Before this PR
1.The circuit breaker would trip on CompactionError::Shutdown. That's
wrong, we want to ignore those cases.
2. remote timeline client shutdown would not be mapped to
CompactionError::Shutdown in all circumstances.
We observed this in staging, see
https://neondb.slack.com/archives/C033RQ5SPDH/p1721829745384449
This PR fixes (1) with a simple `match` statement, and (2) by switching
a bunch of `anyhow` usage over to distinguished errors that ultimately
get mapped to `CompactionError::Shutdown`.
I removed the implicit `#[from]` conversion from `anyhow::Error` to
`CompactionError::Other` to discover all the places that were mapping
remote timeline client shutdown to `anyhow::Error`.
In my opinion `#[from]` is an antipattern and we should avoid it,
especially for `anyhow::Error`. If some callee is going to return
anyhow, the very least the caller should to is to acknowledge, through a
`map_err(MyError::Other)` that they're conflating different failure
reasons.
## Problem
LayerAccessStats contains a lot of detail that we don't use: short
histories of most recent accesses, specifics on what kind of task
accessed a layer, etc. This is all stored inside a Mutex, which is
locked every time something accesses a layer.
## Summary of changes
- Store timestamps at a very low resolution (to the nearest second),
sufficient for use on the timescales of eviction.
- Pack access time and last residence change time into a single u64
- Use the high bits of the u64 for other flags, including the new layer
visibility concept.
- Simplify the external-facing model for access stats to just include
what we now track.
Note that the `HistoryBufferWithDropCounter` is removed here because it
is no longer used. I do not dislike this type, we just happen not to use
it for anything else at present.
Co-authored-by: Christian Schwarz <christian@neon.tech>
## Problem
As described in https://github.com/neondatabase/neon/issues/8398, layer
visibility is a new hint that will help us manage disk space more
efficiently.
## Summary of changes
- Introduce LayerVisibilityHint and store it as part of access stats
- Automatically mark a layer visible if it is accessed, or when it is
created.
The impact on the access stats size will be reversed in
https://github.com/neondatabase/neon/pull/8431
This is functionally a no-op change: subsequent PRs will add the logic
that sets layers to Covered, and which uses the layer visibility as an
input to eviction and heatmap generation.
---------
Co-authored-by: Joonas Koivunen <joonas@neon.tech>
Part of #8128.
## Problem
Scrubber uses `scan_metadata` command to flag metadata inconsistencies.
To trust it at scale, we need to make sure the errors we emit is a
reflection of real scenario. One check performed in the scrubber is to
see whether layers listed in the latest `index_part.json` is present in
object listing. Currently, the scrubber does not robustly handle the
case where objects are uploaded/deleted during the scan.
## Summary of changes
**Condition for success:** An object in the index is (1) in the object
listing we acquire from S3 or (2) found in a HeadObject request (new
object).
- Add in the `HeadObject` requests for the layers missing from the
object listing.
- Keep the order of first getting the object listing and then
downloading the layers.
- Update check to only consider shards with highest shard count.
- Skip analyzing a timeline if `deleted_at` tombstone is marked in
`index_part.json`.
- Add new test to see if scrubber actually detect the metadata
inconsistency.
_Misc_
- A timeline with no ancestor should always have some layers.
- Removed experimental histograms
_Caveat_
- Ancestor layer is not cleaned until #8308 is implemented. If ancestor
layers reference non-existing layers in the index, the scrubber will
emit false positives.
Signed-off-by: Yuchen Liang <yuchen@neon.tech>
Use the k-merge iterator in the compaction process to reduce memory
footprint.
part of https://github.com/neondatabase/neon/issues/8002
## Summary of changes
* refactor the bottom-most compaction code to use k-merge iterator
* add Send bound on some structs as it is used across the await points
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
Co-authored-by: Arpad Müller <arpad-m@users.noreply.github.com>
## Problem
We lack insight into:
- How much of a tenant's physical size is image vs. delta layers
- Average sizes of image vs. delta layers
- Total layer counts per timeline, indicating size of index_part object
As well as general observability love, this is motivated by
https://github.com/neondatabase/neon/issues/6738, where we need to
define some sensible thresholds for storage amplification, and using
total physical size may not work well (if someone does a lot of DROPs
then it's legitimate for the physical-synthetic ratio to be huge), but
the ratio between image layer size and delta layer size may be a better
indicator of whether we're generating unreasonable quantities of image
layers.
## Summary of changes
- Add pageserver_layer_bytes and pageserver_layer_count metrics,
labelled by timeline and `kind` (delta or image)
- Add & subtract these with LayerInner's lifetime.
I'm intentionally avoiding using a generic metric RAII guard object, to
avoid bloating LayerInner: it already has all the information it needs
to update metric on new+drop.
## Problem
The `evictions_with_low_residence_duration` is used as an indicator of
cache thrashing. However, there are situations where it is quite
legitimate to only have a short residence during compaction, where a
delta is downloaded, used to generate an image layer, and then
discarded. This can lead to false positive alerts.
## Summary of changes
- Only track low residence duration for layers that have been accessed
at least once (compaction doesn't count as an access). This will give us
a metric that indicates thrashing on layers that the _user_ is using,
rather than those we're downloading for housekeeping purposes.
Once we add "layer visibility" as an explicit property of layers, this
can also be used as a cleaner condition (residence of non-visible layers
should never be alertable)
Removes the `ImageCompressionAlgorithm::DisabledNoDecompress` variant.
We now assume any blob with the specific bits set is actually a
compressed blob.
The `ImageCompressionAlgorithm::Disabled` variant still remains and is
the new default.
Reverts large parts of #8238 , as originally intended in that PR.
Part of #5431
PR #8106 was created with the assumption that no blob is larger than
`256 MiB`. Due to #7852 we have checking for *writes* of blobs larger
than that limit, but we didn't have checking for *reads* of such large
blobs: in theory, we could be reading these blobs every day but we just
don't happen to write the blobs for some reason.
Therefore, we now add a warning for *reads* of such large blobs as well.
To make deploying compression less dangerous, we therefore only assume a
blob is compressed if the compression setting is present in the config.
This also means that we can't back out of compression once we enabled
it.
Part of https://github.com/neondatabase/neon/issues/5431
Before this PR, during timeline shutdown, we'd occasionally see
log lines like this one:
```
2024-06-26T18:28:11.063402Z INFO initial_size_calculation{tenant_id=$TENANT,shard_id=0000 timeline_id=$TIMELINE}:logical_size_calculation_task:get_or_maybe_download{layer=000000000000000000000000000000000000-000000067F0001A3950001C1630100000000__0000000D88265898}: layer file download failed, and caller has been cancelled: Cancelled, shutting down
Stack backtrace:
0: <core::result::Result<T,F> as core::ops::try_trait::FromResidual<core::result::Result<core::convert::Infallible,E>>>::from_residual
at /rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/core/src/result.rs:1964:27
pageserver::tenant::remote_timeline_client::RemoteTimelineClient::download_layer_file::{{closure}}
at /home/nonroot/pageserver/src/tenant/remote_timeline_client.rs:531:13
pageserver::tenant::storage_layer::layer::LayerInner::download_and_init::{{closure}}
at /home/nonroot/pageserver/src/tenant/storage_layer/layer.rs:1136:14
pageserver::tenant::storage_layer::layer::LayerInner::download_init_and_wait::{{closure}}::{{closure}}
at /home/nonroot/pageserver/src/tenant/storage_layer/layer.rs:1082:74
```
We can eliminate the anyhow backtrace with no loss of information
because the conversion to anyhow::Error happens in exactly one place.
refs #7427
## Problem
In https://github.com/neondatabase/neon/pull/5299, the new config-v1
tenant config file was added to hold the LocationConf type. We left the
old config file in place for forward compat, and because running without
generations (therefore without LocationConf) as still useful before the
storage controller was ready for prime-time.
Closes: https://github.com/neondatabase/neon/issues/5388
## Summary of changes
- Remove code for reading and writing the legacy config file
- Remove Generation::Broken: it was unused.
- Treat missing config file on disk as an error loading a tenant, rather
than defaulting it. We can now remove LocationConf::default, and thereby
guarantee that we never construct a tenant with a None generation.
- Update some comments + add some assertions to clarify that
Generation::None is only used in layer metadata, not in the state of a
running tenant.
- Update docker compose test to create tenants with a generation
Adds manual compaction trigger; add gc compaction to test_gc_feedback
Part of https://github.com/neondatabase/neon/issues/8002
```
test_gc_feedback[debug-pg15].logical_size: 50 Mb
test_gc_feedback[debug-pg15].physical_size: 2269 Mb
test_gc_feedback[debug-pg15].physical/logical ratio: 44.5302
test_gc_feedback[debug-pg15].max_total_num_of_deltas: 7
test_gc_feedback[debug-pg15].max_num_of_deltas_above_image: 2
test_gc_feedback[debug-pg15].logical_size_after_bottom_most_compaction: 50 Mb
test_gc_feedback[debug-pg15].physical_size_after_bottom_most_compaction: 287 Mb
test_gc_feedback[debug-pg15].physical/logical ratio after bottom_most_compaction: 5.6312
test_gc_feedback[debug-pg15].max_total_num_of_deltas_after_bottom_most_compaction: 4
test_gc_feedback[debug-pg15].max_num_of_deltas_above_image_after_bottom_most_compaction: 1
```
## Summary of changes
* Add the manual compaction trigger
* Use in test_gc_feedback
* Add a guard to avoid running it with retain_lsns
* Fix: Do `schedule_compaction_update` after compaction
* Fix: Supply deltas in the correct order to reconstruct value
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
part of https://github.com/neondatabase/neon/issues/8002
## Summary of changes
This pull request adds the image layer iterator. It buffers a fixed
amount of key-value pairs in memory, and give developer an iterator
abstraction over the image layer. Once the buffer is exhausted, it will
issue 1 I/O to fetch the next batch.
Due to the Rust lifetime mysteries, the `get_stream_from` API has been
refactored to `into_stream` and consumes `self`.
Delta layer iterator implementation will be similar, therefore I'll add
it after this pull request gets merged.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
A demo for a building block for compaction. The GC-compaction operation
iterates all layers below/intersect with the GC horizon, and do a full
layer rewrite of all of them. The end result will be image layer
covering the full keyspace at GC-horizon, and a bunch of delta layers
above the GC-horizon. This helps us collect the garbages of the
test_gc_feedback test case to reduce space amplification.
This operation can be manually triggered using an HTTP API or be
triggered based on some metrics. Actual method TBD.
The test is very basic and it's very likely that most part of the
algorithm will be rewritten. I would like to get this merged so that I
can have a basic skeleton for the algorithm and then make incremental
changes.
<img width="924" alt="image"
src="https://github.com/neondatabase/neon/assets/4198311/f3d49f4e-634f-4f56-986d-bfefc6ae6ee2">
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
fixes https://github.com/neondatabase/neon/issues/7790 (duplicating most
of the issue description here for posterity)
# Background
From the time before always-authoritative `index_part.json`, we had to
handle duplicate layers. See the RFC for an illustration of how
duplicate layers could happen:
a8e6d259cb/docs/rfcs/027-crash-consistent-layer-map-through-index-part.md (L41-L50)
As of #5198 , we should not be exposed to that problem anymore.
# Problem 1
We still have
1. [code in
Pageserver](82960b2175/pageserver/src/tenant/timeline.rs (L4502-L4521))
than handles duplicate layers
2. [tests in the test
suite](d9dcbffac3/test_runner/regress/test_duplicate_layers.py (L15))
that demonstrates the problem using a failpoint
However, the test in the test suite doesn't use the failpoint to induce
a crash that could legitimately happen in production.
What is does instead is to return early with an `Ok()`, so that the code
in Pageserver that handles duplicate layers (item 1) actually gets
exercised.
That "return early" would be a bug in the routine if it happened in
production.
So, the tests in the test suite are tests for their own sake, but don't
serve to actually regress-test any production behavior.
# Problem 2
Further, if production code _did_ (it nowawdays doesn't!) create a
duplicate layer, the code in Pageserver that handles the condition (item
1 above) is too little and too late:
* the code handles it by discarding the newer `struct Layer`; that's
good.
* however, on disk, we have already overwritten the old with the new
layer file
* the fact that we do it atomically doesn't matter because ...
* if the new layer file is not bit-identical, then we have a cache
coherency problem
* PS PageCache block cache: caches old bit battern
* blob_io offsets stored in variables, based on pre-overwrite bit
pattern / offsets
* => reading based on these offsets from the new file might yield
different data than before
# Solution
- Remove the test suite code pertaining to Problem 1
- Move & rename test suite code that actually tests RFC-27
crash-consistent layer map.
- Remove the Pageserver code that handles duplicate layers too late
(Problem 1)
- Use `RENAME_NOREPLACE` to prevent over-rename the file during
`.finish()`, bail with an error if it happens (Problem 2)
- This bailing prevents the caller from even trying to insert into the
layer map, as they don't even get a `struct Layer` at hand.
- Add `abort`s in the place where we have the layer map lock and check
for duplicates (Problem 2)
- Note again, we can't reach there because we bail from `.finish()` much
earlier in the code.
- Share the logic to clean up after failed `.finish()` between image
layers and delta layers (drive-by cleanup)
- This exposed that test `image_layer_rewrite` was overwriting layer
files in place. Fix the test.
# Future Work
This PR adds a new failure scenario that was previously "papered over"
by the overwriting of layers:
1. Start a compaction that will produce 3 layers: A, B, C
2. Layer A is `finish()`ed successfully.
3. Layer B fails mid-way at some `put_value()`.
4. Compaction bails out, sleeps 20s.
5. Some disk space gets freed in the meantime.
6. Compaction wakes from sleep, another iteration starts, it attempts to
write Layer A again. But the `.finish()` **fails because A already
exists on disk**.
The failure in step 5 is new with this PR, and it **causes the
compaction to get stuck**.
Before, it would silently overwrite the file and "successfully" complete
the second iteration.
The mitigation for this is to `/reset` the tenant.
## Problem
Looking at several noisy shutdown logs:
- In https://github.com/neondatabase/neon/issues/7861 we're hitting a
log error with `InternalServerError(timeline shutting down\n'` on the
checkpoint API handler.
- In the field, we see initial_logical_size_calculation errors on
shutdown, via DownloadError
- In the field, we see errors logged from layer download code
(independent of the error propagated) during shutdown
Closes: https://github.com/neondatabase/neon/issues/7861
## Summary of changes
The theme of these changes is to avoid propagating anyhow::Errors for
cases that aren't really unexpected error cases that we might want a
stacktrace for, and avoid "Other" error variants unless we really do
have unexpected error cases to propagate.
- On the flush_frozen_layers path, use the `FlushLayerError` type
throughout, rather than munging it into an anyhow::Error. Give
FlushLayerError an explicit from_anyhow helper that checks for timeline
cancellation, and uses it to give a Cancelled error instead of an Other
error when the timeline is shutting down.
- In logical size calculation, remove BackgroundCalculationError (this
type was just a Cancelled variant and an Other variant), and instead use
CalculateLogicalSizeError throughout. This can express a
PageReconstructError, and has a From impl that translates cancel-like
page reconstruct errors to Cancelled.
- Replace CalculateLogicalSizeError's Other(anyhow::Error) variant case
with a Decode(DeserializeError) variant, as this was the only kind of
error we actually used in the Other case.
- During layer download, drop out early if the timeline is shutting
down, so that we don't do an `error!()` log of the shutdown error in
this case.
## Problem
- After a shard split of a large existing tenant, child tenants can end
up with oversized historic layers indefinitely, if those layers are
prevented from being GC'd by branchpoints.
This PR follows https://github.com/neondatabase/neon/pull/7531, and adds
rewriting of layers that contain a mixture of needed & un-needed
contents, in addition to dropping un-needed layers.
Closes: https://github.com/neondatabase/neon/issues/7504
## Summary of changes
- Add methods to ImageLayer for reading back existing layers
- Extend `compact_shard_ancestors` to rewrite layer files that contain a
mixture of keys that we want and keys we do not, if unwanted keys are
the majority of those in the file.
- Amend initialization code to handle multiple layers with the same
LayerName properly
- Get rid of of renaming bad layer files to `.old` since that's now
expected on restarts during rewrites.
Once upon a time, we used to have duplicated types for runtime IndexPart
and whatever we stored. Because of the serde fixes in #5335 we have no
need for duplicated IndexPart type anymore, but the `IndexLayerMetadata`
stayed.
- remove the type
- remove LayerFileMetadata::file_size() in favor of direct field access
Split off from #7833. Cc: #3072.
the gate was accidentially being dropped before the final blocking
phase, possibly explaining the resident physical size global problems
during deletions.
it could had caused more harm as well, but the path is not actively
being tested because cplane no longer puts locationconfigs with higher
generation number during normal operation which prompted the last wave
of fixes.
Cc: #7341.
detaching a timeline from its ancestor can leave the resulting timeline
with more L0 layers than the compaction threshold. most of the time, the
detached timeline has made progress, and next time the L0 -> L1
compaction happens near the original branch point and not near the
last_record_lsn.
add a test to ensure that inheriting the historical L0s does not change
fullbackup. additionally:
- add `wait_until_completed` to test-only timeline checkpoint and
compact HTTP endpoints. with `?wait_until_completed=true` the endpoints
will wait until the remote client has completed uploads.
- for delta layers, describe L0-ness with the `/layer` endpoint
Cc: #6994
## Problem
- When a layer with legacy local path format is evicted and then
re-downloaded, a panic happened because the path downloaded by remote
storage didn't match the path stored in Layer.
- While investigating, I also realized that secondary locations would
have a similar issue with evictions.
Closes: #7783
## Summary of changes
- Make remote timeline client take local paths as an input: it should
not have its own ideas about local paths, instead it just uses the layer
path that the Layer has.
- Make secondary state store an explicit local path, populated on scan
of local disk at startup. This provides the same behavior as for Layer,
that our local_layer_path is a _default_, but the layer path can
actually be anything (e.g. an old style one).
- Add tests for both cases.
## Problem
This is historical baggage from when the pageserver could be run with
local disk only: we had a bunch of places where we had to treat remote
storage as optional.
Closes: https://github.com/neondatabase/neon/issues/6890
## Changes
- Remove Option<> around remote storage (in
https://github.com/neondatabase/neon/pull/7722 we made remote storage
clearly mandatory)
- Remove code for deleting old metadata files: they're all gone now.
- Remove other references to metadata files when loading directories, as
none exist.
I checked last 14 days of logs for "found legacy metadata", there are no
instances.
- Rename "filename" types which no longer map directly to a filename
(LayerFileName -> LayerName)
- Add a -v1- part to local layer paths to smooth the path to future
updates (we anticipate a -v2- that uses checksums later)
- Rename methods that refer to the string-ized version of a LayerName to
no longer be called "filename"
- Refactor reconcile() function to use a LocalLayerFileMetadata type
that includes the local path, rather than carrying local path separately
in a tuple and unwrap()'ing it later.
## Problem
In https://github.com/neondatabase/neon/pull/7531, we would like to be
able to rewrite layers safely. One option is to make `Layer` able to
rewrite files in place safely (e.g. by blocking evictions/deletions for
an old Layer while a new one is created), but that's relatively fragile.
It's more robust in general if we simply never overwrite the same local
file: we can do that by putting the generation number in the filename.
## Summary of changes
- Add `local_layer_path` (counterpart to `remote_layer_path`) and
convert all locations that manually constructed a local layer path by
joining LayerFileName to timeline path
- In the layer upload path, construct remote paths with
`remote_layer_path` rather than trying to build them out of a local
path.
- During startup, carry the full path to layer files through
`init::reconcile`, and pass it into `Layer::for_resident`
- Add a test to make sure we handle upgrades properly.
- Comment out the generation part of `local_layer_path`, since we need
to maintain forward compatibility for one release. A tiny followup PR
will enable it afterwards.
We could make this a bit simpler if we bulk renamed existing layers on
startup instead of carrying literal paths through init, but that is
operationally risky on existing servers with millions of layer files. We
can always do a renaming change in future if it becomes annoying, but
for the moment it's kind of nice to have a structure that enables us to
change local path names again in future quite easily.
We should rename `LayerFileName` to `LayerName` or somesuch, to make it
more obvious that it's not a literal filename: this was already a bit
confusing where that type is used in remote paths. That will be a
followup, to avoid polluting this PR's diff.
## Problem
Timelines cannot be deleted if they have children. In many production
cases, a branch or a timeline has been created off the main branch for
various reasons to the effect of having now a "new main" branch. This
feature will make it possible to detach a timeline from its ancestor by
inheriting all of the data before the branchpoint to the detached
timeline and by also reparenting all of the ancestor's earlier branches
to the detached timeline.
## Summary of changes
- Earlier added copy_lsn_prefix functionality is used
- RemoteTimelineClient learns to adopt layers by copying them from
another timeline
- LayerManager adds support for adding adopted layers
-
`timeline::Timeline::{prepare_to_detach,complete_detaching}_from_ancestor`
and `timeline::detach_ancestor` are added
- HTTP PUT handler
Cc: #6994
Co-authored-by: Christian Schwarz <christian@neon.tech>
Instead of showing the full path of layer traversal, we now only show
tenant (in tracing context)+timeline+filename.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
We are currently supporting two read paths. No bueno.
## Summary of changes
High level: use vectored read path to serve get page requests - gated by
`get_impl` config
Low level:
1. Add ps config, `get_impl` to specify which read path to use when
serving get page requests
2. Fix base cached image handling for the vectored read path. This was
subtly broken: previously we
would not mark keys that went past their cached lsn as complete. This is
a self standing change which
could be its own PR, but I've included it here because writing separate
tests for it is tricky.
3. Fork get page to use either the legacy or vectored implementation
4. Validate the use of vectored read path when serving get page requests
against the legacy implementation.
Controlled by `validate_vectored_get` ps config.
5. Use the vectored read path to serve get page requests in tests (with
validation).
## Note
Since the vectored read path does not go through the page cache to read
buffers, this change also amounts to a removal of the buffer page cache. Materialized page cache
is still used.
As part of https://github.com/neondatabase/neon/pull/7375 and to improve
the current vectored get implementation, we separate the missing key
error out. This also saves us several Box allocations in the get page
implementation.
## Summary of changes
* Create a caching field of layer traversal id for each of the layer.
* Remove box allocations for layer traversal id retrieval and implement
MissingKey error message as before. This should be a little bit faster.
* Do not format error message until `Display`.
* For in-mem layer, the descriptor is different before/after frozen. I'm
using once lock for that.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
#7030 introduced an annoying papercut, deeming a failure to acquire a
strong reference to `LayerInner` from `DownloadedLayer::drop` as a
canceled eviction. Most of the time, it wasn't that, but just timeline
deletion or tenant detach with the layer not wanting to be deleted or
evicted.
When a Layer is dropped as part of a normal shutdown, the `Layer` is
dropped first, and the `DownloadedLayer` the second. Because of this, we
cannot detect eviction being canceled from the `DownloadedLayer::drop`.
We can detect it from `LayerInner::drop`, which this PR adds.
Test case is added which before had 1 started eviction, 2 canceled. Now
it accurately finds 1 started, 1 canceled.
For "timeline ancestor merge" or "timeline detach," we need to "cut"
delta layers at particular LSN. The name "truncate" is not used as it
would imply that a layer file changes, instead of what happens: we copy
keys with Lsn less than a "cut point".
Cc: #6994
Add the "copy delta layer prefix" operation to DeltaLayerInner, re-using
some of the vectored read internals. The code is `cfg(test)` until it
will be used later with a more complete integration test.
## Problem
During this week's deployment we observed panics due to the blobs
for certain keys not fitting in the vectored read buffers. The likely
cause of this is a bloated AUX_FILE_KEY caused by logical replication.
## Summary of changes
This pr fixes the issue by allocating a buffer big enough to fit
the widest read. It also has the benefit of saving space if all keys
in the read have blobs smaller than the max vectored read size.
If the soft limit for the max size of a vectored read is violated,
we print a warning which includes the offending key and lsn.
A randomised (but deterministic) end to end test is also added for
vectored reads on the delta layer.
Before this PR, each core had 3 executor threads from 3 different
runtimes. With this PR, we just have one runtime, with one thread per
core. Switching to a single tokio runtime should reduce that effective
over-commit of CPU and in theory help with tail latencies -- iff all
tokio tasks are well-behaved and yield to the runtime regularly.
Are All Tasks Well-Behaved? Are We Ready?
-----------------------------------------
Sadly there doesn't seem to be good out-of-the box tokio tooling to
answer this question.
We *believe* all tasks are well behaved in today's code base, as of the
switch to `virtual_file_io_engine = "tokio-epoll-uring"` in production
(https://github.com/neondatabase/aws/pull/1121).
The only remaining executor-thread-blocking code is walredo and some
filesystem namespace operations.
Filesystem namespace operations work is being tracked in #6663 and not
considered likely to actually block at this time.
Regarding walredo, it currently does a blocking `poll` for read/write to
the pipe file descriptors we use for IPC with the walredo process.
There is an ongoing experiment to make walredo async (#6628), but it
needs more time because there are surprisingly tricky trade-offs that
are articulated in that PR's description (which itself is still WIP).
What's relevant for *this* PR is that
1. walredo is always CPU-bound
2. production tail latencies for walredo request-response
(`pageserver_wal_redo_seconds_bucket`) are
- p90: with few exceptions, low hundreds of micro-seconds
- p95: except on very packed pageservers, below 1ms
- p99: all below 50ms, vast majority below 1ms
- p99.9: almost all around 50ms, rarely at >= 70ms
- [Dashboard
Link](https://neonprod.grafana.net/d/edgggcrmki3uof/2024-03-walredo-latency?orgId=1&var-ds=ZNX49CDVz&var-pXX_by_instance=0.9&var-pXX_by_instance=0.99&var-pXX_by_instance=0.95&var-adhoc=instance%7C%21%3D%7Cpageserver-30.us-west-2.aws.neon.tech&var-per_instance_pXX_max_seconds=0.0005&from=1711049688777&to=1711136088777)
The ones below 1ms are below our current threshold for when we start
thinking about yielding to the executor.
The tens of milliseconds stalls aren't great, but, not least because of
the implicit overcommit of CPU by the three runtimes, we can't be sure
whether these tens of milliseconds are inherently necessary to do the
walredo work or whether we could be faster if there was less contention
for CPU.
On the first item (walredo being always CPU-bound work): it means that
walredo processes will always compete with the executor threads.
We could yield, using async walredo, but then we hit the trade-offs
explained in that PR.
tl;dr: the risk of stalling executor threads through blocking walredo
seems low, and switching to one runtime cleans up one potential source
for higher-than-necessary stall times (explained in the previous
paragraphs).
Code Changes
------------
- Remove the 3 different runtime definitions.
- Add a new definition called `THE_RUNTIME`.
- Use it in all places that previously used one of the 3 removed
runtimes.
- Remove the argument from `task_mgr`.
- Fix failpoint usage where `pausable_failpoint!` should have been used.
We encountered some actual failures because of this, e.g., hung
`get_metric()` calls during test teardown that would client-timeout
after 300s.
As indicated by the comment above `THE_RUNTIME`, we could take this
clean-up further.
But before we create so much churn, let's first validate that there's no
perf regression.
Performance
-----------
We will test this in staging using the various nightly benchmark runs.
However, the worst-case impact of this change is likely compaction
(=>image layer creation) competing with compute requests.
Image layer creation work can't be easily generated & repeated quickly
by pagebench.
So, we'll simply watch getpage & basebackup tail latencies in staging.
Additionally, I have done manual benchmarking using pagebench.
Report:
https://neondatabase.notion.site/2024-03-23-oneruntime-change-benchmarking-22a399c411e24399a73311115fb703ec?pvs=4
Tail latencies and throughput are marginally better (no regression =
good).
Except in a workload with 128 clients against one tenant.
There, the p99.9 and p99.99 getpage latency is about 2x worse (at
slightly lower throughput).
A dip in throughput every 20s (compaction_period_ is clearly visible,
and probably responsible for that worse tail latency.
This has potential to improve with async walredo, and is an edge case
workload anyway.
Future Work
-----------
1. Once this change has shown satisfying results in production, change
the codebase to use the ambient runtime instead of explicitly
referencing `THE_RUNTIME`.
2. Have a mode where we run with a single-threaded runtime, so we
uncover executor stalls more quickly.
3. Switch or write our own failpoints library that is async-native:
https://github.com/neondatabase/neon/issues/7216
## Problem
The current implementation of struct Layer supports canceled read
requests, but those will leave the internal state such that a following
`Layer::keep_resident` call will need to repair the state. In
pathological cases seen during generation numbers resetting in staging
or with too many in-progress on-demand downloads, this repair activity
will need to wait for the download to complete, which stalls disk
usage-based eviction. Similar stalls have been observed in staging near
disk-full situations, where downloads failed because the disk was full.
Fixes#6028 or the "layer is present on filesystem but not evictable"
problems by:
1. not canceling pending evictions by a canceled
`LayerInner::get_or_maybe_download`
2. completing post-download initialization of the `LayerInner::inner`
from the download task
Not canceling evictions above case (1) and always initializing (2) lead
to plain `LayerInner::inner` always having the up-to-date information,
which leads to the old `Layer::keep_resident` never having to wait for
downloads to complete. Finally, the `Layer::keep_resident` is replaced
with `Layer::is_likely_resident`. These fix#7145.
## Summary of changes
- add a new test showing that a canceled get_or_maybe_download should
not cancel the eviction
- switch to using a `watch` internally rather than a `broadcast` to
avoid hanging eviction while a download is ongoing
- doc changes for new semantics and cleanup
- fix `Layer::keep_resident` to use just `self.0.inner.get()` as truth
as `Layer::is_likely_resident`
- remove `LayerInner::wanted_evicted` boolean as no longer needed
Builds upon: #7185. Cc: #5331.
Before this PR, cancellation for `LayerInner::get_or_maybe_download`
could occur so that we have downloaded the layer file in the filesystem,
but because of the cancellation chance, we have not set the internal
`LayerInner::inner` or initialized the state. With the detached init
support introduced in #7135 and in place in #7152, we can now initialize
the internal state after successfully downloading in the spawned task.
The next PR will fix the remaining problems that this PR leaves:
- `Layer::keep_resident` is still used because
- `Layer::get_or_maybe_download` always cancels an eviction, even when
canceled
Split off from #7030. Stacked on top of #7152. Cc: #5331.
The second part of work towards fixing `Layer::keep_resident` so that it
does not need to repair the internal state. #7135 added a nicer API for
initialization. This PR uses it to remove a few indentation levels and
the loop construction. The next PR #7175 will use the refactorings done
in this PR, and always initialize the internal state after a download.
Cc: #5331
Since #6115 with more often used get_value_reconstruct_data and friends,
we should not have needless INFO level span creation near hot paths. In
our prod configuration, INFO spans are always created, but in practice,
very rarely anything at INFO level is logged underneath.
`ResidentLayer::load_keys` is only used during compaction so it is not
that hot, but this aligns the access paths and their span usage.
PR changes the span level to debug to align with others, and adds the
layer name to the error which was missing.
Split off from #7030.
Split off from #7030:
- each early exit is counted as canceled init, even though it most
likely was just `LayerInner::keep_resident` doing the no-download repair
check
- `downloaded_after` could had been accounted for multiple times, and
also when repairing to match on-disk state
Cc: #5331
This is a follow-up to #7051 where `LayerInner::drop` and
`LayerInner::evict_blocking` were not noticed to require a gate before
the file deletion. The lack of entering a gate opens up a similar
possibility of deleting a layer file which a newer Timeline instance has
already checked out to be resident in a similar case as #7051.
## Problem
Before this PR, it was possible that on-demand downloads were started
after `Timeline::shutdown()`.
For example, we have observed a walreceiver-connection-handler-initiated
on-demand download that was started after `Timeline::shutdown()`s final
`task_mgr::shutdown_tasks()` call.
The underlying issue is that `task_mgr::shutdown_tasks()` isn't sticky,
i.e., new tasks can be spawned during or after
`task_mgr::shutdown_tasks()`.
Cc: https://github.com/neondatabase/neon/issues/4175 in lieu of a more
specific issue for task_mgr. We already decided we want to get rid of it
anyways.
Original investigation:
https://neondb.slack.com/archives/C033RQ5SPDH/p1709824952465949
## Changes
- enter gate while downloading
- use timeline cancellation token for cancelling download
thereby, fixes#7054
Entering the gate might also remove recent "kept the gate from closing"
in staging.
## Problem
Collection of small changes, batched together to reduce CI overhead.
## Summary of changes
- Layer download messages include size -- this is useful when watching a
pageserver hydrate its on disk cache in the log.
- Controller migrate API could put an invalid NodeId into TenantState
- Scheduling errors during tenant create could result in creating some
shards and not others.
- Consistency check could give hard-to-understand failures in tests if a
reconcile was in process: explicitly fail the check if reconciles are in
progress instead.
part of #6663
See that epic for more context & related commits.
Problem
-------
Before this PR, the layer-file-creating code paths were using
VirtualFile, but under the hood these were still blocking system calls.
Generally this meant we'd stall the executor thread, unless the caller
"knew" and used the following pattern instead:
```
spawn_blocking(|| {
Handle::block_on(async {
VirtualFile::....().await;
})
}).await
```
Solution
--------
This PR adopts `tokio-epoll-uring` on the layer-file-creating code paths
in pageserver.
Note that on-demand downloads still use `tokio::fs`, these will be
converted in a future PR.
Design: Avoiding Regressions With `std-fs`
------------------------------------------
If we make the VirtualFile write path truly async using
`tokio-epoll-uring`, should we then remove the `spawn_blocking` +
`Handle::block_on` usage upstack in the same commit?
No, because if we’re still using the `std-fs` io engine, we’d then block
the executor in those places where previously we were protecting us from
that through the `spawn_blocking` .
So, if we want to see benefits from `tokio-epoll-uring` on the write
path while also preserving the ability to switch between
`tokio-epoll-uring` and `std-fs` , where `std-fs` will behave identical
to what we have now, we need to ***conditionally* use `spawn_blocking +
Handle::block_on`** .
I.e., in the places where we use that know, we’ll need to make that
conditional based on the currently configured io engine.
It boils down to investigating all the places where we do
`spawn_blocking(... block_on(... VirtualFile::...))`.
Detailed [write-up of that investigation in
Notion](https://neondatabase.notion.site/Surveying-VirtualFile-write-path-usage-wrt-tokio-epoll-uring-integration-spawn_blocking-Handle-bl-5dc2270dbb764db7b2e60803f375e015?pvs=4
), made publicly accessible.
tl;dr: Preceding PRs addressed the relevant call sites:
- `metadata` file: turns out we could simply remove it (#6777, #6769,
#6775)
- `create_delta_layer()`: made sensitive to `virtual_file_io_engine` in
#6986
NB: once we are switched over to `tokio-epoll-uring` everywhere in
production, we can deprecate `std-fs`; to keep macOS support, we can use
`tokio::fs` instead. That will remove this whole headache.
Code Changes In This PR
-----------------------
- VirtualFile API changes
- `VirtualFile::write_at`
- implement an `ioengine` operation and switch `VirtualFile::write_at`
to it
- `VirtualFile::metadata()`
- curiously, we only use it from the layer writers' `finish()` methods
- introduce a wrapper `Metadata` enum because `std::fs::Metadata` cannot
be constructed by code outside rust std
- `VirtualFile::sync_all()` and for completeness sake, add
`VirtualFile::sync_data()`
Testing & Rollout
-----------------
Before merging this PR, we ran the CI with both io engines.
Additionally, the changes will soak in staging.
We could have a feature gate / add a new io engine
`tokio-epoll-uring-write-path` to do a gradual rollout. However, that's
not part of this PR.
Future Work
-----------
There's still some use of `std::fs` and/or `tokio::fs` for directory
namespace operations, e.g. `std::fs::rename`.
We're not addressing those in this PR, as we'll need to add the support
in tokio-epoll-uring first. Note that rename itself is usually fast if
the directory is in the kernel dentry cache, and only the fsync after
rename is slow. These fsyncs are using tokio-epoll-uring, so, the impact
should be small.
Because of bugs evictions could hang and pause disk usage eviction task.
One such bug is known and fixed#6928. Guard each layer eviction with a
modest timeout deeming timeouted evictions as failures, to be
conservative.
In addition, add logging and metrics recording on each eviction
iteration:
- log collection completed with duration and amount of layers
- per tenant collection time is observed in a new histogram
- per tenant layer count is observed in a new histogram
- record metric for collected, selected and evicted layer counts
- log if eviction takes more than 10s
- log eviction completion with eviction duration
Additionally remove dead code for which no dead code warnings appeared
in earlier PR.
Follow-up to: #6060.
## Problem
The vectored read path proposed in
https://github.com/neondatabase/neon/pull/6576 seems
to be functionally correct, but in my testing (see below) it is about 10-20% slower than the naive
sequential vectored implementation.
## Summary of changes
There's three parts to this PR:
1. Supporting vectored blob reads. This is actually trickier than it
sounds because on disk blobs are prefixed with a variable length size header.
Since the blobs are not necessarily fixed size, we need to juggle the offsets
such that the callers can retrieve the blobs from the resulting buffer.
2. Merge disk read requests issued by the vectored read path up to a
maximum size. Again, the merging is complicated by the fact that blobs
are not fixed size. We keep track of the begin and end offset of each blob
and pass them into the vectored blob reader. In turn, the reader will return
a buffer and the offsets at which the blobs begin and end.
3. A benchmark for basebackup requests against tenant with large SLRU
block counts is added. This required a small change to pagebench and a new config
variable for the pageserver which toggles the vectored get validation.
We can probably optimise things further by adding a little bit of
concurrency for our IO. In principle, it's as simple as spawning a task which deals with issuing
IO and doing the serialisation and handling on the parent task which receives input via a
channel.
Not allowing evicting wanted deleted layers is something I've forgotten
to implement on #5645. This PR makes it possible to evict such layers,
which should reduce the amount of hanging evictions.
Fixes: #6928
Co-authored-by: Christian Schwarz <christian@neon.tech>
This PR introduces a new vectored implementation of the read path.
The search is basically a DFS if you squint at it long enough.
LayerFringe tracks the next layers to visit and acts as our stack.
Vertices are tuples of (layer, keyspace, lsn range). Continuously
pop the top of the stack (most recent layer) and do all the reads
for one layer at once.
The search maintains a fringe (`LayerFringe`) which tracks all the
layers that intersect the current keyspace being searched. Continuously
pop the top of the fringe (layer with highest LSN) and get all the data
required from the layer in one go.
Said search is done on one timeline at a time. If data is still required for
some keys, then search the ancestor timeline.
Apart from the high level layer traversal, vectored variants have been
introduced for grabbing data from each layer type. They still suffer from
read amplification issues and that will be addressed in a different PR.
You might notice that in some places we duplicate the code for the
existing read path. All of that code will be removed when we switch
the non-vectored read path to proxy into the vectored read path.
In the meantime, we'll have to contend with the extra cruft for the sake
of testing and gentle releasing.
