Compare commits

...

5 Commits

Author SHA1 Message Date
Alex Chi Z acb171b8b3 fix(build): warnings in dockerfile
Signed-off-by: Alex Chi Z <chi@neon.tech>
2024-08-09 12:36:38 +08:00
John Spray 4431688dc6 tests: don't require kafka client for regular tests (#8662)
## Problem

We're adding more third party dependencies to support more diverse +
realistic test cases in `test_runner/logical_repl`. I ❤️ these
tests, they are a good thing.

The slight glitch is that python packaging is hard, and some third party
python packages have issues. For example the current kafka dependency
doesn't work on latest python. We can mitigate that by only importing
these more specialized dependencies in the tests that use them.

## Summary of changes

- Move the `kafka` import into a test body, so that folks running the
regular `test_runner/regress` tests don't have to have a working kafka
client package.
2024-08-08 19:24:21 +01:00
John Spray 953b7d4f7e pageserver: remove paranoia double-calculation of retain_lsns (#8617)
## Problem

This code was to mitigate risk in
https://github.com/neondatabase/neon/pull/8427

As expected, we did not hit this code path - the new continuous updates
of gc_info are working fine, we can remove this code now.

## Summary of changes

- Remove block that double-checks retain_lsns
2024-08-08 12:57:48 +01:00
Joonas Koivunen 8561b2c628 fix: stop leaking BackgroundPurges (#8650)
avoid "leaking" the completions of BackgroundPurges by:

1. switching it to TaskTracker for provided close+wait
2. stop using tokio::fs::remove_dir_all which will consume two units of
memory instead of one blocking task

Additionally, use more graceful shutdown in tests which do actually some
background cleanup.
2024-08-08 12:02:53 +01:00
Joonas Koivunen 21638ee96c fix(test): do not fail test for filesystem race (#8643)
evidence:
https://neon-github-public-dev.s3.amazonaws.com/reports/pr-8632/10287641784/index.html#suites/0e58fb04d9998963e98e45fe1880af7d/c7a46335515142b/
2024-08-08 10:34:47 +01:00
8 changed files with 81 additions and 148 deletions
+2 -2
View File
@@ -33,8 +33,8 @@ ARG BUILD_TAG
# Set up cachepot to use an AWS S3 bucket for cache results, to reuse it between `docker build` invocations.
# cachepot falls back to local filesystem if S3 is misconfigured, not failing the build
ARG RUSTC_WRAPPER=cachepot
ENV AWS_REGION=eu-central-1
ENV CACHEPOT_S3_KEY_PREFIX=cachepot
ENV AWS_REGION eu-central-1
ENV CACHEPOT_S3_KEY_PREFIX cachepot
ARG CACHEPOT_BUCKET=neon-github-dev
#ARG AWS_ACCESS_KEY_ID
#ARG AWS_SECRET_ACCESS_KEY
+13 -13
View File
@@ -66,13 +66,13 @@ RUN curl -fsSL "https://github.com/protocolbuffers/protobuf/releases/download/v$
&& rm -rf protoc.zip protoc
# s5cmd
ENV S5CMD_VERSION=2.2.2
ENV S5CMD_VERSION 2.2.2
RUN curl -sL "https://github.com/peak/s5cmd/releases/download/v${S5CMD_VERSION}/s5cmd_${S5CMD_VERSION}_Linux-$(uname -m | sed 's/x86_64/64bit/g' | sed 's/aarch64/arm64/g').tar.gz" | tar zxvf - s5cmd \
&& chmod +x s5cmd \
&& mv s5cmd /usr/local/bin/s5cmd
# LLVM
ENV LLVM_VERSION=18
ENV LLVM_VERSION 18
RUN curl -fsSL 'https://apt.llvm.org/llvm-snapshot.gpg.key' | apt-key add - \
&& echo "deb http://apt.llvm.org/${DEBIAN_VERSION_CODENAME}/ llvm-toolchain-${DEBIAN_VERSION_CODENAME}-${LLVM_VERSION} main" > /etc/apt/sources.list.d/llvm.stable.list \
&& apt update \
@@ -125,8 +125,8 @@ RUN for package in Capture::Tiny DateTime Devel::Cover Digest::MD5 File::Spec JS
&& rm -rf ../lcov.tar.gz
# Compile and install the static OpenSSL library
ENV OPENSSL_VERSION=1.1.1w
ENV OPENSSL_PREFIX=/usr/local/openssl
ENV OPENSSL_VERSION 1.1.1w
ENV OPENSSL_PREFIX /usr/local/openssl
RUN wget -O /tmp/openssl-${OPENSSL_VERSION}.tar.gz https://www.openssl.org/source/openssl-${OPENSSL_VERSION}.tar.gz && \
echo "cf3098950cb4d853ad95c0841f1f9c6d3dc102dccfcacd521d93925208b76ac8 /tmp/openssl-${OPENSSL_VERSION}.tar.gz" | sha256sum --check && \
cd /tmp && \
@@ -145,8 +145,8 @@ RUN wget -O /tmp/openssl-${OPENSSL_VERSION}.tar.gz https://www.openssl.org/sourc
# TODO: at this time, Dockerfile.compute-node uses the debian bullseye libicu
# package, which is 67.1. We're duplicating that knowledge here, and also, technically,
# Debian has a few patches on top of 67.1 that we're not adding here.
ENV ICU_VERSION=67.1
ENV ICU_PREFIX=/usr/local/icu
ENV ICU_VERSION 67.1
ENV ICU_PREFIX /usr/local/icu
# Download and build static ICU
RUN wget -O /tmp/libicu-${ICU_VERSION}.tgz https://github.com/unicode-org/icu/releases/download/release-${ICU_VERSION//./-}/icu4c-${ICU_VERSION//./_}-src.tgz && \
@@ -168,9 +168,9 @@ USER nonroot:nonroot
WORKDIR /home/nonroot
# Python
ENV PYTHON_VERSION=3.9.18 \
PYENV_ROOT=/home/nonroot/.pyenv \
PATH=/home/nonroot/.pyenv/shims:/home/nonroot/.pyenv/bin:/home/nonroot/.poetry/bin:$PATH
ENV PYTHON_VERSION 3.9.18
ENV PYENV_ROOT /home/nonroot/.pyenv
ENV PATH /home/nonroot/.pyenv/shims:/home/nonroot/.pyenv/bin:/home/nonroot/.poetry/bin:$PATH
RUN set -e \
&& cd $HOME \
&& curl -sSO https://raw.githubusercontent.com/pyenv/pyenv-installer/master/bin/pyenv-installer \
@@ -192,9 +192,9 @@ WORKDIR /home/nonroot
# Rust
# Please keep the version of llvm (installed above) in sync with rust llvm (`rustc --version --verbose | grep LLVM`)
ENV RUSTC_VERSION=1.80.0
ENV RUSTUP_HOME="/home/nonroot/.rustup"
ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
ENV RUSTC_VERSION 1.80.0
ENV RUSTUP_HOME "/home/nonroot/.rustup"
ENV PATH "/home/nonroot/.cargo/bin:${PATH}"
RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && whoami && \
chmod +x rustup-init && \
./rustup-init -y --default-toolchain ${RUSTC_VERSION} && \
@@ -211,7 +211,7 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
cargo install cargo-nextest && \
rm -rf /home/nonroot/.cargo/registry && \
rm -rf /home/nonroot/.cargo/git
ENV RUSTC_WRAPPER=cachepot
ENV RUSTC_WRAPPER cachepot
# Show versions
RUN whoami \
+8 -8
View File
@@ -647,8 +647,8 @@ RUN apt-get update && \
apt-get install -y curl libclang-dev cmake && \
useradd -ms /bin/bash nonroot -b /home
ENV HOME=/home/nonroot
ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
ENV HOME /home/nonroot
ENV PATH "/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
USER nonroot
WORKDIR /home/nonroot
ARG PG_VERSION
@@ -873,7 +873,7 @@ RUN make -j $(getconf _NPROCESSORS_ONLN) \
#########################################################################################
FROM $REPOSITORY/$IMAGE:$TAG AS compute-tools
ARG BUILD_TAG
ENV BUILD_TAG=$BUILD_TAG
ENV BUILD_TAG $BUILD_TAG
USER nonroot
# Copy entire project to get Cargo.* files with proper dependencies for the whole project
@@ -968,11 +968,11 @@ RUN patch -p1 < /ext-src/pg_hintplan.patch
COPY --chmod=755 docker-compose/run-tests.sh /run-tests.sh
RUN patch -p1 </ext-src/pg_anon.patch
RUN patch -p1 </ext-src/pg_cron.patch
ENV PATH=/usr/local/pgsql/bin:$PATH
ENV PGHOST=compute
ENV PGPORT=55433
ENV PGUSER=cloud_admin
ENV PGDATABASE=postgres
ENV PATH /usr/local/pgsql/bin:$PATH
ENV PGHOST compute
ENV PGPORT 55433
ENV PGUSER cloud_admin
ENV PGDATABASE postgres
#########################################################################################
#
# Final layer
+20 -5
View File
@@ -56,7 +56,6 @@ impl Statvfs {
}
pub mod mock {
use anyhow::Context;
use camino::Utf8Path;
use regex::Regex;
use tracing::log::info;
@@ -135,14 +134,30 @@ pub mod mock {
{
continue;
}
total += entry
.metadata()
.with_context(|| format!("get metadata of {:?}", entry.path()))?
.len();
let m = match entry.metadata() {
Ok(m) => m,
Err(e) if is_not_found(&e) => {
// some temp file which got removed right as we are walking
continue;
}
Err(e) => {
return Err(anyhow::Error::new(e)
.context(format!("get metadata of {:?}", entry.path())))
}
};
total += m.len();
}
Ok(total)
}
fn is_not_found(e: &walkdir::Error) -> bool {
let Some(io_error) = e.io_error() else {
return false;
};
let kind = io_error.kind();
matches!(kind, std::io::ErrorKind::NotFound)
}
pub struct Statvfs {
pub blocks: u64,
pub blocks_available: u64,
-48
View File
@@ -3012,54 +3012,6 @@ impl Tenant {
// because that will stall branch creation.
let gc_cs = self.gc_cs.lock().await;
// Paranoia check: it is critical that GcInfo's list of child timelines is correct, to avoid incorrectly GC'ing data they
// depend on. So although GcInfo is updated continuously by Timeline::new and Timeline::drop, we also calculate it here
// and fail out if it's inaccurate.
// (this can be removed later, it's a risk mitigation for https://github.com/neondatabase/neon/pull/8427)
{
let mut all_branchpoints: BTreeMap<TimelineId, Vec<(Lsn, TimelineId)>> =
BTreeMap::new();
timelines.iter().for_each(|timeline| {
if let Some(ancestor_timeline_id) = &timeline.get_ancestor_timeline_id() {
let ancestor_children =
all_branchpoints.entry(*ancestor_timeline_id).or_default();
ancestor_children.push((timeline.get_ancestor_lsn(), timeline.timeline_id));
}
});
for timeline in &timelines {
let mut branchpoints: Vec<(Lsn, TimelineId)> = all_branchpoints
.remove(&timeline.timeline_id)
.unwrap_or_default();
branchpoints.sort_by_key(|b| b.0);
let target = timeline.gc_info.read().unwrap();
// We require that retain_lsns contains everything in `branchpoints`, but not that
// they are exactly equal: timeline deletions can race with us, so retain_lsns
// may contain some extra stuff. It is safe to have extra timelines in there, because it
// just means that we retain slightly more data than we otherwise might.
let have_branchpoints = target.retain_lsns.iter().copied().collect::<HashSet<_>>();
for b in &branchpoints {
if !have_branchpoints.contains(b) {
tracing::error!(
"Bug: `retain_lsns` is set incorrectly. Expected be {:?}, but found {:?}",
branchpoints,
target.retain_lsns
);
debug_assert!(false);
// Do not GC based on bad information!
// (ab-use an existing GcError type rather than adding a new one, since this is a
// "should never happen" check that will be removed soon).
return Err(GcError::Remote(anyhow::anyhow!(
"retain_lsns failed validation!"
)));
}
}
}
}
// Ok, we now know all the branch points.
// Update the GC information for each timeline.
let mut gc_timelines = Vec::with_capacity(timelines.len());
+29 -67
View File
@@ -224,21 +224,8 @@ async fn safe_rename_tenant_dir(path: impl AsRef<Utf8Path>) -> std::io::Result<U
}
/// See [`Self::spawn`].
#[derive(Clone)]
pub struct BackgroundPurges(Arc<std::sync::Mutex<BackgroundPurgesInner>>);
enum BackgroundPurgesInner {
Open(tokio::task::JoinSet<()>),
// we use the async mutex for coalescing
ShuttingDown(Arc<tokio::sync::Mutex<tokio::task::JoinSet<()>>>),
}
impl Default for BackgroundPurges {
fn default() -> Self {
Self(Arc::new(std::sync::Mutex::new(
BackgroundPurgesInner::Open(JoinSet::new()),
)))
}
}
#[derive(Clone, Default)]
pub struct BackgroundPurges(tokio_util::task::TaskTracker);
impl BackgroundPurges {
/// When we have moved a tenant's content to a temporary directory, we may delete it lazily in
@@ -247,24 +234,32 @@ impl BackgroundPurges {
/// Although we are cleaning up the tenant, this task is not meant to be bound by the lifetime of the tenant in memory.
/// Thus the [`BackgroundPurges`] type to keep track of these tasks.
pub fn spawn(&self, tmp_path: Utf8PathBuf) {
let mut guard = self.0.lock().unwrap();
let jset = match &mut *guard {
BackgroundPurgesInner::Open(ref mut jset) => jset,
BackgroundPurgesInner::ShuttingDown(_) => {
warn!("trying to spawn background purge during shutdown, ignoring");
return;
// because on shutdown we close and wait, we are misusing TaskTracker a bit.
//
// so first acquire a token, then check if the tracker has been closed. the tracker might get closed
// right after, but at least the shutdown will wait for what we are spawning next.
let token = self.0.token();
if self.0.is_closed() {
warn!(
%tmp_path,
"trying to spawn background purge during shutdown, ignoring"
);
return;
}
let span = info_span!(parent: None, "background_purge", %tmp_path);
let task = move || {
let _token = token;
let _entered = span.entered();
if let Err(error) = std::fs::remove_dir_all(tmp_path.as_path()) {
// should we fatal_io_error here?
warn!(%error, "failed to purge tenant directory");
}
};
jset.spawn_on(
async move {
if let Err(error) = fs::remove_dir_all(tmp_path.as_path()).await {
// should we fatal_io_error here?
warn!(%error, path=%tmp_path, "failed to purge tenant directory");
}
}
.instrument(info_span!(parent: None, "background_purge")),
BACKGROUND_RUNTIME.handle(),
);
BACKGROUND_RUNTIME.spawn_blocking(task);
}
/// When this future completes, all background purges have completed.
@@ -278,42 +273,9 @@ impl BackgroundPurges {
/// instances of this future will continue to be correct.
#[instrument(skip_all)]
pub async fn shutdown(&self) {
let jset = {
let mut guard = self.0.lock().unwrap();
match &mut *guard {
BackgroundPurgesInner::Open(jset) => {
*guard = BackgroundPurgesInner::ShuttingDown(Arc::new(tokio::sync::Mutex::new(
std::mem::take(jset),
)))
}
BackgroundPurgesInner::ShuttingDown(_) => {
// calling shutdown multiple times is most likely a bug in pageserver shutdown code
warn!("already shutting down");
}
};
match &mut *guard {
BackgroundPurgesInner::ShuttingDown(ref mut jset) => jset.clone(),
BackgroundPurgesInner::Open(_) => {
unreachable!("above code transitions into shut down state");
}
}
};
let mut jset = jset.lock().await; // concurrent callers coalesce here
while let Some(res) = jset.join_next().await {
match res {
Ok(()) => {}
Err(e) if e.is_panic() => {
// If it panicked, the error is already logged by the panic hook.
}
Err(e) if e.is_cancelled() => {
unreachable!("we don't cancel the joinset or runtime")
}
Err(e) => {
// No idea when this can happen, but let's log it.
warn!(%e, "background purge task failed or panicked");
}
}
}
// forbid new tasks (can be called many times)
self.0.close();
self.0.wait().await;
}
}
+2 -1
View File
@@ -12,7 +12,6 @@ import requests
from fixtures.log_helper import log
from fixtures.neon_fixtures import RemotePostgres
from fixtures.utils import wait_until
from kafka import KafkaConsumer
class DebeziumAPI:
@@ -95,6 +94,8 @@ def debezium(remote_pg: RemotePostgres):
log.debug("%s %s %s", resp.status_code, resp.ok, resp.text)
assert resp.status_code == 201
assert len(dbz.list_connectors()) == 1
from kafka import KafkaConsumer
consumer = KafkaConsumer(
"dbserver1.inventory.customers",
bootstrap_servers=["kafka:9092"],
+7 -4
View File
@@ -128,6 +128,8 @@ def test_tenant_delete_smoke(
assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "attached"}) == 1
assert ps_http.get_metric_value("pageserver_tenant_manager_slots", {"mode": "inprogress"}) == 0
env.pageserver.stop()
def test_long_timeline_create_cancelled_by_tenant_delete(neon_env_builder: NeonEnvBuilder):
"""Reproduction of 2023-11-23 stuck tenants investigation"""
@@ -200,11 +202,10 @@ def test_long_timeline_create_cancelled_by_tenant_delete(neon_env_builder: NeonE
if deletion is not None:
deletion.join()
env.pageserver.stop()
def test_tenant_delete_races_timeline_creation(
neon_env_builder: NeonEnvBuilder,
pg_bin: PgBin,
):
def test_tenant_delete_races_timeline_creation(neon_env_builder: NeonEnvBuilder):
"""
Validate that timeline creation executed in parallel with deletion works correctly.
@@ -318,6 +319,8 @@ def test_tenant_delete_races_timeline_creation(
# We deleted our only tenant, and the scrubber fails if it detects nothing
neon_env_builder.disable_scrub_on_exit()
env.pageserver.stop()
def test_tenant_delete_scrubber(pg_bin: PgBin, neon_env_builder: NeonEnvBuilder):
"""