## 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
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.
This changes the default value of the `timeline_offloading` pageserver
and tenant configs to true, now that offloading has been rolled out
without problems.
There is also a small fix in the tenant config merge function, where we
applied the `lazy_slru_download` value instead of `timeline_offloading`.
Related issue: https://github.com/neondatabase/cloud/issues/21353
## Problem
The compaction loop currently runs periodically, which can cause it to
wait for up to 20 seconds before starting L0 compaction by default.
Also, when we later separate the semaphores for L0 compaction and image
compaction, we want to give up waiting for the image compaction
semaphore if L0 compaction is needed on any timeline.
Touches #10694.
## Summary of changes
Notify the compaction loop when an L0 flush (on any timeline) exceeds
`compaction_threshold`.
Also do some opportunistic cleanups in the area.
## Problem
helps investigate https://github.com/neondatabase/neon/issues/10482
## Summary of changes
In debug mode and testing mode, we will record all files visited by a
read operation, and print it out when it errors.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
L0 compaction can get starved by other background tasks. It needs to be
responsive to avoid read amp blowing up during heavy write workloads.
Touches #10694.
## Summary of changes
Add a separate semaphore for compaction, configurable via
`use_compaction_semaphore` (disabled by default). This is primarily for
testing in staging; it needs further work (in particular to split
image/L0 compaction jobs) before it can be enabled.
## Problem
Following #10641, let's add a few critical errors.
Resolves#10094.
## Summary of changes
Adds the following critical errors:
* WAL sender read/decode failure.
* WAL record ingestion failure.
* WAL redo failure.
* Missing key during compaction.
We don't add an error for missing keys during GetPage requests, since
we've seen a handful of these in production recently, and the cause is
still unclear (most likely a benign race).
## Problem
The local filesystem backend for remote storage doesn't set a
concurrency limit. While it can't/won't enforce a concurrency limit
itself, this also bounds the upload queue concurrency. Some tests create
thousands of uploads, which slows down the quadratic scheduling of the
upload queue, and there is no point spawning that many Tokio tasks.
Resolves#10409.
## Summary of changes
Set a concurrency limit of 100 for the LocalFS backend.
Before: `test_layer_map[release-pg17].test_query: 68.338 s`
After: `test_layer_map[release-pg17].test_query: 5.209 s`
Given a remote extensions manifest of the following:
```json
{
"public_extensions": [],
"custom_extensions": null,
"library_index": {
"pg_search": "pg_search"
},
"extension_data": {
"pg_search": {
"control_data": {
"pg_search.control": "comment = 'pg_search: Full text search for PostgreSQL using BM25'\ndefault_version = '0.14.1'\nmodule_pathname = '$libdir/pg_search'\nrelocatable = false\nsuperuser = true\nschema = paradedb\ntrusted = true\n"
},
"archive_path": "13117844657/v14/extensions/pg_search.tar.zst"
}
}
}
```
We were allowing a compute to install a remote extension that wasn't
listed in either public_extensions or custom_extensions.
Signed-off-by: Tristan Partin <tristan@neon.tech>
## Problem
Image layer generation could block L0 compactions for a long time.
## Summary of changes
* Refactored the return value of `create_image_layers_for_*` functions
to make it self-explainable.
* Preempt image layer generation in `Try` mode if L0 piles up.
Note that we might potentially run into a state that only the beginning
part of the keyspace gets image coverage. In that case, we either need
to implement something to prioritize some keyspaces with image coverage,
or tune the image_creation_threshold to ensure that the frequency of
image creation could keep up with L0 compaction.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
Co-authored-by: Erik Grinaker <erik@neon.tech>
## Problem
We don't currently have good alerts for critical errors, e.g. data
loss/corruption.
Touches #10094.
## Summary of changes
Add a `critical!` macro and corresponding
`libmetrics_tracing_event_count{level="critical"}` metric. This will:
* Emit an `ERROR` log message with prefix `"CRITICAL:"` and a backtrace.
* Increment `libmetrics_tracing_event_count{level="critical"}`, and
indirectly `level="error"`.
* Trigger a pageable alert (via the metric above).
* In debug builds, panic the process.
I'll add uses of the macro separately.
Logging errors with the debug format specifier causes multi-line errors,
which are sometimes a pain to deal with. Instead, we should use anyhow's
alternate display format, which shows the same information on a single
line.
Also adjusted a couple of error messages that were stale.
Fixesneondatabase/cloud#14710.
## Problem
The approach of having CancelMap as an in-memory structure increases
code complexity,
as well as putting additional load for Redis streams.
## Summary of changes
- Implement a set of KV ops for Redis client;
- Remove cancel notifications code;
- Send KV ops over the bounded channel to the handling background task
for removing and adding the cancel keys.
Closes#9660
## 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>
## Problem
We were logging a warning after a single request timeout, while listing
objects.
Closes: https://github.com/neondatabase/neon/issues/10166
## Summary of changes
- These timeouts are a pretty normal part of life, so back it off to
only log a warning after two in a row.
## 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.
Drop logical replication subscribers
before compute starts on a non-main branch.
Add new compute_ctl spec flag: drop_subscriptions_before_start
If it is set, drop all the subscriptions from the compute node
before it starts.
To avoid race on compute start, use new GUC
neon.disable_logical_replication_subscribers
to temporarily disable logical replication workers until we drop the
subscriptions.
Ensure that we drop subscriptions exactly once when endpoint starts on a
new branch.
It is essential, because otherwise, we may drop not only inherited, but
newly created subscriptions.
We cannot rely only on spec.drop_subscriptions_before_start flag,
because if for some reason compute restarts inside VM,
it will start again with the same spec and flag value.
To handle this, we save the fact of the operation in the database
in the neon.drop_subscriptions_done table.
If the table does not exist, we assume that the operation was never
performed, so we must do it.
If table exists, we check if the operation was performed on the current
timeline.
fixes: https://github.com/neondatabase/neon/issues/8790
## 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
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>
Add an endpoint to obtain the utilization of a safekeeper. Future
changes to the storage controller can use this endpoint to find the most
suitable safekeepers for newly created timelines, analogously to how
it's done for pageservers already.
Initially we just want to assign by timeline count, then we can iterate
from there.
Part of https://github.com/neondatabase/neon/issues/9011
# 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
gc-compaction needs the partitioning data to decide the job split. This
refactor allows concurrent access/computing the partitioning.
## Summary of changes
Make `partitioning` an ArcSwap so that others can access the
partitioning while we compute it. Fully eliminate the `repartition is
called concurrently` warning when gc-compaction is going on.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
Rename the safekeeper scheduling policy "disabled" to "pause".
A rename was requested in
https://github.com/neondatabase/neon/pull/10400#discussion_r1916259124,
as the "disabled" policy is meant to be analogous to the "pause" policy
for pageservers.
Also simplify the `SkSchedulingPolicyArg::from_str` function, relying on
the `from_str` implementation of `SkSchedulingPolicy`. Latter is used
for the database format as well, so it is quite stable. If we ever want
to change the UI, we'll need to duplicate the function again but this is
cheap.
## Problem
When a pageserver is receiving high rates of requests, we don't have a
good way to efficiently discover what the client's access pattern is.
Closes: https://github.com/neondatabase/neon/issues/10275
## Summary of changes
- Add
`/v1/tenant/x/timeline/y/page_trace?size_limit_bytes=...&time_limit_secs=...`
API, which returns a binary buffer.
- Add `pagectl page-trace` tool to decode and analyze the output.
---------
Co-authored-by: Erik Grinaker <erik@neon.tech>
Implementing the last missing endpoint of #9981, this adds support to
set the scheduling policy of an individual safekeeper, as specified in
the RFC. However, unlike in the RFC we call the endpoint
`scheduling_policy` not `status`
Closes#9981.
As for why not use the upsert endpoint for this: we want to have the
safekeeper upsert endpoint be used for testing and for deploying new
safekeepers, but not for changes of the scheduling policy. We don't want
to change any of the other fields when marking a safekeeper as
decommissioned for example, so we'd have to first fetch them only to
then specify them again. Of course one can also design an endpoint where
one can omit any field and it doesn't get modified, but it's still not
great for observability to put everything into one big "change something
about this safekeeper" endpoint.
## Problem
Safekeepers currently decode and interpret WAL for each shard
separately.
This is wasteful in terms of CPU memory usage - we've seen this in
profiles.
## Summary of changes
Fan-out interpreted WAL to multiple shards.
The basic is that wal decoding and interpretation happens in a separate
tokio task and senders
attach to it. Senders only receive batches concerning their shard and
only past the Lsn they've last seen.
Fan-out is gated behind the `wal_reader_fanout` safekeeper flag
(disabled by default for now).
When fan-out is enabled, it might be desirable to control the absolute
delta between the
current position and a new shard's desired position (i.e. how far behind
or ahead a shard may be).
`max_delta_for_fanout` is a new optional safekeeper flag which dictates
whether to create a new
WAL reader or attach to the existing one. By default, this behaviour is
disabled. Let's consider enabling
it if we spot the need for it in the field.
## Testing
Tests passed [here](https://github.com/neondatabase/neon/pull/10301)
with wal reader fanout enabled
as of
34f6a71718.
Related: https://github.com/neondatabase/neon/issues/9337
Epic: https://github.com/neondatabase/neon/issues/9329
## Problem
https://github.com/neondatabase/neon/issues/9965
## Summary of changes
Add to safekeeper http endpoint to switch membership configuration. Also
add it to python client for tests, and add simple test itself.
## Problem
Currently, we call `InterpretedWalRecord::from_bytes_filtered`
from each shard. To serve multiple shards at the same time,
the API needs to allow for enquiring about multiple shards.
## Summary of changes
This commit tweaks it a pretty brute force way. Naively, we could
just generate the shard for a key, but pre and post split shards
may be subscribed at the same time, so doing it efficiently is more
complex.
## Problem
https://github.com/neondatabase/neon/issues/9965
## Summary of changes
Add safekeeper membership configuration struct itself and storing it in
the control file. In passing also add creation timestamp to the control
file (there were cases where I wanted it in the past).
Remove obsolete unused PersistedPeerInfo struct from control file (still
keep it control_file_upgrade.rs to have it in old upgrade code).
Remove the binary representation of cfile in the roundtrip test.
Updating it is annoying, and we still test the actual roundtrip.
Also add configuration to timeline creation http request, currently used
only in one python test. In passing, slightly change LSNs meaning in the
request: normally start_lsn is passed (the same as ancestor_start_lsn in
similar pageserver call), but we allow specifying higher commit_lsn for
manual intervention if needed. Also when given LSN initialize
term_history with it.
## Problem
The upload queue can currently schedule an arbitrary number of tasks.
This can both spawn an unbounded number of Tokio tasks, and also
significantly slow down upload queue scheduling as it's quadratic in
number of operations.
Touches #10096.
## Summary of changes
Limit the number of inprogress tasks to the remote storage upload
concurrency. While this concurrency limit is shared across all tenants,
there's certainly no point in scheduling more than this -- we could even
consider setting the limit lower, but don't for now to avoid
artificially constraining tenants.
## Problem
Since https://github.com/neondatabase/neon/pull/9916, the preferred AZ
of a tenant is much more impactful, and we would like to make it more
visible in tooling.
## Summary of changes
- Include AZ in node describe API
- Include AZ info in node & tenant outputs in CLI
- Add metrics for per-node shard counts, labelled by AZ
- Add a CLI for setting preferred AZ on a tenant
- Extend AZ-setting API+CLI to handle None for clearing preferred AZ
## Problem
We want to do a more robust job of scheduling tenants into their home
AZ: https://github.com/neondatabase/neon/issues/8264.
Closes: https://github.com/neondatabase/neon/issues/8969
## Summary of changes
### Scope
This PR combines prioritizing AZ with a larger rework of how we do
optimisation. The rationale is that just bumping AZ in the order of
Score attributes is a very tiny change: the interesting part is lining
up all the optimisation logic to respect this properly, which means
rewriting it to use the same scores as the scheduler, rather than the
fragile hand-crafted logic that we had before. Separating these changes
out is possible, but would involve doing two rounds of test updates
instead of one.
### Scheduling optimisation
`TenantShard`'s `optimize_attachment` and `optimize_secondary` methods
now both use the scheduler to pick a new "favourite" location. Then
there is some refined logic for whether + how to migrate to it:
- To decide if a new location is sufficiently "better", we generate
scores using some projected ScheduleContexts that exclude the shard
under consideration, so that we avoid migrating from a node with
AffinityScore(2) to a node with AffinityScore(1), only to migrate back
later.
- Score types get a `for_optimization` method so that when we compare
scores, we will only do an optimisation if the scores differ by their
highest-ranking attributes, not just because one pageserver is lower in
utilization. Eventually we _will_ want a mode that does this, but doing
it here would make scheduling logic unstable and harder to test, and to
do this correctly one needs to know the size of the tenant that one is
migrating.
- When we find a new attached location that we would like to move to, we
will create a new secondary location there, even if we already had one
on some other node. This handles the case where we have a home AZ A, and
want to migrate the attachment between pageservers in that AZ while
retaining a secondary location in some other AZ as well.
- A unit test is added for
https://github.com/neondatabase/neon/issues/8969, which is implicitly
fixed by reworking optimisation to use the same scheduling scores as
scheduling.
## Problem
In preparation to https://github.com/neondatabase/neon/issues/9516. We
need to store rel size and directory data in the sparse keyspace, but it
does not support inheritance yet.
## Summary of changes
Add a new type of keyspace "sparse but inherited" into the system.
On the read path: we don't remove the key range when we descend into the
ancestor. The search will stop when (1) the full key range is covered by
image layers (which has already been implemented before), or (2) we
reach the end of the ancestor chain.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
Currently, if we want to move a secondary there isn't a neat way to do
that: we just have migration API for the attached location, and it is
only clean to use that if you've manually created a secondary via
pageserver API in the place you're going to move it to.
Secondary migration API enables:
- Moving the secondary somewhere because we would like to later move the
attached location there.
- Move the secondary location because we just want to reclaim some disk
space from its current location.
## Summary of changes
- Add `/migrate_secondary` API
- Add `tenant-shard-migrate-secondary` CLI
- Add tests for above
Taking continuous profiles every 20 seconds is likely too expensive (in
dollar terms). Let's try 60-second profiles. We can now interrupt
running profiles via `?force=true`, so this should be fine.
With a new beta build of the rust compiler, it's good to check out the
new lints. Either to find false positives, or find flaws in our code.
Additionally, it helps reduce the effort required to update to 1.85 in 6
weeks.
## Problem
It's only possible to take one CPU profile at a time. With Grafana
continuous profiling, a (low-frequency) CPU profile will always be
running, making it hard to take an ad hoc CPU profile at the same time.
Resolves#10072.
## Summary of changes
Add a `force` parameter for `/profile/cpu` which will end and return an
already running CPU profile, starting a new one for the current caller.
