We saw pageserver OOMs
https://github.com/neondatabase/cloud/issues/19715 for tenants doing
large writes. Add log lines around in-memory layers to hopefully collect
some info during my on-call shift next week.
## Summary of changes
* Estimate in-memory size of an in-mem layer.
* Print frozen layer number if there are too many layers accumulated in
memory.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
https://github.com/neondatabase/neon/pull/9524 split the decoding and
interpretation step from ingestion.
The output of the first phase is a `wal_decoder::models::InterpretedWalRecord`.
Before this patch set that struct contained a list of `Value` instances.
We wish to lift the decoding and interpretation step to the safekeeper,
but it would be nice if the safekeeper gave us a batch containing the raw data instead of actual values.
## Summary of changes
Main goal here is to make `InterpretedWalRecord` hold a raw buffer which
contains pre-serialized Values.
For this we do:
1. Add a `SerializedValueBatch` type. This is `inmemory_layer::SerializedBatch` with some
extra functionality for extension, observing values for shard 0 and tests.
2. Replace `inmemory_layer::SerializedBatch` with `SerializedValueBatch`
3. Make `DatadirModification` maintain a `SerializedValueBatch`.
### `DatadirModification` changes
`DatadirModification` now maintains a `SerializedValueBatch` and extends
it as new WAL records come in (to avoid flushing to disk on every
record).
In turn, this cascaded into a number of modifications to
`DatadirModification`:
1. Replace `pending_data_pages` and `pending_zero_data_pages` with `pending_data_batch`.
2. Removal of `pending_zero_data_pages` and its cousin `on_wal_record_end`
3. Rename `pending_bytes` to `pending_metadata_bytes` since this is what it tracks now.
4. Adapting of various utility methods like `len`, `approx_pending_bytes` and `has_dirty_data_pages`.
Removal of `pending_zero_data_pages` and the optimisation associated
with it ((1) and (2)) deserves more detail.
Previously all zero data pages went through `pending_zero_data_pages`.
We wrote zero data pages when filling gaps caused by relation extension
(case A) and when handling special wal records (case B). If it happened
that the same WAL record contained a non zero write for an entry in
`pending_zero_data_pages` we skipped the zero write.
Case A: We handle this differently now. When ingesting the
`SerialiezdValueBatch` associated with one PG WAL record, we identify the gaps and fill the
them in one go. Essentially, we move from a per key process (gaps were filled after each
new key), and replace it with a per record process. Hence, the optimisation is not
required anymore.
Case B: When the handling of a special record needs to zero out a key,
it just adds that to the current batch. I inspected the code, and I
don't think the optimisation kicked in here.
## Problem
We wish to have high level WAL decoding logic in `wal_decoder::decoder`
module.
## Summary of Changes
For this we need the `Value` and `NeonWalRecord` types accessible there, so:
1. Move `Value` and `NeonWalRecord` to `pageserver::value` and
`pageserver::record` respectively.
2. Get rid of `pageserver::repository` (follow up from (1))
3. Move PG specific WAL record types to `postgres_ffi::walrecord`. In
theory they could live in `wal_decoder`, but it would create a circular
dependency between `wal_decoder` and `postgres_ffi`. Long term it makes
sense for those types to be PG version specific, so that will work out nicely.
4. Move higher level WAL record types (to be ingested by pageserver)
into `wal_decoder::models`
Related: https://github.com/neondatabase/neon/issues/9335
Epic: https://github.com/neondatabase/neon/issues/9329
Part of #8130
## Problem
Pageserver previously goes through the kernel page cache for all the
IOs. The kernel page cache makes light-loaded pageserver have deceptive
fast performance. Using direct IO would offer predictable latencies of
our virtual file IO operations.
In particular for reads, the data pages also have an extremely low
temporal locality because the most frequently accessed pages are cached
on the compute side.
## Summary of changes
This PR enables pageserver to use direct IO for delta layer and image
layer reads. We can ship them separately because these layers are
write-once, read-many, so we will not be mixing buffered IO with direct
IO.
- implement `IoBufferMut`, an buffer type with aligned allocation
(currently set to 512).
- use `IoBufferMut` at all places we are doing reads on image + delta
layers.
- leverage Rust type system and use `IoBufAlignedMut` marker trait to
guarantee that the input buffers for the IO operations are aligned.
- page cache allocation is also made aligned.
_* in-memory layer reads and the write path will be shipped separately._
## Testing
Integration test suite run with O_DIRECT enabled:
https://github.com/neondatabase/neon/pull/9350
## Performance
We evaluated performance based on the `get-page-at-latest-lsn`
benchmark. The results demonstrate a decrease in the number of IOps, no
sigificant change in the latency mean, and an slight improvement on the
p99.9 and p99.99 latencies.
[Benchmark](https://www.notion.so/neondatabase/Benchmark-O_DIRECT-for-image-and-delta-layers-2024-10-01-112f189e00478092a195ea5a0137e706?pvs=4)
## Rollout
We will add `virtual_file_io_mode=direct` region by region to enable
direct IO on image + delta layers.
Signed-off-by: Yuchen Liang <yuchen@neon.tech>
This PR simplifies the pageserver configuration parsing as follows:
* introduce the `pageserver_api::config::ConfigToml` type
* implement `Default` for `ConfigToml`
* use serde derive to do the brain-dead leg-work of processing the toml
document
* use `serde(default)` to fill in default values
* in `pageserver` crate:
* use `toml_edit` to deserialize the pageserver.toml string into a
`ConfigToml`
* `PageServerConfig::parse_and_validate` then
* consumes the `ConfigToml`
* destructures it exhaustively into its constituent fields
* constructs the `PageServerConfig`
The rules are:
* in `ConfigToml`, use `deny_unknown_fields` everywhere
* static default values go in `pageserver_api`
* if there cannot be a static default value (e.g. which default IO
engine to use, because it depends on the runtime), make the field in
`ConfigToml` an `Option`
* if runtime-augmentation of a value is needed, do that in
`parse_and_validate`
* a good example is `virtual_file_io_engine` or `l0_flush`, both of
which need to execute code to determine the effective value in
`PageServerConf`
The benefits:
* massive amount of brain-dead repetitive code can be deleted
* "unused variable" compile-time errors when removing a config value,
due to the exhaustive destructuring in `parse_and_validate`
* compile-time errors guide you when adding a new config field
Drawbacks:
* serde derive is sometimes a bit too magical
* `deny_unknown_fields` is easy to miss
Future Work / Benefits:
* make `neon_local` use `pageserver_api` to construct `ConfigToml` and
write it to `pageserver.toml`
* This provides more type safety / coompile-time errors than the current
approach.
### Refs
Fixes#3682
### Future Work
* `remote_storage` deser doesn't reject unknown fields
https://github.com/neondatabase/neon/issues/8915
* clean up `libs/pageserver_api/src/config.rs` further
* break up into multiple files, at least for tenant config
* move `models` as appropriate / refine distinction between config and
API models / be explicit about when it's the same
* use `pub(crate)` visibility on `mod defaults` to detect stale values
## Problem
Currently, DatadirModification keeps a key-indexed map of all pending
writes, even though we (almost) never need to read back dirty pages for
anything other than metadata pages (e.g. relation sizes).
Related: https://github.com/neondatabase/neon/issues/6345
## Summary of changes
- commit() modifications before ingesting database creation wal records,
so that they are guaranteed to be able to get() everything they need
directly from the underlying Timeline.
- Split dirty pages in DatadirModification into pending_metadata_pages
and pending_data_pages. The data ones don't need to be in a
key-addressable format, so they just go in a Vec instead.
- Special case handling of zero-page writes in DatadirModification,
putting them in a map which is flushed on the end of a WAL record. This
handles the case where during ingest, we might first write a zero page,
and then ingest a postgres write to that page. We used to do this via
the key-indexed map of writes, but in this PR we change the data page
write path to not bother indexing these by key.
My least favorite thing about this PR is that I needed to change the
DatadirModification interface to add the on_record_end call. This is not
very invasive because there's really only one place we use it, but it
changes the object's behaviour from being clearly an aggregation of many
records to having some per-record state. I could avoid this by
implicitly doing the work when someone calls set_lsn or commit -- I'm
open to opinions on whether that's cleaner or dirtier.
## Performance
There may be some efficiency improvement here, but the primary
motivation is to enable an earlier stage of ingest to operate without
access to a Timeline. The `pending_data_pages` part is the "fast path"
bulk write data that can in principle be generated without a Timeline,
in parallel with other ingest batches, and ultimately on the safekeeper.
`test_bulk_insert` on AX102 shows approximately the same results as in
the previous PR #8591:
```
------------------------------ Benchmark results -------------------------------
test_bulk_insert[neon-release-pg16].insert: 23.577 s
test_bulk_insert[neon-release-pg16].pageserver_writes: 5,428 MB
test_bulk_insert[neon-release-pg16].peak_mem: 637 MB
test_bulk_insert[neon-release-pg16].size: 0 MB
test_bulk_insert[neon-release-pg16].data_uploaded: 1,922 MB
test_bulk_insert[neon-release-pg16].num_files_uploaded: 8
test_bulk_insert[neon-release-pg16].wal_written: 1,382 MB
test_bulk_insert[neon-release-pg16].wal_recovery: 18.264 s
test_bulk_insert[neon-release-pg16].compaction: 0.052 s
```
Part of [Epic: Bypass PageCache for user data
blocks](https://github.com/neondatabase/neon/issues/7386).
# Problem
`InMemoryLayer` still uses the `PageCache` for all data stored in the
`VirtualFile` that underlies the `EphemeralFile`.
# Background
Before this PR, `EphemeralFile` is a fancy and (code-bloated) buffered
writer around a `VirtualFile` that supports `blob_io`.
The `InMemoryLayerInner::index` stores offsets into the `EphemeralFile`.
At those offset, we find a varint length followed by the serialized
`Value`.
Vectored reads (`get_values_reconstruct_data`) are not in fact vectored
- each `Value` that needs to be read is read sequentially.
The `will_init` bit of information which we use to early-exit the
`get_values_reconstruct_data` for a given key is stored in the
serialized `Value`, meaning we have to read & deserialize the `Value`
from the `EphemeralFile`.
The L0 flushing **also** needs to re-determine the `will_init` bit of
information, by deserializing each value during L0 flush.
# Changes
1. Store the value length and `will_init` information in the
`InMemoryLayer::index`. The `EphemeralFile` thus only needs to store the
values.
2. For `get_values_reconstruct_data`:
- Use the in-memory `index` figures out which values need to be read.
Having the `will_init` stored in the index enables us to do that.
- View the EphemeralFile as a byte array of "DIO chunks", each 512 bytes
in size (adjustable constant). A "DIO chunk" is the minimal unit that we
can read under direct IO.
- Figure out which chunks need to be read to retrieve the serialized
bytes for thes values we need to read.
- Coalesce chunk reads such that each DIO chunk is only read once to
serve all value reads that need data from that chunk.
- Merge adjacent chunk reads into larger
`EphemeralFile::read_exact_at_eof_ok` of up to 128k (adjustable
constant).
3. The new `EphemeralFile::read_exact_at_eof_ok` fills the IO buffer
from the underlying VirtualFile and/or its in-memory buffer.
4. The L0 flush code is changed to use the `index` directly, `blob_io`
5. We can remove the `ephemeral_file::page_caching` construct now.
The `get_values_reconstruct_data` changes seem like a bit overkill but
they are necessary so we issue the equivalent amount of read system
calls compared to before this PR where it was highly likely that even if
the first PageCache access was a miss, remaining reads within the same
`get_values_reconstruct_data` call from the same `EphemeralFile` page
were a hit.
The "DIO chunk" stuff is truly unnecessary for page cache bypass, but,
since we're working on [direct
IO](https://github.com/neondatabase/neon/issues/8130) and
https://github.com/neondatabase/neon/issues/8719 specifically, we need
to do _something_ like this anyways in the near future.
# Alternative Design
The original plan was to use the `vectored_blob_io` code it relies on
the invariant of Delta&Image layers that `index order == values order`.
Further, `vectored_blob_io` code's strategy for merging IOs is limited
to adjacent reads. However, with direct IO, there is another level of
merging that should be done, specifically, if multiple reads map to the
same "DIO chunk" (=alignment-requirement-sized and -aligned region of
the file), then it's "free" to read the chunk into an IO buffer and
serve the two reads from that buffer.
=> https://github.com/neondatabase/neon/issues/8719
# Testing / Performance
Correctness of the IO merging code is ensured by unit tests.
Additionally, minimal tests are added for the `EphemeralFile`
implementation and the bit-packed `InMemoryLayerIndexValue`.
Performance testing results are presented below.
All pref testing done on my M2 MacBook Pro, running a Linux VM.
It's a release build without `--features testing`.
We see definitive improvement in ingest performance microbenchmark and
an ad-hoc microbenchmark for getpage against InMemoryLayer.
```
baseline: commit 7c74112b2a origin/main
HEAD: ef1c55c52e
```
<details>
```
cargo bench --bench bench_ingest -- 'ingest 128MB/100b seq, no delta'
baseline
ingest-small-values/ingest 128MB/100b seq, no delta
time: [483.50 ms 498.73 ms 522.53 ms]
thrpt: [244.96 MiB/s 256.65 MiB/s 264.73 MiB/s]
HEAD
ingest-small-values/ingest 128MB/100b seq, no delta
time: [479.22 ms 482.92 ms 487.35 ms]
thrpt: [262.64 MiB/s 265.06 MiB/s 267.10 MiB/s]
```
</details>
We don't have a micro-benchmark for InMemoryLayer and it's quite
cumbersome to add one. So, I did manual testing in `neon_local`.
<details>
```
./target/release/neon_local stop
rm -rf .neon
./target/release/neon_local init
./target/release/neon_local start
./target/release/neon_local tenant create --set-default
./target/release/neon_local endpoint create foo
./target/release/neon_local endpoint start foo
psql 'postgresql://cloud_admin@127.0.0.1:55432/postgres'
psql (13.16 (Debian 13.16-0+deb11u1), server 15.7)
CREATE TABLE wal_test (
id SERIAL PRIMARY KEY,
data TEXT
);
DO $$
DECLARE
i INTEGER := 1;
BEGIN
WHILE i <= 500000 LOOP
INSERT INTO wal_test (data) VALUES ('data');
i := i + 1;
END LOOP;
END $$;
-- => result is one L0 from initdb and one 137M-sized ephemeral-2
DO $$
DECLARE
i INTEGER := 1;
random_id INTEGER;
random_record wal_test%ROWTYPE;
start_time TIMESTAMP := clock_timestamp();
selects_completed INTEGER := 0;
min_id INTEGER := 1; -- Minimum ID value
max_id INTEGER := 100000; -- Maximum ID value, based on your insert range
iters INTEGER := 100000000; -- Number of iterations to run
BEGIN
WHILE i <= iters LOOP
-- Generate a random ID within the known range
random_id := min_id + floor(random() * (max_id - min_id + 1))::int;
-- Select the row with the generated random ID
SELECT * INTO random_record
FROM wal_test
WHERE id = random_id;
-- Increment the select counter
selects_completed := selects_completed + 1;
-- Check if a second has passed
IF EXTRACT(EPOCH FROM clock_timestamp() - start_time) >= 1 THEN
-- Print the number of selects completed in the last second
RAISE NOTICE 'Selects completed in last second: %', selects_completed;
-- Reset counters for the next second
selects_completed := 0;
start_time := clock_timestamp();
END IF;
-- Increment the loop counter
i := i + 1;
END LOOP;
END $$;
./target/release/neon_local stop
baseline: commit 7c74112b2a origin/main
NOTICE: Selects completed in last second: 1864
NOTICE: Selects completed in last second: 1850
NOTICE: Selects completed in last second: 1851
NOTICE: Selects completed in last second: 1918
NOTICE: Selects completed in last second: 1911
NOTICE: Selects completed in last second: 1879
NOTICE: Selects completed in last second: 1858
NOTICE: Selects completed in last second: 1827
NOTICE: Selects completed in last second: 1933
ours
NOTICE: Selects completed in last second: 1915
NOTICE: Selects completed in last second: 1928
NOTICE: Selects completed in last second: 1913
NOTICE: Selects completed in last second: 1932
NOTICE: Selects completed in last second: 1846
NOTICE: Selects completed in last second: 1955
NOTICE: Selects completed in last second: 1991
NOTICE: Selects completed in last second: 1973
```
NB: the ephemeral file sizes differ by ca 1MiB, ours being 1MiB smaller.
</details>
# Rollout
This PR changes the code in-place and is not gated by a feature flag.
## Problem/Solution
TimelineWriter::put_batch is simply a loop over individual puts. Each
put acquires and releases locks, and checks for potentially starting a
new layer. Batching these is more efficient, but more importantly
unlocks future changes where we can pre-build serialized buffers much
earlier in the ingest process, potentially even on the safekeeper
(imagine a future model where some variant of DatadirModification lives
on the safekeeper).
Ensuring that the values in put_batch are written to one layer also
enables a simplification upstream, where we no longer need to write
values in LSN-order. This saves us a sort, but also simplifies follow-on
refactors to DatadirModification: we can store metadata keys and data
keys separately at that level without needing to zip them together in
LSN order later.
## Why?
In this PR, these changes are simplify optimizations, but they are
motivated by evolving the ingest path in the direction of disentangling
extracting DatadirModification from Timeline. It may not obvious how
right now, but the general idea is that we'll end up with three phases
of ingest:
- A) Decode walrecords and build a datadirmodification with all the
simple data contents already in a big serialized buffer ready to write
to an ephemeral layer **<-- this part can be pipelined and parallelized,
and done on a safekeeper!**
- B) Let that datadirmodification see a Timeline, so that it can also
generate all the metadata updates that require a read-modify-write of
existing pages
- C) Dump the results of B into an ephemeral layer.
Related: https://github.com/neondatabase/neon/issues/8452
## Caveats
Doing a big monolithic buffer of values to write to disk is ordinarily
an anti-pattern: we prefer nice streaming I/O. However:
- In future, when we do this first decode stage on the safekeeper, it
would be inefficient to serialize a Vec of Value, and then later
deserialize it just to add blob size headers while writing into the
ephemeral layer format. The idea is that for bulk write data, we will
serialize exactly once.
- The monolithic buffer is a stepping stone to pipelining more of this:
by seriailizing earlier (rather than at the final put_value), we will be
able to parallelize the wal decoding and bulk serialization of data page
writes.
- The ephemeral layer's buffered writer already stalls writes while it
waits to flush: so while yes we'll stall for a couple milliseconds to
write a couple megabytes, we already have stalls like this, just
distributed across smaller writes.
## Benchmarks
This PR is primarily a stepping stone to safekeeper ingest filtering,
but also provides a modest efficiency improvement to the `wal_recovery`
part of `test_bulk_ingest`.
test_bulk_ingest:
```
test_bulk_insert[neon-release-pg16].insert: 23.659 s
test_bulk_insert[neon-release-pg16].pageserver_writes: 5,428 MB
test_bulk_insert[neon-release-pg16].peak_mem: 626 MB
test_bulk_insert[neon-release-pg16].size: 0 MB
test_bulk_insert[neon-release-pg16].data_uploaded: 1,922 MB
test_bulk_insert[neon-release-pg16].num_files_uploaded: 8
test_bulk_insert[neon-release-pg16].wal_written: 1,382 MB
test_bulk_insert[neon-release-pg16].wal_recovery: 18.981 s
test_bulk_insert[neon-release-pg16].compaction: 0.055 s
vs. tip of main:
test_bulk_insert[neon-release-pg16].insert: 24.001 s
test_bulk_insert[neon-release-pg16].pageserver_writes: 5,428 MB
test_bulk_insert[neon-release-pg16].peak_mem: 604 MB
test_bulk_insert[neon-release-pg16].size: 0 MB
test_bulk_insert[neon-release-pg16].data_uploaded: 1,922 MB
test_bulk_insert[neon-release-pg16].num_files_uploaded: 8
test_bulk_insert[neon-release-pg16].wal_written: 1,382 MB
test_bulk_insert[neon-release-pg16].wal_recovery: 23.586 s
test_bulk_insert[neon-release-pg16].compaction: 0.054 s
```
It's been rolled out everywhere, no configs are referencing it.
All code that's made dead by the removal of the config option is removed
as part of this PR.
The `page_caching::PreWarmingWriter` in `::No` mode is equivalent to a
`size_tracking_writer`, so, use that.
part of https://github.com/neondatabase/neon/issues/7418
The `tokio_epoll_uring::Slice` / `tokio_uring::Slice` type is weird.
The new `FullSlice` newtype is better. See the doc comment for details.
The naming is not ideal, but we'll clean that up in a future refactoring
where we move the `FullSlice` into `tokio_epoll_uring`. Then, we'll do
the following:
* tokio_epoll_uring::Slice is removed
* `FullSlice` becomes `tokio_epoll_uring::IoBufView`
* new type `tokio_epoll_uring::IoBufMutView` for the current
`tokio_epoll_uring::Slice<IoBufMut>`
Context
-------
I did this work in preparation for
https://github.com/neondatabase/neon/pull/8537.
There, I'm changing the type that the `inmemory_layer.rs` passes to
`DeltaLayerWriter::put_value_bytes` and thus it seemed like a good
opportunity to make this cleanup first.
## Problem
This follows a PR that insists all input keys are representable in 16
bytes:
- https://github.com/neondatabase/neon/pull/8648
& a PR that prevents postgres from sending us keys that use the high
bits of field2:
- https://github.com/neondatabase/neon/pull/8657
Motivation for this change:
1. Ingest is bottlenecked on CPU
2. InMemoryLayer can create huge (~1M value) BTreeMap<Key,_> for its
index.
3. Maps over i128 are much faster than maps over an arbitrary 18 byte
struct.
It may still be worthwhile to make the index two-tier to optimize for
the case where only the last 4 bytes (blkno) of the key vary frequently,
but simply using the i128 representation of keys has a big impact for
very little effort.
Related: #8452
## Summary of changes
- Introduce `CompactKey` type which contains an i128
- Use this instead of Key in InMemoryLayer's index, converting back and
forth as needed.
## Performance
All the small-value `bench_ingest` cases show improved throughput.
The one that exercises this index most directly shows a 35% throughput
increase:
```
ingest-small-values/ingest 128MB/100b seq, no delta
time: [374.29 ms 378.56 ms 383.38 ms]
thrpt: [333.88 MiB/s 338.13 MiB/s 341.98 MiB/s]
change:
time: [-26.993% -26.117% -25.111%] (p = 0.00 < 0.05)
thrpt: [+33.531% +35.349% +36.974%]
Performance has improved.
```
Ephemeral files cleanup on drop but did not delay shutdown, leading to
problems with restarting the tenant. The solution is as proposed:
- make ephemeral files carry the gate guard to delay `Timeline::gate`
closing
- flush in-memory layers and strong references to those on
`Timeline::shutdown`
The above are realized by making LayerManager an `enum` with `Open` and
`Closed` variants, and fail requests to modify `LayerMap`.
Additionally:
- fix too eager anyhow conversions in compaction
- unify how we freeze layers and handle errors
- optimize likely_resident_layers to read LayerFileManager hashmap
values instead of bouncing through LayerMap
Fixes: #7830
## Problem
We lack a rust bench for the inmemory layer and delta layer write paths:
it is useful to benchmark these components independent of postgres & WAL
decoding.
Related: https://github.com/neondatabase/neon/issues/8452
## Summary of changes
- Refactor DeltaLayerWriter to avoid carrying a Timeline, so that it can
be cleanly tested + benched without a Tenant/Timeline test harness. It
only needed the Timeline for building `Layer`, so this can be done in a
separate step.
- Add `bench_ingest`, which exercises a variety of workload "shapes"
(big values, small values, sequential keys, random keys)
- Include a small uncontroversial optimization: in `freeze`, only
exhaustively walk values to assert ordering relative to end_lsn in debug
mode.
These benches are limited by drive performance on a lot of machines, but
still useful as a local tool for iterating on CPU/memory improvements
around this code path.
Anecdotal measurements on Hetzner AX102 (Ryzen 7950xd):
```
ingest-small-values/ingest 128MB/100b seq
time: [1.1160 s 1.1230 s 1.1289 s]
thrpt: [113.38 MiB/s 113.98 MiB/s 114.70 MiB/s]
Found 1 outliers among 10 measurements (10.00%)
1 (10.00%) low mild
Benchmarking ingest-small-values/ingest 128MB/100b rand: Warming up for 3.0000 s
Warning: Unable to complete 10 samples in 10.0s. You may wish to increase target time to 18.9s.
ingest-small-values/ingest 128MB/100b rand
time: [1.9001 s 1.9056 s 1.9110 s]
thrpt: [66.982 MiB/s 67.171 MiB/s 67.365 MiB/s]
Benchmarking ingest-small-values/ingest 128MB/100b rand-1024keys: Warming up for 3.0000 s
Warning: Unable to complete 10 samples in 10.0s. You may wish to increase target time to 11.0s.
ingest-small-values/ingest 128MB/100b rand-1024keys
time: [1.0715 s 1.0828 s 1.0937 s]
thrpt: [117.04 MiB/s 118.21 MiB/s 119.46 MiB/s]
ingest-small-values/ingest 128MB/100b seq, no delta
time: [425.49 ms 429.07 ms 432.04 ms]
thrpt: [296.27 MiB/s 298.32 MiB/s 300.83 MiB/s]
Found 1 outliers among 10 measurements (10.00%)
1 (10.00%) low mild
ingest-big-values/ingest 128MB/8k seq
time: [373.03 ms 375.84 ms 379.17 ms]
thrpt: [337.58 MiB/s 340.57 MiB/s 343.13 MiB/s]
Found 1 outliers among 10 measurements (10.00%)
1 (10.00%) high mild
ingest-big-values/ingest 128MB/8k seq, no delta
time: [81.534 ms 82.811 ms 83.364 ms]
thrpt: [1.4994 GiB/s 1.5095 GiB/s 1.5331 GiB/s]
Found 1 outliers among 10 measurements (10.00%)
```
## Problem
We have been maintaining two read paths (legacy and vectored) for a
while now. The legacy read-path was only used for cross validation in some tests.
## Summary of changes
* Tweak all tests that were using the legacy read path to use the
vectored read path instead
* Remove the read path dispatching based on the pageserver configs
* Remove the legacy read path code
We will be able to remove the single blob io code in
`pageserver/src/tenant/blob_io.rs` when https://github.com/neondatabase/neon/issues/7386 is complete.
Closes https://github.com/neondatabase/neon/issues/8005
## Problem
The in-memory layer vectored read was very slow in some conditions
(walingest::test_large_rel) test. Upon profiling, I realised that 80% of
the time was spent building up the binary heap of reads. This stage
isn't actually needed.
## Summary of changes
Remove the planning stage as we never took advantage of it in order to
merge reads. There should be no functional change from this patch.
## Problem
Slack thread:
https://neondb.slack.com/archives/C033RQ5SPDH/p1720511577862519
We're seeing OOMs in staging on a pageserver that has
l0_flush.mode=Direct enabled.
There's a strong correlation between jumps in `maxrss_kb` and
`pageserver_timeline_ephemeral_bytes`, so, it's quite likely that
l0_flush.mode=Direct is the culprit.
Notably, the expected max memory usage on that staging server by the
l0_flush.mode=Direct is ~2GiB but we're seeing as much as 24GiB max RSS
before the OOM kill.
One hypothesis is that we're dropping the semaphore permit before all
the dirtied pages have been flushed to disk. (The flushing to disk
likely happens in the fsync inside the `.finish()` call, because we're
using ext4 in data=ordered mode).
## Summary of changes
Hold the permit until after we're done with `.finish()`.
part of https://github.com/neondatabase/neon/issues/7418
# Motivation
(reproducing #7418)
When we do an `InMemoryLayer::write_to_disk`, there is a tremendous
amount of random read I/O, as deltas from the ephemeral file (written in
LSN order) are written out to the delta layer in key order.
In benchmarks (https://github.com/neondatabase/neon/pull/7409) we can
see that this delta layer writing phase is substantially more expensive
than the initial ingest of data, and that within the delta layer write a
significant amount of the CPU time is spent traversing the page cache.
# High-Level Changes
Add a new mode for L0 flush that works as follows:
* Read the full ephemeral file into memory -- layers are much smaller
than total memory, so this is afforable
* Do all the random reads directly from this in memory buffer instead of
using blob IO/page cache/disk reads.
* Add a semaphore to limit how many timelines may concurrently do this
(limit peak memory).
* Make the semaphore configurable via PS config.
# Implementation Details
The new `BlobReaderRef::Slice` is a temporary hack until we can ditch
`blob_io` for `InMemoryLayer` => Plan for this is laid out in
https://github.com/neondatabase/neon/issues/8183
# Correctness
The correctness of this change is quite obvious to me: we do what we did
before (`blob_io`) but read from memory instead of going to disk.
The highest bug potential is in doing owned-buffers IO. I refactored the
API a bit in preliminary PR
https://github.com/neondatabase/neon/pull/8186 to make it less
error-prone, but still, careful review is requested.
# Performance
I manually measured single-client ingest performance from `pgbench -i
...`.
Full report:
https://neondatabase.notion.site/2024-06-28-benchmarking-l0-flush-performance-e98cff3807f94cb38f2054d8c818fe84?pvs=4
tl;dr:
* no speed improvements during ingest, but
* significantly lower pressure on PS PageCache (eviction rate drops to
1/3)
* (that's why I'm working on this)
* noticable but modestly lower CPU time
This is good enough for merging this PR because the changes require
opt-in.
We'll do more testing in staging & pre-prod.
# Stability / Monitoring
**memory consumption**: there's no _hard_ limit on max `InMemoryLayer`
size (aka "checkpoint distance") , hence there's no hard limit on the
memory allocation we do for flushing. In practice, we a) [log a
warning](23827c6b0d/pageserver/src/tenant/timeline.rs (L5741-L5743))
when we flush oversized layers, so we'd know which tenant is to blame
and b) if we were to put a hard limit in place, we would have to decide
what to do if there is an InMemoryLayer that exceeds the limit.
It seems like a better option to guarantee a max size for frozen layer,
dependent on `checkpoint_distance`. Then limit concurrency based on
that.
**metrics**: we do have the
[flush_time_histo](23827c6b0d/pageserver/src/tenant/timeline.rs (L3725-L3726)),
but that includes the wait time for the semaphore. We could add a
separate metric for the time spent after acquiring the semaphore, so one
can infer the wait time. Seems unnecessary at this point, though.
As seen with the pgvector 0.7.0 index builds, we can receive large
batches of images, leading to very large L0 layers in the range of 1GB.
These large layers are produced because we are only able to roll the
layer after we have witnessed two different Lsns in a single
`DataDirModification::commit`. As the single Lsn batches of images can
span over multiple `DataDirModification` lifespans, we will rarely get
to write two different Lsns in a single `put_batch` currently.
The solution is to remember the TimelineWriterState instead of eagerly
forgetting it until we really open the next layer or someone else
flushes (while holding the write_guard).
Additional changes are test fixes to avoid "initdb image layer
optimization" or ignoring initdb layers for assertion.
Cc: #7197 because small `checkpoint_distance` will now trigger the
"initdb image layer optimization"
extracted (and tested) from
https://github.com/neondatabase/neon/pull/7468, part of
https://github.com/neondatabase/neon/issues/7462.
The current codebase assumes the keyspace is dense -- which means that
if we have a keyspace of 0x00-0x100, we assume every key (e.g., 0x00,
0x01, 0x02, ...) exists in the storage engine. However, the assumption
does not hold any more in metadata keyspace. The metadata keyspace is
sparse. It is impossible to do per-key check.
Ideally, we should not have the assumption of dense keyspace at all, but
this would incur a lot of refactors. Therefore, we split the keyspaces
we have to dense/sparse and handle them differently in the code for now.
At some point in the future, we should assume all keyspaces are sparse.
## Summary of changes
* Split collect_keyspace to return dense+sparse keyspace.
* Do not allow generating image layers for sparse keyspace (for now --
will fix this next week, we need image layers anyways).
* Generate delta layers for sparse keyspace.
---------
Signed-off-by: Alex Chi Z <chi@neon.tech>
## Problem
Followup to https://github.com/neondatabase/neon/pull/6776
While #6776 makes compaction safe on sharded tenants, the logic for
keyspace partitioning remains inefficient: it assumes that the size of
data on a pageserver can be calculated simply as the range between start
and end of a Range -- this is not the case in sharded tenants, where
data within a range belongs to a variety of shards.
Closes: https://github.com/neondatabase/neon/issues/6774
## Summary of changes
I experimented with using a sharding-aware range type in KeySpace to
replace all the Range<Key> uses, but the impact on other code was quite
large (many places use the ranges), and not all of them need this
property of being able to approximate the physical size of data within a
key range.
So I compromised on expressing this as a ShardedRange type, but only
using that type selctively: during keyspace repartition, and in tiered
compaction when accumulating key ranges.
- keyspace partitioning methods take sharding parameters as an input
- new `ShardedRange` type wraps a Range<Key> and a shard identity
- ShardedRange::page_count is the shard-aware replacement for
key_range_size
- Callers that don't need to be shard-aware (e.g. vectored get code that
just wants to count the number of keys in a keyspace) can use
ShardedRange::raw_size to get the faster, shard-naive code (same as old
`key_range_size`)
- Compaction code is updated to carry a shard identity so that it can
use shard aware calculations
- Unit tests for the new fragmentation logic.
- Add a test for compaction on sharded tenants, that validates that we
generate appropriately sized image layers (this fails before fixing
keyspace partitioning)
previously in https://github.com/neondatabase/neon/pull/7375, we
observed that for in-memory layers, we will need to iterate every key in
the key space in order to get the result. The operation can be more
efficient if we use BTreeMap as the in-memory layer representation, even
if we are doing vectored get in a dense keyspace. Imagine a case that
the in-memory layer covers a very little part of the keyspace, and most
of the keys need to be found in lower layers. Using a BTreeMap can
significantly reduce probes for nonexistent keys.
## Summary of changes
* Use BTreeMap as in-memory layer representation.
* Optimize the vectored get flow to utilize the range scan functionality
of BTreeMap.
Signed-off-by: Alex Chi Z <chi@neon.tech>
part of https://github.com/neondatabase/neon/issues/7124
Changes
-------
This PR replaces the `EphemeralFile::write_blob`-specifc `struct Writer`
with re-use of `owned_buffers_io::write::BufferedWriter`.
Further, it restructures the code to cleanly separate
* the high-level aspect of EphemeralFile's write_blob / read_blk API
* the page-caching aspect
* the aspect of IO
* performing buffered write IO to an underlying VirtualFile
* serving reads from either the VirtualFile or the buffer if it hasn't
been flushed yet
* the annoying "feature" that reads past the end of the written range
are allowed and expected to return zeroed memory, as long as one remains
within one PAGE_SZ
## 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>
Problem
Currently, we base our time based layer rolling decision on the last
time we froze a layer. This means that if we roll a layer and then go
idle for longer than the checkpoint timeout the next layer will be
rolled after the first write. This is of course not desirable.
Summary of changes
Record the timepoint of the first write to an open layer and use that
for time based layer rolling decisions. Note that I had to keep
`Timeline::last_freeze_ts` for the sharded tenant disk consistent lsn
skip hack.
Fixes#7241
## Problem
The vectored read path holds the layer map lock while visiting a
timeline.
## Summary of changes
* Rework the fringe order to hold `Layer` on `Arc<InMemoryLayer>`
handles instead of descriptions that are resolved by the layer map at
the time of read. Note that previously `get_values_reconstruct_data` was
implemented for the layer description which already knew the lsn range
for the read. Now it is implemented on the new `ReadableLayer` handle
and needs to get the lsn range as an argument.
* Drop the layer map lock after updating the fringe.
Related https://github.com/neondatabase/neon/issues/6833
## Problem
Follows: https://github.com/neondatabase/neon/pull/7182
- Sufficient concurrent writes could OOM a pageserver from the size of
indices on all the InMemoryLayer instances.
- Enforcement of checkpoint_period only happened if there were some
writes.
Closes: https://github.com/neondatabase/neon/issues/6916
## Summary of changes
- Add `ephemeral_bytes_per_memory_kb` config property. This controls the
ratio of ephemeral layer capacity to memory capacity. The weird unit is
to enable making the ratio less than 1:1 (set this property to 1024 to
use 1MB of ephemeral layers for every 1MB of RAM, set it smaller to get
a fraction).
- Implement background layer rolling checks in
Timeline::compaction_iteration -- this ensures we apply layer rolling
policy in the absence of writes.
- During background checks, if the total ephemeral layer size has
exceeded the limit, then roll layers whose size is greater than the mean
size of all ephemeral layers.
- Remove the tick() path from walreceiver: it isn't needed any more now
that we do equivalent checks from compaction_iteration.
- Add tests for the above.
---------
Co-authored-by: Arpad Müller <arpad-m@users.noreply.github.com>
## Problem
Large quantities of ephemeral layer data can lead to excessive memory
consumption (https://github.com/neondatabase/neon/issues/6939). We
currently don't have a way to know how much ephemeral layer data is
present on a pageserver.
Before we can add new behaviors to proactively roll layers in response
to too much ephemeral data, we must calculate that total.
Related: https://github.com/neondatabase/neon/issues/6916
## Summary of changes
- Create GlobalResources and GlobalResourceUnits types, where timelines
carry a GlobalResourceUnits in their TimelineWriterState.
- Periodically update the size in GlobalResourceUnits:
- During tick()
- During layer roll
- During put() if the latest value has drifted more than 10MB since our
last update
- Expose the value of the global ephemeral layer bytes counter as a
prometheus metric.
- Extend the lifetime of TimelineWriterState:
- Instead of dropping it in TimelineWriter::drop, let it remain.
- Drop TimelineWriterState in roll_layer: this drops our guard on the
global byte count to reflect the fact that we're freezing the layer.
- Ensure the validity of the later in the writer state by clearing the
state in the same place we freeze layers, and asserting on the
write-ability of the layer in `writer()`
- Add a 'context' parameter to `get_open_layer_action` so that it can
skip the prev_lsn==lsn check when called in tick() -- this is needed
because now tick is called with a populated state, where
prev_lsn==Some(lsn) is true for an idle timeline.
- Extend layer rolling test to use this metric
## Problem
We reverted https://github.com/neondatabase/neon/pull/6661 a few days
ago. The change led to OOMs in
benchmarks followed by large WAL reingests.
The issue was that we removed [this
code](d04af08567/pageserver/src/tenant/timeline/walreceiver/walreceiver_connection.rs (L409-L417)).
That call may trigger a roll of the open layer due to
the keepalive messages received from the safekeeper. Removing it meant
that enforcing
of checkpoint timeout became even more lax and led to using up large
amounts of memory
for the in memory layer indices.
## Summary of changes
Piggyback on keep alive messages to enforce checkpoint timeout. This is
a hack, but it's exactly what
the current code is doing.
## Alternatives
Christhian, Joonas and myself sketched out a timer based approach
[here](https://github.com/neondatabase/neon/pull/6940). While discussing
it further, it became obvious that's also a bit of a hack and not the
desired end state. I chose not
to take that further since it's not what we ultimately want and it'll be
harder to rip out.
Right now it's unclear what the ideal system behaviour is:
* early flushing on memory pressure, or ...
* detaching tenants on memory pressure
This reverts commits 587cb705b8 (PR #6661)
and fcbe9fb184 (PR #6842).
Conflicts:
pageserver/src/tenant.rs
pageserver/src/tenant/timeline.rs
The conflicts were with
* pageserver: adjust checkpoint distance for sharded tenants (#6852)
* pageserver: add vectored get implementation (#6576)
Also we had to keep the `allowed_errors` to make `test_forward_compatibility` happy,
see the PR thread on GitHub for details.
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.
## Problem
One WAL record can actually produce an arbitrary amount of key value pairs.
This is problematic since it might cause our frozen layers to bloat past the
max allowed size of S3 single shot uploads.
[#6639](https://github.com/neondatabase/neon/pull/6639) introduced a "should roll"
check after every batch of `ingest_batch_size` (100 WAL records by default). This helps,
but the original problem still exists.
## Summary of changes
This patch moves the responsibility of rolling the currently open layer
to the `TimelineWriter`. Previously, this was done ad-hoc via calls
to `check_checkpoint_distance`. The advantages of this approach are:
* ability to split one batch over multiple open layers
* less layer map locking
* remove ad-hoc check_checkpoint_distance calls
More specifically, we track the current size of the open layer in the
writer. On each `put` check whether the current layer should be closed
and a new one opened. Keeping track of the currently open layer results
in less contention on the layer map lock. It only needs to be acquired
on the first write and on writes that require a roll afterwards.
Rolling the open layer can be triggered by:
1. The distance from the last LSN we rolled at. This bounds the amount
of WAL that the safekeepers need to store.
2. The size of the currently open layer.
3. The time since the last roll. It helps safekeepers to regard
pageserver as caught up and suspend activity.
Closes#6624
This PR refactors the `blob_io` code away from using slices towards
taking owned buffers and return them after use.
Using owned buffers will eventually allow us to use io_uring for writes.
part of https://github.com/neondatabase/neon/issues/6663
Depends on https://github.com/neondatabase/tokio-epoll-uring/pull/43
The high level scheme is as follows:
- call writing functions with the `BoundedBuf`
- return the underlying `BoundedBuf::Buf` for potential reuse in the
caller
NB: Invoking `BoundedBuf::slice(..)` will return a slice that _includes
the uninitialized portion of `BoundedBuf`_.
I.e., the portion between `bytes_init()` and `bytes_total()`.
It's a safe API that actually permits access to uninitialized memory.
Not great.
Another wrinkle is that it panics if the range has length 0.
However, I don't want to switch away from the `BoundedBuf` API, since
it's what tokio-uring uses.
We can always weed this out later by replacing `BoundedBuf` with our own
type.
Created an issue so we don't forget:
https://github.com/neondatabase/tokio-epoll-uring/issues/46
## Problem
For context, this problem was observed in a research project where we
try to make neon run in multiple regions and I was asked by @hlinnaka to
make this PR.
In our project, we use the pageserver in a non-conventional way such
that we would send a larger number of requests to the pageserver than
normal (imagine postgres without the buffer pool). I measured the time
from the moment a WAL record left the safekeeper to when it reached the
pageserver
([code](e593db1f5a/pageserver/src/tenant/timeline/walreceiver/walreceiver_connection.rs (L282-L287)))
and observed that when the number of get_page_at_lsn requests was high,
the wal receiving time increased significantly (see the left side of the
graphs below).
Upon further investigation, I found that the delay was caused by this
line
d2ca410919/pageserver/src/tenant/timeline.rs (L2348)
The `get_layer_for_write` method is called for every value during WAL
ingestion and it tries to acquire layers write lock every time, thus
this results in high contention when read lock is acquired more
frequently.
## Summary of changes
It is unnecessary to call `get_layer_for_write` repeatedly for all
values in a WAL message since they would end up in the same memory layer
anyway, so I created the batched versions of `InMemoryLayer::put_value`,
`InMemoryLayer ::put_tombstone`, `Timeline::put_value`, and
`Timeline::put_tombstone`, that acquire the locks once for a batch of
values.
Additionally, `DatadirModification` is changed to store multiple
versions of uncommitted values, and `WalIngest::ingest_record()` can now
ingest records without immediately committing them.
With these new APIs, the new ingestion loop can be changed to commit for
every `ingest_batch_size` records. The `ingest_batch_size` variable is
exposed as a config. If it is set to 1 then we get the same behavior
before this change. I found that setting this value to 100 seems to work
the best, and you can see its effect on the right side of the above
graphs.
---------
Co-authored-by: John Spray <john@neon.tech>
(includes two preparatory commits from
https://github.com/neondatabase/neon/pull/5960)
## Problem
To accommodate multiple shards in the same tenant on the same
pageserver, we must include the full TenantShardId in local paths. That
means that all code touching local storage needs to see the
TenantShardId.
## Summary of changes
- Replace `tenant_id: TenantId` with `tenant_shard_id: TenantShardId` on
Tenant, Timeline and RemoteTimelineClient.
- Use TenantShardId in helpers for building local paths.
- Update all the relevant call sites.
This doesn't update absolutely everything: things like PageCache,
TaskMgr, WalRedo are still shard-naive. The purpose of this PR is to
update the core types so that others code can be added/updated
incrementally without churning the most central shared types.
Implement a new `struct Layer` abstraction which manages downloadness
internally, requiring no LayerMap locking or rewriting to download or
evict providing a property "you have a layer, you can read it". The new
`struct Layer` provides ability to keep the file resident via a RAII
structure for new layers which still need to be uploaded. Previous
solution solved this `RemoteTimelineClient::wait_completion` which lead
to bugs like #5639. Evicting or the final local deletion after garbage
collection is done using Arc'd value `Drop`.
With a single `struct Layer` the closed open ended `trait Layer`, `trait
PersistentLayer` and `struct RemoteLayer` are removed following noting
that compaction could be simplified by simply not using any of the
traits in between: #4839.
The new `struct Layer` is a preliminary to remove
`Timeline::layer_removal_cs` documented in #4745.
Preliminaries: #4936, #4937, #5013, #5014, #5022, #5033, #5044, #5058,
#5059, #5061, #5074, #5103, epic #5172, #5645, #5649. Related split off:
#5057, #5134.
The TaskKind dimension added in #5339 is insufficient to understand what
kind of data causes the cache hits.
Regarding performance considerations: I'm not too worried because we're
moving from 3 to 4 one-byte sized fields; likely the space now used by
the new field was padding before. Didn't check this, though, and it
doesn't matter, we need the data.
What I don't like about this PR is that we have an `Unknown` content
type, and I also don't like that there's no compile-time way to assert
that it's set to something != `Unknown` when calling the page cache.
But, this is what I could come up with before tomorrow’s release, and I
think it covers the hot paths.
This PR adds a `task_kind` label to page cache access metrics.
These are to validate our hypothesis that the high hit page cache rate
we observe in prod is due to internal tasks, not getpage requests from
compute.
We believe the latter should near-always be a pageserver-page-cache
_miss_ because compute has it's own page cache, and hence there is no
locality of reference for its accesses to pageserver page cache.
Before this PR, we didn't have `RequestContext` propagation to any code
below the on-demand downloader.
The vast majority of changes in this PR is concerned with adding that
propagation.
## Problem
Once we use async file system APIs for `VirtualFile`, these functions
will also need to be async fn.
## Summary of changes
Makes the functions `open, open_with_options, create,sync_all,with_file`
of `VirtualFile` async fn, including all functions that call it. Like in
the prior PRs, the actual I/O operations are not using async APIs yet,
as per request in the #4743 epic.
We switch towards not using `VirtualFile` in the par_fsync module,
hopefully this is only temporary until we can actually do fully async
I/O in `VirtualFile`. This might cause us to exhaust fd limits in the
tests, but it should only be an issue for the local developer as we have
high ulimits in prod.
This PR is a follow-up of #5189, #5190, #5195, and #5203. Part of #4743.
## Problem
We want to convert the `VirtualFile` APIs to async fn so that we can
adopt one of the async I/O solutions.
## Summary of changes
Convert the following APIs of `VirtualFile` to async fn (as well as all
of the APIs calling it):
* `VirtualFile::seek`
* `VirtualFile::metadata`
* Also, prepare for deletion of the write impl by writing the summary to
a buffer before writing it to disk, as suggested in
https://github.com/neondatabase/neon/issues/4743#issuecomment-1700663864
. This change adds an additional warning for the case when the summary
exceeds a block. Previously, we'd have silently corrupted data in this
(unlikely) case.
* `WriteBlobWriter::write_blob`, in preparation for making
`VirtualFile::write_all` async.
## Problem
The `EphemeralFile::write_blob` function accesses the page cache
internally. We want to require `async` for these accesses in #5023.
## Summary of changes
This removes the implementaiton of the `BlobWriter` trait for
`EphemeralFile` and turns the `write_blob` function into an inherent
function. We can then make it async as well as the `push_bytes`
function. We move the `SER_BUFFER` thread-local into the
`InMemoryLayerInner` so that the same buffer can be accessed by
different threads as the async is (potentially) moved between threads.
Part of #4743, preparation for #5023.
(This PR is the successor of https://github.com/neondatabase/neon/pull/4984 )
## Summary
The current way in which `EphemeralFile` uses `PageCache` complicates
the Pageserver code base to a degree that isn't worth it.
This PR refactors how we cache `EphemeralFile` contents, by exploiting
the append-only nature of `EphemeralFile`.
The result is that `PageCache` only holds `ImmutableFilePage` and
`MaterializedPage`.
These types of pages are read-only and evictable without write-back.
This allows us to remove the writeback code from `PageCache`, also
eliminating an entire failure mode.
Futher, many great open-source libraries exist to solve the problem of a
read-only cache,
much better than our `page_cache.rs` (e.g., better replacement policy,
less global locking).
With this PR, we can now explore using them.
## Problem & Analysis
Before this PR, `PageCache` had three types of pages:
* `ImmutableFilePage`: caches Delta / Image layer file contents
* `MaterializedPage`: caches results of Timeline::get (page
materialization)
* `EphemeralPage`: caches `EphemeralFile` contents
`EphemeralPage` is quite different from `ImmutableFilePage` and
`MaterializedPage`:
* Immutable and materialized pages are for the acceleration of (future)
reads of the same data using `PAGE_CACHE_SIZE * PAGE_SIZE` bytes of
DRAM.
* Ephemeral pages are a write-back cache of `EphemeralFile` contents,
i.e., if there is pressure in the page cache, we spill `EphemeralFile`
contents to disk.
`EphemeralFile` is only used by `InMemoryLayer`, for the following
purposes:
* **write**: when filling up the `InMemoryLayer`, via `impl BlobWriter
for EphemeralFile`
* **read**: when doing **page reconstruction** for a page@lsn that isn't
written to disk
* **read**: when writing L0 layer files, we re-read the `InMemoryLayer`
and put the contents into the L0 delta writer
(**`create_delta_layer`**). This happens every 10min or when
InMemoryLayer reaches 256MB in size.
The access patterns of the `InMemoryLayer` use case are as follows:
* **write**: via `BlobWriter`, strictly append-only
* **read for page reconstruction**: via `BlobReader`, random
* **read for `create_delta_layer`**: via `BlobReader`, dependent on
data, but generally random. Why?
* in classical LSM terms, this function is what writes the
memory-resident `C0` tree into the disk-resident `C1` tree
* in our system, though, the values of InMemoryLayer are stored in an
EphemeralFile, and hence they are not guaranteed to be memory-resident
* the function reads `Value`s in `Key, LSN` order, which is `!=` insert
order
What do these `EphemeralFile`-level access patterns mean for the page
cache?
* **write**:
* the common case is that `Value` is a WAL record, and if it isn't a
full-page-image WAL record, then it's smaller than `PAGE_SIZE`
* So, the `EphemeralPage` pages act as a buffer for these `< PAGE_CACHE`
sized writes.
* If there's no page cache eviction between subsequent
`InMemoryLayer::put_value` calls, the `EphemeralPage` is still resident,
so the page cache avoids doing a `write` system call.
* In practice, a busy page server will have page cache evictions because
we only configure 64MB of page cache size.
* **reads for page reconstruction**: read acceleration, just as for the
other page types.
* **reads for `create_delta_layer`**:
* The `Value` reads happen through a `BlockCursor`, which optimizes the
case of repeated reads from the same page.
* So, the best case is that subsequent values are located on the same
page; hence `BlockCursor`s buffer is maximally effective.
* The worst case is that each `Value` is on a different page; hence the
`BlockCursor`'s 1-page-sized buffer is ineffective.
* The best case translates into `256MB/PAGE_SIZE` page cache accesses,
one per page.
* the worst case translates into `#Values` page cache accesses
* again, the page cache accesses must be assumed to be random because
the `Value`s aren't accessed in insertion order but `Key, LSN` order.
## Summary of changes
Preliminaries for this PR were:
- #5003
- #5004
- #5005
- uncommitted microbenchmark in #5011
Based on the observations outlined above, this PR makes the following
changes:
* Rip out `EphemeralPage` from `page_cache.rs`
* Move the `block_io::FileId` to `page_cache::FileId`
* Add a `PAGE_SIZE`d buffer to the `EphemeralPage` struct.
It's called `mutable_tail`.
* Change `write_blob` to use `mutable_tail` for the write buffering
instead of a page cache page.
* if `mutable_tail` is full, it writes it out to disk, zeroes it out,
and re-uses it.
* There is explicitly no double-buffering, so that memory allocation per
`EphemeralFile` instance is fixed.
* Change `read_blob` to return different `BlockLease` variants depending
on `blknum`
* for the `blknum` that corresponds to the `mutable_tail`, return a ref
to it
* Rust borrowing rules prevent `write_blob` calls while refs are
outstanding.
* for all non-tail blocks, return a page-cached `ImmutablePage`
* It is safe to page-cache these as ImmutablePage because EphemeralFile
is append-only.
## Performance
How doe the changes above affect performance?
M claim is: not significantly.
* **write path**:
* before this PR, the `EphemeralFile::write_blob` didn't issue its own
`write` system calls.
* If there were enough free pages, it didn't issue *any* `write` system
calls.
* If it had to evict other `EphemeralPage`s to get pages a page for its
writes (`get_buf_for_write`), the page cache code would implicitly issue
the writeback of victim pages as needed.
* With this PR, `EphemeralFile::write_blob` *always* issues *all* of its
*own* `write` system calls.
* Also, the writes are explicit instead of implicit through page cache
write back, which will help #4743
* The perf impact of always doing the writes is the CPU overhead and
syscall latency.
* Before this PR, we might have never issued them if there were enough
free pages.
* We don't issue `fsync` and can expect the writes to only hit the
kernel page cache.
* There is also an advantage in issuing the writes directly: the perf
impact is paid by the tenant that caused the writes, instead of whatever
tenant evicts the `EphemeralPage`.
* **reads for page reconstruction**: no impact.
* The `write_blob` function pre-warms the page cache when it writes the
`mutable_tail` to disk.
* So, the behavior is the same as with the EphemeralPages before this
PR.
* **reads for `create_delta_layer`**: no impact.
* Same argument as for page reconstruction.
* Note for the future:
* going through the page cache likely causes read amplification here.
Why?
* Due to the `Key,Lsn`-ordered access pattern, we don't read all the
values in the page before moving to the next page. In the worst case, we
might read the same page multiple times to read different `Values` from
it.
* So, it might be better to bypass the page cache here.
* Idea drafts:
* bypass PS page cache + prefetch pipeline + iovec-based IO
* bypass PS page cache + use `copy_file_range` to copy from ephemeral
file into the L0 delta file, without going through user space
## Problem
The `BlockCursor::read_blob` and `BlockCursor::read_blob_into_buf`
functions are calling `read_blk` internally, so if we want to make that
function async fn, they need to be async themselves.
## Summary of changes
* We first turn `ValueRef::load` into an async fn.
* Then, we switch the `RwLock` implementation in `InMemoryLayer` to use
the one from `tokio`.
* Last, we convert the `read_blob` and `read_blob_into_buf` functions
into async fn.
In three instances we use `Handle::block_on`:
* one use is in compaction code, which currently isn't async. We put the
entire loop into an `async` block to prevent the potentially hot loop
from doing cross-thread operations.
* one use is in dumping code for `DeltaLayer`. The "proper" way to
address this would be to enable the visit function to take async
closures, but then we'd need to be generic over async fs non async,
which [isn't supported by rust right
now](https://blog.rust-lang.org/inside-rust/2022/07/27/keyword-generics.html).
The other alternative would be to do a first pass where we cache the
data into memory, and only then to dump it.
* the third use is in writing code, inside a loop that copies from one
file to another. It is is synchronous and we'd like to keep it that way
(for now?).
Part of #4743
## Problem
In some places, the lock on `InMemoryLayerInner` is only created to
obtain `end_lsn`. This is not needed however, if we move `end_lsn` to
`InMemoryLayer` instead.
## Summary of changes
Make `end_lsn` a member of `InMemoryLayer`, and do less locking of
`InMemoryLayerInner`. `end_lsn` is changed from `Option<Lsn>` into an
`OnceLock<Lsn>`. Thanks to this change, we don't need to lock any more
in three functions.
Part of #4743 . Suggested in
https://github.com/neondatabase/neon/pull/4905#issuecomment-1666458428 .
