Compare commits

..

46 Commits

Author SHA1 Message Date
Alexander Lakhin
6368777778 Save coverage data upon completion of regress-tests; Remove legacy step "Build coverage report" 2025-07-03 13:50:45 +03:00
Suhas Thalanki
5f3532970e [compute] fix: background worker that collects installed extension metrics now updates collection interval (#12277)
## Problem

Previously, the background worker that collects the list of installed
extensions across DBs had a timeout set to 1 hour. This cause a problem
with computes that had a `suspend_timeout` > 1 hour as this collection
was treated as activity, preventing compute shutdown.

Issue: https://github.com/neondatabase/cloud/issues/30147

## Summary of changes

Passing the `suspend_timeout` as part of the `ComputeSpec` so that any
updates to this are taken into account by the background worker and
updates its collection interval.
2025-06-30 22:12:37 +00:00
Arpad Müller
2e681e0ef8 detach_ancestor: delete the right layer when hardlink fails (#12397)
If a hardlink operation inside `detach_ancestor` fails due to the layer
already existing, we delete the layer to make sure the source is one we
know about, and then retry.

But we deleted the wrong file, namely, the one we wanted to use as the
source of the hardlink. As a result, the follow up hard link operation
failed. Our PR corrects this mistake.
2025-06-30 21:36:15 +00:00
Dmitrii Kovalkov
8e216a3a59 storcon: notify cplane on safekeeper membership change (#12390)
## Problem
We don't notify cplane about safekeeper membership change yet. Without
the notification the compute needs to know all the safekeepers on the
cluster to be able to speak to them. Change notifications will allow to
avoid it.

- Closes: https://github.com/neondatabase/neon/issues/12188

## Summary of changes
- Implement `notify_safekeepers` method in `ComputeHook`
- Notify cplane about safekeepers in `safekeeper_migrate` handler.
- Update the test to make sure notifications work.

## Out of scope
- There is `cplane_notified_generation` field in `timelines` table in
strocon's database. It's not needed now, so it's not updated in the PR.
Probably we can remove it.
- e2e tests to make sure it works with a production cplane
2025-06-30 14:09:50 +00:00
Erik Grinaker
d0a4ae3e8f pageserver: add gRPC LSN lease support (#12384)
## Problem

The gRPC API does not provide LSN leases.

## Summary of changes

* Add LSN lease support to the gRPC API.
* Use gRPC LSN leases for static computes with `grpc://` connstrings.
* Move `PageserverProtocol` into the `compute_api::spec` module and
reuse it.
2025-06-30 12:44:17 +00:00
Erik Grinaker
a384d7d501 pageserver: assert no changes to shard identity (#12379)
## Problem

Location config changes can currently result in changes to the shard
identity. Such changes will cause data corruption, as seen with #12217.

Resolves #12227.
Requires #12377.

## Summary of changes

Assert that the shard identity does not change on location config
updates and on (re)attach.

This is currently asserted with `critical!`, in case it misfires in
production. Later, we should reject such requests with an error and turn
this into a proper assertion.
2025-06-30 12:36:45 +00:00
Christian Schwarz
66f53d9d34 refactor(pageserver): force explicit mapping to CreateImageLayersError::Other (#12382)
Implicit mapping to an `anyhow::Error` when we do `?` is discouraged
because tooling to find those places isn't great.

As a drive-by, also make SplitImageLayerWriter::new infallible and sync.
I think we should also make ImageLayerWriter::new completely lazy,
then `BatchLayerWriter:new` infallible and async.
2025-06-30 11:03:48 +00:00
Busra Kugler
2af9380962 Revert "Replace step-security maintained actions" (#12386)
Reverts neondatabase/neon#11663 and
https://github.com/neondatabase/neon/pull/11265/

Step Security is not yet approved by Databricks team, in order to
prevent issues during Github org migration, I'll revert this PR to use
the previous action instead of Step Security maintained action.
2025-06-30 10:15:10 +00:00
Ivan Efremov
620d50432c Fix path issue in the proxy-bernch CI workflow (#12388) 2025-06-30 09:33:57 +00:00
Erik Grinaker
1d43f3bee8 pageserver: fix stripe size persistence in legacy HTTP handlers (#12377)
## Problem

Similarly to #12217, the following endpoints may result in a stripe size
mismatch between the storage controller and Pageserver if an unsharded
tenant has a different stripe size set than the default. This can lead
to data corruption if the tenant is later manually split without
specifying an explicit stripe size, since the storage controller and
Pageserver will apply different defaults. This commonly happens with
tenants that were created before the default stripe size was changed
from 32k to 2k.

* `PUT /v1/tenant/config`
* `PATCH /v1/tenant/config`

These endpoints are no longer in regular production use (they were used
when cplane still managed Pageserver directly), but can still be called
manually or by tests.

## Summary of changes

Retain the current shard parameters when updating the location config in
`PUT | PATCH /v1/tenant/config`.

Also opportunistically derive `Copy` for `ShardParameters`.
2025-06-30 09:08:44 +00:00
Dmitrii Kovalkov
c746678bbc storcon: implement safekeeper_migrate handler (#11849)
This PR implements a safekeeper migration algorithm from RFC-035


https://github.com/neondatabase/neon/blob/main/docs/rfcs/035-safekeeper-dynamic-membership-change.md#change-algorithm

- Closes: https://github.com/neondatabase/neon/issues/11823

It is not production-ready yet, but I think it's good enough to commit
and start testing.

There are some known issues which will be addressed in later PRs:
- https://github.com/neondatabase/neon/issues/12186
- https://github.com/neondatabase/neon/issues/12187
- https://github.com/neondatabase/neon/issues/12188
- https://github.com/neondatabase/neon/issues/12189
- https://github.com/neondatabase/neon/issues/12190
- https://github.com/neondatabase/neon/issues/12191
- https://github.com/neondatabase/neon/issues/12192

## Summary of changes
- Implement `tenant_timeline_safekeeper_migrate` handler to drive the
migration
- Add possibility to specify number of safekeepers per timeline in tests
(`timeline_safekeeper_count`)
- Add `term` and `flush_lsn` to `TimelineMembershipSwitchResponse`
- Implement compare-and-swap (CAS) operation over timeline in DB for
updating membership configuration safely.
- Write simple test to verify that migration code works
2025-06-30 08:30:05 +00:00
Aleksandr Sarantsev
9bb4688c54 storcon: Remove testing feature from kick_secondary_downloads (#12383)
## Problem

Some of the design decisions in PR #12256 were influenced by the
requirements of consistency tests. These decisions introduced
intermediate logic that is no longer needed and should be cleaned up.

## Summary of Changes
- Remove the `feature("testing")` flag related to
`kick_secondary_download`.
- Set the default value of `kick_secondary_download` back to false,
reflecting the intended production behavior.

Co-authored-by: Aleksandr Sarantsev <aleksandr.sarantsev@databricks.com>
2025-06-30 05:41:05 +00:00
Dmitrii Kovalkov
47553dbaf9 neon_local: set timeline_safekeeper_count if we have less than 3 safekeepers (#12378)
## Problem
- Closes: https://github.com/neondatabase/neon/issues/12298

## Summary of changes
- Set `timeline_safekeeper_count` in `neon_local` if we have less than 3
safekeepers
- Remove `cfg!(feature = "testing")` code from
`safekeepers_for_new_timeline`
- Change `timeline_safekeeper_count` type to `usize`
2025-06-28 12:59:29 +00:00
Erik Grinaker
e50b914a8e compute_tools: support gRPC base backups in compute_ctl (#12244)
## Problem

`compute_ctl` should support gRPC base backups.

Requires #12111.
Requires #12243.
Touches #11926.

## Summary of changes

Support `grpc://` connstrings for `compute_ctl` base backups.
2025-06-27 16:39:00 +00:00
Christian Schwarz
e33e109403 fix(pageserver): buffered writer cancellation error handling (#12376)
## Problem

The problem has been well described in already-commited PR #11853.
tl;dr: BufferedWriter is sensitive to cancellation, which the previous
approach was not.

The write path was most affected (ingest & compaction), which was mostly
fixed in #11853:
it introduced `PutError` and mapped instances of `PutError` that were
due to cancellation of underlying buffered writer into
`CreateImageLayersError::Cancelled`.

However, there is a long tail of remaining errors that weren't caught by
#11853 that result in `CompactionError::Other`s, which we log with great
noise.

## Solution

The stack trace logging for CompactionError::Other added in #11853
allows us to chop away at that long tail using the following pattern:
- look at the stack trace
- from leaf up, identify the place where we incorrectly map from the
distinguished variant X indicating cancellation to an `anyhow::Error`
- follow that anyhow further up, ensuring it stays the same anyhow all
the way up in the `CompactionError::Other`
- since it stayed one anyhow chain all the way up, root_cause() will
yield us X
- so, in `log_compaction_error`, add an additional `downcast_ref` check
for X

This PR specifically adds checks for
- the flush task cancelling (FlushTaskError, BlobWriterError)
- opening of the layer writer (GateError)

That should cover all the reports in issues 
- https://github.com/neondatabase/cloud/issues/29434
- https://github.com/neondatabase/neon/issues/12162

## Refs
- follow-up to #11853
- fixup of / fixes https://github.com/neondatabase/neon/issues/11762
- fixes https://github.com/neondatabase/neon/issues/12162
- refs https://github.com/neondatabase/cloud/issues/29434
2025-06-27 15:26:00 +00:00
Folke Behrens
0ee15002fc proxy: Move client connection accept and handshake to pglb (#12380)
* This must be a no-op.
* Move proxy::task_main to pglb::task_main.
* Move client accept, TLS and handshake to pglb.
* Keep auth and wake in proxy.
2025-06-27 15:20:23 +00:00
Arpad Müller
4c7956fa56 Fix hang deleting offloaded timelines (#12366)
We don't have cancellation support for timeline deletions. In other
words, timeline deletion might still go on in an older generation while
we are attaching it in a newer generation already, because the
cancellation simply hasn't reached the deletion code.

This has caused us to hit a situation with offloaded timelines in which
the timeline was in an unrecoverable state: always returning an accepted
response, but never a 404 like it should be.

The detailed description can be found in
[here](https://github.com/neondatabase/cloud/issues/30406#issuecomment-3008667859)
(private repo link).

TLDR:

1. we ask to delete timeline on old pageserver/generation, starts
process in background
2. the storcon migrates the tenant to a different pageserver.
- during attach, the pageserver still finds an index part, so it adds it
to `offloaded_timelines`
4. the timeline deletion finishes, removing the index part in S3
5. there is a retry of the timeline deletion endpoint, sent to the new
pageserver location. it is bound to fail however:
- as the index part is gone, we print `Timeline already deleted in
remote storage`.
- the problem is that we then return an accepted response code, and not
a 404.
- this confuses the code calling us. it thinks the timeline is not
deleted, so keeps retrying.
- this state never gets recovered from until a reset/detach, because of
the `offloaded_timelines` entry staying there.

This is where this PR fixes things: if no index part can be found, we
can safely assume that the timeline is gone in S3 (it's the last thing
to be deleted), so we can remove it from `offloaded_timelines` and
trigger a reupload of the manifest. Subsequent retries will pick that
up.

Why not improve the cancellation support? It is a more disruptive code
change, that might have its own risks. So we don't do it for now.

Fixes https://github.com/neondatabase/cloud/issues/30406
2025-06-27 15:14:55 +00:00
Heikki Linnakangas
5a82182c48 impr(ci): Refactor postgres Makefile targets to a separate makefile (#12363)
Mainly for general readability. Some notable changes:

- Postgres can be built without the rest of the repository, and in
particular without any of the Rust bits. Some CI scripts took advantage
of that, so let's make that more explicit by separating those parts.
Also add an explicit comment about that in the new postgres.mk file.

- Add a new PG_INSTALL_CACHED variable. If it's set, `make all` and
other top-Makefile targets skip checking if Postgres is up-to-date. This
is also to be used in CI scripts that build and cache Postgres as
separate steps. (It is currently only used in the macos walproposer-lib
rule, but stay tuned for more.)

- Introduce a POSTGRES_VERSIONS variable that lists all supported
PostgreSQL versions. Refactor a few Makefile rules to use that.
2025-06-27 14:49:52 +00:00
Arpad Müller
37e181af8a Update rust to 1.88.0 (#12364)
We keep the practice of keeping the compiler up to date, pointing to the
latest release. This is done by many other projects in the Rust
ecosystem as well.

[Announcement blog
post](https://blog.rust-lang.org/2025/06/26/Rust-1.88.0/)

Prior update was in https://github.com/neondatabase/neon/pull/11938
2025-06-27 13:51:59 +00:00
Peter Bendel
6f4198c78a treat strategy flag test_maintenance as boolean data type (#12373)
## Problem

In large oltp test run
https://github.com/neondatabase/neon/actions/runs/15905488707/job/44859116742
we see that the `Benchmark database maintenance` step is skipped in all
3 strategy variants, however it should be executed in two.

This is due to treating the `test_maintenance` boolean type in the
strategy in the condition of the `Benchmark database maintenance` step

## Summary of changes
Use a boolean condition instead of a string comparison

## Test run from this pull request branch

https://github.com/neondatabase/neon/actions/runs/15923605412
2025-06-27 13:49:26 +00:00
Vlad Lazar
cc1664ef93 pageserver: allow flush task cancelled error in sharding autosplit test (#12374)
## Problem

Test is failing due to compaction shutdown noise (see
https://github.com/neondatabase/neon/issues/12162).

## Summary of changes

Allow list the noise.
2025-06-27 13:13:11 +00:00
Vlad Lazar
ebb6e26a64 pageserver: handle multiple attached children in shard resolution (#12336)
## Problem

When resolving a shard during a split we might have multiple attached
shards with the old shard count (i.e. not all of them are marked in
progress and ignored). Hence, we can compute the desired shard number
based on the old shard count and misroute the request.

## Summary of Changes

Recompute the desired shard every time the shard count changes during
the iteration
2025-06-27 12:46:18 +00:00
Mikhail
ebc12a388c fix: endpoint_storage_addr as String (#12359)
It's not a SocketAddr as we use k8s DNS
https://github.com/neondatabase/cloud/issues/19011
2025-06-27 11:06:27 +00:00
Conrad Ludgate
abc1efd5a6 [proxy] fix connect_to_compute retry handling (#12351)
# Problem

In #12335 I moved the `authenticate` method outside of the
`connect_to_compute` loop. This triggered [e2e tests to become
flaky](https://github.com/neondatabase/cloud/pull/30533). This
highlighted an edge case we forgot to consider with that change.

When we connect to compute, the compute IP might be cached. This cache
hit might however be stale. Because we can't validate the IP is
associated with a specific compute-id☨, we will succeed the
connect_to_compute operation and fail when it comes to password
authentication☨☨. Before the change, we were invalidating the cache and
triggering wake_compute if the authentication failed.

Additionally, I noticed some faulty logic I introduced 1 year ago
https://github.com/neondatabase/neon/pull/8141/files#diff-5491e3afe62d8c5c77178149c665603b29d88d3ec2e47fc1b3bb119a0a970afaL145-R147

☨ We can when we roll out TLS, as the certificate common name includes
the compute-id.

☨☨ Technically password authentication could pass for the wrong compute,
but I think this would only happen in the very very rare event that the
IP got reused **and** the compute's endpoint happened to be a
branch/replica.

# Solution

1. Fix the broken logic
2. Simplify cache invalidation (I don't know why it was so convoluted)
3. Add a loop around connect_to_compute + authenticate to re-introduce
the wake_compute invalidation we accidentally removed.

I went with this approach to try and avoid interfering with
https://github.com/neondatabase/neon/compare/main...cloneable/proxy-pglb-connect-compute-split.
The changes made in commit 3 will move into `handle_client_request` I
suspect,
2025-06-27 10:36:27 +00:00
Dmitrii Kovalkov
6fa1562b57 pageserver: increase default max_size_entries limit for basebackup cache (#12343)
## Problem
Some pageservers hit `max_size_entries` limit in staging with only ~25
MiB storage used by basebackup cache. The limit is too strict. It should
be safe to relax it.

- Part of https://github.com/neondatabase/cloud/issues/29353

## Summary of changes
- Increase the default `max_size_entries` from 1000 to 10000
2025-06-27 09:18:18 +00:00
Heikki Linnakangas
10afac87e7 impr(ci): Remove unnecessary 'make postgres-headers' build step (#12354)
The 'make postgres' step includes installation of the headers, no need
to do that separately.
2025-06-26 16:45:34 +00:00
Vlad Lazar
72b3c9cd11 pageserver: fix wal receiver hang on remote client shutdown (#12348)
## Problem

Druing shard splits we shut down the remote client early and allow the
parent shard to keep ingesting data. While ingesting data, the wal
receiver task may wait for the current flush to complete in order to
apply backpressure. Notifications are delivered via
`Timeline::layer_flush_done_tx`.

When the remote client was being shut down the flush loop exited
whithout delivering a notification. This left
`Timeline::wait_flush_completion` hanging indefinitely which blocked the
shutdown of the wal receiver task, and, hence, the shard split.

## Summary of Changes

Deliver a final notification when the flush loop is shutting down
without the timeline cancel cancellation token having fired. I tried
writing a test for this, but got stuck in failpoint hell and decided
it's not worth it.

`test_sharding_autosplit`, which reproduces this reliably in CI, passed
with the proposed fix in
https://github.com/neondatabase/neon/pull/12304.

Closes https://github.com/neondatabase/neon/issues/12060
2025-06-26 16:35:34 +00:00
Arpad Müller
232f2447d4 Support pull_timeline of timelines without writes (#12028)
Make the safekeeper `pull_timeline` endpoint support timelines that
haven't had any writes yet. In the storcon managed sk timelines world,
if a safekeeper goes down temporarily, the storcon will schedule a
`pull_timeline` call. There is no guarantee however that by when the
safekeeper is online again, there have been writes to the timeline yet.

The `snapshot` endpoint gives an error if the timeline hasn't had
writes, so we avoid calling it if `timeline_start_lsn` indicates a
freshly created timeline.

Fixes #11422
Part of #11670
2025-06-26 16:29:03 +00:00
Erik Grinaker
a2d2108e6a pageserver: use base backup cache with gRPC (#12352)
## Problem

gRPC base backups do not use the base backup cache.

Touches https://github.com/neondatabase/neon/issues/11728.

## Summary of changes

Integrate gRPC base backups with the base backup cache.

Also fixes a bug where the base backup cache did not differentiate
between primary/replica base backups (at least I think that's a bug?).
2025-06-26 15:52:15 +00:00
Alex Chi Z.
33c0d5e2f4 fix(pageserver): make posthog config parsing more robust (#12356)
## Problem

In our infra config, we have to split server_api_key and other fields in
two files: the former one in the sops file, and the latter one in the
normal config. It creates the situation that we might misconfigure some
regions that it only has part of the fields available, causing
storcon/pageserver refuse to start.

## Summary of changes

Allow PostHog config to have part of the fields available. Parse it
later.

Signed-off-by: Alex Chi Z <chi@neon.tech>
2025-06-26 15:49:08 +00:00
Dmitrii Kovalkov
605fb04f89 pageserver: use bounded sender for basebackup cache (#12342)
## Problem
Basebackup cache now uses unbounded channel for prepare requests. In
theory it can grow large if the cache is hung and does not process the
requests.

- Part of https://github.com/neondatabase/cloud/issues/29353

## Summary of changes
- Replace an unbounded channel with a bounded one, the size is
configurable.
- Add `pageserver_basebackup_cache_prepare_queue_size` to observe the
size of the queue.
- Refactor a bit to move all metrics logic to `basebackup_cache.rs`
2025-06-26 13:26:24 +00:00
Conrad Ludgate
fd1e8ec257 [proxy] review and cleanup CLI args (#12167)
I was looking at how we could expose our proxy config as toml again, and
as I was writing out the schema format, I noticed some cruft in our CLI
args that no longer seem to be in use.

The redis change is the most complex, but I am pretty sure it's sound.
Since https://github.com/neondatabase/cloud/pull/15613 cplane longer
publishes to the global redis instance.
2025-06-26 11:25:41 +00:00
Konstantin Knizhnik
be23eae3b6 Mark pages as avaiable in LFC only after generation check (#12350)
## Problem

If LFC generation is changed then `lfc_readv_select` will return -1 but
pages are still marked as available in bitmap.

## Summary of changes

Update bitmap after generation check.

Co-authored-by: Kosntantin Knizhnik <konstantin.knizhnik@databricks.com>
2025-06-26 07:06:27 +00:00
Alex Chi Z.
6f70885e11 fix(pageserver): allow refresh_interval to be empty (#12349)
## Problem

Fix for https://github.com/neondatabase/neon/pull/12324

## Summary of changes

Need `serde(default)` to allow this field not present in the config,
otherwise there will be a config deserialization error.

---------

Signed-off-by: Alex Chi Z <chi@neon.tech>
2025-06-25 22:15:03 +00:00
Erik Grinaker
f755979102 pageserver: payload compression for gRPC base backups (#12346)
## Problem

gRPC base backups use gRPC compression. However, this has two problems:

* Base backup caching will cache compressed base backups (making gRPC
compression pointless).
* Tonic does not support varying the compression level, and zstd default
level is 10% slower than gzip fastest level.

Touches https://github.com/neondatabase/neon/issues/11728.
Touches https://github.com/neondatabase/cloud/issues/29353.

## Summary of changes

This patch adds a gRPC parameter `BaseBackupRequest::compression`
specifying the compression algorithm. It also moves compression into
`send_basebackup_tarball` to reduce code duplication.

A follow-up PR will integrate the base backup cache with gRPC.
2025-06-25 18:16:23 +00:00
Matthias van de Meent
1d49eefbbb RFC: Endpoint Persistent Unlogged Files Storage (#9661)
## Summary
A design for a storage system that allows storage of files required to
make
Neon's Endpoints have a better experience at or after a reboot.

## Motivation
Several systems inside PostgreSQL (and Neon) need some persistent
storage for
optimal workings across reboots and restarts, but still work without.
Examples are the cumulative statistics file in `pg_stat/global.stat`,
`pg_stat_statements`' `pg_stat/pg_stat_statements.stat`, and
`pg_prewarm`'s
`autoprewarm.blocks`. We need a storage system that can store and manage
these files for each Endpoint.

[GH rendered
file](https://github.com/neondatabase/neon/blob/MMeent/rfc-unlogged-file/docs/rfcs/040-Endpoint-Persistent-Unlogged-Files-Storage.md)

Part of https://github.com/neondatabase/cloud/issues/24225
2025-06-25 16:25:57 +00:00
Alex Chi Z.
6c77638ea1 feat(storcon): retrieve feature flag and pass to pageservers (#12324)
## Problem

part of https://github.com/neondatabase/neon/issues/11813

## Summary of changes

It costs $$$ to directly retrieve the feature flags from the pageserver.
Therefore, this patch adds new APIs to retrieve the spec from the
storcon and updates it via pageserver.

* Storcon retrieves the feature flag and send it to the pageservers.
* If the feature flag gets updated outside of the normal refresh loop of
the pageserver, pageserver won't fetch the flags on its own as long as
the last updated time <= refresh_period.

Signed-off-by: Alex Chi Z <chi@neon.tech>
2025-06-25 14:58:18 +00:00
Conrad Ludgate
517a3d0d86 [proxy]: BatchQueue::call is not cancel safe - make it directly cancellation aware (#12345)
## Problem

https://github.com/neondatabase/cloud/issues/30539

If the current leader cancels the `call` function, then it has removed
the jobs from the queue, but will never finish sending the responses.
Because of this, it is not cancellation safe.

## Summary of changes

Document these functions as not cancellation safe. Move cancellation of
the queued jobs into the queue itself.

## Alternatives considered

1. We could spawn the task that runs the batch, since that won't get
cancelled.
* This requires `fn call(self: Arc<Self>)` or `fn call(&'static self)`.
2. We could add another scopeguard and return the requests back to the
queue.
* This requires that requests are always retry safe, and also requires
requests to be `Clone`.
2025-06-25 14:19:20 +00:00
Conrad Ludgate
27ca1e21be [console_redirect_proxy]: fix channel binding (#12238)
## Problem

While working more on TLS to compute, I realised that Console Redirect
-> pg-sni-router -> compute would break if channel binding was set to
prefer. This is because the channel binding data would differ between
Console Redirect -> pg-sni-router vs pg-sni-router -> compute.

I also noticed that I actually disabled channel binding in #12145, since
`connect_raw` would think that the connection didn't support TLS.

## Summary of changes

Make sure we specify the channel binding.
Make sure that `connect_raw` can see if we have TLS support.
2025-06-25 13:41:30 +00:00
Arpad Müller
1dc01c9bed Support cancellations of timelines with hanging ondemand downloads (#12330)
In `test_layer_download_cancelled_by_config_location`, we simulate hung
downloads via the `before-downloading-layer-stream-pausable` failpoint.
Then, we cancel a timeline via the `location_config` endpoint.

With the new default as of
https://github.com/neondatabase/neon/pull/11712, we would be creating
the timeline on safekeepers regardless if there have been writes or not,
and it turns out the test relied on the timeline not existing on
safekeepers, due to a cancellation bug:

* as established before, the test makes the read path hang
* the timeline cancellation function first cancels the walreceiver, and
only then cancels the timeline's token
* `WalIngest::new` is requesting a checkpoint, which hits the read path
* at cancellation time, we'd be hanging inside the read, not seeing the
cancellation of the walreceiver
* the test would time out due to the hang

This is probably also reproducible in the wild when there is S3
unavailabilies or bottlenecks. So we thought that it's worthwhile to fix
the hang issue. The approach chosen in the end involves the
`tokio::select` macro.

In PR 11712, we originally punted on the test due to the hang and opted
it out from the new default, but now we can use the new default.

Part of https://github.com/neondatabase/neon/issues/12299
2025-06-25 13:40:38 +00:00
Heikki Linnakangas
7c4c36f5ac Remove unnecessary separate installation of libpq (#12287)
`make install` compiles and installs libpq. Remove redundant separate
step to compile and install it.
2025-06-25 10:47:56 +00:00
Tristan Partin
a2d623696c Update pgaudit to latest versions (#12328)
These updates contain some bug fixes and are completely backwards
compatible with what we currently support in Neon.

Link: https://github.com/pgaudit/pgaudit/compare/1.6.2...1.6.3
Link: https://github.com/pgaudit/pgaudit/compare/1.7.0...1.7.1
Link: https://github.com/pgaudit/pgaudit/compare/16.0...16.1
Link: https://github.com/pgaudit/pgaudit/compare/17.0...17.1
Signed-off-by: Tristan Partin <tristan.partin@databricks.com>

Signed-off-by: Tristan Partin <tristan.partin@databricks.com>
2025-06-25 09:03:02 +00:00
Tristan Partin
aa75722010 Set pgaudit.log=none for monitoring connections (#12137)
pgaudit can spam logs due to all the monitoring that we do. Logs from
these connections are not necessary for HIPPA compliance, so we can stop
logging from those connections.

Part-of: https://github.com/neondatabase/cloud/issues/29574

Signed-off-by: Tristan Partin <tristan@neon.tech>
2025-06-24 17:42:23 +00:00
Matthias van de Meent
6c6de6382a Use enum-typed PG versions (#12317)
This makes it possible for the compiler to validate that a match block
matched all PostgreSQL versions we support.

## Problem
We did not have a complete picture about which places we had to test
against PG versions, and what format these versions were: The full PG
version ID format (Major/minor/bugfix `MMmmbb`) as transfered in
protocol messages, or only the Major release version (`MM`). This meant
type confusion was rampant.

With this change, it becomes easier to develop new version-dependent
features, by making type and niche confusion impossible.

## Summary of changes
Every use of `pg_version` is now typed as either `PgVersionId` (u32,
valued in decimal `MMmmbb`) or PgMajorVersion (an enum, with a value for
every major version we support, serialized and stored like a u32 with
the value of that major version)

---------

Co-authored-by: Arpad Müller <arpad-m@users.noreply.github.com>
2025-06-24 17:25:31 +00:00
Dmitry Savelev
158d84ea30 Switch the billing metrics storage format to ndjson. (#12338)
## Problem

The billing team wants to change the billing events pipeline and use a
common events format in S3 buckets across different event producers.

## Summary of changes

Change the events storage format for billing events from JSON to NDJSON.

Resolves: https://github.com/neondatabase/cloud/issues/29994
2025-06-24 15:36:36 +00:00
Conrad Ludgate
4dd9ca7b04 [proxy]: authenticate to compute after connect_to_compute (#12335)
## Problem

PGLB will do the connect_to_compute logic, neonkeeper will do the
session establishment logic. We should split it.

## Summary of changes

Moves postgres authentication to compute to a separate routine that
happens after connect_to_compute.
2025-06-24 14:15:36 +00:00
170 changed files with 5098 additions and 2100 deletions

View File

@@ -4,6 +4,7 @@
!Cargo.lock
!Cargo.toml
!Makefile
!postgres.mk
!rust-toolchain.toml
!scripts/ninstall.sh
!docker-compose/run-tests.sh

View File

@@ -391,10 +391,6 @@ jobs:
PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
USE_LFC: ${{ matrix.lfc_state == 'with-lfc' && 'true' || 'false' }}
# Temporary disable this step until we figure out why it's so flaky
# Ref https://github.com/neondatabase/neon/issues/4540
- name: Merge and upload coverage data
if: |
false &&
inputs.build-type == 'debug' && matrix.pg_version == 'v16'
if: inputs.build-type == 'debug'
uses: ./.github/actions/save-coverage-data

View File

@@ -94,11 +94,6 @@ jobs:
run: |
make "neon-pg-ext-${{ matrix.postgres-version }}" -j$(sysctl -n hw.ncpu)
- name: Get postgres headers ${{ matrix.postgres-version }}
if: steps.cache_pg.outputs.cache-hit != 'true'
run: |
make postgres-headers-${{ matrix.postgres-version }} -j$(sysctl -n hw.ncpu)
- name: Upload "pg_install/${{ matrix.postgres-version }}" artifact
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
with:
@@ -140,6 +135,12 @@ jobs:
name: pg_install--v17
path: pg_install/v17
# `actions/download-artifact` doesn't preserve permissions:
# https://github.com/actions/download-artifact?tab=readme-ov-file#permission-loss
- name: Make pg_install/v*/bin/* executable
run: |
chmod +x pg_install/v*/bin/*
- name: Cache walproposer-lib
id: cache_walproposer_lib
uses: actions/cache@5a3ec84eff668545956fd18022155c47e93e2684 # v4.2.3
@@ -167,7 +168,7 @@ jobs:
- name: Build walproposer-lib (only for v17)
if: steps.cache_walproposer_lib.outputs.cache-hit != 'true'
run:
make walproposer-lib -j$(sysctl -n hw.ncpu)
make walproposer-lib -j$(sysctl -n hw.ncpu) PG_INSTALL_CACHED=1
- name: Upload "build/walproposer-lib" artifact
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2

View File

@@ -69,7 +69,7 @@ jobs:
submodules: true
- name: Check for file changes
uses: step-security/paths-filter@v3
uses: dorny/paths-filter@de90cc6fb38fc0963ad72b210f1f284cd68cea36 # v3.0.2
id: files-changed
with:
token: ${{ secrets.GITHUB_TOKEN }}
@@ -484,21 +484,6 @@ jobs:
- name: Merge coverage data
run: scripts/coverage "--profraw-prefix=$GITHUB_JOB" --dir=/tmp/coverage merge
- name: Build coverage report
env:
COMMIT_URL: ${{ github.server_url }}/${{ github.repository }}/commit/${{ github.event.pull_request.head.sha || github.sha }}
run: |
scripts/coverage --dir=/tmp/coverage \
report \
--input-objects=/tmp/coverage/binaries.list \
--commit-url=${COMMIT_URL} \
--format=github
scripts/coverage --dir=/tmp/coverage \
report \
--input-objects=/tmp/coverage/binaries.list \
--format=lcov
- name: Build coverage report NEW
id: upload-coverage-report-new
env:
@@ -511,6 +496,13 @@ jobs:
CURRENT="${COMMIT_SHA}"
BASELINE="$(git merge-base $BASE_SHA $CURRENT)"
mkdir /tmp/coverage/report
scripts/coverage --dir=/tmp/coverage \
report \
--input-objects=/tmp/coverage/binaries.list \
--format=lcov
cp /tmp/coverage/report/lcov.info ./${CURRENT}.info
GENHTML_ARGS="--ignore-errors path,unmapped,empty --synthesize-missing --demangle-cpp rustfilt --output-directory lcov-html ${CURRENT}.info"

View File

@@ -153,7 +153,7 @@ jobs:
PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
- name: Benchmark database maintenance
if: ${{ matrix.test_maintenance == 'true' }}
if: ${{ matrix.test_maintenance }}
uses: ./.github/actions/run-python-test-set
with:
build_type: ${{ env.BUILD_TYPE }}

View File

@@ -53,7 +53,7 @@ jobs:
submodules: true
- name: Check for Postgres changes
uses: step-security/paths-filter@v3
uses: dorny/paths-filter@1441771bbfdd59dcd748680ee64ebd8faab1a242 #v3
id: files_changed
with:
token: ${{ github.token }}

View File

@@ -34,7 +34,7 @@ jobs:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
- uses: step-security/changed-files@3dbe17c78367e7d60f00d78ae6781a35be47b4a1 # v45.0.1
- uses: tj-actions/changed-files@ed68ef82c095e0d48ec87eccea555d944a631a4c # v46.0.5
id: python-src
with:
files: |
@@ -45,7 +45,7 @@ jobs:
poetry.lock
pyproject.toml
- uses: step-security/changed-files@3dbe17c78367e7d60f00d78ae6781a35be47b4a1 # v45.0.1
- uses: tj-actions/changed-files@ed68ef82c095e0d48ec87eccea555d944a631a4c # v46.0.5
id: rust-src
with:
files: |

View File

@@ -60,22 +60,23 @@ jobs:
} >> "$GITHUB_ENV"
- name: Run proxy-bench
run: ./${PROXY_BENCH_PATH}/run.sh
run: ${PROXY_BENCH_PATH}/run.sh
- name: Ingest Bench Results # neon repo script
if: success()
if: always()
run: |
mkdir -p $TEST_OUTPUT
python $NEON_DIR/scripts/proxy_bench_results_ingest.py --out $TEST_OUTPUT
- name: Push Metrics to Proxy perf database
if: success()
if: always()
env:
PERF_TEST_RESULT_CONNSTR: "${{ secrets.PROXY_TEST_RESULT_CONNSTR }}"
REPORT_FROM: $TEST_OUTPUT
run: $NEON_DIR/scripts/generate_and_push_perf_report.sh
- name: Docker cleanup
if: always()
run: docker compose down
- name: Notify Failure

52
Cargo.lock generated
View File

@@ -1279,6 +1279,7 @@ dependencies = [
"remote_storage",
"serde",
"serde_json",
"url",
"utils",
]
@@ -1316,8 +1317,10 @@ dependencies = [
"opentelemetry",
"opentelemetry_sdk",
"p256 0.13.2",
"pageserver_page_api",
"postgres",
"postgres_initdb",
"postgres_versioninfo",
"regex",
"remote_storage",
"reqwest",
@@ -1334,6 +1337,7 @@ dependencies = [
"tokio-postgres",
"tokio-stream",
"tokio-util",
"tonic 0.13.1",
"tower 0.5.2",
"tower-http",
"tower-otel",
@@ -4406,6 +4410,8 @@ dependencies = [
"once_cell",
"postgres_backend",
"postgres_ffi_types",
"postgres_versioninfo",
"posthog_client_lite",
"rand 0.8.5",
"remote_storage",
"reqwest",
@@ -4416,6 +4422,7 @@ dependencies = [
"strum",
"strum_macros",
"thiserror 1.0.69",
"tracing",
"tracing-utils",
"utils",
]
@@ -4429,6 +4436,7 @@ dependencies = [
"futures",
"http-utils",
"pageserver_api",
"postgres_versioninfo",
"reqwest",
"serde",
"thiserror 1.0.69",
@@ -4471,12 +4479,14 @@ dependencies = [
"bytes",
"futures",
"pageserver_api",
"postgres_ffi",
"postgres_ffi_types",
"prost 0.13.5",
"prost-types 0.13.5",
"strum",
"strum_macros",
"thiserror 1.0.69",
"tokio",
"tokio-util",
"tonic 0.13.1",
"tonic-build",
"utils",
@@ -4897,6 +4907,7 @@ dependencies = [
"once_cell",
"postgres",
"postgres_ffi_types",
"postgres_versioninfo",
"pprof",
"regex",
"serde",
@@ -4919,11 +4930,23 @@ version = "0.1.0"
dependencies = [
"anyhow",
"camino",
"postgres_versioninfo",
"thiserror 1.0.69",
"tokio",
"workspace_hack",
]
[[package]]
name = "postgres_versioninfo"
version = "0.1.0"
dependencies = [
"anyhow",
"serde",
"serde_repr",
"thiserror 1.0.69",
"workspace_hack",
]
[[package]]
name = "posthog_client_lite"
version = "0.1.0"
@@ -5136,7 +5159,7 @@ dependencies = [
"petgraph",
"prettyplease",
"prost 0.13.5",
"prost-types 0.13.3",
"prost-types 0.13.5",
"regex",
"syn 2.0.100",
"tempfile",
@@ -5179,9 +5202,9 @@ dependencies = [
[[package]]
name = "prost-types"
version = "0.13.3"
version = "0.13.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4759aa0d3a6232fb8dbdb97b61de2c20047c68aca932c7ed76da9d788508d670"
checksum = "52c2c1bf36ddb1a1c396b3601a3cec27c2462e45f07c386894ec3ccf5332bd16"
dependencies = [
"prost 0.13.5",
]
@@ -6115,6 +6138,7 @@ dependencies = [
"postgres-protocol",
"postgres_backend",
"postgres_ffi",
"postgres_versioninfo",
"pprof",
"pq_proto",
"rand 0.8.5",
@@ -6159,6 +6183,7 @@ dependencies = [
"const_format",
"pageserver_api",
"postgres_ffi",
"postgres_versioninfo",
"pq_proto",
"serde",
"serde_json",
@@ -6481,6 +6506,17 @@ dependencies = [
"thiserror 1.0.69",
]
[[package]]
name = "serde_repr"
version = "0.1.20"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "175ee3e80ae9982737ca543e96133087cbd9a485eecc3bc4de9c1a37b47ea59c"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.100",
]
[[package]]
name = "serde_spanned"
version = "0.6.6"
@@ -6775,6 +6811,7 @@ dependencies = [
"chrono",
"clap",
"clashmap",
"compute_api",
"control_plane",
"cron",
"diesel",
@@ -6786,6 +6823,7 @@ dependencies = [
"hex",
"http-utils",
"humantime",
"humantime-serde",
"hyper 0.14.30",
"itertools 0.10.5",
"json-structural-diff",
@@ -6796,6 +6834,7 @@ dependencies = [
"pageserver_api",
"pageserver_client",
"postgres_connection",
"posthog_client_lite",
"rand 0.8.5",
"regex",
"reqwest",
@@ -7606,7 +7645,7 @@ dependencies = [
"prettyplease",
"proc-macro2",
"prost-build 0.13.3",
"prost-types 0.13.3",
"prost-types 0.13.5",
"quote",
"syn 2.0.100",
]
@@ -7618,7 +7657,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f9687bd5bfeafebdded2356950f278bba8226f0b32109537c4253406e09aafe1"
dependencies = [
"prost 0.13.5",
"prost-types 0.13.3",
"prost-types 0.13.5",
"tokio",
"tokio-stream",
"tonic 0.13.1",
@@ -8647,7 +8686,6 @@ dependencies = [
"num-iter",
"num-rational",
"num-traits",
"once_cell",
"p256 0.13.2",
"parquet",
"prettyplease",

View File

@@ -23,6 +23,7 @@ members = [
"libs/pageserver_api",
"libs/postgres_ffi",
"libs/postgres_ffi_types",
"libs/postgres_versioninfo",
"libs/safekeeper_api",
"libs/desim",
"libs/neon-shmem",
@@ -151,6 +152,7 @@ pprof = { version = "0.14", features = ["criterion", "flamegraph", "frame-pointe
procfs = "0.16"
prometheus = {version = "0.13", default-features=false, features = ["process"]} # removes protobuf dependency
prost = "0.13.5"
prost-types = "0.13.5"
rand = "0.8"
redis = { version = "0.29.2", features = ["tokio-rustls-comp", "keep-alive"] }
regex = "1.10.2"
@@ -174,6 +176,7 @@ serde_json = "1"
serde_path_to_error = "0.1"
serde_with = { version = "3", features = [ "base64" ] }
serde_assert = "0.5.0"
serde_repr = "0.1.20"
sha2 = "0.10.2"
signal-hook = "0.3"
smallvec = "1.11"
@@ -197,7 +200,7 @@ tokio-postgres-rustls = "0.12.0"
tokio-rustls = { version = "0.26.0", default-features = false, features = ["tls12", "ring"]}
tokio-stream = "0.1"
tokio-tar = "0.3"
tokio-util = { version = "0.7.10", features = ["io", "rt"] }
tokio-util = { version = "0.7.10", features = ["io", "io-util", "rt"] }
toml = "0.8"
toml_edit = "0.22"
tonic = { version = "0.13.1", default-features = false, features = ["channel", "codegen", "gzip", "prost", "router", "server", "tls-ring", "tls-native-roots", "zstd"] }
@@ -261,6 +264,7 @@ postgres_backend = { version = "0.1", path = "./libs/postgres_backend/" }
postgres_connection = { version = "0.1", path = "./libs/postgres_connection/" }
postgres_ffi = { version = "0.1", path = "./libs/postgres_ffi/" }
postgres_ffi_types = { version = "0.1", path = "./libs/postgres_ffi_types/" }
postgres_versioninfo = { version = "0.1", path = "./libs/postgres_versioninfo/" }
postgres_initdb = { path = "./libs/postgres_initdb" }
posthog_client_lite = { version = "0.1", path = "./libs/posthog_client_lite" }
pq_proto = { version = "0.1", path = "./libs/pq_proto/" }

View File

@@ -40,6 +40,7 @@ COPY --chown=nonroot vendor/postgres-v16 vendor/postgres-v16
COPY --chown=nonroot vendor/postgres-v17 vendor/postgres-v17
COPY --chown=nonroot pgxn pgxn
COPY --chown=nonroot Makefile Makefile
COPY --chown=nonroot postgres.mk postgres.mk
COPY --chown=nonroot scripts/ninstall.sh scripts/ninstall.sh
ENV BUILD_TYPE=release

131
Makefile
View File

@@ -4,11 +4,14 @@ ROOT_PROJECT_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
# managers.
POSTGRES_INSTALL_DIR ?= $(ROOT_PROJECT_DIR)/pg_install/
# Supported PostgreSQL versions
POSTGRES_VERSIONS = v17 v16 v15 v14
# CARGO_BUILD_FLAGS: Extra flags to pass to `cargo build`. `--locked`
# and `--features testing` are popular examples.
#
# CARGO_PROFILE: You can also set to override the cargo profile to
# use. By default, it is derived from BUILD_TYPE.
# CARGO_PROFILE: Set to override the cargo profile to use. By default,
# it is derived from BUILD_TYPE.
# All intermediate build artifacts are stored here.
BUILD_DIR := build
@@ -95,93 +98,24 @@ CACHEDIR_TAG_CONTENTS := "Signature: 8a477f597d28d172789f06886806bc55"
# Top level Makefile to build Neon and PostgreSQL
#
.PHONY: all
all: neon postgres neon-pg-ext
all: neon postgres-install neon-pg-ext
### Neon Rust bits
#
# The 'postgres_ffi' depends on the Postgres headers.
.PHONY: neon
neon: postgres-headers walproposer-lib cargo-target-dir
neon: postgres-headers-install walproposer-lib cargo-target-dir
+@echo "Compiling Neon"
$(CARGO_CMD_PREFIX) cargo build $(CARGO_BUILD_FLAGS) $(CARGO_PROFILE)
.PHONY: cargo-target-dir
cargo-target-dir:
# https://github.com/rust-lang/cargo/issues/14281
mkdir -p target
test -e target/CACHEDIR.TAG || echo "$(CACHEDIR_TAG_CONTENTS)" > target/CACHEDIR.TAG
### PostgreSQL parts
# Some rules are duplicated for Postgres v14 and 15. We may want to refactor
# to avoid the duplication in the future, but it's tolerable for now.
#
$(BUILD_DIR)/%/config.status:
mkdir -p $(BUILD_DIR)
test -e $(BUILD_DIR)/CACHEDIR.TAG || echo "$(CACHEDIR_TAG_CONTENTS)" > $(BUILD_DIR)/CACHEDIR.TAG
+@echo "Configuring Postgres $* build"
@test -s $(ROOT_PROJECT_DIR)/vendor/postgres-$*/configure || { \
echo "\nPostgres submodule not found in $(ROOT_PROJECT_DIR)/vendor/postgres-$*/, execute "; \
echo "'git submodule update --init --recursive --depth 2 --progress .' in project root.\n"; \
exit 1; }
mkdir -p $(BUILD_DIR)/$*
VERSION=$*; \
EXTRA_VERSION=$$(cd $(ROOT_PROJECT_DIR)/vendor/postgres-$$VERSION && git rev-parse HEAD); \
(cd $(BUILD_DIR)/$$VERSION && \
env PATH="$(EXTRA_PATH_OVERRIDES):$$PATH" $(ROOT_PROJECT_DIR)/vendor/postgres-$$VERSION/configure \
CFLAGS='$(PG_CFLAGS)' LDFLAGS='$(PG_LDFLAGS)' \
$(PG_CONFIGURE_OPTS) --with-extra-version=" ($$EXTRA_VERSION)" \
--prefix=$(abspath $(POSTGRES_INSTALL_DIR))/$$VERSION > configure.log)
# nicer alias to run 'configure'
# Note: I've been unable to use templates for this part of our configuration.
# I'm not sure why it wouldn't work, but this is the only place (apart from
# the "build-all-versions" entry points) where direct mention of PostgreSQL
# versions is used.
.PHONY: postgres-configure-v17
postgres-configure-v17: $(BUILD_DIR)/v17/config.status
.PHONY: postgres-configure-v16
postgres-configure-v16: $(BUILD_DIR)/v16/config.status
.PHONY: postgres-configure-v15
postgres-configure-v15: $(BUILD_DIR)/v15/config.status
.PHONY: postgres-configure-v14
postgres-configure-v14: $(BUILD_DIR)/v14/config.status
# Install the PostgreSQL header files into $(POSTGRES_INSTALL_DIR)/<version>/include
.PHONY: postgres-headers-%
postgres-headers-%: postgres-configure-%
+@echo "Installing PostgreSQL $* headers"
$(MAKE) -C $(BUILD_DIR)/$*/src/include MAKELEVEL=0 install
# Compile and install PostgreSQL
.PHONY: postgres-%
postgres-%: postgres-configure-% \
postgres-headers-% # to prevent `make install` conflicts with neon's `postgres-headers`
+@echo "Compiling PostgreSQL $*"
$(MAKE) -C $(BUILD_DIR)/$* MAKELEVEL=0 install
+@echo "Compiling libpq $*"
$(MAKE) -C $(BUILD_DIR)/$*/src/interfaces/libpq install
+@echo "Compiling pg_prewarm $*"
$(MAKE) -C $(BUILD_DIR)/$*/contrib/pg_prewarm install
+@echo "Compiling pg_buffercache $*"
$(MAKE) -C $(BUILD_DIR)/$*/contrib/pg_buffercache install
+@echo "Compiling pg_visibility $*"
$(MAKE) -C $(BUILD_DIR)/$*/contrib/pg_visibility install
+@echo "Compiling pageinspect $*"
$(MAKE) -C $(BUILD_DIR)/$*/contrib/pageinspect install
+@echo "Compiling pg_trgm $*"
$(MAKE) -C $(BUILD_DIR)/$*/contrib/pg_trgm install
+@echo "Compiling amcheck $*"
$(MAKE) -C $(BUILD_DIR)/$*/contrib/amcheck install
+@echo "Compiling test_decoding $*"
$(MAKE) -C $(BUILD_DIR)/$*/contrib/test_decoding install
.PHONY: postgres-check-%
postgres-check-%: postgres-%
$(MAKE) -C $(BUILD_DIR)/$* MAKELEVEL=0 check
.PHONY: neon-pg-ext-%
neon-pg-ext-%: postgres-%
neon-pg-ext-%: postgres-install-%
+@echo "Compiling neon-specific Postgres extensions for $*"
mkdir -p $(BUILD_DIR)/pgxn-$*
$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config COPT='$(COPT)' \
@@ -220,39 +154,14 @@ ifeq ($(UNAME_S),Linux)
pg_crc32c.o
endif
# Shorthand to call neon-pg-ext-% target for all Postgres versions
.PHONY: neon-pg-ext
neon-pg-ext: \
neon-pg-ext-v14 \
neon-pg-ext-v15 \
neon-pg-ext-v16 \
neon-pg-ext-v17
# shorthand to build all Postgres versions
.PHONY: postgres
postgres: \
postgres-v14 \
postgres-v15 \
postgres-v16 \
postgres-v17
.PHONY: postgres-headers
postgres-headers: \
postgres-headers-v14 \
postgres-headers-v15 \
postgres-headers-v16 \
postgres-headers-v17
.PHONY: postgres-check
postgres-check: \
postgres-check-v14 \
postgres-check-v15 \
postgres-check-v16 \
postgres-check-v17
neon-pg-ext: $(foreach pg_version,$(POSTGRES_VERSIONS),neon-pg-ext-$(pg_version))
# This removes everything
.PHONY: distclean
distclean:
$(RM) -r $(POSTGRES_INSTALL_DIR)
$(RM) -r $(POSTGRES_INSTALL_DIR) $(BUILD_DIR)
$(CARGO_CMD_PREFIX) cargo clean
.PHONY: fmt
@@ -300,3 +209,19 @@ neon-pgindent: postgres-v17-pg-bsd-indent neon-pg-ext-v17
.PHONY: setup-pre-commit-hook
setup-pre-commit-hook:
ln -s -f $(ROOT_PROJECT_DIR)/pre-commit.py .git/hooks/pre-commit
# Targets for building PostgreSQL are defined in postgres.mk.
#
# But if the caller has indicated that PostgreSQL is already
# installed, by setting the PG_INSTALL_CACHED variable, skip it.
ifdef PG_INSTALL_CACHED
postgres-install: skip-install
$(foreach pg_version,$(POSTGRES_VERSIONS),postgres-install-$(pg_version)): skip-install
postgres-headers-install:
+@echo "Skipping installation of PostgreSQL headers because PG_INSTALL_CACHED is set"
skip-install:
+@echo "Skipping PostgreSQL installation because PG_INSTALL_CACHED is set"
else
include postgres.mk
endif

View File

@@ -165,6 +165,7 @@ RUN curl -fsSL \
&& rm sql_exporter.tar.gz
# protobuf-compiler (protoc)
# Keep the version the same as in compute/compute-node.Dockerfile
ENV PROTOC_VERSION=25.1
RUN curl -fsSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOC_VERSION}/protoc-${PROTOC_VERSION}-linux-$(uname -m | sed 's/aarch64/aarch_64/g').zip" -o "protoc.zip" \
&& unzip -q protoc.zip -d protoc \
@@ -179,7 +180,7 @@ RUN curl -sL "https://github.com/peak/s5cmd/releases/download/v${S5CMD_VERSION}/
&& mv s5cmd /usr/local/bin/s5cmd
# LLVM
ENV LLVM_VERSION=19
ENV LLVM_VERSION=20
RUN curl -fsSL 'https://apt.llvm.org/llvm-snapshot.gpg.key' | apt-key add - \
&& echo "deb http://apt.llvm.org/${DEBIAN_VERSION}/ llvm-toolchain-${DEBIAN_VERSION}-${LLVM_VERSION} main" > /etc/apt/sources.list.d/llvm.stable.list \
&& apt update \
@@ -292,7 +293,7 @@ WORKDIR /home/nonroot
# Rust
# Please keep the version of llvm (installed above) in sync with rust llvm (`rustc --version --verbose | grep LLVM`)
ENV RUSTC_VERSION=1.87.0
ENV RUSTC_VERSION=1.88.0
ENV RUSTUP_HOME="/home/nonroot/.rustup"
ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
ARG RUSTFILT_VERSION=0.2.1

View File

@@ -22,7 +22,7 @@ sql_exporter.yml: $(jsonnet_files)
--output-file etc/$@ \
--tla-str collector_name=neon_collector \
--tla-str collector_file=neon_collector.yml \
--tla-str 'connection_string=postgresql://cloud_admin@127.0.0.1:5432/postgres?sslmode=disable&application_name=sql_exporter' \
--tla-str 'connection_string=postgresql://cloud_admin@127.0.0.1:5432/postgres?sslmode=disable&application_name=sql_exporter&pgaudit.log=none' \
etc/sql_exporter.jsonnet
sql_exporter_autoscaling.yml: $(jsonnet_files)
@@ -30,7 +30,7 @@ sql_exporter_autoscaling.yml: $(jsonnet_files)
--output-file etc/$@ \
--tla-str collector_name=neon_collector_autoscaling \
--tla-str collector_file=neon_collector_autoscaling.yml \
--tla-str 'connection_string=postgresql://cloud_admin@127.0.0.1:5432/postgres?sslmode=disable&application_name=sql_exporter_autoscaling' \
--tla-str 'connection_string=postgresql://cloud_admin@127.0.0.1:5432/postgres?sslmode=disable&application_name=sql_exporter_autoscaling&pgaudit.log=none' \
etc/sql_exporter.jsonnet
.PHONY: clean

View File

@@ -115,6 +115,9 @@ ARG EXTENSIONS=all
FROM $BASE_IMAGE_SHA AS build-deps
ARG DEBIAN_VERSION
# Keep in sync with build-tools.Dockerfile
ENV PROTOC_VERSION=25.1
# Use strict mode for bash to catch errors early
SHELL ["/bin/bash", "-euo", "pipefail", "-c"]
@@ -149,8 +152,14 @@ RUN case $DEBIAN_VERSION in \
libclang-dev \
jsonnet \
$VERSION_INSTALLS \
&& apt clean && rm -rf /var/lib/apt/lists/* && \
useradd -ms /bin/bash nonroot -b /home
&& apt clean && rm -rf /var/lib/apt/lists/* \
&& useradd -ms /bin/bash nonroot -b /home \
# Install protoc from binary release, since Debian's versions are too old.
&& curl -fsSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOC_VERSION}/protoc-${PROTOC_VERSION}-linux-$(uname -m | sed 's/aarch64/aarch_64/g').zip" -o "protoc.zip" \
&& unzip -q protoc.zip -d protoc \
&& mv protoc/bin/protoc /usr/local/bin/protoc \
&& mv protoc/include/google /usr/local/include/google \
&& rm -rf protoc.zip protoc
#########################################################################################
#
@@ -171,9 +180,6 @@ RUN cd postgres && \
eval $CONFIGURE_CMD && \
make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s install && \
make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C contrib/ install && \
# Install headers
make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/include install && \
make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/interfaces/libpq install && \
# Enable some of contrib extensions
echo 'trusted = true' >> /usr/local/pgsql/share/extension/autoinc.control && \
echo 'trusted = true' >> /usr/local/pgsql/share/extension/dblink.control && \
@@ -1173,7 +1179,7 @@ COPY --from=pgrag-src /ext-src/ /ext-src/
# Install it using virtual environment, because Python 3.11 (the default version on Debian 12 (Bookworm)) complains otherwise
WORKDIR /ext-src/onnxruntime-src
RUN apt update && apt install --no-install-recommends --no-install-suggests -y \
python3 python3-pip python3-venv protobuf-compiler && \
python3 python3-pip python3-venv && \
apt clean && rm -rf /var/lib/apt/lists/* && \
python3 -m venv venv && \
. venv/bin/activate && \
@@ -1568,20 +1574,20 @@ ARG PG_VERSION
WORKDIR /ext-src
RUN case "${PG_VERSION}" in \
"v14") \
export PGAUDIT_VERSION=1.6.2 \
export PGAUDIT_CHECKSUM=1f350d70a0cbf488c0f2b485e3a5c9b11f78ad9e3cbb95ef6904afa1eb3187eb \
export PGAUDIT_VERSION=1.6.3 \
export PGAUDIT_CHECKSUM=37a8f5a7cc8d9188e536d15cf0fdc457fcdab2547caedb54442c37f124110919 \
;; \
"v15") \
export PGAUDIT_VERSION=1.7.0 \
export PGAUDIT_CHECKSUM=8f4a73e451c88c567e516e6cba7dc1e23bc91686bb6f1f77f8f3126d428a8bd8 \
export PGAUDIT_VERSION=1.7.1 \
export PGAUDIT_CHECKSUM=e9c8e6e092d82b2f901d72555ce0fe7780552f35f8985573796cd7e64b09d4ec \
;; \
"v16") \
export PGAUDIT_VERSION=16.0 \
export PGAUDIT_CHECKSUM=d53ef985f2d0b15ba25c512c4ce967dce07b94fd4422c95bd04c4c1a055fe738 \
export PGAUDIT_VERSION=16.1 \
export PGAUDIT_CHECKSUM=3bae908ab70ba0c6f51224009dbcfff1a97bd6104c6273297a64292e1b921fee \
;; \
"v17") \
export PGAUDIT_VERSION=17.0 \
export PGAUDIT_CHECKSUM=7d0d08d030275d525f36cd48b38c6455f1023da863385badff0cec44965bfd8c \
export PGAUDIT_VERSION=17.1 \
export PGAUDIT_CHECKSUM=9c5f37504d393486cc75d2ced83f75f5899be64fa85f689d6babb833b4361e6c \
;; \
*) \
echo "pgaudit is not supported on this PostgreSQL version" && exit 1;; \

View File

@@ -26,7 +26,7 @@ commands:
- name: postgres-exporter
user: nobody
sysvInitAction: respawn
shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter pgaudit.log=none" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
- name: pgbouncer-exporter
user: postgres
sysvInitAction: respawn
@@ -59,7 +59,7 @@ files:
# the rules use ALL as the hostname. Avoid the pointless lookups and the "unable to
# resolve host" log messages that they generate.
Defaults !fqdn
# Allow postgres user (which is what compute_ctl runs as) to run /neonvm/bin/resize-swap
# and /neonvm/bin/set-disk-quota as root without requiring entering a password (NOPASSWD),
# regardless of hostname (ALL)

View File

@@ -26,7 +26,7 @@ commands:
- name: postgres-exporter
user: nobody
sysvInitAction: respawn
shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter pgaudit.log=none" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
- name: pgbouncer-exporter
user: postgres
sysvInitAction: respawn
@@ -59,7 +59,7 @@ files:
# the rules use ALL as the hostname. Avoid the pointless lookups and the "unable to
# resolve host" log messages that they generate.
Defaults !fqdn
# Allow postgres user (which is what compute_ctl runs as) to run /neonvm/bin/resize-swap
# and /neonvm/bin/set-disk-quota as root without requiring entering a password (NOPASSWD),
# regardless of hostname (ALL)

View File

@@ -38,6 +38,7 @@ once_cell.workspace = true
opentelemetry.workspace = true
opentelemetry_sdk.workspace = true
p256 = { version = "0.13", features = ["pem"] }
pageserver_page_api.workspace = true
postgres.workspace = true
regex.workspace = true
reqwest = { workspace = true, features = ["json"] }
@@ -53,6 +54,7 @@ tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
tokio-postgres.workspace = true
tokio-util.workspace = true
tokio-stream.workspace = true
tonic.workspace = true
tower-otel.workspace = true
tracing.workspace = true
tracing-opentelemetry.workspace = true
@@ -64,6 +66,7 @@ uuid.workspace = true
walkdir.workspace = true
x509-cert.workspace = true
postgres_versioninfo.workspace = true
postgres_initdb.workspace = true
compute_api.workspace = true
utils.workspace = true

View File

@@ -36,6 +36,8 @@
use std::ffi::OsString;
use std::fs::File;
use std::process::exit;
use std::sync::Arc;
use std::sync::atomic::AtomicU64;
use std::sync::mpsc;
use std::thread;
use std::time::Duration;
@@ -190,7 +192,9 @@ fn main() -> Result<()> {
cgroup: cli.cgroup,
#[cfg(target_os = "linux")]
vm_monitor_addr: cli.vm_monitor_addr,
installed_extensions_collection_interval: cli.installed_extensions_collection_interval,
installed_extensions_collection_interval: Arc::new(AtomicU64::new(
cli.installed_extensions_collection_interval,
)),
},
config,
)?;

View File

@@ -29,7 +29,7 @@ use anyhow::{Context, bail};
use aws_config::BehaviorVersion;
use camino::{Utf8Path, Utf8PathBuf};
use clap::{Parser, Subcommand};
use compute_tools::extension_server::{PostgresMajorVersion, get_pg_version};
use compute_tools::extension_server::get_pg_version;
use nix::unistd::Pid;
use std::ops::Not;
use tracing::{Instrument, error, info, info_span, warn};
@@ -179,12 +179,8 @@ impl PostgresProcess {
.await
.context("create pgdata directory")?;
let pg_version = match get_pg_version(self.pgbin.as_ref()) {
PostgresMajorVersion::V14 => 14,
PostgresMajorVersion::V15 => 15,
PostgresMajorVersion::V16 => 16,
PostgresMajorVersion::V17 => 17,
};
let pg_version = get_pg_version(self.pgbin.as_ref());
postgres_initdb::do_run_initdb(postgres_initdb::RunInitdbArgs {
superuser: initdb_user,
locale: DEFAULT_LOCALE, // XXX: this shouldn't be hard-coded,

View File

@@ -6,7 +6,7 @@ use compute_api::responses::{
LfcPrewarmState, TlsConfig,
};
use compute_api::spec::{
ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, ExtVersion, PgIdent,
ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, ExtVersion, PageserverProtocol, PgIdent,
};
use futures::StreamExt;
use futures::future::join_all;
@@ -15,17 +15,17 @@ use itertools::Itertools;
use nix::sys::signal::{Signal, kill};
use nix::unistd::Pid;
use once_cell::sync::Lazy;
use pageserver_page_api::{self as page_api, BaseBackupCompression};
use postgres;
use postgres::NoTls;
use postgres::error::SqlState;
use remote_storage::{DownloadError, RemotePath};
use std::collections::{HashMap, HashSet};
use std::net::SocketAddr;
use std::os::unix::fs::{PermissionsExt, symlink};
use std::path::Path;
use std::process::{Command, Stdio};
use std::str::FromStr;
use std::sync::atomic::{AtomicU32, Ordering};
use std::sync::atomic::{AtomicU32, AtomicU64, Ordering};
use std::sync::{Arc, Condvar, Mutex, RwLock};
use std::time::{Duration, Instant};
use std::{env, fs};
@@ -36,6 +36,7 @@ use utils::id::{TenantId, TimelineId};
use utils::lsn::Lsn;
use utils::measured_stream::MeasuredReader;
use utils::pid_file;
use utils::shard::{ShardCount, ShardIndex, ShardNumber};
use crate::configurator::launch_configurator;
use crate::disk_quota::set_disk_quota;
@@ -69,6 +70,7 @@ pub static BUILD_TAG: Lazy<String> = Lazy::new(|| {
.unwrap_or(BUILD_TAG_DEFAULT)
.to_string()
});
const DEFAULT_INSTALLED_EXTENSIONS_COLLECTION_INTERVAL: u64 = 3600;
/// Static configuration params that don't change after startup. These mostly
/// come from the CLI args, or are derived from them.
@@ -102,7 +104,7 @@ pub struct ComputeNodeParams {
pub remote_ext_base_url: Option<Url>,
/// Interval for installed extensions collection
pub installed_extensions_collection_interval: u64,
pub installed_extensions_collection_interval: Arc<AtomicU64>,
}
/// Compute node info shared across several `compute_ctl` threads.
@@ -125,6 +127,9 @@ pub struct ComputeNode {
// key: ext_archive_name, value: started download time, download_completed?
pub ext_download_progress: RwLock<HashMap<String, (DateTime<Utc>, bool)>>,
pub compute_ctl_config: ComputeCtlConfig,
/// Handle to the extension stats collection task
extension_stats_task: Mutex<Option<tokio::task::JoinHandle<()>>>,
}
// store some metrics about download size that might impact startup time
@@ -218,7 +223,8 @@ pub struct ParsedSpec {
pub pageserver_connstr: String,
pub safekeeper_connstrings: Vec<String>,
pub storage_auth_token: Option<String>,
pub endpoint_storage_addr: Option<SocketAddr>,
/// k8s dns name and port
pub endpoint_storage_addr: Option<String>,
pub endpoint_storage_token: Option<String>,
}
@@ -313,13 +319,10 @@ impl TryFrom<ComputeSpec> for ParsedSpec {
.or(Err("invalid timeline id"))?
};
let endpoint_storage_addr: Option<SocketAddr> = spec
let endpoint_storage_addr: Option<String> = spec
.endpoint_storage_addr
.clone()
.or_else(|| spec.cluster.settings.find("neon.endpoint_storage_addr"))
.unwrap_or_default()
.parse()
.ok();
.or_else(|| spec.cluster.settings.find("neon.endpoint_storage_addr"));
let endpoint_storage_token = spec
.endpoint_storage_token
.clone()
@@ -405,7 +408,7 @@ impl ComputeNode {
// that can affect `compute_ctl` and prevent it from properly configuring the database schema.
// Unset them via connection string options before connecting to the database.
// N.B. keep it in sync with `ZENITH_OPTIONS` in `get_maintenance_client()`.
const EXTRA_OPTIONS: &str = "-c role=cloud_admin -c default_transaction_read_only=off -c search_path=public -c statement_timeout=0";
const EXTRA_OPTIONS: &str = "-c role=cloud_admin -c default_transaction_read_only=off -c search_path=public -c statement_timeout=0 -c pgaudit.log=none";
let options = match conn_conf.get_options() {
// Allow the control plane to override any options set by the
// compute
@@ -429,6 +432,7 @@ impl ComputeNode {
state_changed: Condvar::new(),
ext_download_progress: RwLock::new(HashMap::new()),
compute_ctl_config: config.compute_ctl_config,
extension_stats_task: Mutex::new(None),
})
}
@@ -516,6 +520,9 @@ impl ComputeNode {
None
};
// Terminate the extension stats collection task
this.terminate_extension_stats_task();
// Terminate the vm_monitor so it releases the file watcher on
// /sys/fs/cgroup/neon-postgres.
// Note: the vm-monitor only runs on linux because it requires cgroups.
@@ -998,13 +1005,80 @@ impl ComputeNode {
Ok(())
}
// Get basebackup from the libpq connection to pageserver using `connstr` and
// unarchive it to `pgdata` directory overriding all its previous content.
/// Fetches a basebackup from the Pageserver using the compute state's Pageserver connstring and
/// unarchives it to `pgdata` directory, replacing any existing contents.
#[instrument(skip_all, fields(%lsn))]
fn try_get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
let spec = compute_state.pspec.as_ref().expect("spec must be set");
let start_time = Instant::now();
let shard0_connstr = spec.pageserver_connstr.split(',').next().unwrap();
let started = Instant::now();
let (connected, size) = match PageserverProtocol::from_connstring(shard0_connstr)? {
PageserverProtocol::Libpq => self.try_get_basebackup_libpq(spec, lsn)?,
PageserverProtocol::Grpc => self.try_get_basebackup_grpc(spec, lsn)?,
};
let mut state = self.state.lock().unwrap();
state.metrics.pageserver_connect_micros =
connected.duration_since(started).as_micros() as u64;
state.metrics.basebackup_bytes = size as u64;
state.metrics.basebackup_ms = started.elapsed().as_millis() as u64;
Ok(())
}
/// Fetches a basebackup via gRPC. The connstring must use grpc://. Returns the timestamp when
/// the connection was established, and the (compressed) size of the basebackup.
fn try_get_basebackup_grpc(&self, spec: &ParsedSpec, lsn: Lsn) -> Result<(Instant, usize)> {
let shard0_connstr = spec
.pageserver_connstr
.split(',')
.next()
.unwrap()
.to_string();
let shard_index = match spec.pageserver_connstr.split(',').count() as u8 {
0 | 1 => ShardIndex::unsharded(),
count => ShardIndex::new(ShardNumber(0), ShardCount(count)),
};
let (reader, connected) = tokio::runtime::Handle::current().block_on(async move {
let mut client = page_api::Client::new(
shard0_connstr,
spec.tenant_id,
spec.timeline_id,
shard_index,
spec.storage_auth_token.clone(),
None, // NB: base backups use payload compression
)
.await?;
let connected = Instant::now();
let reader = client
.get_base_backup(page_api::GetBaseBackupRequest {
lsn: (lsn != Lsn(0)).then_some(lsn),
compression: BaseBackupCompression::Gzip,
replica: spec.spec.mode != ComputeMode::Primary,
full: false,
})
.await?;
anyhow::Ok((reader, connected))
})?;
let mut reader = MeasuredReader::new(tokio_util::io::SyncIoBridge::new(reader));
// Set `ignore_zeros` so that unpack() reads the entire stream and doesn't just stop at the
// end-of-archive marker. If the server errors, the tar::Builder drop handler will write an
// end-of-archive marker before the error is emitted, and we would not see the error.
let mut ar = tar::Archive::new(flate2::read::GzDecoder::new(&mut reader));
ar.set_ignore_zeros(true);
ar.unpack(&self.params.pgdata)?;
Ok((connected, reader.get_byte_count()))
}
/// Fetches a basebackup via libpq. The connstring must use postgresql://. Returns the timestamp
/// when the connection was established, and the (compressed) size of the basebackup.
fn try_get_basebackup_libpq(&self, spec: &ParsedSpec, lsn: Lsn) -> Result<(Instant, usize)> {
let shard0_connstr = spec.pageserver_connstr.split(',').next().unwrap();
let mut config = postgres::Config::from_str(shard0_connstr)?;
@@ -1018,16 +1092,14 @@ impl ComputeNode {
}
config.application_name("compute_ctl");
if let Some(spec) = &compute_state.pspec {
config.options(&format!(
"-c neon.compute_mode={}",
spec.spec.mode.to_type_str()
));
}
config.options(&format!(
"-c neon.compute_mode={}",
spec.spec.mode.to_type_str()
));
// Connect to pageserver
let mut client = config.connect(NoTls)?;
let pageserver_connect_micros = start_time.elapsed().as_micros() as u64;
let connected = Instant::now();
let basebackup_cmd = match lsn {
Lsn(0) => {
@@ -1064,16 +1136,13 @@ impl ComputeNode {
// 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.
// The tar::Builder drop handler will write an end-of-archive marker
// before emitting the error, and we would not see it otherwise.
let mut ar = tar::Archive::new(flate2::read::GzDecoder::new(&mut bufreader));
ar.set_ignore_zeros(true);
ar.unpack(&self.params.pgdata)?;
// Report metrics
let mut state = self.state.lock().unwrap();
state.metrics.pageserver_connect_micros = pageserver_connect_micros;
state.metrics.basebackup_bytes = measured_reader.get_byte_count() as u64;
state.metrics.basebackup_ms = start_time.elapsed().as_millis() as u64;
Ok(())
Ok((connected, measured_reader.get_byte_count()))
}
// Gets the basebackup in a retry loop
@@ -1610,6 +1679,8 @@ impl ComputeNode {
tls_config = self.compute_ctl_config.tls.clone();
}
self.update_installed_extensions_collection_interval(&spec);
let max_concurrent_connections = self.max_service_connections(compute_state, &spec);
// Merge-apply spec & changes to PostgreSQL state.
@@ -1674,6 +1745,8 @@ impl ComputeNode {
let tls_config = self.tls_config(&spec);
self.update_installed_extensions_collection_interval(&spec);
if let Some(ref pgbouncer_settings) = spec.pgbouncer_settings {
info!("tuning pgbouncer");
@@ -2278,10 +2351,20 @@ LIMIT 100",
}
pub fn spawn_extension_stats_task(&self) {
// Cancel any existing task
if let Some(handle) = self.extension_stats_task.lock().unwrap().take() {
handle.abort();
}
let conf = self.tokio_conn_conf.clone();
let installed_extensions_collection_interval =
self.params.installed_extensions_collection_interval;
tokio::spawn(async move {
let atomic_interval = self.params.installed_extensions_collection_interval.clone();
let mut installed_extensions_collection_interval =
2 * atomic_interval.load(std::sync::atomic::Ordering::SeqCst);
info!(
"[NEON_EXT_SPAWN] Spawning background installed extensions worker with Timeout: {}",
installed_extensions_collection_interval
);
let handle = tokio::spawn(async move {
// An initial sleep is added to ensure that two collections don't happen at the same time.
// The first collection happens during compute startup.
tokio::time::sleep(tokio::time::Duration::from_secs(
@@ -2294,8 +2377,48 @@ LIMIT 100",
loop {
interval.tick().await;
let _ = installed_extensions(conf.clone()).await;
// Acquire a read lock on the compute spec and then update the interval if necessary
interval = tokio::time::interval(tokio::time::Duration::from_secs(std::cmp::max(
installed_extensions_collection_interval,
2 * atomic_interval.load(std::sync::atomic::Ordering::SeqCst),
)));
installed_extensions_collection_interval = interval.period().as_secs();
}
});
// Store the new task handle
*self.extension_stats_task.lock().unwrap() = Some(handle);
}
fn terminate_extension_stats_task(&self) {
if let Some(handle) = self.extension_stats_task.lock().unwrap().take() {
handle.abort();
}
}
fn update_installed_extensions_collection_interval(&self, spec: &ComputeSpec) {
// Update the interval for collecting installed extensions statistics
// If the value is -1, we never suspend so set the value to default collection.
// If the value is 0, it means default, we will just continue to use the default.
if spec.suspend_timeout_seconds == -1 || spec.suspend_timeout_seconds == 0 {
info!(
"[NEON_EXT_INT_UPD] Spec Timeout: {}, New Timeout: {}",
spec.suspend_timeout_seconds, DEFAULT_INSTALLED_EXTENSIONS_COLLECTION_INTERVAL
);
self.params.installed_extensions_collection_interval.store(
DEFAULT_INSTALLED_EXTENSIONS_COLLECTION_INTERVAL,
std::sync::atomic::Ordering::SeqCst,
);
} else {
info!(
"[NEON_EXT_INT_UPD] Spec Timeout: {}",
spec.suspend_timeout_seconds
);
self.params.installed_extensions_collection_interval.store(
spec.suspend_timeout_seconds as u64,
std::sync::atomic::Ordering::SeqCst,
);
}
}
}

View File

@@ -74,9 +74,11 @@ More specifically, here is an example ext_index.json
use std::path::Path;
use std::str;
use crate::metrics::{REMOTE_EXT_REQUESTS_TOTAL, UNKNOWN_HTTP_STATUS};
use anyhow::{Context, Result, bail};
use bytes::Bytes;
use compute_api::spec::RemoteExtSpec;
use postgres_versioninfo::PgMajorVersion;
use regex::Regex;
use remote_storage::*;
use reqwest::StatusCode;
@@ -86,8 +88,6 @@ use tracing::log::warn;
use url::Url;
use zstd::stream::read::Decoder;
use crate::metrics::{REMOTE_EXT_REQUESTS_TOTAL, UNKNOWN_HTTP_STATUS};
fn get_pg_config(argument: &str, pgbin: &str) -> String {
// gives the result of `pg_config [argument]`
// where argument is a flag like `--version` or `--sharedir`
@@ -106,7 +106,7 @@ fn get_pg_config(argument: &str, pgbin: &str) -> String {
.to_string()
}
pub fn get_pg_version(pgbin: &str) -> PostgresMajorVersion {
pub fn get_pg_version(pgbin: &str) -> PgMajorVersion {
// pg_config --version returns a (platform specific) human readable string
// such as "PostgreSQL 15.4". We parse this to v14/v15/v16 etc.
let human_version = get_pg_config("--version", pgbin);
@@ -114,25 +114,11 @@ pub fn get_pg_version(pgbin: &str) -> PostgresMajorVersion {
}
pub fn get_pg_version_string(pgbin: &str) -> String {
match get_pg_version(pgbin) {
PostgresMajorVersion::V14 => "v14",
PostgresMajorVersion::V15 => "v15",
PostgresMajorVersion::V16 => "v16",
PostgresMajorVersion::V17 => "v17",
}
.to_owned()
get_pg_version(pgbin).v_str()
}
#[derive(Copy, Clone, Debug, PartialEq, Eq)]
pub enum PostgresMajorVersion {
V14,
V15,
V16,
V17,
}
fn parse_pg_version(human_version: &str) -> PostgresMajorVersion {
use PostgresMajorVersion::*;
fn parse_pg_version(human_version: &str) -> PgMajorVersion {
use PgMajorVersion::*;
// Normal releases have version strings like "PostgreSQL 15.4". But there
// are also pre-release versions like "PostgreSQL 17devel" or "PostgreSQL
// 16beta2" or "PostgreSQL 17rc1". And with the --with-extra-version
@@ -143,10 +129,10 @@ fn parse_pg_version(human_version: &str) -> PostgresMajorVersion {
.captures(human_version)
{
Some(captures) if captures.len() == 2 => match &captures["major"] {
"14" => return V14,
"15" => return V15,
"16" => return V16,
"17" => return V17,
"14" => return PG14,
"15" => return PG15,
"16" => return PG16,
"17" => return PG17,
_ => {}
},
_ => {}
@@ -343,25 +329,25 @@ mod tests {
#[test]
fn test_parse_pg_version() {
use super::PostgresMajorVersion::*;
assert_eq!(parse_pg_version("PostgreSQL 15.4"), V15);
assert_eq!(parse_pg_version("PostgreSQL 15.14"), V15);
use postgres_versioninfo::PgMajorVersion::*;
assert_eq!(parse_pg_version("PostgreSQL 15.4"), PG15);
assert_eq!(parse_pg_version("PostgreSQL 15.14"), PG15);
assert_eq!(
parse_pg_version("PostgreSQL 15.4 (Ubuntu 15.4-0ubuntu0.23.04.1)"),
V15
PG15
);
assert_eq!(parse_pg_version("PostgreSQL 14.15"), V14);
assert_eq!(parse_pg_version("PostgreSQL 14.0"), V14);
assert_eq!(parse_pg_version("PostgreSQL 14.15"), PG14);
assert_eq!(parse_pg_version("PostgreSQL 14.0"), PG14);
assert_eq!(
parse_pg_version("PostgreSQL 14.9 (Debian 14.9-1.pgdg120+1"),
V14
PG14
);
assert_eq!(parse_pg_version("PostgreSQL 16devel"), V16);
assert_eq!(parse_pg_version("PostgreSQL 16beta1"), V16);
assert_eq!(parse_pg_version("PostgreSQL 16rc2"), V16);
assert_eq!(parse_pg_version("PostgreSQL 16extra"), V16);
assert_eq!(parse_pg_version("PostgreSQL 16devel"), PG16);
assert_eq!(parse_pg_version("PostgreSQL 16beta1"), PG16);
assert_eq!(parse_pg_version("PostgreSQL 16rc2"), PG16);
assert_eq!(parse_pg_version("PostgreSQL 16extra"), PG16);
}
#[test]

View File

@@ -4,7 +4,9 @@ use std::thread;
use std::time::{Duration, SystemTime};
use anyhow::{Result, bail};
use compute_api::spec::ComputeMode;
use compute_api::spec::{ComputeMode, PageserverProtocol};
use itertools::Itertools as _;
use pageserver_page_api as page_api;
use postgres::{NoTls, SimpleQueryMessage};
use tracing::{info, warn};
use utils::id::{TenantId, TimelineId};
@@ -76,25 +78,17 @@ fn acquire_lsn_lease_with_retry(
loop {
// Note: List of pageservers is dynamic, need to re-read configs before each attempt.
let configs = {
let (connstrings, auth) = {
let state = compute.state.lock().unwrap();
let spec = state.pspec.as_ref().expect("spec must be set");
let conn_strings = spec.pageserver_connstr.split(',');
conn_strings
.map(|connstr| {
let mut config = postgres::Config::from_str(connstr).expect("Invalid connstr");
if let Some(storage_auth_token) = &spec.storage_auth_token {
config.password(storage_auth_token.clone());
}
config
})
.collect::<Vec<_>>()
(
spec.pageserver_connstr.clone(),
spec.storage_auth_token.clone(),
)
};
let result = try_acquire_lsn_lease(tenant_id, timeline_id, lsn, &configs);
let result =
try_acquire_lsn_lease(&connstrings, auth.as_deref(), tenant_id, timeline_id, lsn);
match result {
Ok(Some(res)) => {
return Ok(res);
@@ -116,68 +110,104 @@ fn acquire_lsn_lease_with_retry(
}
}
/// Tries to acquire an LSN lease through PS page_service API.
/// Tries to acquire LSN leases on all Pageserver shards.
fn try_acquire_lsn_lease(
connstrings: &str,
auth: Option<&str>,
tenant_id: TenantId,
timeline_id: TimelineId,
lsn: Lsn,
configs: &[postgres::Config],
) -> Result<Option<SystemTime>> {
fn get_valid_until(
config: &postgres::Config,
tenant_shard_id: TenantShardId,
timeline_id: TimelineId,
lsn: Lsn,
) -> Result<Option<SystemTime>> {
let mut client = config.connect(NoTls)?;
let cmd = format!("lease lsn {tenant_shard_id} {timeline_id} {lsn} ");
let res = client.simple_query(&cmd)?;
let msg = match res.first() {
Some(msg) => msg,
None => bail!("empty response"),
};
let row = match msg {
SimpleQueryMessage::Row(row) => row,
_ => bail!("error parsing lsn lease response"),
let connstrings = connstrings.split(',').collect_vec();
let shard_count = connstrings.len();
let mut leases = Vec::new();
for (shard_number, &connstring) in connstrings.iter().enumerate() {
let tenant_shard_id = match shard_count {
0 | 1 => TenantShardId::unsharded(tenant_id),
shard_count => TenantShardId {
tenant_id,
shard_number: ShardNumber(shard_number as u8),
shard_count: ShardCount::new(shard_count as u8),
},
};
// Note: this will be None if a lease is explicitly not granted.
let valid_until_str = row.get("valid_until");
let valid_until = valid_until_str.map(|s| {
SystemTime::UNIX_EPOCH
.checked_add(Duration::from_millis(u128::from_str(s).unwrap() as u64))
.expect("Time larger than max SystemTime could handle")
});
Ok(valid_until)
let lease = match PageserverProtocol::from_connstring(connstring)? {
PageserverProtocol::Libpq => {
acquire_lsn_lease_libpq(connstring, auth, tenant_shard_id, timeline_id, lsn)?
}
PageserverProtocol::Grpc => {
acquire_lsn_lease_grpc(connstring, auth, tenant_shard_id, timeline_id, lsn)?
}
};
leases.push(lease);
}
let shard_count = configs.len();
Ok(leases.into_iter().min().flatten())
}
let valid_until = if shard_count > 1 {
configs
.iter()
.enumerate()
.map(|(shard_number, config)| {
let tenant_shard_id = TenantShardId {
tenant_id,
shard_count: ShardCount::new(shard_count as u8),
shard_number: ShardNumber(shard_number as u8),
};
get_valid_until(config, tenant_shard_id, timeline_id, lsn)
})
.collect::<Result<Vec<Option<SystemTime>>>>()?
.into_iter()
.min()
.unwrap()
} else {
get_valid_until(
&configs[0],
TenantShardId::unsharded(tenant_id),
timeline_id,
lsn,
)?
/// Acquires an LSN lease on a single shard, using the libpq API. The connstring must use a
/// postgresql:// scheme.
fn acquire_lsn_lease_libpq(
connstring: &str,
auth: Option<&str>,
tenant_shard_id: TenantShardId,
timeline_id: TimelineId,
lsn: Lsn,
) -> Result<Option<SystemTime>> {
let mut config = postgres::Config::from_str(connstring)?;
if let Some(auth) = auth {
config.password(auth);
}
let mut client = config.connect(NoTls)?;
let cmd = format!("lease lsn {tenant_shard_id} {timeline_id} {lsn} ");
let res = client.simple_query(&cmd)?;
let msg = match res.first() {
Some(msg) => msg,
None => bail!("empty response"),
};
let row = match msg {
SimpleQueryMessage::Row(row) => row,
_ => bail!("error parsing lsn lease response"),
};
// Note: this will be None if a lease is explicitly not granted.
let valid_until_str = row.get("valid_until");
let valid_until = valid_until_str.map(|s| {
SystemTime::UNIX_EPOCH
.checked_add(Duration::from_millis(u128::from_str(s).unwrap() as u64))
.expect("Time larger than max SystemTime could handle")
});
Ok(valid_until)
}
/// Acquires an LSN lease on a single shard, using the gRPC API. The connstring must use a
/// grpc:// scheme.
fn acquire_lsn_lease_grpc(
connstring: &str,
auth: Option<&str>,
tenant_shard_id: TenantShardId,
timeline_id: TimelineId,
lsn: Lsn,
) -> Result<Option<SystemTime>> {
tokio::runtime::Handle::current().block_on(async move {
let mut client = page_api::Client::new(
connstring.to_string(),
tenant_shard_id.tenant_id,
timeline_id,
tenant_shard_id.to_index(),
auth.map(String::from),
None,
)
.await?;
let req = page_api::LeaseLsnRequest { lsn };
match client.lease_lsn(req).await {
Ok(expires) => Ok(Some(expires)),
// Lease couldn't be acquired because the LSN has been garbage collected.
Err(err) if err.code() == tonic::Code::FailedPrecondition => Ok(None),
Err(err) => Err(err.into()),
}
})
}

View File

@@ -3,7 +3,8 @@
"timestamp": "2021-05-23T18:25:43.511Z",
"operation_uuid": "0f657b36-4b0f-4a2d-9c2e-1dcd615e7d8b",
"suspend_timeout_seconds": 3600,
"cluster": {
"cluster_id": "test-cluster-42",
"name": "Zenith Test",

View File

@@ -16,9 +16,9 @@ use std::time::Duration;
use anyhow::{Context, Result, anyhow, bail};
use clap::Parser;
use compute_api::requests::ComputeClaimsScope;
use compute_api::spec::ComputeMode;
use compute_api::spec::{ComputeMode, PageserverProtocol};
use control_plane::broker::StorageBroker;
use control_plane::endpoint::{ComputeControlPlane, EndpointTerminateMode, PageserverProtocol};
use control_plane::endpoint::{ComputeControlPlane, EndpointTerminateMode};
use control_plane::endpoint_storage::{ENDPOINT_STORAGE_DEFAULT_ADDR, EndpointStorage};
use control_plane::local_env;
use control_plane::local_env::{
@@ -48,7 +48,7 @@ use postgres_connection::parse_host_port;
use safekeeper_api::membership::{SafekeeperGeneration, SafekeeperId};
use safekeeper_api::{
DEFAULT_HTTP_LISTEN_PORT as DEFAULT_SAFEKEEPER_HTTP_PORT,
DEFAULT_PG_LISTEN_PORT as DEFAULT_SAFEKEEPER_PG_PORT,
DEFAULT_PG_LISTEN_PORT as DEFAULT_SAFEKEEPER_PG_PORT, PgMajorVersion, PgVersionId,
};
use storage_broker::DEFAULT_LISTEN_ADDR as DEFAULT_BROKER_ADDR;
use tokio::task::JoinSet;
@@ -64,7 +64,7 @@ const DEFAULT_PAGESERVER_ID: NodeId = NodeId(1);
const DEFAULT_BRANCH_NAME: &str = "main";
project_git_version!(GIT_VERSION);
const DEFAULT_PG_VERSION: u32 = 17;
const DEFAULT_PG_VERSION: PgMajorVersion = PgMajorVersion::PG17;
const DEFAULT_PAGESERVER_CONTROL_PLANE_API: &str = "http://127.0.0.1:1234/upcall/v1/";
@@ -169,7 +169,7 @@ struct TenantCreateCmdArgs {
#[arg(default_value_t = DEFAULT_PG_VERSION)]
#[clap(long, help = "Postgres version to use for the initial timeline")]
pg_version: u32,
pg_version: PgMajorVersion,
#[clap(
long,
@@ -292,7 +292,7 @@ struct TimelineCreateCmdArgs {
#[arg(default_value_t = DEFAULT_PG_VERSION)]
#[clap(long, help = "Postgres version")]
pg_version: u32,
pg_version: PgMajorVersion,
}
#[derive(clap::Args)]
@@ -324,7 +324,7 @@ struct TimelineImportCmdArgs {
#[arg(default_value_t = DEFAULT_PG_VERSION)]
#[clap(long, help = "Postgres version of the backup being imported")]
pg_version: u32,
pg_version: PgMajorVersion,
}
#[derive(clap::Subcommand)]
@@ -603,7 +603,7 @@ struct EndpointCreateCmdArgs {
#[arg(default_value_t = DEFAULT_PG_VERSION)]
#[clap(long, help = "Postgres version")]
pg_version: u32,
pg_version: PgMajorVersion,
/// Use gRPC to communicate with Pageservers, by generating grpc:// connstrings.
///
@@ -1295,7 +1295,7 @@ async fn handle_timeline(cmd: &TimelineCmd, env: &mut local_env::LocalEnv) -> Re
},
new_members: None,
};
let pg_version = args.pg_version * 10000;
let pg_version = PgVersionId::from(args.pg_version);
let req = safekeeper_api::models::TimelineCreateRequest {
tenant_id,
timeline_id,
@@ -1649,7 +1649,9 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
// If --safekeepers argument is given, use only the listed
// safekeeper nodes; otherwise all from the env.
let safekeepers = parse_safekeepers(&args.safekeepers)?;
endpoint.reconfigure(pageservers, None, safekeepers).await?;
endpoint
.reconfigure(Some(pageservers), None, safekeepers, None)
.await?;
}
EndpointCmd::Stop(args) => {
let endpoint_id = &args.endpoint_id;

View File

@@ -56,8 +56,8 @@ use compute_api::responses::{
TlsConfig,
};
use compute_api::spec::{
Cluster, ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, Database, PgIdent,
RemoteExtSpec, Role,
Cluster, ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, Database, PageserverProtocol,
PgIdent, RemoteExtSpec, Role,
};
use jsonwebtoken::jwk::{
AlgorithmParameters, CommonParameters, EllipticCurve, Jwk, JwkSet, KeyAlgorithm, KeyOperations,
@@ -67,6 +67,7 @@ use nix::sys::signal::{Signal, kill};
use pageserver_api::shard::ShardStripeSize;
use pem::Pem;
use reqwest::header::CONTENT_TYPE;
use safekeeper_api::PgMajorVersion;
use safekeeper_api::membership::SafekeeperGeneration;
use serde::{Deserialize, Serialize};
use sha2::{Digest, Sha256};
@@ -89,7 +90,7 @@ pub struct EndpointConf {
pg_port: u16,
external_http_port: u16,
internal_http_port: u16,
pg_version: u32,
pg_version: PgMajorVersion,
grpc: bool,
skip_pg_catalog_updates: bool,
reconfigure_concurrency: usize,
@@ -192,7 +193,7 @@ impl ComputeControlPlane {
pg_port: Option<u16>,
external_http_port: Option<u16>,
internal_http_port: Option<u16>,
pg_version: u32,
pg_version: PgMajorVersion,
mode: ComputeMode,
grpc: bool,
skip_pg_catalog_updates: bool,
@@ -312,7 +313,7 @@ pub struct Endpoint {
pub internal_http_address: SocketAddr,
// postgres major version in the format: 14, 15, etc.
pg_version: u32,
pg_version: PgMajorVersion,
// These are not part of the endpoint as such, but the environment
// the endpoint runs in.
@@ -372,29 +373,6 @@ impl std::fmt::Display for EndpointTerminateMode {
}
}
/// Protocol used to connect to a Pageserver.
#[derive(Clone, Copy, Debug)]
pub enum PageserverProtocol {
Libpq,
Grpc,
}
impl PageserverProtocol {
/// Returns the URL scheme for the protocol, used in connstrings.
pub fn scheme(&self) -> &'static str {
match self {
Self::Libpq => "postgresql",
Self::Grpc => "grpc",
}
}
}
impl Display for PageserverProtocol {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.write_str(self.scheme())
}
}
impl Endpoint {
fn from_dir_entry(entry: std::fs::DirEntry, env: &LocalEnv) -> Result<Endpoint> {
if !entry.file_type()?.is_dir() {
@@ -557,7 +535,7 @@ impl Endpoint {
conf.append("hot_standby", "on");
// prefetching of blocks referenced in WAL doesn't make sense for us
// Neon hot standby ignores pages that are not in the shared_buffers
if self.pg_version >= 15 {
if self.pg_version >= PgMajorVersion::PG15 {
conf.append("recovery_prefetch", "off");
}
}
@@ -802,6 +780,7 @@ impl Endpoint {
endpoint_storage_addr: Some(endpoint_storage_addr),
endpoint_storage_token: Some(endpoint_storage_token),
autoprewarm: false,
suspend_timeout_seconds: -1, // Only used in neon_local.
};
// this strange code is needed to support respec() in tests
@@ -996,12 +975,11 @@ impl Endpoint {
pub async fn reconfigure(
&self,
pageservers: Vec<(PageserverProtocol, Host, u16)>,
pageservers: Option<Vec<(PageserverProtocol, Host, u16)>>,
stripe_size: Option<ShardStripeSize>,
safekeepers: Option<Vec<NodeId>>,
safekeeper_generation: Option<SafekeeperGeneration>,
) -> Result<()> {
anyhow::ensure!(!pageservers.is_empty(), "no pageservers provided");
let (mut spec, compute_ctl_config) = {
let config_path = self.endpoint_path().join("config.json");
let file = std::fs::File::open(config_path)?;
@@ -1013,16 +991,24 @@ impl Endpoint {
let postgresql_conf = self.read_postgresql_conf()?;
spec.cluster.postgresql_conf = Some(postgresql_conf);
let pageserver_connstr = Self::build_pageserver_connstr(&pageservers);
spec.pageserver_connstring = Some(pageserver_connstr);
if stripe_size.is_some() {
spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
// If pageservers are not specified, don't change them.
if let Some(pageservers) = pageservers {
anyhow::ensure!(!pageservers.is_empty(), "no pageservers provided");
let pageserver_connstr = Self::build_pageserver_connstr(&pageservers);
spec.pageserver_connstring = Some(pageserver_connstr);
if stripe_size.is_some() {
spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
}
}
// If safekeepers are not specified, don't change them.
if let Some(safekeepers) = safekeepers {
let safekeeper_connstrings = self.build_safekeepers_connstrs(safekeepers)?;
spec.safekeeper_connstrings = safekeeper_connstrings;
if let Some(g) = safekeeper_generation {
spec.safekeepers_generation = Some(g.into_inner());
}
}
let client = reqwest::Client::builder()
@@ -1060,6 +1046,24 @@ impl Endpoint {
}
}
pub async fn reconfigure_pageservers(
&self,
pageservers: Vec<(PageserverProtocol, Host, u16)>,
stripe_size: Option<ShardStripeSize>,
) -> Result<()> {
self.reconfigure(Some(pageservers), stripe_size, None, None)
.await
}
pub async fn reconfigure_safekeepers(
&self,
safekeepers: Vec<NodeId>,
generation: SafekeeperGeneration,
) -> Result<()> {
self.reconfigure(None, None, Some(safekeepers), Some(generation))
.await
}
pub async fn stop(
&self,
mode: EndpointTerminateMode,

View File

@@ -12,9 +12,11 @@ use std::{env, fs};
use anyhow::{Context, bail};
use clap::ValueEnum;
use pageserver_api::config::PostHogConfig;
use pem::Pem;
use postgres_backend::AuthType;
use reqwest::{Certificate, Url};
use safekeeper_api::PgMajorVersion;
use serde::{Deserialize, Serialize};
use utils::auth::encode_from_key_file;
use utils::id::{NodeId, TenantId, TenantTimelineId, TimelineId};
@@ -210,7 +212,9 @@ pub struct NeonStorageControllerConf {
pub use_local_compute_notifications: bool,
pub timeline_safekeeper_count: Option<i64>,
pub timeline_safekeeper_count: Option<usize>,
pub posthog_config: Option<PostHogConfig>,
pub kick_secondary_downloads: Option<bool>,
}
@@ -244,6 +248,7 @@ impl Default for NeonStorageControllerConf {
use_https_safekeeper_api: false,
use_local_compute_notifications: true,
timeline_safekeeper_count: None,
posthog_config: None,
kick_secondary_downloads: None,
}
}
@@ -424,25 +429,21 @@ impl LocalEnv {
self.pg_distrib_dir.clone()
}
pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
pub fn pg_distrib_dir(&self, pg_version: PgMajorVersion) -> anyhow::Result<PathBuf> {
let path = self.pg_distrib_dir.clone();
#[allow(clippy::manual_range_patterns)]
match pg_version {
14 | 15 | 16 | 17 => Ok(path.join(format!("v{pg_version}"))),
_ => bail!("Unsupported postgres version: {}", pg_version),
}
Ok(path.join(pg_version.v_str()))
}
pub fn pg_dir(&self, pg_version: u32, dir_name: &str) -> anyhow::Result<PathBuf> {
pub fn pg_dir(&self, pg_version: PgMajorVersion, dir_name: &str) -> anyhow::Result<PathBuf> {
Ok(self.pg_distrib_dir(pg_version)?.join(dir_name))
}
pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
pub fn pg_bin_dir(&self, pg_version: PgMajorVersion) -> anyhow::Result<PathBuf> {
self.pg_dir(pg_version, "bin")
}
pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
pub fn pg_lib_dir(&self, pg_version: PgMajorVersion) -> anyhow::Result<PathBuf> {
self.pg_dir(pg_version, "lib")
}

View File

@@ -22,6 +22,7 @@ use pageserver_api::shard::TenantShardId;
use pageserver_client::mgmt_api;
use postgres_backend::AuthType;
use postgres_connection::{PgConnectionConfig, parse_host_port};
use safekeeper_api::PgMajorVersion;
use utils::auth::{Claims, Scope};
use utils::id::{NodeId, TenantId, TimelineId};
use utils::lsn::Lsn;
@@ -607,7 +608,7 @@ impl PageServerNode {
timeline_id: TimelineId,
base: (Lsn, PathBuf),
pg_wal: Option<(Lsn, PathBuf)>,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<()> {
// Init base reader
let (start_lsn, base_tarfile_path) = base;

View File

@@ -6,6 +6,8 @@ use std::str::FromStr;
use std::sync::OnceLock;
use std::time::{Duration, Instant};
use crate::background_process;
use crate::local_env::{LocalEnv, NeonStorageControllerConf};
use camino::{Utf8Path, Utf8PathBuf};
use hyper0::Uri;
use nix::unistd::Pid;
@@ -22,6 +24,7 @@ use pageserver_client::mgmt_api::ResponseErrorMessageExt;
use pem::Pem;
use postgres_backend::AuthType;
use reqwest::{Method, Response};
use safekeeper_api::PgMajorVersion;
use serde::de::DeserializeOwned;
use serde::{Deserialize, Serialize};
use tokio::process::Command;
@@ -31,9 +34,6 @@ use utils::auth::{Claims, Scope, encode_from_key_file};
use utils::id::{NodeId, TenantId};
use whoami::username;
use crate::background_process;
use crate::local_env::{LocalEnv, NeonStorageControllerConf};
pub struct StorageController {
env: LocalEnv,
private_key: Option<Pem>,
@@ -48,7 +48,7 @@ pub struct StorageController {
const COMMAND: &str = "storage_controller";
const STORAGE_CONTROLLER_POSTGRES_VERSION: u32 = 16;
const STORAGE_CONTROLLER_POSTGRES_VERSION: PgMajorVersion = PgMajorVersion::PG16;
const DB_NAME: &str = "storage_controller";
@@ -184,9 +184,15 @@ impl StorageController {
/// to other versions if that one isn't found. Some automated tests create circumstances
/// where only one version is available in pg_distrib_dir, such as `test_remote_extensions`.
async fn get_pg_dir(&self, dir_name: &str) -> anyhow::Result<Utf8PathBuf> {
let prefer_versions = [STORAGE_CONTROLLER_POSTGRES_VERSION, 16, 15, 14];
const PREFER_VERSIONS: [PgMajorVersion; 5] = [
STORAGE_CONTROLLER_POSTGRES_VERSION,
PgMajorVersion::PG16,
PgMajorVersion::PG15,
PgMajorVersion::PG14,
PgMajorVersion::PG17,
];
for v in prefer_versions {
for v in PREFER_VERSIONS {
let path = Utf8PathBuf::from_path_buf(self.env.pg_dir(v, dir_name)?).unwrap();
if tokio::fs::try_exists(&path).await? {
return Ok(path);
@@ -632,10 +638,28 @@ impl StorageController {
args.push("--timelines-onto-safekeepers".to_string());
}
if let Some(sk_cnt) = self.config.timeline_safekeeper_count {
// neon_local is used in test environments where we often have less than 3 safekeepers.
if self.config.timeline_safekeeper_count.is_some() || self.env.safekeepers.len() < 3 {
let sk_cnt = self
.config
.timeline_safekeeper_count
.unwrap_or(self.env.safekeepers.len());
args.push(format!("--timeline-safekeeper-count={sk_cnt}"));
}
let mut envs = vec![
("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
];
if let Some(posthog_config) = &self.config.posthog_config {
envs.push((
"POSTHOG_CONFIG".to_string(),
serde_json::to_string(posthog_config)?,
));
}
println!("Starting storage controller");
background_process::start_process(
@@ -643,10 +667,7 @@ impl StorageController {
&instance_dir,
&self.env.storage_controller_bin(),
args,
vec![
("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
],
envs,
background_process::InitialPidFile::Create(self.pid_file(start_args.instance_id)),
&start_args.start_timeout,
|| async {

View File

@@ -4,6 +4,7 @@
"timestamp": "2022-10-12T18:00:00.000Z",
"operation_uuid": "0f657b36-4b0f-4a2d-9c2e-1dcd615e7d8c",
"suspend_timeout_seconds": -1,
"cluster": {
"cluster_id": "docker_compose",

View File

@@ -0,0 +1,396 @@
# Memo: Endpoint Persistent Unlogged Files Storage
Created on 2024-11-05
Implemented on N/A
## Summary
A design for a storage system that allows storage of files required to make
Neon's Endpoints have a better experience at or after a reboot.
## Motivation
Several systems inside PostgreSQL (and Neon) need some persistent storage for
optimal workings across reboots and restarts, but still work without.
Examples are the query-level statistics files of `pg_stat_statements` in
`pg_stat/pg_stat_statements.stat`, and `pg_prewarm`'s `autoprewarm.blocks`.
We need a storage system that can store and manage these files for each
Endpoint, without necessarily granting users access to an unlimited storage
device.
## Goals
- Store known files for Endpoints with reasonable persistence.
_Data loss in this service, while annoying and bad for UX, won't lose any
customer's data._
## Non Goals (if relevant)
- This storage system does not need branching, file versioning, or other such
features. The files are as ephemeral to the timeline of the data as the
Endpoints that host the data.
- This storage system does not need to store _all_ user files, only 'known'
user files.
- This storage system does not need to be hosted fully inside Computes.
_Instead, this will be a separate component similar to Pageserver,
SafeKeeper, the S3 proxy used for dynamically loaded extensions, etc._
## Impacted components
- Compute needs new code to load and store these files in its lifetime.
- Control Plane needs to consider this new storage system when signalling
the deletion of an Endpoint, Timeline, or Tenant.
- Control Plane needs to consider this new storage system when it resets
or re-assigns an endpoint's timeline/branch state.
A new service is created: the Endpoint Persistent Unlogged Files Storage
service. This could be integrated in e.g. Pageserver or Control Plane, or a
separately hosted service.
## Proposed implementation
Endpoint-related data files are managed by a newly designed service (which
optionally is integrated in an existing service like Pageserver or Control
Plane), which stores data directly into S3 or any blob storage of choice.
Upon deletion of the Endpoint, or reassignment of the endpoint to a different
branch, this ephemeral data is dropped: the data stored may not match the
state of the branch's data after reassignment, and on endpoint deletion the
data won't have any use to the user.
Compute gets credentials (JWT token with Tenant, Timeline & Endpoint claims)
which it can use to authenticate to this new service and retrieve and store
data associated with this endpoint. This limited scope reduces leaks of data
across endpoints and timeline resets, and limits the ability of endpoints to
mess with other endpoints' data.
The path of this endpoint data in S3 is initially as follows:
s3://<regional-epufs-bucket>/
tenants/
<hex-tenant-id>/
tenants/
<hex-timeline-id>/
endpoints/
<endpoint-id>/
pgdata/
<file_path_in_pgdatadir>
For other blob storages an equivalent or similar path can be constructed.
### Reliability, failure modes and corner cases (if relevant)
Reliability is important, but not critical to the workings of Neon. The data
stored in this service will, when lost, reduce performance, but won't be a
cause of permanent data loss - only operational metadata is stored.
Most, if not all, blob storage services have sufficiently high persistence
guarantees to cater our need for persistence and uptime. The only concern with
blob storages is that the access latency is generally higher than local disk,
but for the object types stored (cache state, ...) I don't think this will be
much of an issue.
### Interaction/Sequence diagram (if relevant)
In these diagrams you can replace S3 with any persistent storage device of
choice, but S3 is chosen as representative name: The well-known and short name
of AWS' blob storage. Azure Blob Storage should work too, but it has a much
longer name making it less practical for the diagrams.
Write data:
```http
POST /tenants/<tenant-id>/timelines/<tl-id>/endpoints/<endpoint-id>/pgdata/<the-pgdata-path>
Host: epufs.svc.neon.local
<<<
200 OK
{
"version": "<opaque>", # opaque file version token, changes when the file contents change
"size": <bytes>,
}
```
```mermaid
sequenceDiagram
autonumber
participant co as Compute
participant ep as EPUFS
participant s3 as Blob Storage
co-->ep: Connect with credentials
co->>+ep: Store Unlogged Persistent File
opt is authenticated
ep->>s3: Write UPF to S3
end
ep->>-co: OK / Failure / Auth Failure
co-->ep: Cancel connection
```
Read data: (optional with cache-relevant request parameters, e.g. If-Modified-Since)
```http
GET /tenants/<tenant-id>/timelines/<tl-id>/endpoints/<endpoint-id>/pgdata/<the-pgdata-path>
Host: epufs.svc.neon.local
<<<
200 OK
<file data>
```
```mermaid
sequenceDiagram
autonumber
participant co as Compute
participant ep as EPUFS
participant s3 as Blob Storage
co->>+ep: Read Unlogged Persistent File
opt is authenticated
ep->>+s3: Request UPF from storage
s3->>-ep: Receive UPF from storage
end
ep->>-co: OK(response) / Failure(storage, auth, ...)
```
Compute Startup:
```mermaid
sequenceDiagram
autonumber
participant co as Compute
participant ps as Pageserver
participant ep as EPUFS
participant es as Extension server
note over co: Bind endpoint ep-xxx
par Get basebackup
co->>+ps: Request basebackup @ LSN
ps-)ps: Construct basebackup
ps->>-co: Receive basebackup TAR @ LSN
and Get startup-critical Unlogged Persistent Files
co->>+ep: Get all UPFs of endpoint ep-xxx
ep-)ep: Retrieve and gather all UPFs
ep->>-co: TAR of UPFs
and Get startup-critical extensions
loop For every startup-critical extension
co->>es: Get critical extension
es->>co: Receive critical extension
end
end
note over co: Start compute
```
CPlane ops:
```http
DELETE /tenants/<tenant-id>/timelines/<timeline-id>/endpoints/<endpoint-id>
Host: epufs.svc.neon.local
<<<
200 OK
{
"tenant": "<tenant-id>",
"timeline": "<timeline-id>",
"endpoint": "<endpoint-id>",
"deleted": {
"files": <count>,
"bytes": <count>,
},
}
```
```http
DELETE /tenants/<tenant-id>/timelines/<timeline-id>
Host: epufs.svc.neon.local
<<<
200 OK
{
"tenant": "<tenant-id>",
"timeline": "<timeline-id>",
"deleted": {
"files": <count>,
"bytes": <count>,
},
}
```
```http
DELETE /tenants/<tenant-id>
Host: epufs.svc.neon.local
<<<
200 OK
{
"tenant": "<tenant-id>",
"deleted": {
"files": <count>,
"bytes": <count>,
},
}
```
```mermaid
sequenceDiagram
autonumber
participant cp as Control Plane
participant ep as EPUFS
participant s3 as Blob Storage
alt Tenant deleted
cp-)ep: Tenant deleted
loop For every object associated with removed tenant
ep->>s3: Remove data of deleted tenant from Storage
end
opt
ep-)cp: Tenant cleanup complete
end
alt Timeline deleted
cp-)ep: Timeline deleted
loop For every object associated with removed timeline
ep->>s3: Remove data of deleted timeline from Storage
end
opt
ep-)cp: Timeline cleanup complete
end
else Endpoint reassigned or removed
cp->>+ep: Endpoint reassigned
loop For every object associated with reassigned/removed endpoint
ep->>s3: Remove data from Storage
end
ep->>-cp: Cleanup complete
end
```
### Scalability (if relevant)
Provisionally: As this service is going to be part of compute startup, this
service should be able to quickly respond to all requests. Therefore this
service is deployed to every AZ we host Computes in, and Computes communicate
(generally) only to the EPUFS endpoint of the AZ they're hosted in.
Local caching of frequently restarted endpoints' data or metadata may be
needed for best performance. However, due to the regional nature of stored
data but zonal nature of the service deployment, we should be careful when we
implement any local caching, as it is possible that computes in AZ 1 will
update data originally written and thus cached by AZ 2. Cache version tests
and invalidation is therefore required if we want to roll out caching to this
service, which is too broad a scope for an MVC. This is why caching is left
out of scope for this RFC, and should be considered separately after this RFC
is implemented.
### Security implications (if relevant)
This service must be able to authenticate users at least by Tenant ID,
Timeline ID and Endpoint ID. This will use the existing JWT infrastructure of
Compute, which will be upgraded to the extent needed to support Timeline- and
Endpoint-based claims.
The service requires unlimited access to (a prefix of) a blob storage bucket,
and thus must be hosted outside the Compute VM sandbox.
A service that generates pre-signed request URLs for Compute to download the
data from that URL is likely problematic, too: Compute would be able to write
unlimited data to the bucket, or exfiltrate this signed URL to get read/write
access to specific objects in this bucket, which would still effectively give
users access to the S3 bucket (but with improved access logging).
There may be a use case for transferring data associated with one endpoint to
another endpoint (e.g. to make one endpoint warm its caches with the state of
another endpoint), but that's not currently in scope, and specific needs may
be solved through out-of-line communication of data or pre-signed URLs.
### Unresolved questions (if relevant)
Caching of files is not in the implementation scope of the document, but
should at some future point be considered to maximize performance.
## Alternative implementation (if relevant)
Several ideas have come up to solve this issue:
### Use AUXfile
One prevalent idea was to WAL-log the files using our AUXfile mechanism.
Benefits:
+ We already have this storage mechanism
Demerits:
- It isn't available on read replicas
- Additional WAL will be consumed during shutdown and after the shutdown
checkpoint, which needs PG modifications to work without panics.
- It increases the data we need to manage in our versioned storage, thus
causing higher storage costs with higher retention due to duplication at
the storage layer.
### Sign URLs for read/write operations, instead of proxying them
Benefits:
+ The service can be implemented with a much reduced IO budget
Demerits:
- Users could get access to these signed credentials
- Not all blob storage services may implement URL signing
### Give endpoints each their own directly accessed block volume
Benefits:
+ Easier to integrate for PostgreSQL
Demerits:
- Little control on data size and contents
- Potentially problematic as we'd need to store data all across the pgdata
directory.
- EBS is not a good candidate
- Attaches in 10s of seconds, if not more; i.e. too cold to start
- Shared EBS volumes are a no-go, as you'd have to schedule the endpoint
with users of the same EBS volumes, which can't work with VM migration
- EBS storage costs are very high (>80$/kilotenant when using a
volume/tenant)
- EBS volumes can't be mounted across AZ boundaries
- Bucket per endpoint is unfeasible
- S3 buckets are priced at $20/month per 1k, which we could better spend
on developers.
- Allocating service accounts takes time (100s of ms), and service accounts
are a limited resource, too; so they're not a good candidate to allocate
on a per-endpoint basis.
- Giving credentials limited to prefix has similar issues as the pre-signed
URL approach.
- Bucket DNS lookup will fill DNS caches and put pressure on DNS lookup
much more than our current systems would.
- Volumes bound by hypervisor are unlikely
- This requires significant investment and increased software on the
hypervisor.
- It is unclear if we can attach volumes after boot, i.e. for pooled
instances.
### Put the files into a table
Benefits:
+ Mostly already available in PostgreSQL
Demerits:
- Uses WAL
- Can't be used after shutdown checkpoint
- Needs a RW endpoint, and table & catalog access to write to this data
- Gets hit with DB size limitations
- Depending on user acces:
- Inaccessible:
The user doesn't have control over database size caused by
these systems.
- Accessible:
The user can corrupt these files and cause the system to crash while
user-corrupted files are present, thus increasing on-call overhead.
## Definition of Done (if relevant)
This project is done if we have:
- One S3 bucket equivalent per region, which stores this per-endpoint data.
- A new service endpoint in at least every AZ, which indirectly grants
endpoints access to the data stored for these endpoints in these buckets.
- Compute writes & reads temp-data at shutdown and startup, respectively, for
at least the pg_prewarm or lfc_prewarm state files.
- Cleanup of endpoint data is triggered when the endpoint is deleted or is
detached from its current timeline.

View File

@@ -12,6 +12,7 @@ jsonwebtoken.workspace = true
serde.workspace = true
serde_json.workspace = true
regex.workspace = true
url.workspace = true
utils = { path = "../utils" }
remote_storage = { version = "0.1", path = "../remote_storage/" }

View File

@@ -4,11 +4,14 @@
//! provide it by calling the compute_ctl's `/compute_ctl` endpoint, or
//! compute_ctl can fetch it by calling the control plane's API.
use std::collections::HashMap;
use std::fmt::Display;
use anyhow::anyhow;
use indexmap::IndexMap;
use regex::Regex;
use remote_storage::RemotePath;
use serde::{Deserialize, Serialize};
use url::Url;
use utils::id::{TenantId, TimelineId};
use utils::lsn::Lsn;
@@ -181,6 +184,11 @@ pub struct ComputeSpec {
/// Download LFC state from endpoint_storage and pass it to Postgres on startup
#[serde(default)]
pub autoprewarm: bool,
/// Suspend timeout in seconds.
///
/// We use this value to derive other values, such as the installed extensions metric.
pub suspend_timeout_seconds: i64,
}
/// Feature flag to signal `compute_ctl` to enable certain experimental functionality.
@@ -429,6 +437,47 @@ pub struct JwksSettings {
pub jwt_audience: Option<String>,
}
/// Protocol used to connect to a Pageserver. Parsed from the connstring scheme.
#[derive(Clone, Copy, Debug, Default)]
pub enum PageserverProtocol {
/// The original protocol based on libpq and COPY. Uses postgresql:// or postgres:// scheme.
#[default]
Libpq,
/// A newer, gRPC-based protocol. Uses grpc:// scheme.
Grpc,
}
impl PageserverProtocol {
/// Parses the protocol from a connstring scheme. Defaults to Libpq if no scheme is given.
/// Errors if the connstring is an invalid URL.
pub fn from_connstring(connstring: &str) -> anyhow::Result<Self> {
let scheme = match Url::parse(connstring) {
Ok(url) => url.scheme().to_lowercase(),
Err(url::ParseError::RelativeUrlWithoutBase) => return Ok(Self::default()),
Err(err) => return Err(anyhow!("invalid connstring URL: {err}")),
};
match scheme.as_str() {
"postgresql" | "postgres" => Ok(Self::Libpq),
"grpc" => Ok(Self::Grpc),
scheme => Err(anyhow!("invalid protocol scheme: {scheme}")),
}
}
/// Returns the URL scheme for the protocol, for use in connstrings.
pub fn scheme(&self) -> &'static str {
match self {
Self::Libpq => "postgresql",
Self::Grpc => "grpc",
}
}
}
impl Display for PageserverProtocol {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.write_str(self.scheme())
}
}
#[cfg(test)]
mod tests {
use std::fs::File;

View File

@@ -3,6 +3,7 @@
"timestamp": "2021-05-23T18:25:43.511Z",
"operation_uuid": "0f657b36-4b0f-4a2d-9c2e-1dcd615e7d8b",
"suspend_timeout_seconds": 3600,
"cluster": {
"cluster_id": "test-cluster-42",

View File

@@ -18,6 +18,8 @@ bytes.workspace = true
byteorder.workspace = true
utils.workspace = true
postgres_ffi_types.workspace = true
postgres_versioninfo.workspace = true
posthog_client_lite.workspace = true
enum-map.workspace = true
strum.workspace = true
strum_macros.workspace = true
@@ -28,12 +30,13 @@ humantime-serde.workspace = true
chrono = { workspace = true, features = ["serde"] }
itertools.workspace = true
storage_broker.workspace = true
camino = {workspace = true, features = ["serde1"]}
camino = { workspace = true, features = ["serde1"] }
remote_storage.workspace = true
postgres_backend.workspace = true
nix = {workspace = true, optional = true}
nix = { workspace = true, optional = true }
reqwest.workspace = true
rand.workspace = true
tracing.workspace = true
tracing-utils.workspace = true
once_cell.workspace = true

View File

@@ -4,6 +4,7 @@ use camino::Utf8PathBuf;
mod tests;
use const_format::formatcp;
use posthog_client_lite::PostHogClientConfig;
pub const DEFAULT_PG_LISTEN_PORT: u16 = 64000;
pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");
pub const DEFAULT_HTTP_LISTEN_PORT: u16 = 9898;
@@ -63,25 +64,66 @@ impl Display for NodeMetadata {
}
}
/// PostHog integration config.
/// PostHog integration config. This is used in pageserver, storcon, and neon_local.
/// Ensure backward compatibility when adding new fields.
#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
pub struct PostHogConfig {
/// PostHog project ID
pub project_id: String,
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
pub project_id: Option<String>,
/// Server-side (private) API key
pub server_api_key: String,
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
pub server_api_key: Option<String>,
/// Client-side (public) API key
pub client_api_key: String,
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
pub client_api_key: Option<String>,
/// Private API URL
pub private_api_url: String,
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
pub private_api_url: Option<String>,
/// Public API URL
pub public_api_url: String,
/// Refresh interval for the feature flag spec
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
pub public_api_url: Option<String>,
/// Refresh interval for the feature flag spec.
/// The storcon will push the feature flag spec to the pageserver. If the pageserver does not receive
/// the spec for `refresh_interval`, it will fetch the spec from the PostHog API.
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
#[serde(with = "humantime_serde")]
pub refresh_interval: Option<Duration>,
}
impl PostHogConfig {
pub fn try_into_posthog_config(self) -> Result<PostHogClientConfig, &'static str> {
let Some(project_id) = self.project_id else {
return Err("project_id is required");
};
let Some(server_api_key) = self.server_api_key else {
return Err("server_api_key is required");
};
let Some(client_api_key) = self.client_api_key else {
return Err("client_api_key is required");
};
let Some(private_api_url) = self.private_api_url else {
return Err("private_api_url is required");
};
let Some(public_api_url) = self.public_api_url else {
return Err("public_api_url is required");
};
Ok(PostHogClientConfig {
project_id,
server_api_key,
client_api_key,
private_api_url,
public_api_url,
})
}
}
/// `pageserver.toml`
///
/// We use serde derive with `#[serde(default)]` to generate a deserializer
@@ -367,6 +409,9 @@ pub struct BasebackupCacheConfig {
// TODO(diko): support max_entry_size_bytes.
// pub max_entry_size_bytes: u64,
pub max_size_entries: usize,
/// Size of the channel used to send prepare requests to the basebackup cache worker.
/// If exceeded, new prepare requests will be dropped.
pub prepare_channel_size: usize,
}
impl Default for BasebackupCacheConfig {
@@ -375,7 +420,8 @@ impl Default for BasebackupCacheConfig {
cleanup_period: Duration::from_secs(60),
max_total_size_bytes: 1024 * 1024 * 1024, // 1 GiB
// max_entry_size_bytes: 16 * 1024 * 1024, // 16 MiB
max_size_entries: 1000,
max_size_entries: 10000,
prepare_channel_size: 100,
}
}
}

View File

@@ -546,6 +546,11 @@ pub struct TimelineImportRequest {
pub sk_set: Vec<NodeId>,
}
#[derive(serde::Serialize, serde::Deserialize, Clone)]
pub struct TimelineSafekeeperMigrateRequest {
pub new_sk_set: Vec<NodeId>,
}
#[cfg(test)]
mod test {
use serde_json;

View File

@@ -11,6 +11,7 @@ use std::time::{Duration, SystemTime};
#[cfg(feature = "testing")]
use camino::Utf8PathBuf;
use postgres_versioninfo::PgMajorVersion;
use serde::{Deserialize, Deserializer, Serialize, Serializer};
use serde_with::serde_as;
pub use utilization::PageserverUtilization;
@@ -20,7 +21,9 @@ use utils::{completion, serde_system_time};
use crate::config::Ratio;
use crate::key::{CompactKey, Key};
use crate::shard::{DEFAULT_STRIPE_SIZE, ShardCount, ShardStripeSize, TenantShardId};
use crate::shard::{
DEFAULT_STRIPE_SIZE, ShardCount, ShardIdentity, ShardStripeSize, TenantShardId,
};
/// The state of a tenant in this pageserver.
///
@@ -398,7 +401,7 @@ pub enum TimelineCreateRequestMode {
// inherits the ancestor's pg_version. Earlier code wasn't
// using a flattened enum, so, it was an accepted field, and
// we continue to accept it by having it here.
pg_version: Option<u32>,
pg_version: Option<PgMajorVersion>,
#[serde(default, skip_serializing_if = "std::ops::Not::not")]
read_only: bool,
},
@@ -410,7 +413,7 @@ pub enum TimelineCreateRequestMode {
Bootstrap {
#[serde(default)]
existing_initdb_timeline_id: Option<TimelineId>,
pg_version: Option<u32>,
pg_version: Option<PgMajorVersion>,
},
}
@@ -474,7 +477,7 @@ pub struct TenantShardSplitResponse {
}
/// Parameters that apply to all shards in a tenant. Used during tenant creation.
#[derive(Serialize, Deserialize, Debug)]
#[derive(Clone, Copy, Serialize, Deserialize, Debug)]
#[serde(deny_unknown_fields)]
pub struct ShardParameters {
pub count: ShardCount,
@@ -496,6 +499,15 @@ impl Default for ShardParameters {
}
}
impl From<ShardIdentity> for ShardParameters {
fn from(identity: ShardIdentity) -> Self {
Self {
count: identity.count,
stripe_size: identity.stripe_size,
}
}
}
#[derive(Debug, Default, Clone, Eq, PartialEq)]
pub enum FieldPatch<T> {
Upsert(T),
@@ -1573,7 +1585,7 @@ pub struct TimelineInfo {
pub last_received_msg_lsn: Option<Lsn>,
/// the timestamp (in microseconds) of the last received message
pub last_received_msg_ts: Option<u128>,
pub pg_version: u32,
pub pg_version: PgMajorVersion,
pub state: TimelineState,

View File

@@ -37,6 +37,7 @@ use std::hash::{Hash, Hasher};
pub use ::utils::shard::*;
use postgres_ffi_types::forknum::INIT_FORKNUM;
use serde::{Deserialize, Serialize};
use utils::critical;
use crate::key::Key;
use crate::models::ShardParameters;
@@ -179,7 +180,7 @@ impl ShardIdentity {
/// For use when creating ShardIdentity instances for new shards, where a creation request
/// specifies the ShardParameters that apply to all shards.
pub fn from_params(number: ShardNumber, params: &ShardParameters) -> Self {
pub fn from_params(number: ShardNumber, params: ShardParameters) -> Self {
Self {
number,
count: params.count,
@@ -188,6 +189,17 @@ impl ShardIdentity {
}
}
/// Asserts that the given shard identities are equal. Changes to shard parameters will likely
/// result in data corruption.
pub fn assert_equal(&self, other: ShardIdentity) {
if self != &other {
// TODO: for now, we're conservative and just log errors in production. Turn this into a
// real assertion when we're confident it doesn't misfire, and also reject requests that
// attempt to change it with an error response.
critical!("shard identity mismatch: {self:?} != {other:?}");
}
}
fn is_broken(&self) -> bool {
self.layout == LAYOUT_BROKEN
}

View File

@@ -19,6 +19,7 @@ serde.workspace = true
postgres_ffi_types.workspace = true
utils.workspace = true
tracing.workspace = true
postgres_versioninfo.workspace = true
[dev-dependencies]
env_logger.workspace = true

View File

@@ -4,6 +4,7 @@ use criterion::{Bencher, Criterion, criterion_group, criterion_main};
use postgres_ffi::v17::wal_generator::LogicalMessageGenerator;
use postgres_ffi::v17::waldecoder_handler::WalStreamDecoderHandler;
use postgres_ffi::waldecoder::WalStreamDecoder;
use postgres_versioninfo::PgMajorVersion;
use pprof::criterion::{Output, PProfProfiler};
use utils::lsn::Lsn;
@@ -32,7 +33,7 @@ fn bench_complete_record(c: &mut Criterion) {
let value_size = LogicalMessageGenerator::make_value_size(size, PREFIX);
let value = vec![1; value_size];
let mut decoder = WalStreamDecoder::new(Lsn(0), 170000);
let mut decoder = WalStreamDecoder::new(Lsn(0), PgMajorVersion::PG17);
let msg = LogicalMessageGenerator::new(PREFIX, &value)
.next()
.unwrap()

View File

@@ -14,6 +14,8 @@ use bytes::Bytes;
use utils::bin_ser::SerializeError;
use utils::lsn::Lsn;
pub use postgres_versioninfo::PgMajorVersion;
macro_rules! postgres_ffi {
($version:ident) => {
#[path = "."]
@@ -91,21 +93,22 @@ macro_rules! dispatch_pgversion {
$version => $code,
default = $invalid_pgver_handling,
pgversions = [
14 : v14,
15 : v15,
16 : v16,
17 : v17,
$crate::PgMajorVersion::PG14 => v14,
$crate::PgMajorVersion::PG15 => v15,
$crate::PgMajorVersion::PG16 => v16,
$crate::PgMajorVersion::PG17 => v17,
]
)
};
($pgversion:expr => $code:expr,
default = $default:expr,
pgversions = [$($sv:literal : $vsv:ident),+ $(,)?]) => {
match ($pgversion) {
pgversions = [$($sv:pat => $vsv:ident),+ $(,)?]) => {
match ($pgversion.clone().into()) {
$($sv => {
use $crate::$vsv as pgv;
$code
},)+
#[allow(unreachable_patterns)]
_ => {
$default
}
@@ -179,9 +182,9 @@ macro_rules! enum_pgversion {
$($variant ( $crate::$md::$t )),+
}
impl self::$name {
pub fn pg_version(&self) -> u32 {
pub fn pg_version(&self) -> PgMajorVersion {
enum_pgversion_dispatch!(self, $name, _ign, {
pgv::bindings::PG_MAJORVERSION_NUM
pgv::bindings::MY_PGVERSION
})
}
}
@@ -195,15 +198,15 @@ macro_rules! enum_pgversion {
};
{name = $name:ident,
path = $p:ident,
typ = $t:ident,
$(typ = $t:ident,)?
pgversions = [$($variant:ident : $md:ident),+ $(,)?]} => {
pub enum $name {
$($variant ($crate::$md::$p::$t)),+
$($variant $(($crate::$md::$p::$t))?),+
}
impl $name {
pub fn pg_version(&self) -> u32 {
pub fn pg_version(&self) -> PgMajorVersion {
enum_pgversion_dispatch!(self, $name, _ign, {
pgv::bindings::PG_MAJORVERSION_NUM
pgv::bindings::MY_PGVERSION
})
}
}
@@ -249,22 +252,21 @@ pub use v14::xlog_utils::{
try_from_pg_timestamp,
};
pub fn bkpimage_is_compressed(bimg_info: u8, version: u32) -> bool {
pub fn bkpimage_is_compressed(bimg_info: u8, version: PgMajorVersion) -> bool {
dispatch_pgversion!(version, pgv::bindings::bkpimg_is_compressed(bimg_info))
}
pub fn generate_wal_segment(
segno: u64,
system_id: u64,
pg_version: u32,
pg_version: PgMajorVersion,
lsn: Lsn,
) -> Result<Bytes, SerializeError> {
assert_eq!(segno, lsn.segment_number(WAL_SEGMENT_SIZE));
dispatch_pgversion!(
pg_version,
pgv::xlog_utils::generate_wal_segment(segno, system_id, lsn),
Err(SerializeError::BadInput)
pgv::xlog_utils::generate_wal_segment(segno, system_id, lsn)
)
}
@@ -272,7 +274,7 @@ pub fn generate_pg_control(
pg_control_bytes: &[u8],
checkpoint_bytes: &[u8],
lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<(Bytes, u64, bool)> {
dispatch_pgversion!(
pg_version,
@@ -352,6 +354,7 @@ pub fn fsm_logical_to_physical(addr: BlockNumber) -> BlockNumber {
pub mod waldecoder {
use std::num::NonZeroU32;
use crate::PgMajorVersion;
use bytes::{Buf, Bytes, BytesMut};
use thiserror::Error;
use utils::lsn::Lsn;
@@ -369,7 +372,7 @@ pub mod waldecoder {
pub struct WalStreamDecoder {
pub lsn: Lsn,
pub pg_version: u32,
pub pg_version: PgMajorVersion,
pub inputbuf: BytesMut,
pub state: State,
}
@@ -382,7 +385,7 @@ pub mod waldecoder {
}
impl WalStreamDecoder {
pub fn new(lsn: Lsn, pg_version: u32) -> WalStreamDecoder {
pub fn new(lsn: Lsn, pg_version: PgMajorVersion) -> WalStreamDecoder {
WalStreamDecoder {
lsn,
pg_version,

View File

@@ -1,3 +1,7 @@
use crate::PgMajorVersion;
pub const MY_PGVERSION: PgMajorVersion = PgMajorVersion::PG14;
pub const XLOG_DBASE_CREATE: u8 = 0x00;
pub const XLOG_DBASE_DROP: u8 = 0x10;

View File

@@ -1,3 +1,7 @@
use crate::PgMajorVersion;
pub const MY_PGVERSION: PgMajorVersion = PgMajorVersion::PG15;
pub const XACT_XINFO_HAS_DROPPED_STATS: u32 = 1u32 << 8;
pub const XLOG_DBASE_CREATE_FILE_COPY: u8 = 0x00;

View File

@@ -1,3 +1,7 @@
use crate::PgMajorVersion;
pub const MY_PGVERSION: PgMajorVersion = PgMajorVersion::PG16;
pub const XACT_XINFO_HAS_DROPPED_STATS: u32 = 1u32 << 8;
pub const XLOG_DBASE_CREATE_FILE_COPY: u8 = 0x00;

View File

@@ -1,3 +1,7 @@
use crate::PgMajorVersion;
pub const MY_PGVERSION: PgMajorVersion = PgMajorVersion::PG17;
pub const XACT_XINFO_HAS_DROPPED_STATS: u32 = 1u32 << 8;
pub const XLOG_DBASE_CREATE_FILE_COPY: u8 = 0x00;

View File

@@ -9,8 +9,8 @@ use utils::bin_ser::DeserializeError;
use utils::lsn::Lsn;
use crate::{
BLCKSZ, BlockNumber, MultiXactId, MultiXactOffset, MultiXactStatus, Oid, RepOriginId,
TimestampTz, TransactionId, XLOG_SIZE_OF_XLOG_RECORD, XLogRecord, pg_constants,
BLCKSZ, BlockNumber, MultiXactId, MultiXactOffset, MultiXactStatus, Oid, PgMajorVersion,
RepOriginId, TimestampTz, TransactionId, XLOG_SIZE_OF_XLOG_RECORD, XLogRecord, pg_constants,
};
#[repr(C)]
@@ -199,20 +199,17 @@ impl DecodedWALRecord {
/// Check if this WAL record represents a legacy "copy" database creation, which populates new relations
/// by reading other existing relations' data blocks. This is more complex to apply than new-style database
/// creations which simply include all the desired blocks in the WAL, so we need a helper function to detect this case.
pub fn is_dbase_create_copy(&self, pg_version: u32) -> bool {
pub fn is_dbase_create_copy(&self, pg_version: PgMajorVersion) -> bool {
if self.xl_rmid == pg_constants::RM_DBASE_ID {
let info = self.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
match pg_version {
14 => {
PgMajorVersion::PG14 => {
// Postgres 14 database creations are always the legacy kind
info == crate::v14::bindings::XLOG_DBASE_CREATE
}
15 => info == crate::v15::bindings::XLOG_DBASE_CREATE_FILE_COPY,
16 => info == crate::v16::bindings::XLOG_DBASE_CREATE_FILE_COPY,
17 => info == crate::v17::bindings::XLOG_DBASE_CREATE_FILE_COPY,
_ => {
panic!("Unsupported postgres version {pg_version}")
}
PgMajorVersion::PG15 => info == crate::v15::bindings::XLOG_DBASE_CREATE_FILE_COPY,
PgMajorVersion::PG16 => info == crate::v16::bindings::XLOG_DBASE_CREATE_FILE_COPY,
PgMajorVersion::PG17 => info == crate::v17::bindings::XLOG_DBASE_CREATE_FILE_COPY,
}
} else {
false
@@ -248,7 +245,7 @@ impl DecodedWALRecord {
pub fn decode_wal_record(
record: Bytes,
decoded: &mut DecodedWALRecord,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<()> {
let mut rnode_spcnode: u32 = 0;
let mut rnode_dbnode: u32 = 0;
@@ -1106,9 +1103,9 @@ pub struct XlClogTruncate {
}
impl XlClogTruncate {
pub fn decode(buf: &mut Bytes, pg_version: u32) -> XlClogTruncate {
pub fn decode(buf: &mut Bytes, pg_version: PgMajorVersion) -> XlClogTruncate {
XlClogTruncate {
pageno: if pg_version < 17 {
pageno: if pg_version < PgMajorVersion::PG17 {
buf.get_u32_le()
} else {
buf.get_u64_le() as u32

View File

@@ -11,9 +11,9 @@ use super::super::waldecoder::WalStreamDecoder;
use super::bindings::{
CheckPoint, ControlFileData, DBState_DB_SHUTDOWNED, FullTransactionId, TimeLineID, TimestampTz,
XLogLongPageHeaderData, XLogPageHeaderData, XLogRecPtr, XLogRecord, XLogSegNo, XLOG_PAGE_MAGIC,
MY_PGVERSION
};
use super::wal_generator::LogicalMessageGenerator;
use super::PG_MAJORVERSION;
use crate::pg_constants;
use crate::PG_TLI;
use crate::{uint32, uint64, Oid};
@@ -233,7 +233,7 @@ pub fn find_end_of_wal(
let mut result = start_lsn;
let mut curr_lsn = start_lsn;
let mut buf = [0u8; XLOG_BLCKSZ];
let pg_version = PG_MAJORVERSION[1..3].parse::<u32>().unwrap();
let pg_version = MY_PGVERSION;
debug!("find_end_of_wal PG_VERSION: {}", pg_version);
let mut decoder = WalStreamDecoder::new(start_lsn, pg_version);

View File

@@ -4,6 +4,7 @@ use std::str::FromStr;
use anyhow::*;
use clap::{Arg, ArgMatches, Command, value_parser};
use postgres::Client;
use postgres_ffi::PgMajorVersion;
use wal_craft::*;
fn main() -> Result<()> {
@@ -48,7 +49,7 @@ fn main() -> Result<()> {
Some(("with-initdb", arg_matches)) => {
let cfg = Conf {
pg_version: *arg_matches
.get_one::<u32>("pg-version")
.get_one::<PgMajorVersion>("pg-version")
.context("'pg-version' is required")?,
pg_distrib_dir: arg_matches
.get_one::<PathBuf>("pg-distrib-dir")

View File

@@ -9,8 +9,8 @@ use log::*;
use postgres::Client;
use postgres::types::PgLsn;
use postgres_ffi::{
WAL_SEGMENT_SIZE, XLOG_BLCKSZ, XLOG_SIZE_OF_XLOG_LONG_PHD, XLOG_SIZE_OF_XLOG_RECORD,
XLOG_SIZE_OF_XLOG_SHORT_PHD,
PgMajorVersion, WAL_SEGMENT_SIZE, XLOG_BLCKSZ, XLOG_SIZE_OF_XLOG_LONG_PHD,
XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD,
};
macro_rules! xlog_utils_test {
@@ -29,7 +29,7 @@ macro_rules! xlog_utils_test {
postgres_ffi::for_all_postgres_versions! { xlog_utils_test }
pub struct Conf {
pub pg_version: u32,
pub pg_version: PgMajorVersion,
pub pg_distrib_dir: PathBuf,
pub datadir: PathBuf,
}
@@ -52,11 +52,7 @@ impl Conf {
pub fn pg_distrib_dir(&self) -> anyhow::Result<PathBuf> {
let path = self.pg_distrib_dir.clone();
#[allow(clippy::manual_range_patterns)]
match self.pg_version {
14 | 15 | 16 | 17 => Ok(path.join(format!("v{}", self.pg_version))),
_ => bail!("Unsupported postgres version: {}", self.pg_version),
}
Ok(path.join(self.pg_version.v_str()))
}
fn pg_bin_dir(&self) -> anyhow::Result<PathBuf> {

View File

@@ -24,7 +24,7 @@ fn init_logging() {
fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
use crate::*;
let pg_version = PG_MAJORVERSION[1..3].parse::<u32>().unwrap();
let pg_version = MY_PGVERSION;
// Craft some WAL
let top_path = PathBuf::from(env!("CARGO_MANIFEST_DIR"))

View File

@@ -9,4 +9,5 @@ anyhow.workspace = true
tokio.workspace = true
camino.workspace = true
thiserror.workspace = true
postgres_versioninfo.workspace = true
workspace_hack = { version = "0.1", path = "../../workspace_hack" }

View File

@@ -7,12 +7,13 @@
use std::fmt;
use camino::Utf8Path;
use postgres_versioninfo::PgMajorVersion;
pub struct RunInitdbArgs<'a> {
pub superuser: &'a str,
pub locale: &'a str,
pub initdb_bin: &'a Utf8Path,
pub pg_version: u32,
pub pg_version: PgMajorVersion,
pub library_search_path: &'a Utf8Path,
pub pgdata: &'a Utf8Path,
}
@@ -79,12 +80,16 @@ pub async fn do_run_initdb(args: RunInitdbArgs<'_>) -> Result<(), Error> {
.stderr(std::process::Stdio::piped());
// Before version 14, only the libc provide was available.
if pg_version > 14 {
if pg_version > PgMajorVersion::PG14 {
// Version 17 brought with it a builtin locale provider which only provides
// C and C.UTF-8. While being safer for collation purposes since it is
// guaranteed to be consistent throughout a major release, it is also more
// performant.
let locale_provider = if pg_version >= 17 { "builtin" } else { "libc" };
let locale_provider = if pg_version >= PgMajorVersion::PG17 {
"builtin"
} else {
"libc"
};
initdb_command.args(["--locale-provider", locale_provider]);
}

View File

@@ -0,0 +1,12 @@
[package]
name = "postgres_versioninfo"
version = "0.1.0"
edition = "2024"
license.workspace = true
[dependencies]
anyhow.workspace = true
thiserror.workspace = true
serde.workspace = true
serde_repr.workspace = true
workspace_hack = { version = "0.1", path = "../../workspace_hack" }

View File

@@ -0,0 +1,175 @@
use serde::{Deserialize, Deserializer, Serialize, Serializer};
use serde_repr::{Deserialize_repr, Serialize_repr};
use std::fmt::{Display, Formatter};
use std::str::FromStr;
/// An enum with one variant for each major version of PostgreSQL that we support.
///
#[derive(Debug, Clone, Copy, Ord, PartialOrd, Eq, PartialEq, Deserialize_repr, Serialize_repr)]
#[repr(u32)]
pub enum PgMajorVersion {
PG14 = 14,
PG15 = 15,
PG16 = 16,
PG17 = 17,
// !!! When you add a new PgMajorVersion, don't forget to update PgMajorVersion::ALL
}
/// A full PostgreSQL version ID, in MMmmbb numerical format (Major/minor/bugfix)
#[derive(Debug, Copy, Clone, Ord, PartialOrd, Eq, PartialEq)]
#[repr(transparent)]
pub struct PgVersionId(u32);
impl PgVersionId {
pub const UNKNOWN: PgVersionId = PgVersionId(0);
pub fn from_full_pg_version(version: u32) -> PgVersionId {
match version {
0 => PgVersionId(version), // unknown version
140000..180000 => PgVersionId(version),
_ => panic!("Invalid full PostgreSQL version ID {version}"),
}
}
}
impl Display for PgVersionId {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
u32::fmt(&self.0, f)
}
}
impl Serialize for PgVersionId {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: Serializer,
{
u32::serialize(&self.0, serializer)
}
}
impl<'de> Deserialize<'de> for PgVersionId {
fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
where
D: Deserializer<'de>,
{
u32::deserialize(deserializer).map(PgVersionId)
}
fn deserialize_in_place<D>(deserializer: D, place: &mut Self) -> Result<(), D::Error>
where
D: Deserializer<'de>,
{
u32::deserialize_in_place(deserializer, &mut place.0)
}
}
impl PgMajorVersion {
/// Get the numerical representation of the represented Major Version
pub const fn major_version_num(&self) -> u32 {
match self {
PgMajorVersion::PG14 => 14,
PgMajorVersion::PG15 => 15,
PgMajorVersion::PG16 => 16,
PgMajorVersion::PG17 => 17,
}
}
/// Get the contents of this version's PG_VERSION file.
///
/// The PG_VERSION file is used to determine the PostgreSQL version that currently
/// owns the data in a PostgreSQL data directory.
pub fn versionfile_string(&self) -> &'static str {
match self {
PgMajorVersion::PG14 => "14",
PgMajorVersion::PG15 => "15",
PgMajorVersion::PG16 => "16\x0A",
PgMajorVersion::PG17 => "17\x0A",
}
}
/// Get the v{version} string of this major PostgreSQL version.
///
/// Because this was hand-coded in various places, this was moved into a shared
/// implementation.
pub fn v_str(&self) -> String {
match self {
PgMajorVersion::PG14 => "v14",
PgMajorVersion::PG15 => "v15",
PgMajorVersion::PG16 => "v16",
PgMajorVersion::PG17 => "v17",
}
.to_string()
}
/// All currently supported major versions of PostgreSQL.
pub const ALL: &'static [PgMajorVersion] = &[
PgMajorVersion::PG14,
PgMajorVersion::PG15,
PgMajorVersion::PG16,
PgMajorVersion::PG17,
];
}
impl Display for PgMajorVersion {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
f.write_str(match self {
PgMajorVersion::PG14 => "PgMajorVersion::PG14",
PgMajorVersion::PG15 => "PgMajorVersion::PG15",
PgMajorVersion::PG16 => "PgMajorVersion::PG16",
PgMajorVersion::PG17 => "PgMajorVersion::PG17",
})
}
}
#[derive(Debug, thiserror::Error)]
#[allow(dead_code)]
pub struct InvalidPgVersion(u32);
impl Display for InvalidPgVersion {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
write!(f, "InvalidPgVersion({})", self.0)
}
}
impl TryFrom<PgVersionId> for PgMajorVersion {
type Error = InvalidPgVersion;
fn try_from(value: PgVersionId) -> Result<Self, Self::Error> {
Ok(match value.0 / 10000 {
14 => PgMajorVersion::PG14,
15 => PgMajorVersion::PG15,
16 => PgMajorVersion::PG16,
17 => PgMajorVersion::PG17,
_ => return Err(InvalidPgVersion(value.0)),
})
}
}
impl From<PgMajorVersion> for PgVersionId {
fn from(value: PgMajorVersion) -> Self {
PgVersionId((value as u32) * 10000)
}
}
#[derive(Debug, PartialEq, Eq, thiserror::Error)]
pub struct PgMajorVersionParseError(String);
impl Display for PgMajorVersionParseError {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
write!(f, "PgMajorVersionParseError({})", self.0)
}
}
impl FromStr for PgMajorVersion {
type Err = PgMajorVersionParseError;
fn from_str(s: &str) -> Result<Self, Self::Err> {
Ok(match s {
"14" => PgMajorVersion::PG14,
"15" => PgMajorVersion::PG15,
"16" => PgMajorVersion::PG16,
"17" => PgMajorVersion::PG17,
_ => return Err(PgMajorVersionParseError(s.to_string())),
})
}
}

View File

@@ -1,17 +1,22 @@
//! A background loop that fetches feature flags from PostHog and updates the feature store.
use std::{sync::Arc, time::Duration};
use std::{
sync::Arc,
time::{Duration, SystemTime},
};
use arc_swap::ArcSwap;
use tokio_util::sync::CancellationToken;
use tracing::{Instrument, info_span};
use crate::{CaptureEvent, FeatureStore, PostHogClient, PostHogClientConfig};
use crate::{
CaptureEvent, FeatureStore, LocalEvaluationResponse, PostHogClient, PostHogClientConfig,
};
/// A background loop that fetches feature flags from PostHog and updates the feature store.
pub struct FeatureResolverBackgroundLoop {
posthog_client: PostHogClient,
feature_store: ArcSwap<FeatureStore>,
feature_store: ArcSwap<(SystemTime, Arc<FeatureStore>)>,
cancel: CancellationToken,
}
@@ -19,11 +24,35 @@ impl FeatureResolverBackgroundLoop {
pub fn new(config: PostHogClientConfig, shutdown_pageserver: CancellationToken) -> Self {
Self {
posthog_client: PostHogClient::new(config),
feature_store: ArcSwap::new(Arc::new(FeatureStore::new())),
feature_store: ArcSwap::new(Arc::new((
SystemTime::UNIX_EPOCH,
Arc::new(FeatureStore::new()),
))),
cancel: shutdown_pageserver,
}
}
/// Update the feature store with a new feature flag spec bypassing the normal refresh loop.
pub fn update(&self, spec: String) -> anyhow::Result<()> {
let resp: LocalEvaluationResponse = serde_json::from_str(&spec)?;
self.update_feature_store_nofail(resp, "http_propagate");
Ok(())
}
fn update_feature_store_nofail(&self, resp: LocalEvaluationResponse, source: &'static str) {
let project_id = self.posthog_client.config.project_id.parse::<u64>().ok();
match FeatureStore::new_with_flags(resp.flags, project_id) {
Ok(feature_store) => {
self.feature_store
.store(Arc::new((SystemTime::now(), Arc::new(feature_store))));
tracing::info!("Feature flag updated from {}", source);
}
Err(e) => {
tracing::warn!("Cannot process feature flag spec from {}: {}", source, e);
}
}
}
pub fn spawn(
self: Arc<Self>,
handle: &tokio::runtime::Handle,
@@ -47,6 +76,17 @@ impl FeatureResolverBackgroundLoop {
_ = ticker.tick() => {}
_ = cancel.cancelled() => break
}
{
let last_update = this.feature_store.load().0;
if let Ok(elapsed) = last_update.elapsed() {
if elapsed < refresh_period {
tracing::debug!(
"Skipping feature flag refresh because it's too soon"
);
continue;
}
}
}
let resp = match this
.posthog_client
.get_feature_flags_local_evaluation()
@@ -58,16 +98,7 @@ impl FeatureResolverBackgroundLoop {
continue;
}
};
let project_id = this.posthog_client.config.project_id.parse::<u64>().ok();
match FeatureStore::new_with_flags(resp.flags, project_id) {
Ok(feature_store) => {
this.feature_store.store(Arc::new(feature_store));
tracing::info!("Feature flag updated");
}
Err(e) => {
tracing::warn!("Cannot process feature flag spec: {}", e);
}
}
this.update_feature_store_nofail(resp, "refresh_loop");
}
tracing::info!("PostHog feature resolver stopped");
}
@@ -92,6 +123,6 @@ impl FeatureResolverBackgroundLoop {
}
pub fn feature_store(&self) -> Arc<FeatureStore> {
self.feature_store.load_full()
self.feature_store.load().1.clone()
}
}

View File

@@ -544,17 +544,8 @@ impl PostHogClient {
self.config.server_api_key.starts_with("phs_")
}
/// Fetch the feature flag specs from the server.
///
/// This is unfortunately an undocumented API at:
/// - <https://posthog.com/docs/api/feature-flags#get-api-projects-project_id-feature_flags-local_evaluation>
/// - <https://posthog.com/docs/feature-flags/local-evaluation>
///
/// The handling logic in [`FeatureStore`] mostly follows the Python API implementation.
/// See `_compute_flag_locally` in <https://github.com/PostHog/posthog-python/blob/master/posthog/client.py>
pub async fn get_feature_flags_local_evaluation(
&self,
) -> anyhow::Result<LocalEvaluationResponse> {
/// Get the raw JSON spec, same as `get_feature_flags_local_evaluation` but without parsing.
pub async fn get_feature_flags_local_evaluation_raw(&self) -> anyhow::Result<String> {
// BASE_URL/api/projects/:project_id/feature_flags/local_evaluation
// with bearer token of self.server_api_key
// OR
@@ -588,7 +579,22 @@ impl PostHogClient {
body
));
}
Ok(serde_json::from_str(&body)?)
Ok(body)
}
/// Fetch the feature flag specs from the server.
///
/// This is unfortunately an undocumented API at:
/// - <https://posthog.com/docs/api/feature-flags#get-api-projects-project_id-feature_flags-local_evaluation>
/// - <https://posthog.com/docs/feature-flags/local-evaluation>
///
/// The handling logic in [`FeatureStore`] mostly follows the Python API implementation.
/// See `_compute_flag_locally` in <https://github.com/PostHog/posthog-python/blob/master/posthog/client.py>
pub async fn get_feature_flags_local_evaluation(
&self,
) -> Result<LocalEvaluationResponse, anyhow::Error> {
let raw = self.get_feature_flags_local_evaluation_raw().await?;
Ok(serde_json::from_str(&raw)?)
}
/// Capture an event. This will only be used to report the feature flag usage back to PostHog, though

View File

@@ -12,7 +12,9 @@ use tokio::net::TcpStream;
use crate::connect::connect;
use crate::connect_raw::{RawConnection, connect_raw};
use crate::tls::{MakeTlsConnect, TlsConnect};
use crate::connect_tls::connect_tls;
use crate::maybe_tls_stream::MaybeTlsStream;
use crate::tls::{MakeTlsConnect, TlsConnect, TlsStream};
use crate::{Client, Connection, Error};
/// TLS configuration.
@@ -238,7 +240,7 @@ impl Config {
connect(tls, self).await
}
pub async fn connect_raw<S, T>(
pub async fn tls_and_authenticate<S, T>(
&self,
stream: S,
tls: T,
@@ -247,7 +249,19 @@ impl Config {
S: AsyncRead + AsyncWrite + Unpin,
T: TlsConnect<S>,
{
connect_raw(stream, tls, self).await
let stream = connect_tls(stream, self.ssl_mode, tls).await?;
connect_raw(stream, self).await
}
pub async fn authenticate<S, T>(
&self,
stream: MaybeTlsStream<S, T>,
) -> Result<RawConnection<S, T>, Error>
where
S: AsyncRead + AsyncWrite + Unpin,
T: TlsStream + Unpin,
{
connect_raw(stream, self).await
}
}

View File

@@ -9,6 +9,7 @@ use crate::codec::BackendMessage;
use crate::config::Host;
use crate::connect_raw::connect_raw;
use crate::connect_socket::connect_socket;
use crate::connect_tls::connect_tls;
use crate::tls::{MakeTlsConnect, TlsConnect};
use crate::{Client, Config, Connection, Error, RawConnection};
@@ -44,13 +45,14 @@ where
T: TlsConnect<TcpStream>,
{
let socket = connect_socket(host_addr, host, port, config.connect_timeout).await?;
let stream = connect_tls(socket, config.ssl_mode, tls).await?;
let RawConnection {
stream,
parameters,
delayed_notice,
process_id,
secret_key,
} = connect_raw(socket, tls, config).await?;
} = connect_raw(stream, config).await?;
let socket_config = SocketConfig {
host_addr,

View File

@@ -16,9 +16,8 @@ use tokio_util::codec::Framed;
use crate::Error;
use crate::codec::{BackendMessage, BackendMessages, FrontendMessage, PostgresCodec};
use crate::config::{self, AuthKeys, Config};
use crate::connect_tls::connect_tls;
use crate::maybe_tls_stream::MaybeTlsStream;
use crate::tls::{TlsConnect, TlsStream};
use crate::tls::TlsStream;
pub struct StartupStream<S, T> {
inner: Framed<MaybeTlsStream<S, T>, PostgresCodec>,
@@ -87,16 +86,13 @@ pub struct RawConnection<S, T> {
}
pub async fn connect_raw<S, T>(
stream: S,
tls: T,
stream: MaybeTlsStream<S, T>,
config: &Config,
) -> Result<RawConnection<S, T::Stream>, Error>
) -> Result<RawConnection<S, T>, Error>
where
S: AsyncRead + AsyncWrite + Unpin,
T: TlsConnect<S>,
T: TlsStream + Unpin,
{
let stream = connect_tls(stream, config.ssl_mode, tls).await?;
let mut stream = StartupStream {
inner: Framed::new(stream, PostgresCodec),
buf: BackendMessages::empty(),

View File

@@ -10,6 +10,7 @@ const_format.workspace = true
serde.workspace = true
serde_json.workspace = true
postgres_ffi.workspace = true
postgres_versioninfo.workspace = true
pq_proto.workspace = true
tokio.workspace = true
utils.workspace = true

View File

@@ -8,6 +8,8 @@ pub mod membership;
/// Public API types
pub mod models;
pub use postgres_versioninfo::{PgMajorVersion, PgVersionId};
/// Consensus logical timestamp. Note: it is a part of sk control file.
pub type Term = u64;
/// With this term timeline is created initially. It
@@ -20,7 +22,7 @@ pub const INITIAL_TERM: Term = 0;
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct ServerInfo {
/// Postgres server version
pub pg_version: u32,
pub pg_version: PgVersionId,
pub system_id: SystemId,
pub wal_seg_size: u32,
}

View File

@@ -4,6 +4,7 @@ use std::net::SocketAddr;
use pageserver_api::shard::ShardIdentity;
use postgres_ffi::TimestampTz;
use postgres_versioninfo::PgVersionId;
use serde::{Deserialize, Serialize};
use tokio::time::Instant;
use utils::id::{NodeId, TenantId, TenantTimelineId, TimelineId};
@@ -23,8 +24,7 @@ pub struct TimelineCreateRequest {
pub tenant_id: TenantId,
pub timeline_id: TimelineId,
pub mconf: Configuration,
/// In the PG_VERSION_NUM macro format, like 140017.
pub pg_version: u32,
pub pg_version: PgVersionId,
pub system_id: Option<u64>,
// By default WAL_SEGMENT_SIZE
pub wal_seg_size: Option<u32>,
@@ -210,7 +210,7 @@ pub struct TimelineStatus {
}
/// Request to switch membership configuration.
#[derive(Serialize, Deserialize)]
#[derive(Clone, Serialize, Deserialize)]
#[serde(transparent)]
pub struct TimelineMembershipSwitchRequest {
pub mconf: Configuration,
@@ -221,6 +221,8 @@ pub struct TimelineMembershipSwitchRequest {
pub struct TimelineMembershipSwitchResponse {
pub previous_conf: Configuration,
pub current_conf: Configuration,
pub term: Term,
pub flush_lsn: Lsn,
}
#[derive(Clone, Copy, Serialize, Deserialize)]

View File

@@ -86,6 +86,14 @@ pub enum GateError {
GateClosed,
}
impl GateError {
pub fn is_cancel(&self) -> bool {
match self {
GateError::GateClosed => true,
}
}
}
impl Default for Gate {
fn default() -> Self {
Self {

View File

@@ -10,7 +10,7 @@ use futures::StreamExt;
use futures::stream::FuturesUnordered;
use pageserver_api::shard::{ShardIdentity, ShardStripeSize};
use postgres_ffi::waldecoder::WalStreamDecoder;
use postgres_ffi::{MAX_SEND_SIZE, WAL_SEGMENT_SIZE};
use postgres_ffi::{MAX_SEND_SIZE, PgMajorVersion, WAL_SEGMENT_SIZE};
use pprof::criterion::{Output, PProfProfiler};
use remote_storage::{
DownloadOpts, GenericRemoteStorage, ListingMode, RemoteStorageConfig, RemoteStorageKind,
@@ -115,7 +115,7 @@ struct BenchmarkData {
#[derive(Deserialize)]
struct BenchmarkMetadata {
pg_version: u32,
pg_version: PgMajorVersion,
start_lsn: Lsn,
}

View File

@@ -7,8 +7,8 @@ use bytes::{Buf, Bytes};
use pageserver_api::key::rel_block_to_key;
use pageserver_api::reltag::{RelTag, SlruKind};
use pageserver_api::shard::ShardIdentity;
use postgres_ffi::pg_constants;
use postgres_ffi::walrecord::*;
use postgres_ffi::{PgMajorVersion, pg_constants};
use postgres_ffi_types::forknum::VISIBILITYMAP_FORKNUM;
use utils::lsn::Lsn;
@@ -24,7 +24,7 @@ impl InterpretedWalRecord {
buf: Bytes,
shards: &[ShardIdentity],
next_record_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<HashMap<ShardIdentity, InterpretedWalRecord>> {
let mut decoded = DecodedWALRecord::default();
decode_wal_record(buf, &mut decoded, pg_version)?;
@@ -78,7 +78,7 @@ impl MetadataRecord {
decoded: &DecodedWALRecord,
shard_records: &mut HashMap<ShardIdentity, InterpretedWalRecord>,
next_record_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<()> {
// Note: this doesn't actually copy the bytes since
// the [`Bytes`] type implements it via a level of indirection.
@@ -193,7 +193,7 @@ impl MetadataRecord {
fn decode_heapam_record(
buf: &mut Bytes,
decoded: &DecodedWALRecord,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<Option<MetadataRecord>> {
// Handle VM bit updates that are implicitly part of heap records.
@@ -205,7 +205,7 @@ impl MetadataRecord {
let mut flags = pg_constants::VISIBILITYMAP_VALID_BITS;
match pg_version {
14 => {
PgMajorVersion::PG14 => {
if decoded.xl_rmid == pg_constants::RM_HEAP_ID {
let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
@@ -272,7 +272,7 @@ impl MetadataRecord {
anyhow::bail!("Unknown RMGR {} for Heap decoding", decoded.xl_rmid);
}
}
15 => {
PgMajorVersion::PG15 => {
if decoded.xl_rmid == pg_constants::RM_HEAP_ID {
let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
@@ -339,7 +339,7 @@ impl MetadataRecord {
anyhow::bail!("Unknown RMGR {} for Heap decoding", decoded.xl_rmid);
}
}
16 => {
PgMajorVersion::PG16 => {
if decoded.xl_rmid == pg_constants::RM_HEAP_ID {
let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
@@ -406,7 +406,7 @@ impl MetadataRecord {
anyhow::bail!("Unknown RMGR {} for Heap decoding", decoded.xl_rmid);
}
}
17 => {
PgMajorVersion::PG17 => {
if decoded.xl_rmid == pg_constants::RM_HEAP_ID {
let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
@@ -473,7 +473,6 @@ impl MetadataRecord {
anyhow::bail!("Unknown RMGR {} for Heap decoding", decoded.xl_rmid);
}
}
_ => {}
}
if new_heap_blkno.is_some() || old_heap_blkno.is_some() {
@@ -500,7 +499,7 @@ impl MetadataRecord {
fn decode_neonmgr_record(
buf: &mut Bytes,
decoded: &DecodedWALRecord,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<Option<MetadataRecord>> {
// Handle VM bit updates that are implicitly part of heap records.
@@ -514,7 +513,7 @@ impl MetadataRecord {
assert_eq!(decoded.xl_rmid, pg_constants::RM_NEON_ID);
match pg_version {
16 | 17 => {
PgMajorVersion::PG16 | PgMajorVersion::PG17 => {
let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
match info {
@@ -574,7 +573,7 @@ impl MetadataRecord {
info => anyhow::bail!("Unknown WAL record type for Neon RMGR: {}", info),
}
}
_ => anyhow::bail!(
PgMajorVersion::PG15 | PgMajorVersion::PG14 => anyhow::bail!(
"Neon RMGR has no known compatibility with PostgreSQL version {}",
pg_version
),
@@ -629,116 +628,121 @@ impl MetadataRecord {
fn decode_dbase_record(
buf: &mut Bytes,
decoded: &DecodedWALRecord,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<Option<MetadataRecord>> {
// TODO: Refactor this to avoid the duplication between postgres versions.
let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
tracing::debug!(%info, %pg_version, "handle RM_DBASE_ID");
if pg_version == 14 {
if info == postgres_ffi::v14::bindings::XLOG_DBASE_CREATE {
let createdb = XlCreateDatabase::decode(buf);
tracing::debug!("XLOG_DBASE_CREATE v14");
match pg_version {
PgMajorVersion::PG14 => {
if info == postgres_ffi::v14::bindings::XLOG_DBASE_CREATE {
let createdb = XlCreateDatabase::decode(buf);
tracing::debug!("XLOG_DBASE_CREATE v14");
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
return Ok(Some(record));
} else if info == postgres_ffi::v14::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
return Ok(Some(record));
} else if info == postgres_ffi::v14::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
return Ok(Some(record));
return Ok(Some(record));
}
}
} else if pg_version == 15 {
if info == postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_WAL_LOG {
tracing::debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
} else if info == postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_FILE_COPY {
// The XLOG record was renamed between v14 and v15,
// but the record format is the same.
// So we can reuse XlCreateDatabase here.
tracing::debug!("XLOG_DBASE_CREATE_FILE_COPY");
PgMajorVersion::PG15 => {
if info == postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_WAL_LOG {
tracing::debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
} else if info == postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_FILE_COPY {
// The XLOG record was renamed between v14 and v15,
// but the record format is the same.
// So we can reuse XlCreateDatabase here.
tracing::debug!("XLOG_DBASE_CREATE_FILE_COPY");
let createdb = XlCreateDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
let createdb = XlCreateDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
return Ok(Some(record));
} else if info == postgres_ffi::v15::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
return Ok(Some(record));
} else if info == postgres_ffi::v15::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
return Ok(Some(record));
return Ok(Some(record));
}
}
} else if pg_version == 16 {
if info == postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_WAL_LOG {
tracing::debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
} else if info == postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_FILE_COPY {
// The XLOG record was renamed between v14 and v15,
// but the record format is the same.
// So we can reuse XlCreateDatabase here.
tracing::debug!("XLOG_DBASE_CREATE_FILE_COPY");
PgMajorVersion::PG16 => {
if info == postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_WAL_LOG {
tracing::debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
} else if info == postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_FILE_COPY {
// The XLOG record was renamed between v14 and v15,
// but the record format is the same.
// So we can reuse XlCreateDatabase here.
tracing::debug!("XLOG_DBASE_CREATE_FILE_COPY");
let createdb = XlCreateDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
let createdb = XlCreateDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
return Ok(Some(record));
} else if info == postgres_ffi::v16::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
return Ok(Some(record));
} else if info == postgres_ffi::v16::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
return Ok(Some(record));
return Ok(Some(record));
}
}
} else if pg_version == 17 {
if info == postgres_ffi::v17::bindings::XLOG_DBASE_CREATE_WAL_LOG {
tracing::debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
} else if info == postgres_ffi::v17::bindings::XLOG_DBASE_CREATE_FILE_COPY {
// The XLOG record was renamed between v14 and v15,
// but the record format is the same.
// So we can reuse XlCreateDatabase here.
tracing::debug!("XLOG_DBASE_CREATE_FILE_COPY");
PgMajorVersion::PG17 => {
if info == postgres_ffi::v17::bindings::XLOG_DBASE_CREATE_WAL_LOG {
tracing::debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
} else if info == postgres_ffi::v17::bindings::XLOG_DBASE_CREATE_FILE_COPY {
// The XLOG record was renamed between v14 and v15,
// but the record format is the same.
// So we can reuse XlCreateDatabase here.
tracing::debug!("XLOG_DBASE_CREATE_FILE_COPY");
let createdb = XlCreateDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
let createdb = XlCreateDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Create(DbaseCreate {
db_id: createdb.db_id,
tablespace_id: createdb.tablespace_id,
src_db_id: createdb.src_db_id,
src_tablespace_id: createdb.src_tablespace_id,
}));
return Ok(Some(record));
} else if info == postgres_ffi::v17::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
return Ok(Some(record));
} else if info == postgres_ffi::v17::bindings::XLOG_DBASE_DROP {
let dropdb = XlDropDatabase::decode(buf);
let record = MetadataRecord::Dbase(DbaseRecord::Drop(DbaseDrop {
db_id: dropdb.db_id,
tablespace_ids: dropdb.tablespace_ids,
}));
return Ok(Some(record));
return Ok(Some(record));
}
}
}
@@ -748,12 +752,12 @@ impl MetadataRecord {
fn decode_clog_record(
buf: &mut Bytes,
decoded: &DecodedWALRecord,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<Option<MetadataRecord>> {
let info = decoded.xl_info & !pg_constants::XLR_INFO_MASK;
if info == pg_constants::CLOG_ZEROPAGE {
let pageno = if pg_version < 17 {
let pageno = if pg_version < PgMajorVersion::PG17 {
buf.get_u32_le()
} else {
buf.get_u64_le() as u32
@@ -765,7 +769,7 @@ impl MetadataRecord {
ClogZeroPage { segno, rpageno },
))))
} else {
assert!(info == pg_constants::CLOG_TRUNCATE);
assert_eq!(info, pg_constants::CLOG_TRUNCATE);
let xlrec = XlClogTruncate::decode(buf, pg_version);
Ok(Some(MetadataRecord::Clog(ClogRecord::Truncate(
@@ -838,14 +842,14 @@ impl MetadataRecord {
fn decode_multixact_record(
buf: &mut Bytes,
decoded: &DecodedWALRecord,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<Option<MetadataRecord>> {
let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
if info == pg_constants::XLOG_MULTIXACT_ZERO_OFF_PAGE
|| info == pg_constants::XLOG_MULTIXACT_ZERO_MEM_PAGE
{
let pageno = if pg_version < 17 {
let pageno = if pg_version < PgMajorVersion::PG17 {
buf.get_u32_le()
} else {
buf.get_u64_le() as u32

View File

@@ -13,7 +13,7 @@ use pageserver_api::keyspace::KeySpace;
use pageserver_api::reltag::RelTag;
use pageserver_api::shard::ShardIdentity;
use postgres_ffi::walrecord::{DecodedBkpBlock, DecodedWALRecord};
use postgres_ffi::{BLCKSZ, page_is_new, page_set_lsn, pg_constants};
use postgres_ffi::{BLCKSZ, PgMajorVersion, page_is_new, page_set_lsn, pg_constants};
use serde::{Deserialize, Serialize};
use utils::bin_ser::BeSer;
use utils::lsn::Lsn;
@@ -139,7 +139,7 @@ impl SerializedValueBatch {
decoded: DecodedWALRecord,
shard_records: &mut HashMap<ShardIdentity, InterpretedWalRecord>,
next_record_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
) -> anyhow::Result<()> {
// First determine how big the buffers need to be and allocate it up-front.
// This duplicates some of the work below, but it's empirically much faster.
@@ -267,7 +267,7 @@ impl SerializedValueBatch {
fn estimate_buffer_size(
decoded: &DecodedWALRecord,
shard: &ShardIdentity,
pg_version: u32,
pg_version: PgMajorVersion,
) -> usize {
let mut estimate: usize = 0;
@@ -303,7 +303,11 @@ impl SerializedValueBatch {
estimate
}
fn block_is_image(decoded: &DecodedWALRecord, blk: &DecodedBkpBlock, pg_version: u32) -> bool {
fn block_is_image(
decoded: &DecodedWALRecord,
blk: &DecodedBkpBlock,
pg_version: PgMajorVersion,
) -> bool {
blk.apply_image
&& blk.has_image
&& decoded.xl_rmid == pg_constants::RM_XLOG_ID

File diff suppressed because one or more lines are too long

View File

@@ -18,6 +18,7 @@ workspace_hack = { version = "0.1", path = "../../workspace_hack" }
tokio-postgres.workspace = true
tokio-stream.workspace = true
tokio.workspace = true
postgres_versioninfo.workspace = true
futures.workspace = true
tokio-util.workspace = true
anyhow.workspace = true

View File

@@ -7,6 +7,7 @@ use detach_ancestor::AncestorDetached;
use http_utils::error::HttpErrorBody;
use pageserver_api::models::*;
use pageserver_api::shard::TenantShardId;
use postgres_versioninfo::PgMajorVersion;
pub use reqwest::Body as ReqwestBody;
use reqwest::{IntoUrl, Method, StatusCode, Url};
use utils::id::{TenantId, TimelineId};
@@ -745,9 +746,11 @@ impl Client {
timeline_id: TimelineId,
base_lsn: Lsn,
end_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
basebackup_tarball: ReqwestBody,
) -> Result<()> {
let pg_version = pg_version.major_version_num();
let uri = format!(
"{}/v1/tenant/{tenant_id}/timeline/{timeline_id}/import_basebackup?base_lsn={base_lsn}&end_lsn={end_lsn}&pg_version={pg_version}",
self.mgmt_api_endpoint,
@@ -841,4 +844,13 @@ impl Client {
.await
.map_err(Error::ReceiveBody)
}
pub async fn update_feature_flag_spec(&self, spec: String) -> Result<()> {
let uri = format!("{}/v1/feature_flag_spec", self.mgmt_api_endpoint);
self.request(Method::POST, uri, spec)
.await?
.json()
.await
.map_err(Error::ReceiveBody)
}
}

View File

@@ -9,12 +9,14 @@ anyhow.workspace = true
bytes.workspace = true
futures.workspace = true
pageserver_api.workspace = true
postgres_ffi.workspace = true
postgres_ffi_types.workspace = true
prost.workspace = true
prost-types.workspace = true
strum.workspace = true
strum_macros.workspace = true
thiserror.workspace = true
tokio.workspace = true
tokio-util.workspace = true
tonic.workspace = true
utils.workspace = true
workspace_hack.workspace = true

View File

@@ -35,6 +35,8 @@
syntax = "proto3";
package page_api;
import "google/protobuf/timestamp.proto";
service PageService {
// Returns whether a relation exists.
rpc CheckRelExists(CheckRelExistsRequest) returns (CheckRelExistsResponse);
@@ -64,6 +66,10 @@ service PageService {
// Fetches an SLRU segment.
rpc GetSlruSegment (GetSlruSegmentRequest) returns (GetSlruSegmentResponse);
// Acquires or extends a lease on the given LSN. This guarantees that the Pageserver won't garbage
// collect the LSN until the lease expires. Must be acquired on all relevant shards.
rpc LeaseLsn (LeaseLsnRequest) returns (LeaseLsnResponse);
}
// The LSN a request should read at.
@@ -110,6 +116,19 @@ message GetBaseBackupRequest {
bool replica = 2;
// If true, include relation files in the base backup. Mainly for debugging and tests.
bool full = 3;
// Compression algorithm to use. Base backups send a compressed payload instead of using gRPC
// compression, so that we can cache compressed backups on the server.
BaseBackupCompression compression = 4;
}
// Base backup compression algorithms.
enum BaseBackupCompression {
// Unknown algorithm. Used when clients send an unsupported algorithm.
BASE_BACKUP_COMPRESSION_UNKNOWN = 0;
// No compression.
BASE_BACKUP_COMPRESSION_NONE = 1;
// GZIP compression.
BASE_BACKUP_COMPRESSION_GZIP = 2;
}
// Base backup response chunk, returned as an ordered stream.
@@ -239,3 +258,17 @@ message GetSlruSegmentRequest {
message GetSlruSegmentResponse {
bytes segment = 1;
}
// Acquires or extends a lease on the given LSN. This guarantees that the Pageserver won't garbage
// collect the LSN until the lease expires. Must be acquired on all relevant shards.
message LeaseLsnRequest {
// The LSN to lease. Can't be 0 or below the current GC cutoff.
uint64 lsn = 1;
}
// Lease acquisition response. If the lease could not be granted because the LSN has already been
// garbage collected, a FailedPrecondition status will be returned instead.
message LeaseLsnResponse {
// The lease expiration time.
google.protobuf.Timestamp expires = 1;
}

View File

@@ -1,8 +1,7 @@
use std::convert::TryInto;
use bytes::Bytes;
use futures::TryStreamExt;
use futures::{Stream, StreamExt};
use anyhow::Result;
use futures::{Stream, StreamExt as _, TryStreamExt as _};
use tokio::io::AsyncRead;
use tokio_util::io::StreamReader;
use tonic::metadata::AsciiMetadataValue;
use tonic::metadata::errors::InvalidMetadataValue;
use tonic::transport::Channel;
@@ -12,8 +11,6 @@ use utils::id::TenantId;
use utils::id::TimelineId;
use utils::shard::ShardIndex;
use anyhow::Result;
use crate::model;
use crate::proto;
@@ -69,6 +66,7 @@ impl tonic::service::Interceptor for AuthInterceptor {
Ok(req)
}
}
#[derive(Clone)]
pub struct Client {
client: proto::PageServiceClient<
@@ -95,7 +93,6 @@ impl Client {
if let Some(compression) = compression {
// TODO: benchmark this (including network latency).
// TODO: consider enabling compression by default.
client = client
.accept_compressed(compression)
.send_compressed(compression);
@@ -121,22 +118,15 @@ impl Client {
pub async fn get_base_backup(
&mut self,
req: model::GetBaseBackupRequest,
) -> Result<impl Stream<Item = Result<Bytes, tonic::Status>> + 'static, tonic::Status> {
let proto_req = proto::GetBaseBackupRequest::from(req);
let response_stream: Streaming<proto::GetBaseBackupResponseChunk> =
self.client.get_base_backup(proto_req).await?.into_inner();
// TODO: Consider dechunking internally
let domain_stream = response_stream.map(|chunk_res| {
chunk_res.and_then(|proto_chunk| {
proto_chunk.try_into().map_err(|e| {
tonic::Status::internal(format!("Failed to convert response chunk: {e}"))
})
})
});
Ok(domain_stream)
) -> Result<impl AsyncRead + use<>, tonic::Status> {
let req = proto::GetBaseBackupRequest::from(req);
let chunks = self.client.get_base_backup(req).await?.into_inner();
let reader = StreamReader::new(
chunks
.map_ok(|resp| resp.chunk)
.map_err(std::io::Error::other),
);
Ok(reader)
}
/// Returns the total size of a database, as # of bytes.
@@ -197,4 +187,17 @@ impl Client {
let response = self.client.get_slru_segment(proto_req).await?;
Ok(response.into_inner().try_into()?)
}
/// Acquires or extends a lease on the given LSN. This guarantees that the Pageserver won't
/// garbage collect the LSN until the lease expires. Must be acquired on all relevant shards.
///
/// Returns the lease expiration time, or a FailedPrecondition status if the lease could not be
/// acquired because the LSN has already been garbage collected.
pub async fn lease_lsn(
&mut self,
req: model::LeaseLsnRequest,
) -> Result<model::LeaseLsnResponse, tonic::Status> {
let req = proto::LeaseLsnRequest::from(req);
Ok(self.client.lease_lsn(req).await?.into_inner().try_into()?)
}
}

View File

@@ -16,10 +16,11 @@
//! stream combinators without dealing with errors, and avoids validating the same message twice.
use std::fmt::Display;
use std::time::{Duration, SystemTime, UNIX_EPOCH};
use bytes::Bytes;
use postgres_ffi::Oid;
// TODO: split out Lsn, RelTag, SlruKind, Oid and other basic types to a separate crate, to avoid
use postgres_ffi_types::Oid;
// TODO: split out Lsn, RelTag, SlruKind and other basic types to a separate crate, to avoid
// pulling in all of their other crate dependencies when building the client.
use utils::lsn::Lsn;
@@ -191,15 +192,21 @@ pub struct GetBaseBackupRequest {
pub replica: bool,
/// If true, include relation files in the base backup. Mainly for debugging and tests.
pub full: bool,
/// Compression algorithm to use. Base backups send a compressed payload instead of using gRPC
/// compression, so that we can cache compressed backups on the server.
pub compression: BaseBackupCompression,
}
impl From<proto::GetBaseBackupRequest> for GetBaseBackupRequest {
fn from(pb: proto::GetBaseBackupRequest) -> Self {
Self {
impl TryFrom<proto::GetBaseBackupRequest> for GetBaseBackupRequest {
type Error = ProtocolError;
fn try_from(pb: proto::GetBaseBackupRequest) -> Result<Self, Self::Error> {
Ok(Self {
lsn: (pb.lsn != 0).then_some(Lsn(pb.lsn)),
replica: pb.replica,
full: pb.full,
}
compression: pb.compression.try_into()?,
})
}
}
@@ -209,10 +216,55 @@ impl From<GetBaseBackupRequest> for proto::GetBaseBackupRequest {
lsn: request.lsn.unwrap_or_default().0,
replica: request.replica,
full: request.full,
compression: request.compression.into(),
}
}
}
/// Base backup compression algorithm.
#[derive(Clone, Copy, Debug)]
pub enum BaseBackupCompression {
None,
Gzip,
}
impl TryFrom<proto::BaseBackupCompression> for BaseBackupCompression {
type Error = ProtocolError;
fn try_from(pb: proto::BaseBackupCompression) -> Result<Self, Self::Error> {
match pb {
proto::BaseBackupCompression::Unknown => Err(ProtocolError::invalid("compression", pb)),
proto::BaseBackupCompression::None => Ok(Self::None),
proto::BaseBackupCompression::Gzip => Ok(Self::Gzip),
}
}
}
impl TryFrom<i32> for BaseBackupCompression {
type Error = ProtocolError;
fn try_from(compression: i32) -> Result<Self, Self::Error> {
proto::BaseBackupCompression::try_from(compression)
.map_err(|_| ProtocolError::invalid("compression", compression))
.and_then(Self::try_from)
}
}
impl From<BaseBackupCompression> for proto::BaseBackupCompression {
fn from(compression: BaseBackupCompression) -> Self {
match compression {
BaseBackupCompression::None => Self::None,
BaseBackupCompression::Gzip => Self::Gzip,
}
}
}
impl From<BaseBackupCompression> for i32 {
fn from(compression: BaseBackupCompression) -> Self {
proto::BaseBackupCompression::from(compression).into()
}
}
pub type GetBaseBackupResponseChunk = Bytes;
impl TryFrom<proto::GetBaseBackupResponseChunk> for GetBaseBackupResponseChunk {
@@ -652,3 +704,54 @@ impl From<GetSlruSegmentResponse> for proto::GetSlruSegmentResponse {
// SlruKind is defined in pageserver_api::reltag.
pub type SlruKind = pageserver_api::reltag::SlruKind;
/// Acquires or extends a lease on the given LSN. This guarantees that the Pageserver won't garbage
/// collect the LSN until the lease expires.
pub struct LeaseLsnRequest {
/// The LSN to lease.
pub lsn: Lsn,
}
impl TryFrom<proto::LeaseLsnRequest> for LeaseLsnRequest {
type Error = ProtocolError;
fn try_from(pb: proto::LeaseLsnRequest) -> Result<Self, Self::Error> {
if pb.lsn == 0 {
return Err(ProtocolError::Missing("lsn"));
}
Ok(Self { lsn: Lsn(pb.lsn) })
}
}
impl From<LeaseLsnRequest> for proto::LeaseLsnRequest {
fn from(request: LeaseLsnRequest) -> Self {
Self { lsn: request.lsn.0 }
}
}
/// Lease expiration time. If the lease could not be granted because the LSN has already been
/// garbage collected, a FailedPrecondition status will be returned instead.
pub type LeaseLsnResponse = SystemTime;
impl TryFrom<proto::LeaseLsnResponse> for LeaseLsnResponse {
type Error = ProtocolError;
fn try_from(pb: proto::LeaseLsnResponse) -> Result<Self, Self::Error> {
let expires = pb.expires.ok_or(ProtocolError::Missing("expires"))?;
UNIX_EPOCH
.checked_add(Duration::new(expires.seconds as u64, expires.nanos as u32))
.ok_or_else(|| ProtocolError::invalid("expires", expires))
}
}
impl From<LeaseLsnResponse> for proto::LeaseLsnResponse {
fn from(response: LeaseLsnResponse) -> Self {
let expires = response.duration_since(UNIX_EPOCH).unwrap_or_default();
Self {
expires: Some(prost_types::Timestamp {
seconds: expires.as_secs() as i64,
nanos: expires.subsec_nanos() as i32,
}),
}
}
}

View File

@@ -317,6 +317,7 @@ impl Client for LibpqClient {
/// A gRPC Pageserver client.
struct GrpcClient {
inner: page_api::Client,
compression: page_api::BaseBackupCompression,
}
impl GrpcClient {
@@ -331,10 +332,14 @@ impl GrpcClient {
ttid.timeline_id,
ShardIndex::unsharded(),
None,
compression.then_some(tonic::codec::CompressionEncoding::Zstd),
None, // NB: uses payload compression
)
.await?;
Ok(Self { inner })
let compression = match compression {
true => page_api::BaseBackupCompression::Gzip,
false => page_api::BaseBackupCompression::None,
};
Ok(Self { inner, compression })
}
}
@@ -348,10 +353,8 @@ impl Client for GrpcClient {
lsn,
replica: false,
full: false,
compression: self.compression,
};
let stream = self.inner.get_base_backup(req).await?;
Ok(Box::pin(StreamReader::new(
stream.map_err(std::io::Error::other),
)))
Ok(Box::pin(self.inner.get_base_backup(req).await?))
}
}

View File

@@ -14,18 +14,19 @@ use std::fmt::Write as FmtWrite;
use std::time::{Instant, SystemTime};
use anyhow::{Context, anyhow};
use async_compression::tokio::write::GzipEncoder;
use bytes::{BufMut, Bytes, BytesMut};
use fail::fail_point;
use pageserver_api::key::{Key, rel_block_to_key};
use pageserver_api::reltag::{RelTag, SlruKind};
use postgres_ffi::pg_constants::{PG_HBA, PGDATA_SPECIAL_FILES};
use postgres_ffi::{
BLCKSZ, PG_TLI, RELSEG_SIZE, WAL_SEGMENT_SIZE, XLogFileName, dispatch_pgversion, pg_constants,
BLCKSZ, PG_TLI, PgMajorVersion, RELSEG_SIZE, WAL_SEGMENT_SIZE, XLogFileName,
dispatch_pgversion, pg_constants,
};
use postgres_ffi_types::constants::{DEFAULTTABLESPACE_OID, GLOBALTABLESPACE_OID};
use postgres_ffi_types::forknum::{INIT_FORKNUM, MAIN_FORKNUM};
use tokio::io;
use tokio::io::AsyncWrite;
use tokio::io::{self, AsyncWrite, AsyncWriteExt as _};
use tokio_tar::{Builder, EntryType, Header};
use tracing::*;
use utils::lsn::Lsn;
@@ -96,6 +97,7 @@ impl From<BasebackupError> for tonic::Status {
/// * When working without safekeepers. In this situation it is important to match the lsn
/// we are taking basebackup on with the lsn that is used in pageserver's walreceiver
/// to start the replication.
#[allow(clippy::too_many_arguments)]
pub async fn send_basebackup_tarball<'a, W>(
write: &'a mut W,
timeline: &'a Timeline,
@@ -103,6 +105,7 @@ pub async fn send_basebackup_tarball<'a, W>(
prev_lsn: Option<Lsn>,
full_backup: bool,
replica: bool,
gzip_level: Option<async_compression::Level>,
ctx: &'a RequestContext,
) -> Result<(), BasebackupError>
where
@@ -121,7 +124,7 @@ where
// prev_lsn value; that happens if the timeline was just branched from
// an old LSN and it doesn't have any WAL of its own yet. We will set
// prev_lsn to Lsn(0) if we cannot provide the correct value.
let (backup_prev, backup_lsn) = if let Some(req_lsn) = req_lsn {
let (backup_prev, lsn) = if let Some(req_lsn) = req_lsn {
// Backup was requested at a particular LSN. The caller should've
// already checked that it's a valid LSN.
@@ -142,7 +145,7 @@ where
};
// Consolidate the derived and the provided prev_lsn values
let prev_lsn = if let Some(provided_prev_lsn) = prev_lsn {
let prev_record_lsn = if let Some(provided_prev_lsn) = prev_lsn {
if backup_prev != Lsn(0) && backup_prev != provided_prev_lsn {
return Err(BasebackupError::Server(anyhow!(
"backup_prev {backup_prev} != provided_prev_lsn {provided_prev_lsn}"
@@ -154,30 +157,55 @@ where
};
info!(
"taking basebackup lsn={}, prev_lsn={} (full_backup={}, replica={})",
backup_lsn, prev_lsn, full_backup, replica
"taking basebackup lsn={lsn}, prev_lsn={prev_record_lsn} \
(full_backup={full_backup}, replica={replica}, gzip={gzip_level:?})",
);
let span = info_span!("send_tarball", backup_lsn=%lsn);
let io_concurrency = IoConcurrency::spawn_from_conf(
timeline.conf.get_vectored_concurrent_io,
timeline
.gate
.enter()
.map_err(|_| BasebackupError::Shutdown)?,
);
let basebackup = Basebackup {
ar: Builder::new_non_terminated(write),
timeline,
lsn: backup_lsn,
prev_record_lsn: prev_lsn,
full_backup,
replica,
ctx,
io_concurrency: IoConcurrency::spawn_from_conf(
timeline.conf.get_vectored_concurrent_io,
timeline
.gate
.enter()
.map_err(|_| BasebackupError::Shutdown)?,
),
};
basebackup
if let Some(gzip_level) = gzip_level {
let mut encoder = GzipEncoder::with_quality(write, gzip_level);
Basebackup {
ar: Builder::new_non_terminated(&mut encoder),
timeline,
lsn,
prev_record_lsn,
full_backup,
replica,
ctx,
io_concurrency,
}
.send_tarball()
.instrument(info_span!("send_tarball", backup_lsn=%backup_lsn))
.await
.instrument(span)
.await?;
encoder
.shutdown()
.await
.map_err(|err| BasebackupError::Client(err, "gzip"))?;
} else {
Basebackup {
ar: Builder::new_non_terminated(write),
timeline,
lsn,
prev_record_lsn,
full_backup,
replica,
ctx,
io_concurrency,
}
.send_tarball()
.instrument(span)
.await?;
}
Ok(())
}
/// This is short-living object only for the time of tarball creation,
@@ -619,10 +647,7 @@ where
};
if spcnode == GLOBALTABLESPACE_OID {
let pg_version_str = match self.timeline.pg_version {
14 | 15 => self.timeline.pg_version.to_string(),
ver => format!("{ver}\x0A"),
};
let pg_version_str = self.timeline.pg_version.versionfile_string();
let header = new_tar_header("PG_VERSION", pg_version_str.len() as u64)?;
self.ar
.append(&header, pg_version_str.as_bytes())
@@ -679,10 +704,7 @@ where
if let Some(img) = relmap_img {
let dst_path = format!("base/{dbnode}/PG_VERSION");
let pg_version_str = match self.timeline.pg_version {
14 | 15 => self.timeline.pg_version.to_string(),
ver => format!("{ver}\x0A"),
};
let pg_version_str = self.timeline.pg_version.versionfile_string();
let header = new_tar_header(&dst_path, pg_version_str.len() as u64)?;
self.ar
.append(&header, pg_version_str.as_bytes())
@@ -713,7 +735,7 @@ where
buf.extend_from_slice(&img[..]);
let crc = crc32c::crc32c(&img[..]);
buf.put_u32_le(crc);
let path = if self.timeline.pg_version < 17 {
let path = if self.timeline.pg_version < PgMajorVersion::PG17 {
format!("pg_twophase/{xid:>08X}")
} else {
format!("pg_twophase/{xid:>016X}")

View File

@@ -1,13 +1,12 @@
use std::{collections::HashMap, sync::Arc};
use anyhow::Context;
use async_compression::tokio::write::GzipEncoder;
use camino::{Utf8Path, Utf8PathBuf};
use metrics::core::{AtomicU64, GenericCounter};
use pageserver_api::{config::BasebackupCacheConfig, models::TenantState};
use tokio::{
io::{AsyncWriteExt, BufWriter},
sync::mpsc::{UnboundedReceiver, UnboundedSender},
sync::mpsc::{Receiver, Sender, error::TrySendError},
};
use tokio_util::sync::CancellationToken;
use utils::{
@@ -20,8 +19,8 @@ use crate::{
basebackup::send_basebackup_tarball,
context::{DownloadBehavior, RequestContext},
metrics::{
BASEBACKUP_CACHE_ENTRIES, BASEBACKUP_CACHE_PREPARE, BASEBACKUP_CACHE_READ,
BASEBACKUP_CACHE_SIZE,
BASEBACKUP_CACHE_ENTRIES, BASEBACKUP_CACHE_PREPARE, BASEBACKUP_CACHE_PREPARE_QUEUE_SIZE,
BASEBACKUP_CACHE_READ, BASEBACKUP_CACHE_SIZE,
},
task_mgr::TaskKind,
tenant::{
@@ -36,8 +35,8 @@ pub struct BasebackupPrepareRequest {
pub lsn: Lsn,
}
pub type BasebackupPrepareSender = UnboundedSender<BasebackupPrepareRequest>;
pub type BasebackupPrepareReceiver = UnboundedReceiver<BasebackupPrepareRequest>;
pub type BasebackupPrepareSender = Sender<BasebackupPrepareRequest>;
pub type BasebackupPrepareReceiver = Receiver<BasebackupPrepareRequest>;
#[derive(Clone)]
struct CacheEntry {
@@ -61,40 +60,65 @@ struct CacheEntry {
/// and ~1 RPS for get requests.
pub struct BasebackupCache {
data_dir: Utf8PathBuf,
config: Option<BasebackupCacheConfig>,
entries: std::sync::Mutex<HashMap<TenantTimelineId, CacheEntry>>,
prepare_sender: BasebackupPrepareSender,
read_hit_count: GenericCounter<AtomicU64>,
read_miss_count: GenericCounter<AtomicU64>,
read_err_count: GenericCounter<AtomicU64>,
prepare_skip_count: GenericCounter<AtomicU64>,
}
impl BasebackupCache {
/// Creates a BasebackupCache and spawns the background task.
/// The initialization of the cache is performed in the background and does not
/// block the caller. The cache will return `None` for any get requests until
/// initialization is complete.
pub fn spawn(
runtime_handle: &tokio::runtime::Handle,
/// Create a new BasebackupCache instance.
/// Also returns a BasebackupPrepareReceiver which is needed to start
/// the background task.
/// The cache is initialized from the data_dir in the background task.
/// The cache will return `None` for any get requests until the initialization is complete.
/// The background task is spawned separately using [`Self::spawn_background_task`]
/// to avoid a circular dependency between the cache and the tenant manager.
pub fn new(
data_dir: Utf8PathBuf,
config: Option<BasebackupCacheConfig>,
prepare_receiver: BasebackupPrepareReceiver,
tenant_manager: Arc<TenantManager>,
cancel: CancellationToken,
) -> Arc<Self> {
) -> (Arc<Self>, BasebackupPrepareReceiver) {
let chan_size = config.as_ref().map(|c| c.max_size_entries).unwrap_or(1);
let (prepare_sender, prepare_receiver) = tokio::sync::mpsc::channel(chan_size);
let cache = Arc::new(BasebackupCache {
data_dir,
config,
entries: std::sync::Mutex::new(HashMap::new()),
prepare_sender,
read_hit_count: BASEBACKUP_CACHE_READ.with_label_values(&["hit"]),
read_miss_count: BASEBACKUP_CACHE_READ.with_label_values(&["miss"]),
read_err_count: BASEBACKUP_CACHE_READ.with_label_values(&["error"]),
prepare_skip_count: BASEBACKUP_CACHE_PREPARE.with_label_values(&["skip"]),
});
if let Some(config) = config {
(cache, prepare_receiver)
}
/// Spawns the background task.
/// The background task initializes the cache from the disk,
/// processes prepare requests, and cleans up outdated cache entries.
/// Noop if the cache is disabled (config is None).
pub fn spawn_background_task(
self: Arc<Self>,
runtime_handle: &tokio::runtime::Handle,
prepare_receiver: BasebackupPrepareReceiver,
tenant_manager: Arc<TenantManager>,
cancel: CancellationToken,
) {
if let Some(config) = self.config.clone() {
let background = BackgroundTask {
c: cache.clone(),
c: self,
config,
tenant_manager,
@@ -109,8 +133,45 @@ impl BasebackupCache {
};
runtime_handle.spawn(background.run(prepare_receiver));
}
}
cache
/// Send a basebackup prepare request to the background task.
/// The basebackup will be prepared asynchronously, it does not block the caller.
/// The request will be skipped if any cache limits are exceeded.
pub fn send_prepare(&self, tenant_shard_id: TenantShardId, timeline_id: TimelineId, lsn: Lsn) {
let req = BasebackupPrepareRequest {
tenant_shard_id,
timeline_id,
lsn,
};
BASEBACKUP_CACHE_PREPARE_QUEUE_SIZE.inc();
let res = self.prepare_sender.try_send(req);
if let Err(e) = res {
BASEBACKUP_CACHE_PREPARE_QUEUE_SIZE.dec();
self.prepare_skip_count.inc();
match e {
TrySendError::Full(_) => {
// Basebackup prepares are pretty rare, normally we should not hit this.
tracing::info!(
tenant_id = %tenant_shard_id.tenant_id,
%timeline_id,
%lsn,
"Basebackup prepare channel is full, skipping the request"
);
}
TrySendError::Closed(_) => {
// Normal during shutdown, not critical.
tracing::info!(
tenant_id = %tenant_shard_id.tenant_id,
%timeline_id,
%lsn,
"Basebackup prepare channel is closed, skipping the request"
);
}
}
}
}
/// Gets a basebackup entry from the cache.
@@ -123,6 +184,10 @@ impl BasebackupCache {
timeline_id: TimelineId,
lsn: Lsn,
) -> Option<tokio::fs::File> {
if !self.is_enabled() {
return None;
}
// Fast path. Check if the entry exists using the in-memory state.
let tti = TenantTimelineId::new(tenant_id, timeline_id);
if self.entries.lock().unwrap().get(&tti).map(|e| e.lsn) != Some(lsn) {
@@ -150,6 +215,10 @@ impl BasebackupCache {
}
}
pub fn is_enabled(&self) -> bool {
self.config.is_some()
}
// Private methods.
fn entry_filename(tenant_id: TenantId, timeline_id: TimelineId, lsn: Lsn) -> String {
@@ -367,6 +436,7 @@ impl BackgroundTask {
loop {
tokio::select! {
Some(req) = prepare_receiver.recv() => {
BASEBACKUP_CACHE_PREPARE_QUEUE_SIZE.dec();
if let Err(err) = self.prepare_basebackup(
req.tenant_shard_id,
req.timeline_id,
@@ -594,13 +664,6 @@ impl BackgroundTask {
let file = tokio::fs::File::create(entry_tmp_path).await?;
let mut writer = BufWriter::new(file);
let mut encoder = GzipEncoder::with_quality(
&mut writer,
// Level::Best because compression is not on the hot path of basebackup requests.
// The decompression is almost not affected by the compression level.
async_compression::Level::Best,
);
// We may receive a request before the WAL record is applied to the timeline.
// Wait for the requested LSN to be applied.
timeline
@@ -613,17 +676,19 @@ impl BackgroundTask {
.await?;
send_basebackup_tarball(
&mut encoder,
&mut writer,
timeline,
Some(req_lsn),
None,
false,
false,
// Level::Best because compression is not on the hot path of basebackup requests.
// The decompression is almost not affected by the compression level.
Some(async_compression::Level::Best),
&ctx,
)
.await?;
encoder.shutdown().await?;
writer.flush().await?;
writer.into_inner().sync_all().await?;

View File

@@ -569,8 +569,10 @@ fn start_pageserver(
pageserver::l0_flush::L0FlushGlobalState::new(conf.l0_flush.clone());
// Scan the local 'tenants/' directory and start loading the tenants
let (basebackup_prepare_sender, basebackup_prepare_receiver) =
tokio::sync::mpsc::unbounded_channel();
let (basebackup_cache, basebackup_prepare_receiver) = BasebackupCache::new(
conf.basebackup_cache_dir(),
conf.basebackup_cache_config.clone(),
);
let deletion_queue_client = deletion_queue.new_client();
let background_purges = mgr::BackgroundPurges::default();
@@ -582,7 +584,7 @@ fn start_pageserver(
remote_storage: remote_storage.clone(),
deletion_queue_client,
l0_flush_global_state,
basebackup_prepare_sender,
basebackup_cache: Arc::clone(&basebackup_cache),
feature_resolver: feature_resolver.clone(),
},
shutdown_pageserver.clone(),
@@ -590,10 +592,8 @@ fn start_pageserver(
let tenant_manager = Arc::new(tenant_manager);
BACKGROUND_RUNTIME.block_on(mgr::init_tenant_mgr(tenant_manager.clone(), order))?;
let basebackup_cache = BasebackupCache::spawn(
basebackup_cache.spawn_background_task(
BACKGROUND_RUNTIME.handle(),
conf.basebackup_cache_dir(),
conf.basebackup_cache_config.clone(),
basebackup_prepare_receiver,
Arc::clone(&tenant_manager),
shutdown_pageserver.child_token(),
@@ -806,7 +806,6 @@ fn start_pageserver(
} else {
None
},
basebackup_cache,
);
// Spawn a Pageserver gRPC server task. It will spawn separate tasks for

View File

@@ -37,7 +37,7 @@ async fn main() -> anyhow::Result<()> {
not_modified_since: Lsn(23),
},
batch_key: 42,
message: format!("message {}", msg),
message: format!("message {msg}"),
}));
let Ok(res) = tokio::time::timeout(Duration::from_secs(10), fut).await else {
eprintln!("pipe seems full");

View File

@@ -11,7 +11,7 @@ use std::num::NonZeroUsize;
use std::sync::Arc;
use std::time::Duration;
use anyhow::{Context, bail, ensure};
use anyhow::{Context, ensure};
use camino::{Utf8Path, Utf8PathBuf};
use once_cell::sync::OnceCell;
use pageserver_api::config::{
@@ -22,6 +22,7 @@ use pageserver_api::models::ImageCompressionAlgorithm;
use pageserver_api::shard::TenantShardId;
use pem::Pem;
use postgres_backend::AuthType;
use postgres_ffi::PgMajorVersion;
use remote_storage::{RemotePath, RemoteStorageConfig};
use reqwest::Url;
use storage_broker::Uri;
@@ -338,20 +339,16 @@ impl PageServerConf {
//
// Postgres distribution paths
//
pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
pub fn pg_distrib_dir(&self, pg_version: PgMajorVersion) -> anyhow::Result<Utf8PathBuf> {
let path = self.pg_distrib_dir.clone();
#[allow(clippy::manual_range_patterns)]
match pg_version {
14 | 15 | 16 | 17 => Ok(path.join(format!("v{pg_version}"))),
_ => bail!("Unsupported postgres version: {}", pg_version),
}
Ok(path.join(pg_version.v_str()))
}
pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
pub fn pg_bin_dir(&self, pg_version: PgMajorVersion) -> anyhow::Result<Utf8PathBuf> {
Ok(self.pg_distrib_dir(pg_version)?.join("bin"))
}
pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
pub fn pg_lib_dir(&self, pg_version: PgMajorVersion) -> anyhow::Result<Utf8PathBuf> {
Ok(self.pg_distrib_dir(pg_version)?.join("lib"))
}
@@ -765,4 +762,40 @@ mod tests {
let result = PageServerConf::parse_and_validate(NodeId(0), config_toml, &workdir);
assert_eq!(result.is_ok(), is_valid);
}
#[test]
fn test_config_posthog_config_is_valid() {
let input = r#"
control_plane_api = "http://localhost:6666"
[posthog_config]
server_api_key = "phs_AAA"
client_api_key = "phc_BBB"
project_id = "000"
private_api_url = "https://us.posthog.com"
public_api_url = "https://us.i.posthog.com"
"#;
let config_toml = toml_edit::de::from_str::<pageserver_api::config::ConfigToml>(input)
.expect("posthogconfig is valid");
let workdir = Utf8PathBuf::from("/nonexistent");
PageServerConf::parse_and_validate(NodeId(0), config_toml, &workdir)
.expect("parse_and_validate");
}
#[test]
fn test_config_posthog_incomplete_config_is_valid() {
let input = r#"
control_plane_api = "http://localhost:6666"
[posthog_config]
server_api_key = "phs_AAA"
private_api_url = "https://us.posthog.com"
public_api_url = "https://us.i.posthog.com"
"#;
let config_toml = toml_edit::de::from_str::<pageserver_api::config::ConfigToml>(input)
.expect("posthogconfig is valid");
let workdir = Utf8PathBuf::from("/nonexistent");
PageServerConf::parse_and_validate(NodeId(0), config_toml, &workdir)
.expect("parse_and_validate");
}
}

View File

@@ -3,7 +3,7 @@ use std::{collections::HashMap, sync::Arc, time::Duration};
use arc_swap::ArcSwap;
use pageserver_api::config::NodeMetadata;
use posthog_client_lite::{
CaptureEvent, FeatureResolverBackgroundLoop, PostHogClientConfig, PostHogEvaluationError,
CaptureEvent, FeatureResolverBackgroundLoop, PostHogEvaluationError,
PostHogFlagFilterPropertyValue,
};
use remote_storage::RemoteStorageKind;
@@ -31,6 +31,13 @@ impl FeatureResolver {
}
}
pub fn update(&self, spec: String) -> anyhow::Result<()> {
if let Some(inner) = &self.inner {
inner.update(spec)?;
}
Ok(())
}
pub fn spawn(
conf: &PageServerConf,
shutdown_pageserver: CancellationToken,
@@ -38,16 +45,24 @@ impl FeatureResolver {
) -> anyhow::Result<Self> {
// DO NOT block in this function: make it return as fast as possible to avoid startup delays.
if let Some(posthog_config) = &conf.posthog_config {
let inner = FeatureResolverBackgroundLoop::new(
PostHogClientConfig {
server_api_key: posthog_config.server_api_key.clone(),
client_api_key: posthog_config.client_api_key.clone(),
project_id: posthog_config.project_id.clone(),
private_api_url: posthog_config.private_api_url.clone(),
public_api_url: posthog_config.public_api_url.clone(),
},
shutdown_pageserver,
);
let posthog_client_config = match posthog_config.clone().try_into_posthog_config() {
Ok(config) => config,
Err(e) => {
tracing::warn!(
"invalid posthog config, skipping posthog integration: {}",
e
);
return Ok(FeatureResolver {
inner: None,
internal_properties: None,
force_overrides_for_testing: Arc::new(ArcSwap::new(Arc::new(
HashMap::new(),
))),
});
}
};
let inner =
FeatureResolverBackgroundLoop::new(posthog_client_config, shutdown_pageserver);
let inner = Arc::new(inner);
// The properties shared by all tenants on this pageserver.

View File

@@ -41,6 +41,7 @@ use pageserver_api::models::{
TopTenantShardItem, TopTenantShardsRequest, TopTenantShardsResponse,
};
use pageserver_api::shard::{ShardCount, TenantShardId};
use postgres_ffi::PgMajorVersion;
use remote_storage::{DownloadError, GenericRemoteStorage, TimeTravelError};
use scopeguard::defer;
use serde_json::json;
@@ -1892,9 +1893,13 @@ async fn update_tenant_config_handler(
let location_conf = LocationConf::attached_single(
new_tenant_conf.clone(),
tenant.get_generation(),
&ShardParameters::default(),
ShardParameters::from(tenant.get_shard_identity()),
);
tenant
.get_shard_identity()
.assert_equal(location_conf.shard); // not strictly necessary since we construct it above
crate::tenant::TenantShard::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
.await
.map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
@@ -1936,9 +1941,13 @@ async fn patch_tenant_config_handler(
let location_conf = LocationConf::attached_single(
updated,
tenant.get_generation(),
&ShardParameters::default(),
ShardParameters::from(tenant.get_shard_identity()),
);
tenant
.get_shard_identity()
.assert_equal(location_conf.shard); // not strictly necessary since we construct it above
crate::tenant::TenantShard::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
.await
.map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
@@ -3385,7 +3394,7 @@ async fn put_tenant_timeline_import_basebackup(
let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
let base_lsn: Lsn = must_parse_query_param(&request, "base_lsn")?;
let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
let pg_version: u32 = must_parse_query_param(&request, "pg_version")?;
let pg_version: PgMajorVersion = must_parse_query_param(&request, "pg_version")?;
check_permission(&request, Some(tenant_id))?;
@@ -3742,6 +3751,20 @@ async fn force_override_feature_flag_for_testing_delete(
json_response(StatusCode::OK, ())
}
async fn update_feature_flag_spec(
mut request: Request<Body>,
_cancel: CancellationToken,
) -> Result<Response<Body>, ApiError> {
check_permission(&request, None)?;
let body = json_request(&mut request).await?;
let state = get_state(&request);
state
.feature_resolver
.update(body)
.map_err(ApiError::InternalServerError)?;
json_response(StatusCode::OK, ())
}
/// Common functionality of all the HTTP API handlers.
///
/// - Adds a tracing span to each request (by `request_span`)
@@ -4127,5 +4150,8 @@ pub fn make_router(
.delete("/v1/feature_flag/:flag_key", |r| {
testing_api_handler("force override feature flag - delete", r, force_override_feature_flag_for_testing_delete)
})
.post("/v1/feature_flag_spec", |r| {
api_handler(r, update_feature_flag_spec)
})
.any(handler_404))
}

View File

@@ -38,6 +38,7 @@ pub mod walredo;
use camino::Utf8Path;
use deletion_queue::DeletionQueue;
use postgres_ffi::PgMajorVersion;
use tenant::mgr::{BackgroundPurges, TenantManager};
use tenant::secondary;
use tracing::{info, info_span};
@@ -51,7 +52,7 @@ use tracing::{info, info_span};
/// backwards-compatible changes to the metadata format.
pub const STORAGE_FORMAT_VERSION: u16 = 3;
pub const DEFAULT_PG_VERSION: u32 = 17;
pub const DEFAULT_PG_VERSION: PgMajorVersion = PgMajorVersion::PG17;
// Magic constants used to identify different kinds of files
pub const IMAGE_FILE_MAGIC: u16 = 0x5A60;

View File

@@ -4439,6 +4439,14 @@ pub(crate) static BASEBACKUP_CACHE_SIZE: Lazy<UIntGauge> = Lazy::new(|| {
.expect("failed to define a metric")
});
pub(crate) static BASEBACKUP_CACHE_PREPARE_QUEUE_SIZE: Lazy<UIntGauge> = Lazy::new(|| {
register_uint_gauge!(
"pageserver_basebackup_cache_prepare_queue_size",
"Number of requests in the basebackup prepare channel"
)
.expect("failed to define a metric")
});
static PAGESERVER_CONFIG_IGNORED_ITEMS: Lazy<UIntGaugeVec> = Lazy::new(|| {
register_uint_gauge_vec!(
"pageserver_config_ignored_items",

View File

@@ -12,9 +12,9 @@ use std::task::{Context, Poll};
use std::time::{Duration, Instant, SystemTime};
use std::{io, str};
use anyhow::{Context as _, anyhow, bail};
use async_compression::tokio::write::GzipEncoder;
use anyhow::{Context as _, bail};
use bytes::{Buf as _, BufMut as _, BytesMut};
use chrono::Utc;
use futures::future::BoxFuture;
use futures::{FutureExt, Stream};
use itertools::Itertools;
@@ -63,7 +63,6 @@ use utils::{failpoint_support, span_record};
use crate::auth::check_permission;
use crate::basebackup::{self, BasebackupError};
use crate::basebackup_cache::BasebackupCache;
use crate::config::PageServerConf;
use crate::context::{
DownloadBehavior, PerfInstrumentFutureExt, RequestContext, RequestContextBuilder,
@@ -138,7 +137,6 @@ pub fn spawn(
perf_trace_dispatch: Option<Dispatch>,
tcp_listener: tokio::net::TcpListener,
tls_config: Option<Arc<rustls::ServerConfig>>,
basebackup_cache: Arc<BasebackupCache>,
) -> Listener {
let cancel = CancellationToken::new();
let libpq_ctx = RequestContext::todo_child(
@@ -160,7 +158,6 @@ pub fn spawn(
conf.pg_auth_type,
tls_config,
conf.page_service_pipelining.clone(),
basebackup_cache,
libpq_ctx,
cancel.clone(),
)
@@ -219,7 +216,6 @@ pub async fn libpq_listener_main(
auth_type: AuthType,
tls_config: Option<Arc<rustls::ServerConfig>>,
pipelining_config: PageServicePipeliningConfig,
basebackup_cache: Arc<BasebackupCache>,
listener_ctx: RequestContext,
listener_cancel: CancellationToken,
) -> Connections {
@@ -263,7 +259,6 @@ pub async fn libpq_listener_main(
auth_type,
tls_config.clone(),
pipelining_config.clone(),
Arc::clone(&basebackup_cache),
connection_ctx,
connections_cancel.child_token(),
gate_guard,
@@ -306,7 +301,6 @@ async fn page_service_conn_main(
auth_type: AuthType,
tls_config: Option<Arc<rustls::ServerConfig>>,
pipelining_config: PageServicePipeliningConfig,
basebackup_cache: Arc<BasebackupCache>,
connection_ctx: RequestContext,
cancel: CancellationToken,
gate_guard: GateGuard,
@@ -372,7 +366,6 @@ async fn page_service_conn_main(
pipelining_config,
conf.get_vectored_concurrent_io,
perf_span_fields,
basebackup_cache,
connection_ctx,
cancel.clone(),
gate_guard,
@@ -426,8 +419,6 @@ struct PageServerHandler {
pipelining_config: PageServicePipeliningConfig,
get_vectored_concurrent_io: GetVectoredConcurrentIo,
basebackup_cache: Arc<BasebackupCache>,
gate_guard: GateGuard,
}
@@ -913,7 +904,6 @@ impl PageServerHandler {
pipelining_config: PageServicePipeliningConfig,
get_vectored_concurrent_io: GetVectoredConcurrentIo,
perf_span_fields: ConnectionPerfSpanFields,
basebackup_cache: Arc<BasebackupCache>,
connection_ctx: RequestContext,
cancel: CancellationToken,
gate_guard: GateGuard,
@@ -927,7 +917,6 @@ impl PageServerHandler {
cancel,
pipelining_config,
get_vectored_concurrent_io,
basebackup_cache,
gate_guard,
}
}
@@ -2613,26 +2602,16 @@ impl PageServerHandler {
prev_lsn,
full_backup,
replica,
None,
&ctx,
)
.await?;
} else {
let mut writer = BufWriter::new(pgb.copyout_writer());
let cached = {
// Basebackup is cached only for this combination of parameters.
if timeline.is_basebackup_cache_enabled()
&& gzip
&& lsn.is_some()
&& prev_lsn.is_none()
{
self.basebackup_cache
.get(tenant_id, timeline_id, lsn.unwrap())
.await
} else {
None
}
};
let cached = timeline
.get_cached_basebackup_if_enabled(lsn, prev_lsn, full_backup, replica, gzip)
.await;
if let Some(mut cached) = cached {
from_cache = true;
@@ -2641,31 +2620,6 @@ impl PageServerHandler {
.map_err(|err| {
BasebackupError::Client(err, "handle_basebackup_request,cached,copy")
})?;
} else if gzip {
let mut encoder = GzipEncoder::with_quality(
&mut writer,
// NOTE using fast compression because it's on the critical path
// for compute startup. For an empty database, we get
// <100KB with this method. The Level::Best compression method
// gives us <20KB, but maybe we should add basebackup caching
// on compute shutdown first.
async_compression::Level::Fastest,
);
basebackup::send_basebackup_tarball(
&mut encoder,
&timeline,
lsn,
prev_lsn,
full_backup,
replica,
&ctx,
)
.await?;
// shutdown the encoder to ensure the gzip footer is written
encoder
.shutdown()
.await
.map_err(|e| QueryError::Disconnected(ConnectionError::Io(e)))?;
} else {
basebackup::send_basebackup_tarball(
&mut writer,
@@ -2674,6 +2628,11 @@ impl PageServerHandler {
prev_lsn,
full_backup,
replica,
// NB: using fast compression because it's on the critical path for compute
// startup. For an empty database, we get <100KB with this method. The
// Level::Best compression method gives us <20KB, but maybe we should add
// basebackup caching on compute shutdown first.
gzip.then_some(async_compression::Level::Fastest),
&ctx,
)
.await?;
@@ -3553,7 +3512,7 @@ impl proto::PageService for GrpcPageServiceHandler {
if timeline.is_archived() == Some(true) {
return Err(tonic::Status::failed_precondition("timeline is archived"));
}
let req: page_api::GetBaseBackupRequest = req.into_inner().into();
let req: page_api::GetBaseBackupRequest = req.into_inner().try_into()?;
span_record!(lsn=?req.lsn);
@@ -3579,20 +3538,50 @@ impl proto::PageService for GrpcPageServiceHandler {
let span = Span::current();
let (mut simplex_read, mut simplex_write) = tokio::io::simplex(CHUNK_SIZE);
let jh = tokio::spawn(async move {
let result = basebackup::send_basebackup_tarball(
&mut simplex_write,
&timeline,
req.lsn,
None,
req.full,
req.replica,
&ctx,
)
.instrument(span) // propagate request span
.await;
simplex_write.shutdown().await.map_err(|err| {
BasebackupError::Server(anyhow!("simplex shutdown failed: {err}"))
})?;
let gzip_level = match req.compression {
page_api::BaseBackupCompression::None => None,
// NB: using fast compression because it's on the critical path for compute
// startup. For an empty database, we get <100KB with this method. The
// Level::Best compression method gives us <20KB, but maybe we should add
// basebackup caching on compute shutdown first.
page_api::BaseBackupCompression::Gzip => Some(async_compression::Level::Fastest),
};
// Check for a cached basebackup.
let cached = timeline
.get_cached_basebackup_if_enabled(
req.lsn,
None,
req.full,
req.replica,
gzip_level.is_some(),
)
.await;
let result = if let Some(mut cached) = cached {
// If we have a cached basebackup, send it.
tokio::io::copy(&mut cached, &mut simplex_write)
.await
.map(|_| ())
.map_err(|err| BasebackupError::Client(err, "cached,copy"))
} else {
basebackup::send_basebackup_tarball(
&mut simplex_write,
&timeline,
req.lsn,
None,
req.full,
req.replica,
gzip_level,
&ctx,
)
.instrument(span) // propagate request span
.await
};
simplex_write
.shutdown()
.await
.map_err(|err| BasebackupError::Client(err, "simplex_write"))?;
result
});
@@ -3772,6 +3761,36 @@ impl proto::PageService for GrpcPageServiceHandler {
let resp: page_api::GetSlruSegmentResponse = resp.segment;
Ok(tonic::Response::new(resp.into()))
}
#[instrument(skip_all, fields(lsn))]
async fn lease_lsn(
&self,
req: tonic::Request<proto::LeaseLsnRequest>,
) -> Result<tonic::Response<proto::LeaseLsnResponse>, tonic::Status> {
let timeline = self.get_request_timeline(&req).await?;
let ctx = self.ctx.with_scope_timeline(&timeline);
// Validate and convert the request, and decorate the span.
let req: page_api::LeaseLsnRequest = req.into_inner().try_into()?;
span_record!(lsn=%req.lsn);
// Attempt to acquire a lease. Return FailedPrecondition if the lease could not be granted.
let lease_length = timeline.get_lsn_lease_length();
let expires = match timeline.renew_lsn_lease(req.lsn, lease_length, &ctx) {
Ok(lease) => lease.valid_until,
Err(err) => return Err(tonic::Status::failed_precondition(format!("{err}"))),
};
// TODO: is this spammy? Move it compute-side?
info!(
"acquired lease for {} until {}",
req.lsn,
chrono::DateTime::<Utc>::from(expires).to_rfc3339()
);
Ok(tonic::Response::new(expires.into()))
}
}
/// gRPC middleware layer that handles observability concerns:

View File

@@ -25,7 +25,7 @@ use pageserver_api::keyspace::{KeySpaceRandomAccum, SparseKeySpace};
use pageserver_api::models::RelSizeMigration;
use pageserver_api::reltag::{BlockNumber, RelTag, SlruKind};
use pageserver_api::shard::ShardIdentity;
use postgres_ffi::{BLCKSZ, TimestampTz, TransactionId};
use postgres_ffi::{BLCKSZ, PgMajorVersion, TimestampTz, TransactionId};
use postgres_ffi_types::forknum::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
use postgres_ffi_types::{Oid, RepOriginId};
use serde::{Deserialize, Serialize};
@@ -1081,7 +1081,7 @@ impl Timeline {
// fetch directory entry
let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
if self.pg_version >= 17 {
if self.pg_version >= PgMajorVersion::PG17 {
Ok(TwoPhaseDirectoryV17::des(&buf)?.xids)
} else {
Ok(TwoPhaseDirectory::des(&buf)?
@@ -1613,7 +1613,7 @@ impl DatadirModification<'_> {
.push((DirectoryKind::Db, MetricsUpdate::Set(0)));
self.put(DBDIR_KEY, Value::Image(buf.into()));
let buf = if self.tline.pg_version >= 17 {
let buf = if self.tline.pg_version >= PgMajorVersion::PG17 {
TwoPhaseDirectoryV17::ser(&TwoPhaseDirectoryV17 {
xids: HashSet::new(),
})
@@ -1967,7 +1967,7 @@ impl DatadirModification<'_> {
) -> Result<(), WalIngestError> {
// Add it to the directory entry
let dirbuf = self.get(TWOPHASEDIR_KEY, ctx).await?;
let newdirbuf = if self.tline.pg_version >= 17 {
let newdirbuf = if self.tline.pg_version >= PgMajorVersion::PG17 {
let mut dir = TwoPhaseDirectoryV17::des(&dirbuf)?;
if !dir.xids.insert(xid) {
Err(WalIngestErrorKind::FileAlreadyExists(xid))?;
@@ -2383,7 +2383,7 @@ impl DatadirModification<'_> {
) -> Result<(), WalIngestError> {
// Remove it from the directory entry
let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
let newdirbuf = if self.tline.pg_version >= 17 {
let newdirbuf = if self.tline.pg_version >= PgMajorVersion::PG17 {
let mut dir = TwoPhaseDirectoryV17::des(&buf)?;
if !dir.xids.remove(&xid) {
@@ -3015,7 +3015,7 @@ mod tests {
// This shard will get the even blocks
let shard = ShardIdentity::from_params(
ShardNumber(0),
&ShardParameters {
ShardParameters {
count: ShardCount(2),
stripe_size: ShardStripeSize(1),
},

View File

@@ -38,6 +38,7 @@ use pageserver_api::models::{
WalRedoManagerStatus,
};
use pageserver_api::shard::{ShardIdentity, ShardStripeSize, TenantShardId};
use postgres_ffi::PgMajorVersion;
use remote_storage::{DownloadError, GenericRemoteStorage, TimeoutOrCancel};
use remote_timeline_client::index::GcCompactionState;
use remote_timeline_client::manifest::{
@@ -79,7 +80,7 @@ use self::timeline::uninit::{TimelineCreateGuard, TimelineExclusionError, Uninit
use self::timeline::{
EvictionTaskTenantState, GcCutoffs, TimelineDeleteProgress, TimelineResources, WaitLsnError,
};
use crate::basebackup_cache::BasebackupPrepareSender;
use crate::basebackup_cache::BasebackupCache;
use crate::config::PageServerConf;
use crate::context;
use crate::context::RequestContextBuilder;
@@ -161,7 +162,7 @@ pub struct TenantSharedResources {
pub remote_storage: GenericRemoteStorage,
pub deletion_queue_client: DeletionQueueClient,
pub l0_flush_global_state: L0FlushGlobalState,
pub basebackup_prepare_sender: BasebackupPrepareSender,
pub basebackup_cache: Arc<BasebackupCache>,
pub feature_resolver: FeatureResolver,
}
@@ -330,7 +331,7 @@ pub struct TenantShard {
deletion_queue_client: DeletionQueueClient,
/// A channel to send async requests to prepare a basebackup for the basebackup cache.
basebackup_prepare_sender: BasebackupPrepareSender,
basebackup_cache: Arc<BasebackupCache>,
/// Cached logical sizes updated updated on each [`TenantShard::gather_size_inputs`].
cached_logical_sizes: tokio::sync::Mutex<HashMap<(TimelineId, Lsn), u64>>,
@@ -497,7 +498,7 @@ impl WalRedoManager {
lsn: Lsn,
base_img: Option<(Lsn, bytes::Bytes)>,
records: Vec<(Lsn, wal_decoder::models::record::NeonWalRecord)>,
pg_version: u32,
pg_version: PgMajorVersion,
redo_attempt_type: RedoAttemptType,
) -> Result<bytes::Bytes, walredo::Error> {
match self {
@@ -933,7 +934,7 @@ pub(crate) enum CreateTimelineParams {
pub(crate) struct CreateTimelineParamsBootstrap {
pub(crate) new_timeline_id: TimelineId,
pub(crate) existing_initdb_timeline_id: Option<TimelineId>,
pub(crate) pg_version: u32,
pub(crate) pg_version: PgMajorVersion,
}
/// NB: See comment on [`CreateTimelineIdempotency::Branch`] for why there's no `pg_version` here.
@@ -971,7 +972,7 @@ pub(crate) enum CreateTimelineIdempotency {
/// NB: special treatment, see comment in [`Self`].
FailWithConflict,
Bootstrap {
pg_version: u32,
pg_version: PgMajorVersion,
},
/// NB: branches always have the same `pg_version` as their ancestor.
/// While [`pageserver_api::models::TimelineCreateRequestMode::Branch::pg_version`]
@@ -1362,7 +1363,7 @@ impl TenantShard {
remote_storage,
deletion_queue_client,
l0_flush_global_state,
basebackup_prepare_sender,
basebackup_cache,
feature_resolver,
} = resources;
@@ -1379,7 +1380,7 @@ impl TenantShard {
remote_storage.clone(),
deletion_queue_client,
l0_flush_global_state,
basebackup_prepare_sender,
basebackup_cache,
feature_resolver,
));
@@ -2541,7 +2542,7 @@ impl TenantShard {
self: &Arc<Self>,
new_timeline_id: TimelineId,
initdb_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
ctx: &RequestContext,
) -> anyhow::Result<(UninitializedTimeline, RequestContext)> {
anyhow::ensure!(
@@ -2593,7 +2594,7 @@ impl TenantShard {
self: &Arc<Self>,
new_timeline_id: TimelineId,
initdb_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
ctx: &RequestContext,
) -> anyhow::Result<Arc<Timeline>> {
let (uninit_tl, ctx) = self
@@ -2632,7 +2633,7 @@ impl TenantShard {
self: &Arc<Self>,
new_timeline_id: TimelineId,
initdb_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
ctx: &RequestContext,
in_memory_layer_desc: Vec<timeline::InMemoryLayerTestDesc>,
delta_layer_desc: Vec<timeline::DeltaLayerTestDesc>,
@@ -2898,7 +2899,7 @@ impl TenantShard {
Lsn(0),
initdb_lsn,
initdb_lsn,
15,
PgMajorVersion::PG15,
);
this.prepare_new_timeline(
new_timeline_id,
@@ -3871,6 +3872,10 @@ impl TenantShard {
&self.tenant_shard_id
}
pub(crate) fn get_shard_identity(&self) -> ShardIdentity {
self.shard_identity
}
pub(crate) fn get_shard_stripe_size(&self) -> ShardStripeSize {
self.shard_identity.stripe_size
}
@@ -4379,7 +4384,7 @@ impl TenantShard {
remote_storage: GenericRemoteStorage,
deletion_queue_client: DeletionQueueClient,
l0_flush_global_state: L0FlushGlobalState,
basebackup_prepare_sender: BasebackupPrepareSender,
basebackup_cache: Arc<BasebackupCache>,
feature_resolver: FeatureResolver,
) -> TenantShard {
assert!(!attached_conf.location.generation.is_none());
@@ -4484,7 +4489,7 @@ impl TenantShard {
ongoing_timeline_detach: std::sync::Mutex::default(),
gc_block: Default::default(),
l0_flush_global_state,
basebackup_prepare_sender,
basebackup_cache,
feature_resolver,
}
}
@@ -4524,6 +4529,10 @@ impl TenantShard {
Ok(toml_edit::de::from_str::<LocationConf>(&config)?)
}
/// Stores a tenant location config to disk.
///
/// NB: make sure to call `ShardIdentity::assert_equal` before persisting a new config, to avoid
/// changes to shard parameters that may result in data corruption.
#[tracing::instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()))]
pub(super) async fn persist_tenant_config(
conf: &'static PageServerConf,
@@ -5090,7 +5099,7 @@ impl TenantShard {
pub(crate) async fn bootstrap_timeline_test(
self: &Arc<Self>,
timeline_id: TimelineId,
pg_version: u32,
pg_version: PgMajorVersion,
load_existing_initdb: Option<TimelineId>,
ctx: &RequestContext,
) -> anyhow::Result<Arc<Timeline>> {
@@ -5232,7 +5241,7 @@ impl TenantShard {
async fn bootstrap_timeline(
self: &Arc<Self>,
timeline_id: TimelineId,
pg_version: u32,
pg_version: PgMajorVersion,
load_existing_initdb: Option<TimelineId>,
ctx: &RequestContext,
) -> Result<CreateTimelineResult, CreateTimelineError> {
@@ -5413,7 +5422,7 @@ impl TenantShard {
pagestream_throttle_metrics: self.pagestream_throttle_metrics.clone(),
l0_compaction_trigger: self.l0_compaction_trigger.clone(),
l0_flush_global_state: self.l0_flush_global_state.clone(),
basebackup_prepare_sender: self.basebackup_prepare_sender.clone(),
basebackup_cache: self.basebackup_cache.clone(),
feature_resolver: self.feature_resolver.clone(),
}
}
@@ -5770,7 +5779,7 @@ impl TenantShard {
async fn run_initdb(
conf: &'static PageServerConf,
initdb_target_dir: &Utf8Path,
pg_version: u32,
pg_version: PgMajorVersion,
cancel: &CancellationToken,
) -> Result<(), InitdbError> {
let initdb_bin_path = conf
@@ -5999,7 +6008,7 @@ pub(crate) mod harness {
) -> anyhow::Result<Arc<TenantShard>> {
let walredo_mgr = Arc::new(WalRedoManager::from(TestRedoManager));
let (basebackup_requst_sender, _) = tokio::sync::mpsc::unbounded_channel();
let (basebackup_cache, _) = BasebackupCache::new(Utf8PathBuf::new(), None);
let tenant = Arc::new(TenantShard::new(
TenantState::Attaching,
@@ -6007,7 +6016,7 @@ pub(crate) mod harness {
AttachedTenantConf::try_from(LocationConf::attached_single(
self.tenant_conf.clone(),
self.generation,
&ShardParameters::default(),
ShardParameters::default(),
))
.unwrap(),
self.shard_identity,
@@ -6017,7 +6026,7 @@ pub(crate) mod harness {
self.deletion_queue.new_client(),
// TODO: ideally we should run all unit tests with both configs
L0FlushGlobalState::new(L0FlushConfig::default()),
basebackup_requst_sender,
basebackup_cache,
FeatureResolver::new_disabled(),
));
@@ -6051,7 +6060,7 @@ pub(crate) mod harness {
lsn: Lsn,
base_img: Option<(Lsn, Bytes)>,
records: Vec<(Lsn, NeonWalRecord)>,
_pg_version: u32,
_pg_version: PgMajorVersion,
_redo_attempt_type: RedoAttemptType,
) -> Result<Bytes, walredo::Error> {
let records_neon = records.iter().all(|r| apply_neon::can_apply_in_neon(&r.1));
@@ -6223,7 +6232,7 @@ mod tests {
async fn randomize_timeline(
tenant: &Arc<TenantShard>,
new_timeline_id: TimelineId,
pg_version: u32,
pg_version: PgMajorVersion,
spec: TestTimelineSpecification,
random: &mut rand::rngs::StdRng,
ctx: &RequestContext,
@@ -11428,11 +11437,11 @@ mod tests {
if left != right {
eprintln!("---LEFT---");
for left in left.iter() {
eprintln!("{}", left);
eprintln!("{left}");
}
eprintln!("---RIGHT---");
for right in right.iter() {
eprintln!("{}", right);
eprintln!("{right}");
}
assert_eq!(left, right);
}

View File

@@ -12,6 +12,7 @@
use pageserver_api::models;
use pageserver_api::shard::{ShardCount, ShardIdentity, ShardNumber, ShardStripeSize};
use serde::{Deserialize, Serialize};
use utils::critical;
use utils::generation::Generation;
#[derive(Debug, Copy, Clone, Serialize, Deserialize, PartialEq, Eq)]
@@ -136,7 +137,7 @@ impl LocationConf {
pub(crate) fn attached_single(
tenant_conf: pageserver_api::models::TenantConfig,
generation: Generation,
shard_params: &models::ShardParameters,
shard_params: models::ShardParameters,
) -> Self {
Self {
mode: LocationMode::Attached(AttachedLocationConfig {
@@ -171,6 +172,16 @@ impl LocationConf {
}
}
// This should never happen.
// TODO: turn this into a proper assertion.
if stripe_size != self.shard.stripe_size {
critical!(
"stripe size mismatch: {} != {}",
self.shard.stripe_size,
stripe_size,
);
}
self.shard.stripe_size = stripe_size;
}

View File

@@ -18,6 +18,7 @@
//! [`IndexPart`]: super::remote_timeline_client::index::IndexPart
use anyhow::ensure;
use postgres_ffi::PgMajorVersion;
use serde::{Deserialize, Serialize};
use utils::bin_ser::{BeSer, SerializeError};
use utils::id::TimelineId;
@@ -136,7 +137,7 @@ struct TimelineMetadataBodyV2 {
latest_gc_cutoff_lsn: Lsn,
initdb_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
}
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
@@ -167,7 +168,7 @@ impl TimelineMetadata {
ancestor_lsn: Lsn,
latest_gc_cutoff_lsn: Lsn,
initdb_lsn: Lsn,
pg_version: u32,
pg_version: PgMajorVersion,
) -> Self {
Self {
hdr: TimelineMetadataHeader {
@@ -215,7 +216,7 @@ impl TimelineMetadata {
ancestor_lsn: body.ancestor_lsn,
latest_gc_cutoff_lsn: body.latest_gc_cutoff_lsn,
initdb_lsn: body.initdb_lsn,
pg_version: 14, // All timelines created before this version had pg_version 14
pg_version: PgMajorVersion::PG14, // All timelines created before this version had pg_version 14
};
hdr.format_version = METADATA_FORMAT_VERSION;
@@ -317,7 +318,7 @@ impl TimelineMetadata {
self.body.initdb_lsn
}
pub fn pg_version(&self) -> u32 {
pub fn pg_version(&self) -> PgMajorVersion {
self.body.pg_version
}
@@ -331,7 +332,7 @@ impl TimelineMetadata {
Lsn::from_hex("00000000").unwrap(),
Lsn::from_hex("00000000").unwrap(),
Lsn::from_hex("00000000").unwrap(),
0,
PgMajorVersion::PG14,
);
let bytes = instance.to_bytes().unwrap();
Self::from_bytes(&bytes).unwrap()
@@ -545,7 +546,7 @@ mod tests {
Lsn(0),
Lsn(0),
Lsn(0),
14, // All timelines created before this version had pg_version 14
PgMajorVersion::PG14, // All timelines created before this version had pg_version 14
);
assert_eq!(
@@ -565,7 +566,7 @@ mod tests {
Lsn(0),
// Updating this version to 17 will cause the test to fail at the
// next assert_eq!().
16,
PgMajorVersion::PG16,
);
let expected_bytes = vec![
/* TimelineMetadataHeader */

View File

@@ -880,6 +880,9 @@ impl TenantManager {
// phase of writing config and/or waiting for flush, before returning.
match fast_path_taken {
Some(FastPathModified::Attached(tenant)) => {
tenant
.shard_identity
.assert_equal(new_location_config.shard);
TenantShard::persist_tenant_config(
self.conf,
&tenant_shard_id,
@@ -914,7 +917,10 @@ impl TenantManager {
return Ok(Some(tenant));
}
Some(FastPathModified::Secondary(_secondary_tenant)) => {
Some(FastPathModified::Secondary(secondary_tenant)) => {
secondary_tenant
.shard_identity
.assert_equal(new_location_config.shard);
TenantShard::persist_tenant_config(
self.conf,
&tenant_shard_id,
@@ -948,6 +954,10 @@ impl TenantManager {
match slot_guard.get_old_value() {
Some(TenantSlot::Attached(tenant)) => {
tenant
.shard_identity
.assert_equal(new_location_config.shard);
// The case where we keep a Tenant alive was covered above in the special case
// for Attached->Attached transitions in the same generation. By this point,
// if we see an attached tenant we know it will be discarded and should be
@@ -981,9 +991,13 @@ impl TenantManager {
// rather than assuming it to be empty.
spawn_mode = SpawnMode::Eager;
}
Some(TenantSlot::Secondary(state)) => {
Some(TenantSlot::Secondary(secondary_tenant)) => {
secondary_tenant
.shard_identity
.assert_equal(new_location_config.shard);
info!("Shutting down secondary tenant");
state.shutdown().await;
secondary_tenant.shutdown().await;
}
Some(TenantSlot::InProgress(_)) => {
// This should never happen: acquire_slot should error out
@@ -2200,7 +2214,7 @@ impl TenantManager {
selector: ShardSelector,
) -> ShardResolveResult {
let tenants = self.tenants.read().unwrap();
let mut want_shard = None;
let mut want_shard: Option<ShardIndex> = None;
let mut any_in_progress = None;
match &*tenants {
@@ -2225,14 +2239,23 @@ impl TenantManager {
return ShardResolveResult::Found(tenant.clone());
}
ShardSelector::Page(key) => {
// First slot we see for this tenant, calculate the expected shard number
// for the key: we will use this for checking if this and subsequent
// slots contain the key, rather than recalculating the hash each time.
if want_shard.is_none() {
want_shard = Some(tenant.shard_identity.get_shard_number(&key));
// Each time we find an attached slot with a different shard count,
// recompute the expected shard number: during shard splits we might
// have multiple shards with the old shard count.
if want_shard.is_none()
|| want_shard.unwrap().shard_count != tenant.shard_identity.count
{
want_shard = Some(ShardIndex {
shard_number: tenant.shard_identity.get_shard_number(&key),
shard_count: tenant.shard_identity.count,
});
}
if Some(tenant.shard_identity.number) == want_shard {
if Some(ShardIndex {
shard_number: tenant.shard_identity.number,
shard_count: tenant.shard_identity.count,
}) == want_shard
{
return ShardResolveResult::Found(tenant.clone());
}
}
@@ -2891,14 +2914,18 @@ mod tests {
use std::collections::BTreeMap;
use std::sync::Arc;
use camino::Utf8PathBuf;
use storage_broker::BrokerClientChannel;
use tracing::Instrument;
use super::super::harness::TenantHarness;
use super::TenantsMap;
use crate::tenant::{
TenantSharedResources,
mgr::{BackgroundPurges, TenantManager, TenantSlot},
use crate::{
basebackup_cache::BasebackupCache,
tenant::{
TenantSharedResources,
mgr::{BackgroundPurges, TenantManager, TenantSlot},
},
};
#[tokio::test(start_paused = true)]
@@ -2924,9 +2951,7 @@ mod tests {
// Invoke remove_tenant_from_memory with a cleanup hook that blocks until we manually
// permit it to proceed: that will stick the tenant in InProgress
let (basebackup_prepare_sender, _) = tokio::sync::mpsc::unbounded_channel::<
crate::basebackup_cache::BasebackupPrepareRequest,
>();
let (basebackup_cache, _) = BasebackupCache::new(Utf8PathBuf::new(), None);
let tenant_manager = TenantManager {
tenants: std::sync::RwLock::new(TenantsMap::Open(tenants)),
@@ -2940,7 +2965,7 @@ mod tests {
l0_flush_global_state: crate::l0_flush::L0FlushGlobalState::new(
h.conf.l0_flush.clone(),
),
basebackup_prepare_sender,
basebackup_cache,
feature_resolver: crate::feature_resolver::FeatureResolver::new_disabled(),
},
cancel: tokio_util::sync::CancellationToken::new(),

View File

@@ -427,8 +427,8 @@ impl GcBlocking {
#[cfg(test)]
mod tests {
use postgres_ffi::PgMajorVersion;
use std::str::FromStr;
use utils::id::TimelineId;
use super::*;
@@ -831,7 +831,7 @@ mod tests {
Lsn::INVALID,
Lsn::from_str("0/1696070").unwrap(),
Lsn::from_str("0/1696070").unwrap(),
14,
PgMajorVersion::PG14,
).with_recalculated_checksum().unwrap(),
deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
archived_at: None,
@@ -893,7 +893,7 @@ mod tests {
Lsn::INVALID,
Lsn::from_str("0/1696070").unwrap(),
Lsn::from_str("0/1696070").unwrap(),
14,
PgMajorVersion::PG14,
).with_recalculated_checksum().unwrap(),
deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
archived_at: Some(parse_naive_datetime("2023-04-29T09:00:00.123000000")),
@@ -957,7 +957,7 @@ mod tests {
Lsn::INVALID,
Lsn::from_str("0/1696070").unwrap(),
Lsn::from_str("0/1696070").unwrap(),
14,
PgMajorVersion::PG14,
).with_recalculated_checksum().unwrap(),
deleted_at: None,
lineage: Default::default(),
@@ -1033,7 +1033,7 @@ mod tests {
Lsn::INVALID,
Lsn::from_str("0/1696070").unwrap(),
Lsn::from_str("0/1696070").unwrap(),
14,
PgMajorVersion::PG14,
).with_recalculated_checksum().unwrap(),
deleted_at: None,
lineage: Default::default(),
@@ -1114,7 +1114,7 @@ mod tests {
Lsn::INVALID,
Lsn::from_str("0/1696070").unwrap(),
Lsn::from_str("0/1696070").unwrap(),
14,
PgMajorVersion::PG14,
).with_recalculated_checksum().unwrap(),
deleted_at: None,
lineage: Default::default(),
@@ -1199,7 +1199,7 @@ mod tests {
Lsn::INVALID,
Lsn::from_str("0/1696070").unwrap(),
Lsn::from_str("0/1696070").unwrap(),
14,
PgMajorVersion::PG14,
).with_recalculated_checksum().unwrap(),
deleted_at: None,
lineage: Default::default(),
@@ -1287,7 +1287,7 @@ mod tests {
Lsn::INVALID,
Lsn::from_str("0/1696070").unwrap(),
Lsn::from_str("0/1696070").unwrap(),
14,
PgMajorVersion::PG14,
).with_recalculated_checksum().unwrap(),
deleted_at: None,
lineage: Default::default(),

View File

@@ -101,7 +101,7 @@ pub(crate) struct SecondaryTenant {
// Secondary mode does not need the full shard identity or the pageserver_api::models::TenantConfig. However,
// storing these enables us to report our full LocationConf, enabling convenient reconciliation
// by the control plane (see [`Self::get_location_conf`])
shard_identity: ShardIdentity,
pub(crate) shard_identity: ShardIdentity,
tenant_conf: std::sync::Mutex<pageserver_api::models::TenantConfig>,
// Internal state used by the Downloader.

View File

@@ -55,11 +55,11 @@ pub struct BatchLayerWriter {
}
impl BatchLayerWriter {
pub async fn new(conf: &'static PageServerConf) -> anyhow::Result<Self> {
Ok(Self {
pub fn new(conf: &'static PageServerConf) -> Self {
Self {
generated_layer_writers: Vec::new(),
conf,
})
}
}
pub fn add_unfinished_image_writer(
@@ -209,6 +209,7 @@ impl<'a> SplitImageLayerWriter<'a> {
) -> anyhow::Result<Self> {
Ok(Self {
target_layer_size,
// XXX make this lazy like in SplitDeltaLayerWriter?
inner: ImageLayerWriter::new(
conf,
timeline_id,
@@ -223,7 +224,7 @@ impl<'a> SplitImageLayerWriter<'a> {
conf,
timeline_id,
tenant_shard_id,
batches: BatchLayerWriter::new(conf).await?,
batches: BatchLayerWriter::new(conf),
lsn,
start_key,
gate,
@@ -319,7 +320,7 @@ pub struct SplitDeltaLayerWriter<'a> {
}
impl<'a> SplitDeltaLayerWriter<'a> {
pub async fn new(
pub fn new(
conf: &'static PageServerConf,
timeline_id: TimelineId,
tenant_shard_id: TenantShardId,
@@ -327,8 +328,8 @@ impl<'a> SplitDeltaLayerWriter<'a> {
target_layer_size: u64,
gate: &'a utils::sync::gate::Gate,
cancel: CancellationToken,
) -> anyhow::Result<Self> {
Ok(Self {
) -> Self {
Self {
target_layer_size,
inner: None,
conf,
@@ -336,10 +337,10 @@ impl<'a> SplitDeltaLayerWriter<'a> {
tenant_shard_id,
lsn_range,
last_key_written: Key::MIN,
batches: BatchLayerWriter::new(conf).await?,
batches: BatchLayerWriter::new(conf),
gate,
cancel,
})
}
}
pub async fn put_value(
@@ -510,9 +511,7 @@ mod tests {
4 * 1024 * 1024,
&tline.gate,
tline.cancel.clone(),
)
.await
.unwrap();
);
image_writer
.put_image(get_key(0), get_img(0), &ctx)
@@ -590,9 +589,7 @@ mod tests {
4 * 1024 * 1024,
&tline.gate,
tline.cancel.clone(),
)
.await
.unwrap();
);
const N: usize = 2000;
for i in 0..N {
let i = i as u32;
@@ -692,9 +689,7 @@ mod tests {
4 * 1024,
&tline.gate,
tline.cancel.clone(),
)
.await
.unwrap();
);
image_writer
.put_image(get_key(0), get_img(0), &ctx)
@@ -770,9 +765,7 @@ mod tests {
4 * 1024 * 1024,
&tline.gate,
tline.cancel.clone(),
)
.await
.unwrap();
);
for i in 0..N {
let i = i as u32;

View File

@@ -1622,11 +1622,6 @@ impl DeltaLayerIterator<'_> {
pub(crate) mod test {
use std::collections::BTreeMap;
use bytes::Bytes;
use itertools::MinMaxResult;
use rand::prelude::{SeedableRng, SliceRandom, StdRng};
use rand::{Rng, RngCore};
use super::*;
use crate::DEFAULT_PG_VERSION;
use crate::context::DownloadBehavior;
@@ -1636,6 +1631,11 @@ pub(crate) mod test {
use crate::tenant::storage_layer::{Layer, ResidentLayer};
use crate::tenant::timeline::layer_manager::LayerManagerLockHolder;
use crate::tenant::{TenantShard, Timeline};
use bytes::Bytes;
use itertools::MinMaxResult;
use postgres_ffi::PgMajorVersion;
use rand::prelude::{SeedableRng, SliceRandom, StdRng};
use rand::{Rng, RngCore};
/// Construct an index for a fictional delta layer and and then
/// traverse in order to plan vectored reads for a query. Finally,
@@ -1995,7 +1995,7 @@ pub(crate) mod test {
let (tenant, ctx) = h.load().await;
let ctx = &ctx;
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, ctx)
.create_test_timeline(TimelineId::generate(), Lsn(0x10), PgMajorVersion::PG14, ctx)
.await
.unwrap();
let ctx = &ctx.with_scope_timeline(&timeline);

View File

@@ -1,6 +1,7 @@
use std::time::UNIX_EPOCH;
use pageserver_api::key::{CONTROLFILE_KEY, Key};
use postgres_ffi::PgMajorVersion;
use tokio::task::JoinSet;
use utils::completion::{self, Completion};
use utils::id::TimelineId;
@@ -45,7 +46,7 @@ async fn smoke_test() {
.create_test_timeline_with_layers(
TimelineId::generate(),
Lsn(0x10),
14,
PgMajorVersion::PG14,
&ctx,
Default::default(), // in-memory layers
Default::default(),
@@ -256,7 +257,12 @@ async fn evict_and_wait_on_wanted_deleted() {
let (tenant, ctx) = h.load().await;
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
.create_test_timeline(
TimelineId::generate(),
Lsn(0x10),
PgMajorVersion::PG14,
&ctx,
)
.await
.unwrap();
@@ -341,7 +347,12 @@ fn read_wins_pending_eviction() {
let download_span = span.in_scope(|| tracing::info_span!("downloading", timeline_id = 1));
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
.create_test_timeline(
TimelineId::generate(),
Lsn(0x10),
PgMajorVersion::PG14,
&ctx,
)
.await
.unwrap();
let ctx = ctx.with_scope_timeline(&timeline);
@@ -474,7 +485,12 @@ fn multiple_pending_evictions_scenario(name: &'static str, in_order: bool) {
let download_span = span.in_scope(|| tracing::info_span!("downloading", timeline_id = 1));
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
.create_test_timeline(
TimelineId::generate(),
Lsn(0x10),
PgMajorVersion::PG14,
&ctx,
)
.await
.unwrap();
let ctx = ctx.with_scope_timeline(&timeline);
@@ -644,7 +660,12 @@ async fn cancelled_get_or_maybe_download_does_not_cancel_eviction() {
let (tenant, ctx) = h.load().await;
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
.create_test_timeline(
TimelineId::generate(),
Lsn(0x10),
PgMajorVersion::PG14,
&ctx,
)
.await
.unwrap();
let ctx = ctx.with_scope_timeline(&timeline);
@@ -730,7 +751,12 @@ async fn evict_and_wait_does_not_wait_for_download() {
let download_span = span.in_scope(|| tracing::info_span!("downloading", timeline_id = 1));
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
.create_test_timeline(
TimelineId::generate(),
Lsn(0x10),
PgMajorVersion::PG14,
&ctx,
)
.await
.unwrap();
let ctx = ctx.with_scope_timeline(&timeline);
@@ -836,7 +862,12 @@ async fn eviction_cancellation_on_drop() {
let (tenant, ctx) = h.load().await;
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
.create_test_timeline(
TimelineId::generate(),
Lsn(0x10),
PgMajorVersion::PG14,
&ctx,
)
.await
.unwrap();

View File

@@ -17,14 +17,17 @@ use tracing::*;
use utils::backoff::exponential_backoff_duration;
use utils::completion::Barrier;
use utils::pausable_failpoint;
use utils::sync::gate::GateError;
use crate::context::{DownloadBehavior, RequestContext};
use crate::metrics::{self, BackgroundLoopSemaphoreMetricsRecorder, TENANT_TASK_EVENTS};
use crate::task_mgr::{self, BACKGROUND_RUNTIME, TOKIO_WORKER_THREADS, TaskKind};
use crate::tenant::blob_io::WriteBlobError;
use crate::tenant::throttle::Stats;
use crate::tenant::timeline::CompactionError;
use crate::tenant::timeline::compaction::CompactionOutcome;
use crate::tenant::{TenantShard, TenantState};
use crate::virtual_file::owned_buffers_io::write::FlushTaskError;
/// Semaphore limiting concurrent background tasks (across all tenants).
///
@@ -313,7 +316,20 @@ pub(crate) fn log_compaction_error(
let timeline = root_cause
.downcast_ref::<PageReconstructError>()
.is_some_and(|e| e.is_stopping());
let is_stopping = upload_queue || timeline;
let buffered_writer_flush_task_canelled = root_cause
.downcast_ref::<FlushTaskError>()
.is_some_and(|e| e.is_cancel());
let write_blob_cancelled = root_cause
.downcast_ref::<WriteBlobError>()
.is_some_and(|e| e.is_cancel());
let gate_closed = root_cause
.downcast_ref::<GateError>()
.is_some_and(|e| e.is_cancel());
let is_stopping = upload_queue
|| timeline
|| buffered_writer_flush_task_canelled
|| write_blob_cancelled
|| gate_closed;
if is_stopping {
Level::INFO

View File

@@ -58,7 +58,7 @@ use pageserver_api::reltag::{BlockNumber, RelTag};
use pageserver_api::shard::{ShardIdentity, ShardIndex, ShardNumber, TenantShardId};
use postgres_connection::PgConnectionConfig;
use postgres_ffi::v14::xlog_utils;
use postgres_ffi::{WAL_SEGMENT_SIZE, to_pg_timestamp};
use postgres_ffi::{PgMajorVersion, WAL_SEGMENT_SIZE, to_pg_timestamp};
use rand::Rng;
use remote_storage::DownloadError;
use serde_with::serde_as;
@@ -95,12 +95,12 @@ use super::storage_layer::{LayerFringe, LayerVisibilityHint, ReadableLayer};
use super::tasks::log_compaction_error;
use super::upload_queue::NotInitialized;
use super::{
AttachedTenantConf, BasebackupPrepareSender, GcError, HeatMapTimeline, MaybeOffloaded,
AttachedTenantConf, GcError, HeatMapTimeline, MaybeOffloaded,
debug_assert_current_span_has_tenant_and_timeline_id,
};
use crate::PERF_TRACE_TARGET;
use crate::aux_file::AuxFileSizeEstimator;
use crate::basebackup_cache::BasebackupPrepareRequest;
use crate::basebackup_cache::BasebackupCache;
use crate::config::PageServerConf;
use crate::context::{
DownloadBehavior, PerfInstrumentFutureExt, RequestContext, RequestContextBuilder,
@@ -201,7 +201,7 @@ pub struct TimelineResources {
pub pagestream_throttle_metrics: Arc<crate::metrics::tenant_throttling::Pagestream>,
pub l0_compaction_trigger: Arc<Notify>,
pub l0_flush_global_state: l0_flush::L0FlushGlobalState,
pub basebackup_prepare_sender: BasebackupPrepareSender,
pub basebackup_cache: Arc<BasebackupCache>,
pub feature_resolver: FeatureResolver,
}
@@ -225,7 +225,7 @@ pub struct Timeline {
/// to shards, and is constant through the lifetime of this Timeline.
shard_identity: ShardIdentity,
pub pg_version: u32,
pub pg_version: PgMajorVersion,
/// The tuple has two elements.
/// 1. `LayerFileManager` keeps track of the various physical representations of the layer files (inmem, local, remote).
@@ -448,7 +448,7 @@ pub struct Timeline {
wait_lsn_log_slow: tokio::sync::Semaphore,
/// A channel to send async requests to prepare a basebackup for the basebackup cache.
basebackup_prepare_sender: BasebackupPrepareSender,
basebackup_cache: Arc<BasebackupCache>,
feature_resolver: FeatureResolver,
}
@@ -763,7 +763,7 @@ pub(crate) enum CreateImageLayersError {
PageReconstructError(#[source] PageReconstructError),
#[error(transparent)]
Other(#[from] anyhow::Error),
Other(anyhow::Error),
}
impl From<layer_manager::Shutdown> for CreateImageLayersError {
@@ -2500,6 +2500,37 @@ impl Timeline {
.unwrap_or(self.conf.default_tenant_conf.basebackup_cache_enabled)
}
/// Try to get a basebackup from the on-disk cache.
pub(crate) async fn get_cached_basebackup(&self, lsn: Lsn) -> Option<tokio::fs::File> {
self.basebackup_cache
.get(self.tenant_shard_id.tenant_id, self.timeline_id, lsn)
.await
}
/// Convenience method to attempt fetching a basebackup for the timeline if enabled and safe for
/// the given request parameters.
///
/// TODO: consider moving this onto GrpcPageServiceHandler once the libpq handler is gone.
pub async fn get_cached_basebackup_if_enabled(
&self,
lsn: Option<Lsn>,
prev_lsn: Option<Lsn>,
full: bool,
replica: bool,
gzip: bool,
) -> Option<tokio::fs::File> {
if !self.is_basebackup_cache_enabled() || !self.basebackup_cache.is_enabled() {
return None;
}
// We have to know which LSN to fetch the basebackup for.
let lsn = lsn?;
// We only cache gzipped, non-full basebackups for primary computes with automatic prev_lsn.
if prev_lsn.is_some() || full || replica || !gzip {
return None;
}
self.get_cached_basebackup(lsn).await
}
/// Prepare basebackup for the given LSN and store it in the basebackup cache.
/// The method is asynchronous and returns immediately.
/// The actual basebackup preparation is performed in the background
@@ -2521,17 +2552,8 @@ impl Timeline {
return;
}
let res = self
.basebackup_prepare_sender
.send(BasebackupPrepareRequest {
tenant_shard_id: self.tenant_shard_id,
timeline_id: self.timeline_id,
lsn,
});
if let Err(e) = res {
// May happen during shutdown, it's not critical.
info!("Failed to send shutdown checkpoint: {e:#}");
}
self.basebackup_cache
.send_prepare(self.tenant_shard_id, self.timeline_id, lsn);
}
}
@@ -2913,7 +2935,7 @@ impl Timeline {
shard_identity: ShardIdentity,
walredo_mgr: Option<Arc<super::WalRedoManager>>,
resources: TimelineResources,
pg_version: u32,
pg_version: PgMajorVersion,
state: TimelineState,
attach_wal_lag_cooldown: Arc<OnceLock<WalLagCooldown>>,
create_idempotency: crate::tenant::CreateTimelineIdempotency,
@@ -3088,7 +3110,7 @@ impl Timeline {
wait_lsn_log_slow: tokio::sync::Semaphore::new(1),
basebackup_prepare_sender: resources.basebackup_prepare_sender,
basebackup_cache: resources.basebackup_cache,
feature_resolver: resources.feature_resolver,
};
@@ -4658,6 +4680,16 @@ impl Timeline {
mut layer_flush_start_rx: tokio::sync::watch::Receiver<(u64, Lsn)>,
ctx: &RequestContext,
) {
// Always notify waiters about the flush loop exiting since the loop might stop
// when the timeline hasn't been cancelled.
let scopeguard_rx = layer_flush_start_rx.clone();
scopeguard::defer! {
let (flush_counter, _) = *scopeguard_rx.borrow();
let _ = self
.layer_flush_done_tx
.send_replace((flush_counter, Err(FlushLayerError::Cancelled)));
}
// Subscribe to L0 delta layer updates, for compaction backpressure.
let mut watch_l0 = match self
.layers
@@ -4687,9 +4719,6 @@ impl Timeline {
let result = loop {
if self.cancel.is_cancelled() {
info!("dropping out of flush loop for timeline shutdown");
// Note: we do not bother transmitting into [`layer_flush_done_tx`], because
// anyone waiting on that will respect self.cancel as well: they will stop
// waiting at the same time we as drop out of this loop.
return;
}
@@ -5561,7 +5590,7 @@ impl Timeline {
self.should_check_if_image_layers_required(lsn)
};
let mut batch_image_writer = BatchLayerWriter::new(self.conf).await?;
let mut batch_image_writer = BatchLayerWriter::new(self.conf);
let mut all_generated = true;
@@ -5665,7 +5694,8 @@ impl Timeline {
self.cancel.clone(),
ctx,
)
.await?;
.await
.map_err(CreateImageLayersError::Other)?;
fail_point!("image-layer-writer-fail-before-finish", |_| {
Err(CreateImageLayersError::Other(anyhow::anyhow!(
@@ -5760,7 +5790,10 @@ impl Timeline {
}
}
let image_layers = batch_image_writer.finish(self, ctx).await?;
let image_layers = batch_image_writer
.finish(self, ctx)
.await
.map_err(CreateImageLayersError::Other)?;
let mut guard = self.layers.write(LayerManagerLockHolder::Compaction).await;
@@ -7593,6 +7626,7 @@ mod tests {
use std::sync::Arc;
use pageserver_api::key::Key;
use postgres_ffi::PgMajorVersion;
use std::iter::Iterator;
use tracing::Instrument;
use utils::id::TimelineId;
@@ -7667,7 +7701,7 @@ mod tests {
.create_test_timeline_with_layers(
TimelineId::generate(),
Lsn(0x10),
14,
PgMajorVersion::PG14,
&ctx,
Vec::new(), // in-memory layers
delta_layers,
@@ -7803,7 +7837,7 @@ mod tests {
.create_test_timeline_with_layers(
TimelineId::generate(),
Lsn(0x10),
14,
PgMajorVersion::PG14,
&ctx,
Vec::new(), // in-memory layers
delta_layers,
@@ -7863,7 +7897,12 @@ mod tests {
let (tenant, ctx) = harness.load().await;
let timeline = tenant
.create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
.create_test_timeline(
TimelineId::generate(),
Lsn(0x10),
PgMajorVersion::PG14,
&ctx,
)
.await
.unwrap();

View File

@@ -3531,10 +3531,7 @@ impl Timeline {
self.get_compaction_target_size(),
&self.gate,
self.cancel.clone(),
)
.await
.context("failed to create delta layer writer")
.map_err(CompactionError::Other)?;
);
#[derive(Default)]
struct RewritingLayers {
@@ -4330,7 +4327,8 @@ impl TimelineAdaptor {
self.timeline.cancel.clone(),
ctx,
)
.await?;
.await
.map_err(CreateImageLayersError::Other)?;
fail_point!("image-layer-writer-fail-before-finish", |_| {
Err(CreateImageLayersError::Other(anyhow::anyhow!(
@@ -4339,7 +4337,10 @@ impl TimelineAdaptor {
});
let keyspace = KeySpace {
ranges: self.get_keyspace(key_range, lsn, ctx).await?,
ranges: self
.get_keyspace(key_range, lsn, ctx)
.await
.map_err(CreateImageLayersError::Other)?,
};
// TODO set proper (stateful) start. The create_image_layer_for_rel_blocks function mostly
let outcome = self
@@ -4358,9 +4359,13 @@ impl TimelineAdaptor {
unfinished_image_layer,
} = outcome
{
let (desc, path) = unfinished_image_layer.finish(ctx).await?;
let (desc, path) = unfinished_image_layer
.finish(ctx)
.await
.map_err(CreateImageLayersError::Other)?;
let image_layer =
Layer::finish_creating(self.timeline.conf, &self.timeline, desc, &path)?;
Layer::finish_creating(self.timeline.conf, &self.timeline, desc, &path)
.map_err(CreateImageLayersError::Other)?;
self.new_images.push(image_layer);
}

View File

@@ -241,8 +241,17 @@ impl DeleteTimelineFlow {
{
Ok(r) => r,
Err(DownloadError::NotFound) => {
// Deletion is already complete
// Deletion is already complete.
// As we came here, we will need to remove the timeline from the tenant though.
tracing::info!("Timeline already deleted in remote storage");
if let TimelineOrOffloaded::Offloaded(_) = &timeline {
// We only supoprt this for offloaded timelines, as we don't know which state non-offloaded timelines are in.
tracing::info!(
"Timeline with gone index part is offloaded timeline. Removing from tenant."
);
remove_maybe_offloaded_timeline_from_tenant(tenant, &timeline, &guard)
.await?;
}
return Ok(());
}
Err(e) => {

View File

@@ -885,7 +885,7 @@ async fn remote_copy(
}
tracing::info!("Deleting orphan layer file to make way for hard linking");
// Delete orphan layer file and try again, to ensure this layer has a well understood source
std::fs::remove_file(adopted_path)
std::fs::remove_file(&adoptee_path)
.map_err(|e| Error::launder(e.into(), Error::Prepare))?;
std::fs::hard_link(adopted_path, &adoptee_path)
.map_err(|e| Error::launder(e.into(), Error::Prepare))?;

Some files were not shown because too many files have changed in this diff Show More