This allows us to error out in the case where we request flush but the
flush loop is not running.
Before, we would only track whether it was started, but not when it
exited.
Better to use an enum with 3 states than a 2-state bool because then
the error message can answer the question whether we ever started
the flush loop or not.
Previously in some cases local metadata was confused with remote one and
there was a check, that we write locally only if remote metadata has
greater disk_consistent_lsn. So because they were equal we didnt write
anything. For attach scenario this ended up in not writing metadata at
all.
Rearrange code so we decide on proper metadata value earlier on and
initialize timeline with correct one without need to update it late in
the initialization process in .reconsile_with_remote
As pointed out in b8488e70a9 (r1024319620)
the following is wrong for the case where the remote storage is empty:
metadata = whatever the local-ONLY metadata is
...
upload_queue.latest_metadata = Some(metadata.clone());
upload_queue.last_uploaded_consistent_lsn = Some(metadata.disk_consistent_lsn());
The reason why it's wrong is that we return last_uploaded_consistent_lsn
to safekeepers. So, we'd be returning an Lsn that is not yet uploaded to
S3.
Before this change, test_pageserver_with_empty_tenants was failing at:
assert loaded_tenant["state"] == {
"Active": {"background_jobs_running": False}
}, "Tenant {tenant_with_empty_timelines_dir} with empty timelines dir should be active and ready for timeline creation"
because background_jobs_running was True instead of False.
Personally I think we should simply always start the background loops
and not bother, but let's punt this until after we've merged this PR.
This change introduces a marker file
$repo/tenants/$tenant_id/attaching
that is present while a tenant is in Attaching state.
When pageserver restarts, we use it to resume the tenant attach operation.
Before this change, a crash during tenant attach would result in one of
the following:
1. crash upon restart due to missing metadata file (IIRC)
2. "successful" loading of the tenant with a subset of timelines
reconcile_with_remote, in this PR, is supposed to download all the layer
files synchronously.
I don't know why, but, download_missing was
1. not doing the download at all for DeltaLayer
2. not using the right RelativePath for image layer
This patch fixes both.
Also improve it so it fails earlier if something is not working
because otherwise it was failing because of the timeout.
And if timeout was big enough test can even pass
This is a part of https://github.com/neondatabase/neon/pull/2595.
It takes out switch to per tenant upload queue and changes to pageserver
startup sequence because these two are highly interleaved with each
other. I'm still not happy with the size of the diff, but splitting
it even more will probably consume even more time. Ideally we should
do it, but this patch isis already a step forward and should be
easier to get this patch in yet still quite difficult.
Mainly because of the size and fixes for existing concerns which
will extend the diff even further
Co-authored-by: Heikki Linnakangas <heikki@neon.tech>
Previously, if the failpoint was not reached for some reason, the test
would only fail because it would reach the 5 minute timeout we have on
all python tests. That's very subtle. Make it fail explicitly, if the
failpoint is not hit on each iteration of the loop.
Extracted from a larger PR, see
https://github.com/neondatabase/neon/pull/2785/files#r1022765794
- Refactor the code a little bit, removing the silly for-loop over a
single element.
- Make it more clear in log messages that the errors are expectd
- Check for a more precise error message "Failed to load delta layer"
instead of just "extracting base backup failed".
If there are any unexpected ERRORs or WARNs in pageserver.log after test
finishes, fail the test. This requires whitelisting the errors that *are*
expected in each test, and there's also a few common errors that are
printed by most tests, which are whitelisted in the fixture itself.
With this, we don't need the special abort() call in testing mode, when
compaction or GC fails. Those failures will print ERRORs to the logs,
which will be picked up by this new mechanisms.
A bunch of errors are currently whitelisted that we probably shouldn't
be emitting in the first place, but fixing those is out of scope for this
commit, so I just left FIXME comments on them.
It's more or less expected from pageserver's point of view. Change the
error kind to ConnectionReset, so that it gets logged at INFO level
instead of ERROR.
We passed the pageserver's libpq endpoint URL as the 'compute_ctl
--connstr' argument, but that was bogus: the --connstr URL is supposed
to be the URL to the *Postgres* instance that compute_ctl launches and
monitors, not to the pageserver. compute_ctl does need the pageserver
URL too, but it is read from the cluster spec JSON, not --connstr.
That was pretty confusing, as you got a lot of "unknown command"
errors in the pageserver log, when compute_tools tries to run regular
SQL commands on the pageserver. The test still passed, however, as it
doesn't require the SQL commands to succeed. But to make this less
confusing, use an invalid hostname instead, so that the queries will
fail to even connect.
- Update vendored PostgreSQL to address prefetch issues
- Make flushed state explicit in PrefetchState
- Move flush logic into prefetch_wait_for, where possible
- Clean up some prefetch state handling code in the various code
elements handling state transitions.
- Fix a race condition in neon_read_at_lsn where a hash entry pointer
was used after the hash table was updated. This could result in
incorrect state transitions and assertion failures after disconnects
during prefetch_wait_for in that neon_read_at_lsn.
Fixes#2780
Before we had separate images for v14 and v15, the compute node image
was called just "neondatabase/compute-node". It has been superseded by
the "neondatabase/compute-node-v14" and "neondatabase/compute-node-v15"
images. The old image is not used by the cloud console build or tests
anymore.
I saw this in 'perf' profile of a sequential scan:
> - 31.93% 0.21% compute request pageserver [.] <pageserver::walredo::PostgresRedoManager as pageserver::walredo::WalRedoManager>::request_redo
> - 31.72% <pageserver::walredo::PostgresRedoManager as pageserver::walredo::WalRedoManager>::request_redo
> - 31.26% pageserver::walredo::PostgresRedoManager::apply_batch_postgres
> + 7.64% <std::process::ChildStdin as std::io::Write>::write
> + 6.17% nix::poll::poll
> + 3.58% <std::process::ChildStderr as std::io::Read>::read
> + 2.96% std::sync::condvar::Condvar::notify_one
> + 2.48% std::sys::unix::locks::futex::Condvar::wait
> + 2.19% alloc::raw_vec::RawVec<T,A>::reserve::do_reserve_and_handle
> + 1.14% std::sys::unix::locks::futex::Mutex::lock_contended
> 0.67% __rust_alloc_zeroed
> 0.62% __stpcpy_ssse3
> 0.56% std::sys::unix::locks::futex::Mutex::wake
Note the 'do_reserve_handle' overhead. That's caused by having to grow
the buffer used to construct the WAL redo request. This commit
eliminates that overhead. It's only about 2% of the overall CPU usage,
but every little helps.
Also reuse the temp buffer when reading records from a DeltaLayer, and
call Vec::reserve to avoid growing a buffer when reading a blob across
pages. I saw a reduction from 2% to 1% of CPU spent in
do_reserve_and_handle in that codepath, but that's such a small change
that it could be just noise. Seems like it shouldn't hurt though.
Closes https://github.com/neondatabase/neon/issues/2697
Example:
https://github.com/neondatabase/neon/actions/runs/3416774593/jobs/5688394855
Adds a set of tests on the storage Docker images before they are pushed
to the public registries:
* tests that pageserver binary has the correct version string (other
binaries are built with the same library, so it should be enough to test
one)
* tests that the compose file set-up works and all components are able
to start and perform a single SQL query (CREATE TABLE)
