## Problem
When a client dropped before a request completed, and a handler returned
an ApiError, we would log that at error severity. That was excessive in
the case of a request erroring on a shutdown, and could cause test
flakes.
example:
https://neon-github-public-dev.s3.amazonaws.com/reports/main/13067651123/index.html#suites/ad9c266207b45eafe19909d1020dd987/6021ce86a0d72ae7/
```
Cancelled request finished with an error: ShuttingDown
```
## Summary of changes
- Log a different info-level on ShuttingDown and ResourceUnavailable API
errors from cancelled requests
## Problem
This assertion is incorrect: it is legal to see another shard's data at
this point, after a shard split.
Closes: https://github.com/neondatabase/neon/issues/10609
## Summary of changes
- Remove faulty assertion
## Problem
If offloading races with normal shutdown, we get a "failed to freeze and
flush: cannot flush frozen layers when flush_loop is not running, state
is Exited". This is harmless but points to it being quite strange to try
and freeze and flush such a timeline. flushing on shutdown for an
archived timeline isn't useful.
Related: https://github.com/neondatabase/neon/issues/10389
## Summary of changes
- During Timeline::shutdown, ignore ShutdownMode::FreezeAndFlush if the
timeline is archived
## Problem
Timeline bootstrap starts a flush loop, but doesn't reliably shut down
the timeline (incl. waiting for flush loop to exit) before destroying
UninitializedTimeline, and that destructor tries to clean up local
storage. If local storage is still being written to, then this is
unsound.
Currently the symptom is that we see a "Directory not empty" error log,
e.g.
https://neon-github-public-dev.s3.amazonaws.com/reports/main/12966756686/index.html#testresult/5523f7d15f46f7f7/retries
## Summary of changes
- Move fallible IO part of bootstrap into a function (notably, this is
fallible in the case of the tenant being shut down while creation is
happening)
- When that function returns an error, call shutdown() on the timeline
## Problem
close https://github.com/neondatabase/neon/issues/10482
## Summary of changes
Add an extra lock on the read path to protect against races. The read
path has an implication that only certain kind of compactions can be
performed. Garbage keys must first have an image layer covering the
range, and then being gc-ed -- they cannot be done in one operation. An
alternative to fix this is to move the layers read guard to be acquired
at the beginning of `get_vectored_reconstruct_data_timeline`, but that
was intentionally optimized out and I don't want to regress.
The race is not limited to image layers. Gc-compaction will consolidate
deltas automatically and produce a flat delta layer (i.e., when we have
retain_lsns below the gc-horizon). The same race would also cause
behaviors like getting an un-replayable key history as in
https://github.com/neondatabase/neon/issues/10049.
Signed-off-by: Alex Chi Z <chi@neon.tech>
Related to #10308, we might have legitimate changes in file size or
generation. Those changes should not cause warn log lines.
In order to detect changes of the generation number while the file size
stayed the same, load the metadata that we store on disk on loading of
the timeline.
Still do a comparison with the on-disk layer sizes to find any
discrepancies that might occur due to race conditions (new metadata file
gets written but layer file has not been updated yet, and PS shuts
down). However, as it's possible to hit it in a race conditon, downgrade
it to a warning.
Also fix a mistake in #10529: we want to compare the old with the new
metadata, not the old metadata with itself.
## Problem
We don't have per-timeline observability for read amplification.
Touches https://github.com/neondatabase/cloud/issues/23283.
## Summary of changes
Add a per-timeline `pageserver_layers_per_read` histogram.
NB: per-timeline histograms are expensive, but probably worth it in this
case.
## Problem
We suspect that Postgres checkpoints will limit the number of page
deltas necessary to reconstruct a page, but don't know for certain.
Touches https://github.com/neondatabase/cloud/issues/23283.
## Summary of changes
Add `pageserver_deltas_per_read_global` metric.
This pairs with `pageserver_layers_per_read_global` from #10573.
## Problem
The current global `pageserver_layers_visited_per_vectored_read_global`
metric does not appear to accurately measure read amplification. It
divides the layer count by the number of reads in a batch, but this
means that e.g. 10 reads with 100 L0 layers will only measure a read amp
of 10 per read, while the actual read amp was 100.
While the cost of layer visits are amortized across the batch, and some
layers may not intersect with a given key, each visited layer
contributes directly to the observed latency for every read in the
batch, which is what we care about.
Touches https://github.com/neondatabase/cloud/issues/23283.
Extracted from #10566.
## Summary of changes
* Count the number of layers visited towards each read in the batch,
instead of the average across the batch.
* Rename `pageserver_layers_visited_per_vectored_read_global` to
`pageserver_layers_per_read_global`.
* Reduce the read amp log warning threshold down from 512 to 100.
## Problem
Follow up of https://github.com/neondatabase/neon/pull/10550 in case the
upper limit is set larger than threshold. It does not make sense for
someone to enforce the behavior like "if there are >= 50 L0s, only
compact 10 of them".
## Summary of changes
Use the maximum of compaction threshold and upper limit when selecting
L0 files to compact.
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
Repartition is slow, but it's only used in image layer creation. We can
skip it if we have a lot of L0 layers to ingest.
## Summary of changes
If L0 compaction is not complete, do not repartition and do not create
image layers.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
We don't have good observability for per-timeline compaction debt,
specifically the number of delta layers in the frozen, L0, and L1
levels.
Touches https://github.com/neondatabase/cloud/issues/23283.
## Summary of changes
* Add a `level` label for `pageserver_layer_{count,size}` with values
`l0`, `l1`, and `frozen`.
* Track metrics for frozen layers.
There is already a `kind={delta,image}` label. `kind=image` is only
possible for `level=l1`.
We don't include the currently open ephemeral layer, only frozen layers.
There is always exactly 1 ephemeral layer, with a dynamic size which is
already tracked in `pageserver_timeline_ephemeral_bytes`.
## Problem
If we are GC-ing because a new image layer was added while traversing
the timeline, then it will remove layers that are required for
fulfilling the current get request (read-path cannot "look back" and
notice the new image layer).
## Summary of Changes
Prevent GC from progressing on the current timeline while it is being
visited for a read.
Epic: https://github.com/neondatabase/neon/issues/9376
Luckily they were the same version, so we didn't spend time compiling
two versions, which could have been the case in the future.
Signed-off-by: Tristan Partin <tristan@neon.tech>
## Problem
Follow-up of the incident, we should not use the same bound on
lower/upper limit of compaction files. This patch adds an upper bound
limit, which is set to 50 for now.
## Summary of changes
Add `compaction_upper_limit`.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
Co-authored-by: Christian Schwarz <christian@neon.tech>
## Problem
We've seen the ingest connection manager get stuck shortly after a
migration.
## Summary of changes
A speculative mitigation is to use the same mechanism as get page
requests for kicking LSN ingest. The connection manager monitors
LSN waits and queries the broker if no updates are received for the
timeline.
Closes https://github.com/neondatabase/neon/issues/10351
## Problem
We need a setting to disable the flush upload wait, to test L0 flush
backpressure in staging.
## Summary of changes
Add `l0_flush_wait_upload` setting.
Only a few things that needed updating:
- async_trait was removed
- Message::Text takes a Utf8Bytes object instead of a String
Signed-off-by: Tristan Partin <tristan@neon.tech>
Co-authored-by: Conrad Ludgate <connor@neon.tech>
In #10308, we noticed many warnings about the local layer having
different sizes on-disk compared to the metadata.
However, the layer downloader would never redownload layer files if the
sizes or generation numbers change. This is obviously a bug, which we
aim to fix with this PR.
This change also moves the code deciding what to do about a layer to a
dedicated function: before we handled the "routing" via control flow,
but now it's become too complicated and it is nicer to have the
different verdicts for a layer spelled out in a list/match.
This reverts commit 9e55d79803.
We'll still need this until we can tune L0 flush backpressure and
compaction. I'll add a setting to disable this separately.
## Problem
This one is fairly embarrassing. Safekeeper node id was used in the
pageserver application name
when connecting to safekeepers.
## Summary of changes
Use the right node id.
Closes https://github.com/neondatabase/neon/issues/10461
During broker deploys, pageservers log this noisy WARN en masse.
I can trivially reproduce the WARN message in neon_local by SIGKILLing
broker during e.g. `pgbench -i`.
I don't understand why tonic is not detecting the error as
`Code::Unavailable`.
Until we find time to understand that / fix upstream, this PR adds the
error message to the existing list of known error messages that get
demoted to INFO level.
Refs:
- refs https://github.com/neondatabase/neon/issues/9562
## Problem
The intent of this parameter is to have pageservers consider themselves
"full" if they've got lots of shards, even if they have plenty of
capacity. It works, but because we typically successfully oversubscribe
capacity up to 200%, the MAX_SHARDS limit is effectively doubled, so
this 20,000 value ends up meaning 40,000, whereas the original intent
was to limit nodes to ~10000 shards.
## Summary of changes
- Change MAX_SHARDS to 5000, so that a node with 5000 will get a 100%
utilization, which is equivalent in practice to being considered "half
full" by the storage controller in capacity terms.
This is all a bit subtle and indiret. Originally the limit was baked
into the pageserver with the idea that the pageserver knows better what
its own resources tolerate than the storage controller does, but in
practice it would be probably be easier to understand all this if we
just did it controller-side. So there's scope to refactor here in
future.
## Problem
In ingest benchmarks, we see L0 compaction delays of over 10 minutes due
to image compaction. We can't stall L0 flushes for that long.
## Summary of changes
Disable L0 flush stalls, and bump the default L0 flush delay threshold
from 20 to 30 L0 layers.
## Problem
If compaction fails, we disable L0 flush stalls to avoid persistent
stalls. However, the logic would unset the failure marker on offload
failures or shutdown. This can lead to sudden L0 flush stalls if we try
and fail to offload a timeline with compaction failures, or if there is
some kind of shutdown race.
Touches #10405.
## Summary of changes
Don't touch the compaction failure marker on offload failures or
shutdown.
## Problem
There is no direct backpressure for compaction and L0 read
amplification. This allows a large buildup of compaction debt and read
amplification.
Resolves#5415.
Requires #10402.
## Summary of changes
Delay layer flushes based on the number of level 0 delta layers:
* `l0_flush_delay_threshold`: delay flushes such that they take 2x as
long (default `2 * compaction_threshold`).
* `l0_flush_stall_threshold`: stall flushes until level 0 delta layers
drop below threshold (default `4 * compaction_threshold`).
If either threshold is reached, ephemeral layer rolls also synchronously
wait for layer flushes to propagate this backpressure up into WAL
ingestion. This will bound the number of frozen layers to 1 once
backpressure kicks in, since all other frozen layers must flush before
the rolled layer.
## Analysis
This will significantly change the compute backpressure characteristics.
Recall the three compute backpressure knobs:
* `max_replication_write_lag`: 500 MB (based on Pageserver
`last_received_lsn`).
* `max_replication_flush_lag`: 10 GB (based on Pageserver
`disk_consistent_lsn`).
* `max_replication_apply_lag`: disabled (based on Pageserver
`remote_consistent_lsn`).
Previously, the Pageserver would keep ingesting WAL and build up
ephemeral layers and L0 layers until the compute hit
`max_replication_flush_lag` at 10 GB and began backpressuring. Now, once
we delay/stall WAL ingestion, the compute will begin backpressuring
after `max_replication_write_lag`, i.e. 500 MB. This is probably a good
thing (we're not building up a ton of compaction debt), but we should
consider tuning these settings.
`max_replication_flush_lag` probably doesn't serve a purpose anymore,
and we should consider removing it.
Furthermore, the removal of the upload barrier in #10402 will mean that
we no longer backpressure flushes based on S3 uploads, since
`max_replication_apply_lag` is disabled. We should consider enabling
this as well.
### When and what do we compact?
Default compaction settings:
* `compaction_threshold`: 10 L0 delta layers.
* `compaction_period`: 20 seconds (between each compaction loop check).
* `checkpoint_distance`: 256 MB (size of L0 delta layers).
* `l0_flush_delay_threshold`: 20 L0 delta layers.
* `l0_flush_stall_threshold`: 40 L0 delta layers.
Compaction characteristics:
* Minimum compaction volume: 10 layers * 256 MB = 2.5 GB.
* Additional compaction volume (assuming 128 MB/s WAL): 128 MB/s * 20
seconds = 2.5 GB (10 L0 layers).
* Required compaction bandwidth: 5.0 GB / 20 seconds = 256 MB/s.
### When do we hit `max_replication_write_lag`?
Depending on how fast compaction and flushes happens, the compute will
backpressure somewhere between `l0_flush_delay_threshold` or
`l0_flush_stall_threshold` + `max_replication_write_lag`.
* Minimum compute backpressure lag: 20 layers * 256 MB + 500 MB = 5.6 GB
* Maximum compute backpressure lag: 40 layers * 256 MB + 500 MB = 10.0
GB
This seems like a reasonable range to me.
This reapplies #10135. Just removing this flush backpressure without
further mitigations caused read amp increases during bulk ingestion
(predictably), so it was reverted. We will replace it by
compaction-based backpressure.
## Problem
In #8550, we made the flush loop wait for uploads after every layer.
This was to avoid unbounded buildup of uploads, and to reduce compaction
debt. However, the approach has several problems:
* It prevents upload parallelism.
* It prevents flush and upload pipelining.
* It slows down ingestion even when there is no need to backpressure.
* It does not directly backpressure based on compaction debt and read
amplification.
We will instead implement compaction-based backpressure in a PR
immediately following this removal (#5415).
Touches #5415.
Touches #10095.
## Summary of changes
Remove waiting on the upload queue in the flush loop.
## Problem
If gc-compaction decides to rewrite an image layer, it will now cause
index_part to lose reference to that layer. In details,
* Assume there's only one image layer of key 0000...AAAA at LSN 0x100
and generation 0xA in the system.
* gc-compaction kicks in at gc-horizon 0x100, and then produce
0000...AAAA at LSN 0x100 and generation 0xB.
* It submits a compaction result update into the index part that unlinks
0000-AAAA-100-A and adds 0000-AAAA-100-B
On the remote storage / local disk side, this is fine -- it unlinks
things correctly and uploads the new file. However, the
`index_part.json` itself doesn't record generations. The buggy procedure
is as follows:
1. upload the new file
2. update the index part to remove the old file and add the new file
3. remove the new file
Therefore, the correct update result process for gc-compaction should be
as follows:
* When modifying the layer map, delete the old one and upload the new
one.
* When updating the index, uploading the new one in the index without
deleting the old one.
## Summary of changes
* Modify `finish_gc_compaction` to correctly order insertions and
deletions.
* Update the way gc-compaction uploads the layer files.
* Add new tests.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
Not really a bug fix, but hopefully can reproduce
https://github.com/neondatabase/neon/issues/10482 more.
If the layer map does not contain layers that end at exactly the end
range of the compaction job, the current split algorithm will produce
the last job that ends at the maximum layer key. This patch extends it
all the way to the compaction job end key.
For example, the user requests a compaction of 0000...FFFF. However, we
only have a layer 0000..3000 in the layer map, and the split job will
have a range of 0000..3000 instead of 0000..FFFF.
This is not a correctness issue but it would be better to fix it so that
we can get consistent job splits.
## Summary of changes
Compaction job split will always cover the full specified key range.
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Refs
- Epic: https://github.com/neondatabase/neon/issues/9378
Co-authored-by: Vlad Lazar <vlad@neon.tech>
Co-authored-by: Christian Schwarz <christian@neon.tech>
## Problem
The read path does its IOs sequentially.
This means that if N values need to be read to reconstruct a page,
we will do N IOs and getpage latency is `O(N*IoLatency)`.
## Solution
With this PR we gain the ability to issue IO concurrently within one
layer visit **and** to move on to the next layer without waiting for IOs
from the previous visit to complete.
This is an evolved version of the work done at the Lisbon hackathon,
cf https://github.com/neondatabase/neon/pull/9002.
## Design
### `will_init` now sourced from disk btree index keys
On the algorithmic level, the only change is that the
`get_values_reconstruct_data`
now sources `will_init` from the disk btree index key (which is
PS-page_cache'd), instead
of from the `Value`, which is only available after the IO completes.
### Concurrent IOs, Submission & Completion
To separate IO submission from waiting for its completion, while
simultaneously
feature-gating the change, we introduce the notion of an `IoConcurrency`
struct
through which IO futures are "spawned".
An IO is an opaque future, and waiting for completions is handled
through
`tokio::sync::oneshot` channels.
The oneshot Receiver's take the place of the `img` and `records` fields
inside `VectoredValueReconstructState`.
When we're done visiting all the layers and submitting all the IOs along
the way
we concurrently `collect_pending_ios` for each value, which means
for each value there is a future that awaits all the oneshot receivers
and then calls into walredo to reconstruct the page image.
Walredo is now invoked concurrently for each value instead of
sequentially.
Walredo itself remains unchanged.
The spawned IO futures are driven to completion by a sidecar tokio task
that
is separate from the task that performs all the layer visiting and
spawning of IOs.
That tasks receives the IO futures via an unbounded mpsc channel and
drives them to completion inside a `FuturedUnordered`.
(The behavior from before this PR is available through
`IoConcurrency::Sequential`,
which awaits the IO futures in place, without "spawning" or "submitting"
them
anywhere.)
#### Alternatives Explored
A few words on the rationale behind having a sidecar *task* and what
alternatives were considered.
One option is to queue up all IO futures in a FuturesUnordered that is
polled
the first time when we `collect_pending_ios`.
Firstly, the IO futures are opaque, compiler-generated futures that need
to be polled at least once to submit their IO. "At least once" because
tokio-epoll-uring may not be able to submit the IO to the kernel on
first
poll right away.
Second, there are deadlocks if we don't drive the IO futures to
completion
independently of the spawning task.
The reason is that both the IO futures and the spawning task may hold
some
_and_ try to acquire _more_ shared limited resources.
For example, both spawning task and IO future may try to acquire
* a VirtualFile file descriptor cache slot async mutex (observed during
impl)
* a tokio-epoll-uring submission slot (observed during impl)
* a PageCache slot (currently this is not the case but we may move more
code into the IO futures in the future)
Another option is to spawn a short-lived `tokio::task` for each IO
future.
We implemented and benchmarked it during development, but found little
throughput improvement and moderate mean & tail latency degradation.
Concerns about pressure on the tokio scheduler made us discard this
variant.
The sidecar task could be obsoleted if the IOs were not arbitrary code
but a well-defined struct.
However,
1. the opaque futures approach taken in this PR allows leaving the
existing
code unchanged, which
2. allows us to implement the `IoConcurrency::Sequential` mode for
feature-gating
the change.
Once the new mode sidecar task implementation is rolled out everywhere,
and `::Sequential` removed, we can think about a descriptive submission
& completion interface.
The problems around deadlocks pointed out earlier will need to be solved
then.
For example, we could eliminate VirtualFile file descriptor cache and
tokio-epoll-uring slots.
The latter has been drafted in
https://github.com/neondatabase/tokio-epoll-uring/pull/63.
See the lengthy doc comment on `spawn_io()` for more details.
### Error handling
There are two error classes during reconstruct data retrieval:
* traversal errors: index lookup, move to next layer, and the like
* value read IO errors
A traversal error fails the entire get_vectored request, as before this
PR.
A value read error only fails that value.
In any case, we preserve the existing behavior that once
`get_vectored` returns, all IOs are done. Panics and failing
to poll `get_vectored` to completion will leave the IOs dangling,
which is safe but shouldn't happen, and so, a rate-limited
log statement will be emitted at warning level.
There is a doc comment on `collect_pending_ios` giving more code-level
details and rationale.
### Feature Gating
The new behavior is opt-in via pageserver config.
The `Sequential` mode is the default.
The only significant change in `Sequential` mode compared to before
this PR is the buffering of results in the `oneshot`s.
## Code-Level Changes
Prep work:
* Make `GateGuard` clonable.
Core Feature:
* Traversal code: track `will_init` in `BlobMeta` and source it from
the Delta/Image/InMemory layer index, instead of determining `will_init`
after we've read the value. This avoids having to read the value to
determine whether traversal can stop.
* Introduce `IoConcurrency` & its sidecar task.
* `IoConcurrency` is the clonable handle.
* It connects to the sidecar task via an `mpsc`.
* Plumb through `IoConcurrency` from high level code to the
individual layer implementations' `get_values_reconstruct_data`.
We piggy-back on the `ValuesReconstructState` for this.
* The sidecar task should be long-lived, so, `IoConcurrency` needs
to be rooted up "high" in the call stack.
* Roots as of this PR:
* `page_service`: outside of pagestream loop
* `create_image_layers`: when it is called
* `basebackup`(only auxfiles + replorigin + SLRU segments)
* Code with no roots that uses `IoConcurrency::sequential`
* any `Timeline::get` call
* `collect_keyspace` is a good example
* follow-up: https://github.com/neondatabase/neon/issues/10460
* `TimelineAdaptor` code used by the compaction simulator, unused in
practive
* `ingest_xlog_dbase_create`
* Transform Delta/Image/InMemoryLayer to
* do their values IO in a distinct `async {}` block
* extend the residence of the Delta/Image layer until the IO is done
* buffer their results in a `oneshot` channel instead of straight
in `ValuesReconstructState`
* the `oneshot` channel is wrapped in `OnDiskValueIo` /
`OnDiskValueIoWaiter`
types that aid in expressiveness and are used to keep track of
in-flight IOs so we can print warnings if we leave them dangling.
* Change `ValuesReconstructState` to hold the receiving end of the
`oneshot` channel aka `OnDiskValueIoWaiter`.
* Change `get_vectored_impl` to `collect_pending_ios` and issue walredo
concurrently, in a `FuturesUnordered`.
Testing / Benchmarking:
* Support queue-depth in pagebench for manual benchmarkinng.
* Add test suite support for setting concurrency mode ps config
field via a) an env var and b) via NeonEnvBuilder.
* Hacky helper to have sidecar-based IoConcurrency in tests.
This will be cleaned up later.
More benchmarking will happen post-merge in nightly benchmarks, plus in
staging/pre-prod.
Some intermediate helpers for manual benchmarking have been preserved in
https://github.com/neondatabase/neon/pull/10466 and will be landed in
later PRs.
(L0 layer stack generator!)
Drive-By:
* test suite actually didn't enable batching by default because
`config.compatibility_neon_binpath` is always Truthy in our CI
environment
=> https://neondb.slack.com/archives/C059ZC138NR/p1737490501941309
* initial logical size calculation wasn't always polled to completion,
which was
surfaced through the added WARN logs emitted when dropping a
`ValuesReconstructState` that still has inflight IOs.
* remove the timing histograms
`pageserver_getpage_get_reconstruct_data_seconds`
and `pageserver_getpage_reconstruct_seconds` because with planning,
value read
IO, and walredo happening concurrently, one can no longer attribute
latency
to any one of them; we'll revisit this when Vlad's work on
tracing/sampling
through RequestContext lands.
* remove code related to `get_cached_lsn()`.
The logic around this has been dead at runtime for a long time,
ever since the removal of the materialized page cache in #8105.
## Testing
Unit tests use the sidecar task by default and run both modes in CI.
Python regression tests and benchmarks also use the sidecar task by
default.
We'll test more in staging and possibly preprod.
# Future Work
Please refer to the parent epic for the full plan.
The next step will be to fold the plumbing of IoConcurrency
into RequestContext so that the function signatures get cleaned up.
Once `Sequential` isn't used anymore, we can take the next
big leap which is replacing the opaque IOs with structs
that have well-defined semantics.
---------
Co-authored-by: Christian Schwarz <christian@neon.tech>
# Refs
- extracted from https://github.com/neondatabase/neon/pull/9353
# Problem
Before this PR, when task_mgr shutdown is signalled, e.g. during
pageserver shutdown or Tenant shutdown, initial logical size calculation
stops polling and drops the future that represents the calculation.
This is against the current policy that we poll all futures to
completion.
This became apparent during development of concurrent IO which warns if
we drop a `Timeline::get_vectored` future that still has in-flight IOs.
We may revise the policy in the future, but, right now initial logical
size calculation is the only part of the codebase that doesn't adhere to
the policy, so let's fix it.
## Code Changes
- make sensitive exclusively to `Timeline::cancel`
- This should be sufficient for all cases of shutdowns; the sensitivity
to task_mgr shutdown is unnecessary.
- this broke the various cancel tests in `test_timeline_size.py`, e.g.,
`test_timeline_initial_logical_size_calculation_cancellation`
- the tests would time out because the await point was not sensitive to
cancellation
- to fix this, refactor `pausable_failpoint` so that it accepts a
cancellation token
- side note: we _really_ should write our own failpoint library; maybe
after we get heap-allocated RequestContext, we can plumb failpoints
through there.
## Problem
Currently, the layer flush loop will continue flushing layers as long as
any are pending, and only notify waiters once there are no further
layers to flush. This can cause waiters to wait longer than necessary,
and potentially starve them if pending layers keep arriving faster than
they can be flushed. The impact of this will increase when we add
compaction backpressure and propagate it up into the WAL receiver.
Extracted from #10405.
## Summary of changes
Break out of the layer flush loop once we've flushed up to the requested
LSN. If further flush requests have arrived in the meanwhile, flushing
will resume immediately after.
## Problem
For compaction backpressure, we need a mechanism to signal when
compaction has reduced the L0 delta layer count below the backpressure
threshold.
Extracted from #10405.
## Summary of changes
Add `LayerMap::watch_level0_deltas()` which returns a
`tokio::sync:⌚:Receiver` signalling the current L0 delta layer
count.
## Problem
It's sometimes useful to obtain the elapsed duration from a
`StorageTimeMetricsTimer` for purposes beyond just recording it in
metrics (e.g. to log it).
Extracted from #10405.
## Summary of changes
Add `StorageTimeMetricsTimer.elapsed()` and return the duration from
`stop_and_record()`.
## Problem
part of https://github.com/neondatabase/neon/issues/9114
The automatic trigger is already implemented at
https://github.com/neondatabase/neon/pull/10221 but I need to write some
tests and finish my experiments in staging before I can merge it with
confidence. Given that I have some other patches that will modify the
config items, I'd like to get the config items merged first to reduce
conflicts.
## Summary of changes
* add `l2_lsn` to index_part.json -- below that LSN, data have been
processed by gc-compaction
* add a set of gc-compaction auto trigger control items into the config
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Refs
- fixes https://github.com/neondatabase/neon/issues/10444
## Problem
We're seeing a panic `handles are only shut down once in their lifetime`
in our performance testbed.
## Hypothesis
Annotated code in
https://github.com/neondatabase/neon/issues/10444#issuecomment-2602286415.
```
T1: drop Cache, executes up to (1)
=> HandleInner is now in state ShutDown
T2: Timeline::shutdown => PerTimelineState::shutdown executes shutdown() again => panics
```
Likely this snuck in the final touches of #10386 where I narrowed down
the locking rules.
## Summary of changes
Make duplicate shutdowns a no-op.
The other test focus on the external interface usage while the tests
added in this PR add some testing around HandleInner's lifecycle,
ensuring we don't leak it once either connection gets dropped or
per-timeline-state is shut down explicitly.
It was requested by review in #10305 to use an enum or something like it
for distinguishing the different modes instead of two parameters,
because two flags allow four combinations, and two of them don't really
make sense/ aren't used.
follow-up of #10305
## Problem
Occasionally, we encounter bugs in test environments that can be
detected at the point of uploading an index, but we proceed to upload it
anyway and leave a tenant in a broken state that's awkward to handle.
## Summary of changes
- Validate index when submitting it for upload, so that we can see the
issue quickly e.g. in an API invoking compaction
- Validate index before executing the upload, so that we have a hard
enforcement that any code path that tries to upload an index will not
overwrite a valid index with an invalid one.
# Refs
- fixes https://github.com/neondatabase/neon/issues/10309
- fixup of batching design, first introduced in
https://github.com/neondatabase/neon/pull/9851
- refinement of https://github.com/neondatabase/neon/pull/8339
# Problem
`Tenant::shutdown` was occasionally taking many minutes (sometimes up to
20) in staging and prod if the
`page_service_pipelining.mode="concurrent-futures"` is enabled.
# Symptoms
The issue happens during shard migration between pageservers.
There is page_service unavailability and hence effectively downtime for
customers in the following case:
1. The source (state `AttachedStale`) gets stuck in `Tenant::shutdown`,
waiting for the gate to close.
2. Cplane/Storcon decides to transition the target `AttachedMulti` to
`AttachedSingle`.
3. That transition comes with a bump of the generation number, causing
the `PUT .../location_config` endpoint to do a full `Tenant::shutdown` /
`Tenant::attach` cycle for the target location.
4. That `Tenant::shutdown` on the target gets stuck, waiting for the
gate to close.
5. Eventually the gate closes (`close completed`), correlating with a
`page_service` connection handler logging that it's exiting because of a
network error (`Connection reset by peer` or `Broken pipe`).
While in (4):
- `Tenant::shutdown` is stuck waiting for all `Timeline::shutdown` calls
to complete.
So, really, this is a `Timeline::shutdown` bug.
- retries from Cplane/Storcon to complete above state transitions, fail
with errors related to the tenant mgr slot being in state
`TenantSlot::InProgress`, the tenant state being
`TenantState::Stopping`, and the timelines being in
`TimelineState::Stopping`, and the `Timeline::cancel` being cancelled.
- Existing (and/or new?) page_service connections log errors `error
reading relation or page version: Not found: Timed out waiting 30s for
tenant active state. Latest state: None`
# Root-Cause
After a lengthy investigation ([internal
write-up](https://www.notion.so/neondatabase/2025-01-09-batching-deadlock-Slow-Log-Analysis-in-Staging-176f189e00478050bc21c1a072157ca4?pvs=4))
I arrived at the following root cause.
The `spsc_fold` channel (`batch_tx`/`batch_rx`) that connects the
Batcher and Executor stages of the pipelined mode was storing a `Handle`
and thus `GateGuard` of the Timeline that was not shutting down.
The design assumption with pipelining was that this would always be a
short transient state.
However, that was incorrect: the Executor was stuck on writing/flushing
an earlier response into the connection to the client, i.e., socket
write being slow because of TCP backpressure.
The probable scenario of how we end up in that case:
1. Compute backend process sends a continuous stream of getpage prefetch
requests into the connection, but never reads the responses (why this
happens: see Appendix section).
2. Batch N is processed by Batcher and Executor, up to the point where
Executor starts flushing the response.
3. Batch N+1 is procssed by Batcher and queued in the `spsc_fold`.
4. Executor is still waiting for batch N flush to finish.
5. Batcher eventually hits the `TimeoutReader` error (10min).
From here on it waits on the
`spsc_fold.send(Err(QueryError(TimeoutReader_error)))`
which will never finish because the batch already inside the `spsc_fold`
is not
being read by the Executor, because the Executor is still stuck in the
flush.
(This state is not observable at our default `info` log level)
6. Eventually, Compute backend process is killed (`close()` on the
socket) or Compute as a whole gets killed (probably no clean TCP
shutdown happening in that case).
7. Eventually, Pageserver TCP stack learns about (6) through RST packets
and the Executor's flush() call fails with an error.
8. The Executor exits, dropping `cancel_batcher` and its end of the
spsc_fold.
This wakes Batcher, causing the `spsc_fold.send` to fail.
Batcher exits.
The pipeline shuts down as intended.
We return from `process_query` and log the `Connection reset by peer` or
`Broken pipe` error.
The following diagram visualizes the wait-for graph at (5)
```mermaid
flowchart TD
Batcher --spsc_fold.send(TimeoutReader_error)--> Executor
Executor --flush batch N responses--> socket.write_end
socket.write_end --wait for TCP window to move forward--> Compute
```
# Analysis
By holding the GateGuard inside the `spsc_fold` open, the pipelining
implementation
violated the principle established in
(https://github.com/neondatabase/neon/pull/8339).
That is, that `Handle`s must only be held across an await point if that
await point
is sensitive to the `<Handle as Deref<Target=Timeline>>::cancel` token.
In this case, we were holding the Handle inside the `spsc_fold` while
awaiting the
`pgb_writer.flush()` future.
One may jump to the conclusion that we should simply peek into the
spsc_fold to get
that Timeline cancel token and be sensitive to it during flush, then.
But that violates another principle of the design from
https://github.com/neondatabase/neon/pull/8339.
That is, that the page_service connection lifecycle and the Timeline
lifecycles must be completely decoupled.
Tt must be possible to shut down one shard without shutting down the
page_service connection, because on that single connection we might be
serving other shards attached to this pageserver.
(The current compute client opens separate connections per shard, but,
there are plans to change that.)
# Solution
This PR adds a `handle::WeakHandle` struct that does _not_ hold the
timeline gate open.
It must be `upgrade()`d to get a `handle::Handle`.
That `handle::Handle` _does_ hold the timeline gate open.
The batch queued inside the `spsc_fold` only holds a `WeakHandle`.
We only upgrade it while calling into the various `handle_` methods,
i.e., while interacting with the `Timeline` via `<Handle as
Deref<Target=Timeline>>`.
All that code has always been required to be (and is!) sensitive to
`Timeline::cancel`, and therefore we're guaranteed to bail from it
quickly when `Timeline::shutdown` starts.
We will drop the `Handle` immediately, before we start
`pgb_writer.flush()`ing the responses.
Thereby letting go of our hold on the `GateGuard`, allowing the timeline
shutdown to complete while the page_service handler remains intact.
# Code Changes
* Reproducer & Regression Test
* Developed and proven to reproduce the issue in
https://github.com/neondatabase/neon/pull/10399
* Add a `Test` message to the pagestream protocol (`cfg(feature =
"testing")`).
* Drive-by minimal improvement to the parsing code, we now have a
`PagestreamFeMessageTag`.
* Refactor `pageserver/client` to allow sending and receiving
`page_service` requests independently.
* Add a Rust helper binary to produce situation (4) from above
* Rationale: (4) and (5) are the same bug class, we're holding a gate
open while `flush()`ing.
* Add a Python regression test that uses the helper binary to
demonstrate the problem.
* Fix
* Introduce and use `WeakHandle` as explained earlier.
* Replace the `shut_down` atomic with two enum states for `HandleInner`,
wrapped in a `Mutex`.
* To make `WeakHandle::upgrade()` and `Handle::downgrade()`
cache-efficient:
* Wrap the `Types::Timeline` in an `Arc`
* Wrap the `GateGuard` in an `Arc`
* The separate `Arc`s enable uncontended cloning of the timeline
reference in `upgrade()` and `downgrade()`.
If instead we were `Arc<Timeline>::clone`, different connection handlers
would be hitting the same cache line on every upgrade()/downgrade(),
causing contention.
* Please read the udpated module-level comment in `mod handle`
module-level comment for details.
# Testing & Performance
The reproducer test that failed before the changes now passes, and
obviously other tests are passing as well.
We'll do more testing in staging, where the issue happens every ~4h if
chaos migrations are enabled in storcon.
Existing perf testing will be sufficient, no perf degradation is
expected.
It's a few more alloctations due to the added Arc's, but, they're low
frequency.
# Appendix: Why Compute Sometimes Doesn't Read Responses
Remember, the whole problem surfaced because flush() was slow because
Compute was not reading responses. Why is that?
In short, the way the compute works, it only advances the page_service
protocol processing when it has an interest in data, i.e., when the
pagestore smgr is called to return pages.
Thus, if compute issues a bunch of requests as part of prefetch but then
it turns out it can service the query without reading those pages, it
may very well happen that these messages stay in the TCP until the next
smgr read happens, either in that session, or possibly in another
session.
If there’s too many unread responses in the TCP, the pageserver kernel
is going to backpressure into userspace, resulting in our stuck flush().
All of this stems from the way vanilla Postgres does prefetching and
"async IO":
it issues `fadvise()` to make the kernel do the IO in the background,
buffering results in the kernel page cache.
It then consumes the results through synchronous `read()` system calls,
which hopefully will be fast because of the `fadvise()`.
If it turns out that some / all of the prefetch results are not needed,
Postgres will not be issuing those `read()` system calls.
The kernel will eventually react to that by reusing page cache pages
that hold completed prefetched data.
Uncompleted prefetch requests may or may not be processed -- it's up to
the kernel.
In Neon, the smgr + Pageserver together take on the role of the kernel
in above paragraphs.
In the current implementation, all prefetches are sent as GetPage
requests to Pageserver.
The responses are only processed in the places where vanilla Postgres
would do the synchronous `read()` system call.
If we never get to that, the responses are queued inside the TCP
connection, which, once buffers run full, will backpressure into
Pageserver's sending code, i.e., the `pgb_writer.flush()` that was the
root cause of the problems we're fixing in this PR.
## Problem
All pageserver have the same application name which makes it hard to
distinguish them.
## Summary of changes
Include the node id in the application name sent to the safekeeper. This
should gives us
more visibility in logs. There's a few metrics that will increase in
cardinality by `pageserver_count`,
but that's fine.
