pageserver: make concurrent offloaded timeline operations safe wrt manifest uploads (#9557)

## Problem

Uploads of the tenant manifest could race between different tasks,
resulting in unexpected results in remote storage.

Closes: https://github.com/neondatabase/neon/issues/9556

## Summary of changes

- Create a central function for uploads that takes a tokio::sync::Mutex
- Store the latest upload in that Mutex, so that when there is lots of
concurrency (e.g. archive 20 timelines at once) we can coalesce their
manifest writes somewhat.
This commit is contained in:
John Spray
2024-10-29 13:54:48 +00:00
committed by GitHub
parent 4ef74215e1
commit 7a1331eee5
5 changed files with 94 additions and 72 deletions

View File

@@ -302,6 +302,13 @@ pub struct Tenant {
/// **Lock order**: if acquiring all (or a subset), acquire them in order `timelines`, `timelines_offloaded`, `timelines_creating`
timelines_offloaded: Mutex<HashMap<TimelineId, Arc<OffloadedTimeline>>>,
/// Serialize writes of the tenant manifest to remote storage. If there are concurrent operations
/// affecting the manifest, such as timeline deletion and timeline offload, they must wait for
/// each other (this could be optimized to coalesce writes if necessary).
///
/// The contents of the Mutex are the last manifest we successfully uploaded
tenant_manifest_upload: tokio::sync::Mutex<Option<TenantManifest>>,
// This mutex prevents creation of new timelines during GC.
// Adding yet another mutex (in addition to `timelines`) is needed because holding
// `timelines` mutex during all GC iteration
@@ -741,6 +748,24 @@ pub enum TimelineArchivalError {
Other(anyhow::Error),
}
#[derive(thiserror::Error, Debug)]
pub(crate) enum TenantManifestError {
#[error("Remote storage error: {0}")]
RemoteStorage(anyhow::Error),
#[error("Cancelled")]
Cancelled,
}
impl From<TenantManifestError> for TimelineArchivalError {
fn from(e: TenantManifestError) -> Self {
match e {
TenantManifestError::RemoteStorage(e) => Self::Other(e),
TenantManifestError::Cancelled => Self::Cancelled,
}
}
}
impl Debug for TimelineArchivalError {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match self {
@@ -1526,18 +1551,7 @@ impl Tenant {
offloaded_timelines_accessor.extend(offloaded_timelines_list.into_iter());
}
if !offloaded_timeline_ids.is_empty() {
let manifest = self.tenant_manifest();
// TODO: generation support
let generation = remote_timeline_client::TENANT_MANIFEST_GENERATION;
upload_tenant_manifest(
&self.remote_storage,
&self.tenant_shard_id,
generation,
&manifest,
&self.cancel,
)
.await
.map_err(TimelineArchivalError::Other)?;
self.store_tenant_manifest().await?;
}
// The local filesystem contents are a cache of what's in the remote IndexPart;
@@ -1918,18 +1932,7 @@ impl Tenant {
};
// Upload new list of offloaded timelines to S3
let manifest = self.tenant_manifest();
// TODO: generation support
let generation = remote_timeline_client::TENANT_MANIFEST_GENERATION;
upload_tenant_manifest(
&self.remote_storage,
&self.tenant_shard_id,
generation,
&manifest,
&cancel,
)
.await
.map_err(TimelineArchivalError::Other)?;
self.store_tenant_manifest().await?;
// Activate the timeline (if it makes sense)
if !(timeline.is_broken() || timeline.is_stopping()) {
@@ -3126,7 +3129,7 @@ impl Tenant {
}
}
let tenant_manifest = self.tenant_manifest();
let tenant_manifest = self.build_tenant_manifest();
// TODO: generation support
let generation = remote_timeline_client::TENANT_MANIFEST_GENERATION;
for child_shard in child_shards {
@@ -3321,7 +3324,8 @@ impl Tenant {
.unwrap_or(self.conf.default_tenant_conf.lsn_lease_length)
}
pub(crate) fn tenant_manifest(&self) -> TenantManifest {
/// Generate an up-to-date TenantManifest based on the state of this Tenant.
fn build_tenant_manifest(&self) -> TenantManifest {
let timelines_offloaded = self.timelines_offloaded.lock().unwrap();
let mut timeline_manifests = timelines_offloaded
@@ -3529,6 +3533,7 @@ impl Tenant {
timelines: Mutex::new(HashMap::new()),
timelines_creating: Mutex::new(HashSet::new()),
timelines_offloaded: Mutex::new(HashMap::new()),
tenant_manifest_upload: Default::default(),
gc_cs: tokio::sync::Mutex::new(()),
walredo_mgr,
remote_storage,
@@ -4708,6 +4713,49 @@ impl Tenant {
.max()
.unwrap_or(0)
}
/// Serialize and write the latest TenantManifest to remote storage.
pub(crate) async fn store_tenant_manifest(&self) -> Result<(), TenantManifestError> {
// Only one manifest write may be done at at time, and the contents of the manifest
// must be loaded while holding this lock. This makes it safe to call this function
// from anywhere without worrying about colliding updates.
let mut guard = tokio::select! {
g = self.tenant_manifest_upload.lock() => {
g
},
_ = self.cancel.cancelled() => {
return Err(TenantManifestError::Cancelled);
}
};
let manifest = self.build_tenant_manifest();
if Some(&manifest) == (*guard).as_ref() {
// Optimisation: skip uploads that don't change anything.
return Ok(());
}
upload_tenant_manifest(
&self.remote_storage,
&self.tenant_shard_id,
self.generation,
&manifest,
&self.cancel,
)
.await
.map_err(|e| {
if self.cancel.is_cancelled() {
TenantManifestError::Cancelled
} else {
TenantManifestError::RemoteStorage(e)
}
})?;
// Store the successfully uploaded manifest, so that future callers can avoid
// re-uploading the same thing.
*guard = Some(manifest);
Ok(())
}
}
/// Create the cluster temporarily in 'initdbpath' directory inside the repository

View File

@@ -249,7 +249,7 @@ pub(crate) use download::{
list_remote_tenant_shards, list_remote_timelines,
};
pub(crate) use index::LayerFileMetadata;
pub(crate) use upload::{upload_initdb_dir, upload_tenant_manifest};
pub(crate) use upload::upload_initdb_dir;
// Occasional network issues and such can cause remote operations to fail, and
// that's expected. If a download fails, we log it at info-level, and retry.

View File

@@ -3,7 +3,7 @@ use serde::{Deserialize, Serialize};
use utils::{id::TimelineId, lsn::Lsn};
/// Tenant-shard scoped manifest
#[derive(Clone, Serialize, Deserialize)]
#[derive(Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct TenantManifest {
/// Debugging aid describing the version of this manifest.
/// Can also be used for distinguishing breaking changes later on.
@@ -23,7 +23,7 @@ pub struct TenantManifest {
/// Very similar to [`pageserver_api::models::OffloadedTimelineInfo`],
/// but the two datastructures serve different needs, this is for a persistent disk format
/// that must be backwards compatible, while the other is only for informative purposes.
#[derive(Clone, Serialize, Deserialize, Copy)]
#[derive(Clone, Serialize, Deserialize, Copy, PartialEq, Eq)]
pub struct OffloadedTimelineManifest {
pub timeline_id: TimelineId,
/// Whether the timeline has a parent it has been branched off from or not

View File

@@ -14,10 +14,9 @@ use crate::{
task_mgr::{self, TaskKind},
tenant::{
metadata::TimelineMetadata,
remote_timeline_client::{
self, MaybeDeletedIndexPart, PersistIndexPartWithDeletedFlagError, RemoteTimelineClient,
},
CreateTimelineCause, DeleteTimelineError, Tenant, TimelineOrOffloaded,
remote_timeline_client::{PersistIndexPartWithDeletedFlagError, RemoteTimelineClient},
CreateTimelineCause, DeleteTimelineError, MaybeDeletedIndexPart, Tenant,
TimelineOrOffloaded,
},
};
@@ -176,32 +175,6 @@ async fn remove_maybe_offloaded_timeline_from_tenant(
Ok(())
}
/// It is important that this gets called when DeletionGuard is being held.
/// For more context see comments in [`DeleteTimelineFlow::prepare`]
async fn upload_new_tenant_manifest(
tenant: &Tenant,
_: &DeletionGuard, // using it as a witness
) -> anyhow::Result<()> {
// This is susceptible to race conditions, i.e. we won't continue deletions if there is a crash
// between the deletion of the index-part.json and reaching of this code.
// So indeed, the tenant manifest might refer to an offloaded timeline which has already been deleted.
// However, we handle this case in tenant loading code so the next time we attach, the issue is
// resolved.
let manifest = tenant.tenant_manifest();
// TODO: generation support
let generation = remote_timeline_client::TENANT_MANIFEST_GENERATION;
remote_timeline_client::upload_tenant_manifest(
&tenant.remote_storage,
&tenant.tenant_shard_id,
generation,
&manifest,
&tenant.cancel,
)
.await?;
Ok(())
}
/// Orchestrates timeline shut down of all timeline tasks, removes its in-memory structures,
/// and deletes its data from both disk and s3.
/// The sequence of steps:
@@ -480,7 +453,15 @@ impl DeleteTimelineFlow {
remove_maybe_offloaded_timeline_from_tenant(tenant, timeline, &guard).await?;
upload_new_tenant_manifest(tenant, &guard).await?;
// This is susceptible to race conditions, i.e. we won't continue deletions if there is a crash
// between the deletion of the index-part.json and reaching of this code.
// So indeed, the tenant manifest might refer to an offloaded timeline which has already been deleted.
// However, we handle this case in tenant loading code so the next time we attach, the issue is
// resolved.
tenant
.store_tenant_manifest()
.await
.map_err(|e| DeleteTimelineError::Other(anyhow::anyhow!(e)))?;
*guard = Self::Finished;

View File

@@ -3,7 +3,7 @@ use std::sync::Arc;
use super::delete::{delete_local_timeline_directory, DeleteTimelineFlow, DeletionGuard};
use super::Timeline;
use crate::span::debug_assert_current_span_has_tenant_and_timeline_id;
use crate::tenant::{remote_timeline_client, OffloadedTimeline, Tenant, TimelineOrOffloaded};
use crate::tenant::{OffloadedTimeline, Tenant, TimelineOrOffloaded};
pub(crate) async fn offload_timeline(
tenant: &Tenant,
@@ -63,17 +63,10 @@ pub(crate) async fn offload_timeline(
// at the next restart attach it again.
// For that to happen, we'd need to make the manifest reflect our *intended* state,
// not our actual state of offloaded timelines.
let manifest = tenant.tenant_manifest();
// TODO: generation support
let generation = remote_timeline_client::TENANT_MANIFEST_GENERATION;
remote_timeline_client::upload_tenant_manifest(
&tenant.remote_storage,
&tenant.tenant_shard_id,
generation,
&manifest,
&tenant.cancel,
)
.await?;
tenant
.store_tenant_manifest()
.await
.map_err(|e| anyhow::anyhow!(e))?;
Ok(())
}