We want to host pg_duckdb (starting with v0.3.1) on Neon.
This PR replaces https://github.com/neondatabase/neon/pull/10350 which
was for older pg_duckdb v0.2.0
Use cases
- faster OLAP queries
- access to datelake files (e.g. parquet) on S3 buckets from Neon
PostgreSQL
Because neon does not provide superuser role to neon customers we need
to grant some additional permissions to neon_superuser:
Note: some grants that we require are already granted to `PUBLIC` in new
release of pg_duckdb
[here](3789e4c509/sql/pg_duckdb--0.2.0--0.3.0.sql (L1054))
```sql
GRANT ALL ON FUNCTION duckdb.install_extension(TEXT) TO neon_superuser;
GRANT ALL ON TABLE duckdb.extensions TO neon_superuser;
GRANT ALL ON SEQUENCE duckdb.extensions_table_seq TO neon_superuser;
```
## Problem
The timestamp prefix of pytest log lines contains milliseconds without
leading zeros, so values of milliseconds less than 100 printed
incorrectly.
For example:
```
2025-02-15 12:02:51.997 INFO [_internal.py:97] 127.0.0.1 - - ...
2025-02-15 12:02:52.4 INFO [_internal.py:97] 127.0.0.1 - - ...
2025-02-15 12:02:52.9 INFO [_internal.py:97] 127.0.0.1 - - ...
2025-02-15 12:02:52.23 INFO [_internal.py:97] 127.0.0.1 - - ...
```
## Summary of changes
Fix log_format for pytest so that milliseconds are printed with leading
zeros.
In commit 9537829ccd I made shared_buffers be derived from the system's
available RAM. However, I failed to remove the old hard-coded
shared_buffers=10GB settings, shared_buffers was set twice. Oopsie.
## Problem
Sometimes, a regression test run gets stuck (taking more than 60
minutes) and is killed by GitHub's `timeout-minutes` without leaving any
traces in the test results database.
I find no correlation between this and either the build type, the
architecture, or the Postgres version.
See: https://neonprod.grafana.net/goto/nM7ih7cHR?orgId=1
## Summary of changes
- Bump `pytest-timeout` to the version that supports `--session-timeout`
- Set `--session-timeout` to (timeout-minutes - 10 minutes) * 60 seconds
in Attempt to stop tests gracefully to generate test reports until they
are forcibly stopped by the stricter `timeout-minutes` limit.
## Problem
Part of https://github.com/neondatabase/neon/issues/9516
## Summary of changes
This patch adds the support for storing reldir in the sparse keyspace.
All logic are guarded with the `rel_size_v2_enabled` flag, so if it's
set to false, the code path is exactly the same as what's currently in
prod.
Note that we did not persist the `rel_size_v2_enabled` flag and the
logic around it will be implemented in the next patch. (i.e., what if we
enabled it, restart the pageserver, and then it gets set to false? we
should still read from v2 using the rel_size_v2_migration_status in the
index_part). The persistence logic I'll implement in the next patch will
disallow switching from v2->v1 via config item.
I also refactored the metrics so that it can work with the new reldir
store. However, this metric is not correctly computed for reldirs (see
the comments) before. With the refactor, the value will be computed only
when we have an initial value for the reldir size. The refactor keeps
the incorrectness of the computation when there are more than 1
database.
For the tests, we currently run all the tests with v2, and I'll set it
to false and add some v2-specific tests before merging, probably also
v1->v2 migration tests.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
I noticed the opportunity to simplify here while working on
https://github.com/neondatabase/neon/pull/9353 .
The only difference is the zero-fill behavior: if one reads past rel
size,
`get_rel_page_at_lsn` returns a zeroed page whereas `Timeline::get`
returns an error.
However, the `endblk` is at most rel size large, because `nblocks` is eq
`get_rel_size`, see a few lines above this change.
We're using the same LSN (`self.lsn`) for everything, so there is no
chance of non-determinism.
Refs:
- Slack discussion debating correctness:
https://neondb.slack.com/archives/C033RQ5SPDH/p1737457010607119
# Summary
In
- https://github.com/neondatabase/neon/pull/10813
we added slow flush logging but it didn't log the TCP send & recv queue
length.
This PR adds that data to the log message.
I believe the implementation to be safe & correct right now, but it's
brittle and thus this PR should be reverted or improved upon once the
investigation is over.
Refs:
- stacked atop https://github.com/neondatabase/neon/pull/10813
- context:
https://neondb.slack.com/archives/C08DE6Q9C3B/p1739464533762049?thread_ts=1739462628.361019&cid=C08DE6Q9C3B
- improves https://github.com/neondatabase/neon/issues/10668
- part of https://github.com/neondatabase/cloud/issues/23515
# How It Works
The trouble is two-fold:
1. getting to the raw socket file descriptor through the many Rust types
that wrap it and
2. integrating with the `measure()` function
Rust wraps it in types to model file descriptor lifetimes and ownership,
and usually one can get access using `as_raw_fd()`.
However, we `split()` the stream and the resulting
[`tokio::io::WriteHalf`](https://docs.rs/tokio/latest/tokio/io/struct.WriteHalf.html)
.
Check the PR commit history for my attempts to do it.
My solution is to get the socket fd before we wrap it in our protocol
types, and to store that fd in the new `PostgresBackend::socket_fd`
field.
I believe it's safe because the lifetime of `PostgresBackend::socket_fd`
value == the lifetime of the `TcpStream` that wrap and store in
`PostgresBackend::framed`.
Specifically, the only place that close()s the socket is the `impl Drop
for TcpStream`.
I think the protocol stack calls `TcpStream::shutdown()`, but, that
doesn't `close()` the file descriptor underneath.
Regarding integration with the `measure()` function, the trouble is that
`flush_fut` is currently a generic `Future` type. So, we just pass in
the `socket_fd` as a separate argument.
A clean implementation would convert the `pgb_writer.flush()` to a named
future that provides an accessor for the socket fd while not being
polled.
I tried (see PR history), but failed to break through the `WriteHalf`.
# Testing
Tested locally by running
```
./target/debug/pagebench get-page-latest-lsn --num-clients=1000 --queue-depth=1000
```
in one terminal, waiting a bit, then
```
pkill -STOP pagebench
```
then wait for slow logs to show up in `pageserver.log`.
Pick one of the slow log message's port pairs, e.g., `127.0.0.1:39500`,
and then checking sockstat output
```
ss -ntp | grep '127.0.0.1:39500'
```
to ensure that send & recv queue size match those in the log message.
The logic might seem a bit intricate / over-optimized, but I recently
spent time benchmarking this code path in the context of a nightly
pagebench regression
(https://github.com/neondatabase/cloud/issues/21759)
and I want to avoid regressing it any further.
Ideally would also log the socket send & recv queue length like we do on
the compute side in
- https://github.com/neondatabase/neon/pull/10673
But that is proving difficult due to the Rust abstractions that wrap the
socket fd.
Work in progress on that is happening in
- https://github.com/neondatabase/neon/pull/10823
Regarding production impact, I am worried at a theoretical level that
the additional logging may cause a downward spiral in the case where a
pageserver is slow to flush because there is not enough CPU. The logging
would consume more CPU and thereby slow down flushes even more. However,
I don't think this matters practically speaking.
# Refs
- context:
https://neondb.slack.com/archives/C08DE6Q9C3B/p1739464533762049?thread_ts=1739462628.361019&cid=C08DE6Q9C3B
- fixes https://github.com/neondatabase/neon/issues/10668
- part of https://github.com/neondatabase/cloud/issues/23515
# Testing
Tested locally by running
```
./target/debug/pagebench get-page-latest-lsn --num-clients=1000 --queue-depth=1000
```
in one terminal, waiting a bit, then
```
pkill -STOP pagebench
```
then wait for slow logs to show up in `pageserver.log`.
To see that the completion log message is logged, run
```
pkill -CONT pagebench
```
## Problem
Some situations may produce a large number of pending reconciles. If we
experience an issue where reconciles are processed more slowly than
expected, that can prevent us responding promptly to user requests like
tenant/timeline CRUD.
This is a cleaner implementation of the hotfix in
https://github.com/neondatabase/neon/pull/10815
## Summary of changes
- Introduce a second semaphore for high priority tasks, with
configurable units (default 256). The intent is that in practical
situations these user-facing requests should never have to wait.
- Use the high priority semaphore for: tenant/timeline CRUD, and shard
splitting operations. Use normal priority for everything else.
## Problem
We run the compatibility tests only if we are upgrading the extension.
An accidental code change may break the test itself, so we have to check
this code as well.
## Summary of changes
The test is scheduled once a day to save time and resources.
---------
Co-authored-by: Alexander Bayandin <alexander@neon.tech>
PR #10305 makes sure that there is no *actual* race, i.e. we will never
attempt to offload a timeline that has just been unarchived, or similar.
However, if a timeline has been unarchived and has children that are
unarchived too, we will get an error log line. Such races can occur as
in compaction we check if the timeline can be offloaded way before we
attempt to offload it: the result might change in the meantime.
This patch checks if the delete guard can't be obtained because the
timeline has unarchived children, and if yes, it does another check for
whether the timeline has become unarchived or not. If it is unarchived,
it just prints an info log msg and integrates itself into the error
suppression logic of the compaction calling into it.
If you squint at it really closely, there is still a possible race in
which we print an error log, but this one is unlikely because the
timeline and its children need to be archived right after the check for
whether the timeline has any unarchived children, and right before the
check whether the timeline is archived. Archival involves a network
operation while nothing between these two checks does that, so it's very
unlikely to happen in real life.
https://github.com/neondatabase/cloud/issues/23979#issuecomment-2651265729
## Problem
We had code for stripping IDs out of proxied paths to reduce cardinality
of metrics, but it was only stripping out tenant IDs, and leaving in
timeline IDs and query parameters (e.g. LSN in lsn->timestamp lookups).
## Summary of changes
- Use a more general regex approach.
There is still some risk that a future pageserver API might include a
parameter in `/the/path/`, but we control that API and it is not often
extended. We will also alert on metrics cardinality in staging so that
if we made that mistake we would notice.
## Problem
See https://github.com/neondatabase/neon/issues/10755
Random access pattern of pgbench leaves sparse chunks, which makes the
on-disk size of file.cache unpredictable.
## Summary of changes
Perform seqscan to fill LFC chunks with data so that on-disk file size
included size of table.
---------
Co-authored-by: Konstantin Knizhnik <knizhnik@neon.tech>
## Problem
`benchmarks` is a long-running and non-blocking job. If, on Staging, a
deploy-blocking job fails, restarting it requires cancelling any running
`benchmarks` jobs, which is a waste of CI resources and requires a
couple of extra clicks for a human to do.
Ref: https://neondb.slack.com/archives/C059ZC138NR/p1739292995400899
## Summary of changes
- Run `benchmarks` after `deploy` job
- Handle `benchmarks` run in PRs with `run-benchmarks` label but without
`deploy` job.
There was a typo in the name of the utilization endpoint URL, fix it.
Also, ensure that the heartbeat mechanism actually works.
Related: #10583, #10429
Part of #9011
Before this PR, an IO error returned from the kernel, e.g., due to a bad
disk, would get bubbled up, all the way to a user-visible query failing.
This is against the IO error handling policy where we have established
and is hence being rectified in this PR.
[[(internal Policy document
link)]](bef44149f7/src/storage/handling_io_and_logical_errors.md (L33-L35))
The practice on the write path seems to be that we call
`maybe_fatal_err()` or `fatal_err()` fairly high up the stack.
That is, regardless of whether std::fs, tokio::fs, or VirtualFile is
used to perform the IO.
For the read path, I choose a centralized approach in this PR by
checking for errors as close to the kernel interface as possible.
I believe this is better for long-term consistency.
To mitigate the problem of missing context if we abort so far down in
the stack, the `on_fatal_io_error` now captures and logs a backtrace.
I grepped the pageserver code base for `fs::read` to convince myself
that all non-VirtualFile reads already handle IO errors according to
policy.
Refs
- fixes https://github.com/neondatabase/neon/issues/10454
## Problem
We didn't catch all client errors causing alerts.
## Summary of changes
Client errors should be wrapped with ClientError so that it doesn't fire
alerts.
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
The test is flaky: WAL in remote storage appears to be corrupted. One of
hypotheses so far is that corruption is the result of local fs
implementation being non atomic, and safekeepers may concurrently PUT
the same segment. That's dubious though because by looking at local_fs
impl I'd expect then early EOF on segment read rather then observed
zeros in test failures, but other directions seem even less probable.
## Summary of changes
Let's add s3 backend as well and see if it is also flaky. Also add some
more logging around segments uploads.
ref https://github.com/neondatabase/neon/issues/10761
There is now a compute_ctl_config field in the response that currently
only contains a JSON Web Key set. compute_ctl currently doesn't do
anything with the keys, but will in the future.
The reasoning for the new field is due to the nature of empty computes.
When an empty compute is created, it does not have a tenant. A compute
spec is the primary means of communicating the details of an attached
tenant. In the empty compute state, there is no spec. Instead we wait
for the control plane to pass us one via /configure. If we were to
include the jwks field in the compute spec, we would have a partial
compute spec, which doesn't logically make sense.
Instead, we can have two means of passing settings to the compute:
- spec: tenant specific config details
- compute_ctl_config: compute specific settings
For instance, the JSON Web Key set passed to the compute is independent
of any tenant. It is a setting of the compute whether it is attached or
not.
Signed-off-by: Tristan Partin <tristan@neon.tech>
pg_search is 46ish MB. All other remote extensions are around hundeds of
KB. 3 seconds is not long enough to download the tarball if the S3
gateway cache doesn't already contain a copy. According to our setup,
the cache is limited to 10 GB in size and anything that has not been
accessed for an hour is purged.
This is really bad for scaling to 0, even more so if you're the only
project actively using the extension in a production Kubernetes cluster.
Signed-off-by: Tristan Partin <tristan@neon.tech>
## Problem
This test occasionally fails while the test teardown tries to do a
graceful shutdown, because the test has quickly written lots of data
into the pageserver.
Closes: #10654
## Summary of changes
- Call `post_checks` at the end of `test_isolation`, as we already do
for test_pg_regress -- this improves our detection of issues, and as a
nice side effect flushes the pageserver.
- Ignore pg_notify files when validating state at end of test, these are
not expected to be the same
## Problem
In #10752 I used an overly-strict regex that only ignored error on a
particular key.
## Summary of changes
- Drop key from regex so it matches all such errors
## Problem
The build of the neon container image is not caching any part of the
rust build, making it fairly slow.
## Summary of changes
Cache dependency building using cargo-chef.
## Problem
Previously, when cutting over to cold secondary locations,
we would clobber the previous, good, heatmap with a cold one.
This is because heatmap generation used to include only resident layers.
Once this merges, we can add an endpoint which triggers full heatmap
hydration on attached locations to heal cold migrations.
## Summary of changes
With this patch, heatmap generation becomes additive. If we have a
heatmap from when this location was secondary, the new uploaded heatmap
will be the result of a reconciliation between the old one and the on
disk resident layers.
More concretely, when we have the previous heatmap:
1. Filter the previous heatmap and keep layers that are (a) present
in the current layer map, (b) visible, (c) not resident. Call this set
of layers `visible_non_resident`.
2. From the layer map, select all layers that are resident and visible.
Call this set of layers `resident`.
3. The new heatmap is the result of merging the two disjoint sets.
Related https://github.com/neondatabase/neon/issues/10541
In #9011, we want to schedule timelines to safekeepers. In order to do
such scheduling, we need information about how utilized a safekeeper is
and if it's available or not.
Therefore, send constant heartbeats to the safekeepers and try to figure
out if they are online or not.
Includes some code from #10440.
## Problem
We expose `latest_gc_cutoff` in our API, and callers understandably were
using that to validate LSNs for branch creation. However, this is _not_
the true GC cutoff from a user's point of view: it's just the point at
which we last actually did GC. The actual cutoff used when validating
branch creations and page_service reads is the min() of latest_gc_cutoff
and the planned GC lsn in GcInfo.
Closes: https://github.com/neondatabase/neon/issues/10639
## Summary of changes
- Expose the more useful min() of GC cutoffs as `gc_cutoff_lsn` in the
API, so that the most obviously named field is really the one people
should use.
- Retain the ability to read the LSN at which GC was actually done, in
an `applied_gc_cutoff_lsn` field.
- Internally rename `latest_gc_cutoff_lsn` to `applied_gc_cutoff_lsn`
("latest" was a confusing name, as the value in GcInfo is more up to
date in terms of what a user experiences)
- Temporarily preserve the old `latest_gc_cutoff_lsn` field for compat
with control plane until we update it to use the new field.
---------
Co-authored-by: Arpad Müller <arpad-m@users.noreply.github.com>
## Problem
https://github.com/neondatabase/neon/pull/10613 changed how images are
pushed, and therefore also how we have to wait for images to be pushed
in `trigger-e2e-tests`. The `trigger-e2e-tests` workflow is triggered in
three different ways:
- When a pull request is pushed to that is already ready to review, here
we call the workflow from `build_and_test`
- When a pull request is marked ready for review, then the workflow is
triggered directly
- When a push to `main` or `release(-.*)?` triggers `build_and_test` and
that indirectly calls `trigger-e2e-tests`.
The second of these paths had a bug, which was not tested in the PR,
because this path being different wasn't clear to me.
## Summary of changes
Fix the jq statement that caused the bug.
## Problem
https://github.com/neondatabase/neon/pull/10613 changed how images are
pushed, and there was a small mismatch between the github workflow and
the script generating what to push where. This resulted in the workflow
trying to push images to prod registries from the main branch, even
though we don't do that and therefore didn't generate a mapping for
those registries in the script that decides what to push where.
This misconception happened because promote-images-dev pushed to dev
registries, and promote-images-prod pushed to prod registries, but
promote-images-prod also updated the latest tag in the dev registries if
and only if we are on the main branch. This last bit is why the
push-<component>-image-prod jobs were trying to run on the main branch.
## Summary of changes
Don't try pushing to prod registries from the main branch.
## Problem
Retagging container images and pushing container images taken from one
registry to another is very tangled up with artifact building and not
separated by component. This makes not building compute for storage
releases and vice versa pretty tricky. To enable that, I want to clean
up retagging and pushing of container images and then continue on making
the pipelines for releases leaner by not building unnecessary things.
## Summary of changes
- Add a reusable workflow that can push to ACR, ECR and Docker Hub,
while being very flexible in terms of source and target images. This
allows for retagging and pushing images between container registries.
- Stop pushing images to registries aside of docker hub in the jobs that
build the images
- Split image pushing into 4 different jobs (not mentioning special
cases):
- neon-dev
- neon-prod
- compute-dev
- compute-prod
## TODO
- Consider also using this for `pin-build-tools-image`, as it's
basically another instance of the same thing.
## Known limitations
- The ECR part of this workflow supports authenticating to multiple AWS
accounts and therefore multiple ECR endpoints, but the ACR part only
supports one Azure Account. If someone with more knowledge on Azure can
tell me whether an equivalent to
https://github.com/aws-actions/amazon-ecr-login?tab=readme-ov-file#login-to-ecr-on-multiple-aws-accounts
is easily possible, that'd be great.
- The `image_map` input is a bit complex. It expects something along the
lines of
```
{
"docker.io/neondatabase/compute-node-v14:13196061314": [
"docker.io/neondatabase/compute-node-v14:13196061314",
"369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:13196061314",
"neoneastus2.azurecr.io/neondatabase/compute-node-v14:13196061314"
],
"docker.io/neondatabase/compute-node-v15:13196061314": [
"docker.io/neondatabase/compute-node-v15:13196061314",
"369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:13196061314",
"neoneastus2.azurecr.io/neondatabase/compute-node-v15:13196061314"
]
}
```
to map from source to target image. We have a small python step to
generate this map for the 4 main image pushing jobs. The concrete
example is taken from
https://github.com/neondatabase/neon/actions/runs/13196061314/job/36838584098?pr=10613#step:3:6
and shortened to two images.
## Problem
We respect `skip_pg_catalog_updates` at the initial start, but ignore at
the follow-up `/configure`. Yet, it's used for storage->cplane->compute
notify requests after migrations, shard split, etc. So every time we get
them, applying the new config takes much longer than it should because
we go through Postgres catalog checks. Cplane sets this flag, when it
does serves notify attach call
9068c7d743
Related to `inc-403`, for example
## Summary of changes
Look at `skip_pg_catalog_updates` in `compute.reconfigure()`
## Problem
PRs created by external contributors, in some cases might list failed
jobs
- `Trigger E2E Tests / cancel-previous-e2e-tests`
- `Trigger E2E Tests / tag`
They don't block the merge, and tests in fact pass (their counterparts
in internal PR), but because jobs are triggered from an external PR (and
not from the corresponding internal one) they still present as red
marks.
For example https://github.com/neondatabase/neon/pull/10778
## Summary of changes
- Check permissions before triggering e2e tests
## Problem
L0 compaction frequently gets starved out by other background tasks and
image/GC compaction. L0 compaction must be responsive to keep read
amplification under control.
Touches #10694.
Resolves#10689.
## Summary of changes
Use a separate semaphore for the L0-only compaction pass.
* Add a `CONCURRENT_L0_COMPACTION_TASKS` semaphore and
`BackgroundLoopKind::L0Compaction`.
* Add a setting `compaction_l0_semaphore` (default off via
`compaction_l0_first`).
* Use the L0 semaphore when doing an `OnlyL0Compaction` pass.
* Use the background semaphore when doing a regular compaction pass
(which includes an initial L0 pass).
* While waiting for the background semaphore, yield for L0 compaction if
triggered.
* Add `CompactFlags::NoYield` to disable L0 yielding, and set it for the
HTTP API route.
* Remove the old `use_compaction_semaphore` setting and
compaction-scoped semaphore.
* Remove the warning when waiting for a semaphore; it's noisy and we
have metrics.
## Problem
With current approach for the base images in `Dockerfiles`, it's hard to
track when image is updated, and as they are base, than update will
invalidate all the layers, as base image changed.
That also becomes more complicated, as we have a number of runners, and
they may have different images with the tag `bookworm-slim`, so that
will lead to invalidate caches, when image build on one runner will be
used on another runners.
To fix that problem, we could pin our base images to the specific sha,
and that not only align images across runners, and also will allow us to
have reproducible build and don't depend on any spontaneous changes in
upstream.
Fix: https://github.com/neondatabase/cloud/issues/24084
## Summary of changes
Beside of the main goal, that PR also included some small changes around
Dockerfiles:
1. Main change: use `SHA` for `bookworm-slim` and `bullseye-slim` debian
images
2. For the layers requiring `curl` we could add `curl` and `unzip` to
the `build-deps` image, and use it as a base image for all the steps,
removing extra dependency on `alpine/curl`
3. added `retry-on-host-error=on` for the `wgetrc` as it happened to me:
fail to resolve hostname
## Problem
Previously, Workload was reconfiguring the compute before each run of
writes, which was meant to be a no-op when nothing changed, but was
actually writing extra data due to an issue being fixed in
https://github.com/neondatabase/neon/pull/10696.
The row counts in tests were too low in some cases, these tests were
only working because of those extra writes that shouldn't have been
happening, and moreover were relying on checkpoints happening.
## Summary of changes
- Only reconfigure compute if the attached pageserver actually changed.
If pageserver is set to None, that means controller is managing
everything, so never reconfigure compute.
- Update tests that wrote too few rows.
---------
Co-authored-by: Alexey Kondratov <kondratov.aleksey@gmail.com>
The old values assumed that you have at least about 18 GB of RAM
available (shared_buffers=10GB and maintenance_work_mem=8GB). That's a
lot when testing locally. Make it configurable, and make the default
assumption much smaller: 256 MB.
This is nice for local testing, but it's also in preparation for
starting to use VMs to run these jobs. When launched in a VM, the
control plane can set these env variables according to the max size of
the VM.
Also change the formula for how RAM is distributed: use 10% of RAM for
shared_buffers, and 70% for maintenance_work_mem. That leaves a good
amount for misc. other stuff and the OS. A very large shared_buffers
setting won't typically help with bulk loading. It won't help with the
network and I/O of processing all the tables, unless maybe if the whole
database fits in shared buffers, but even then it's not much faster than
using local disk. Bulk loading is all sequential I/O. It also won't help
much with index creation, which is also sequential I/O. A large
maintenance_work_mem can be quite useful, however, so that's where we
put most of the RAM.
Resolves: https://github.com/neondatabase/neon/issues/5734
When we query pageserver and it's unavailable after some retries,
postgres sigabrt's. This is intended behavior so I've added tests
checking it
## Problem
When image compaction yields for L0 compaction, it may not immediately
schedule L0 compaction, because it just goes on to compact the next
pending timeline.
Touches #10694.
Requires #10744.
## Summary of changes
Extend `CompactionOutcome` with `YieldForL0` and `Skipped` variants, and
immediately schedule an L0 compaction pass in the `YieldForL0` case.
## Problem
Image compaction can starve out L0 compaction if a tenant has several
timelines with L0 debt.
Touches #10694.
Requires #10740.
## Summary of changes
* Add an initial L0 compaction pass, in order of L0 count.
* Add a tenant option `compaction_l0_first` to control the L0 pass
(disabled by default).
* Add `CompactFlags::OnlyL0Compaction` to run an L0-only compaction
pass.
* Clean up the compaction iteration logic.
A later PR will use separate semaphores for the L0 and image compaction
passes to avoid cross-tenant L0 starvation. That PR will also make image
compaction yield if _any_ of the tenant's timelines have pending L0
compaction to further avoid starvation.
Avoids compiling the crate and its dependencies into binaries that don't
need them. Shrinks the compute_ctl binary from about 31MB to 28MB in the
release-line-debug-size-lto profile.
