
||||
wait_until(
|
||||
2, 0.5, lambda: assert_timeline_detail_404(ps_http, env.initial_tenant, timeline_id)
|
||||
)
|
||||
|
||||
assert_prefix_empty(
|
||||
neon_env_builder,
|
||||
prefix="/".join(
|
||||
(
|
||||
"tenants",
|
||||
str(env.initial_tenant),
|
||||
"timelines",
|
||||
str(timeline_id),
|
||||
)
|
||||
),
|
||||
)
|
||||
|
||||
# for some reason the check above doesnt immediately take effect for the below.
|
||||
# Assume it is mock server incosistency and check twice.
|
||||
wait_until(
|
||||
2,
|
||||
0.5,
|
||||
lambda: assert_prefix_empty(neon_env_builder),
|
||||
)
|
||||
|
||||
|
||||
def test_concurrent_timeline_delete_if_first_stuck_at_index_upload(
|
||||
neon_env_builder: NeonEnvBuilder,
|
||||
@@ -558,87 +457,3 @@ def test_delete_timeline_client_hangup(neon_env_builder: NeonEnvBuilder):
|
||||
ps_http.timeline_detail(env.initial_tenant, child_timeline_id)
|
||||
|
||||
assert exc.value.status_code == 404
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"remote_storage_kind",
|
||||
list(
|
||||
filter(
|
||||
lambda s: s in (RemoteStorageKind.MOCK_S3, RemoteStorageKind.REAL_S3),
|
||||
available_remote_storages(),
|
||||
)
|
||||
),
|
||||
)
|
||||
def test_timeline_delete_works_for_remote_smoke(
|
||||
neon_env_builder: NeonEnvBuilder,
|
||||
remote_storage_kind: RemoteStorageKind,
|
||||
):
|
||||
neon_env_builder.enable_remote_storage(
|
||||
remote_storage_kind=remote_storage_kind,
|
||||
test_name="test_timeline_delete_works_for_remote_smoke",
|
||||
)
|
||||
|
||||
env = neon_env_builder.init_start()
|
||||
|
||||
ps_http = env.pageserver.http_client()
|
||||
pg = env.endpoints.create_start("main")
|
||||
|
||||
tenant_id = TenantId(pg.safe_psql("show neon.tenant_id")[0][0])
|
||||
main_timeline_id = TimelineId(pg.safe_psql("show neon.timeline_id")[0][0])
|
||||
|
||||
assert tenant_id == env.initial_tenant
|
||||
assert main_timeline_id == env.initial_timeline
|
||||
|
||||
timeline_ids = [env.initial_timeline]
|
||||
for i in range(2):
|
||||
branch_timeline_id = env.neon_cli.create_branch(f"new{i}", "main")
|
||||
pg = env.endpoints.create_start(f"new{i}")
|
||||
|
||||
with pg.cursor() as cur:
|
||||
cur.execute("CREATE TABLE f (i integer);")
|
||||
cur.execute("INSERT INTO f VALUES (generate_series(1,1000));")
|
||||
current_lsn = Lsn(query_scalar(cur, "SELECT pg_current_wal_flush_lsn()"))
|
||||
|
||||
# wait until pageserver receives that data
|
||||
wait_for_last_record_lsn(ps_http, tenant_id, branch_timeline_id, current_lsn)
|
||||
|
||||
# run checkpoint manually to be sure that data landed in remote storage
|
||||
ps_http.timeline_checkpoint(tenant_id, branch_timeline_id)
|
||||
|
||||
# wait until pageserver successfully uploaded a checkpoint to remote storage
|
||||
log.info("waiting for checkpoint upload")
|
||||
wait_for_upload(ps_http, tenant_id, branch_timeline_id, current_lsn)
|
||||
log.info("upload of checkpoint is done")
|
||||
timeline_id = TimelineId(pg.safe_psql("show neon.timeline_id")[0][0])
|
||||
|
||||
timeline_ids.append(timeline_id)
|
||||
|
||||
for timeline_id in reversed(timeline_ids):
|
||||
# note that we need to finish previous deletion before scheduling next one
|
||||
# otherwise we can get an "HasChildren" error if deletion is not fast enough (real_s3)
|
||||
ps_http.timeline_delete(tenant_id=tenant_id, timeline_id=timeline_id)
|
||||
|
||||
env.pageserver.allowed_errors.append(
|
||||
f".*Timeline {env.initial_tenant}/{timeline_id} was not found.*"
|
||||
)
|
||||
wait_until(2, 0.5, lambda: assert_timeline_detail_404(ps_http, tenant_id, timeline_id))
|
||||
|
||||
assert_prefix_empty(
|
||||
neon_env_builder,
|
||||
prefix="/".join(
|
||||
(
|
||||
"tenants",
|
||||
str(env.initial_tenant),
|
||||
"timelines",
|
||||
str(timeline_id),
|
||||
)
|
||||
),
|
||||
)
|
||||
|
||||
# for some reason the check above doesnt immediately take effect for the below.
|
||||
# Assume it is mock server incosistency and check twice.
|
||||
wait_until(
|
||||
2,
|
||||
0.5,
|
||||
lambda: assert_prefix_empty(neon_env_builder),
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user