pageserver: direct I/O #8130

jcsp · 2024-06-21T15:36:54Z

Project Slack: #proj-pageserver-direct-io

Currently, we do buffered reads of data pages. Direct reads would be a better fit, because:

Pageserver data pages have an extremely low temporal locality on reads, because any repeatedly accessed pages are cached inside postgres. This makes it largely a waste of memory, which we could be using for other things.
The kernel page cache gives deceptively fast read performance on lightly loaded pageservers, making performance less consistent as pageservers are packed with larger numbers of tenants.

Tasks

Give feedback

predicting/testing perf impact of direct IO (inject delays into reads that currently hit RAM?) -- test the assumption that disks are fast enough and our bottleneck is mostly software.
changing code base to use 4k-aligned-buffers everywhere, and validate this by having VirtualFile panic!() if it sees an unaligned buffer
feature-flagged switch to issue IOs with O_DIRECT
Improve userspace PageCache monitoring, where we sort index pages & expect to have a high hit rate
Reap the benefits of no longer using so much RAM for kernel page cache: make PageCache (and any other userspace caches we choose to add) larger.
Options

Backpointer to the Slack DMs between John and Christian about this: https://neondb.slack.com/archives/D05KTCVS40H/p1718977335312439

Phase 1 Impl

Give feedback

feat(pageserver): add direct io pageserver config #8622

c/storage/pageserver t/feature
vectored read: only read each stx_dio_offset_align/stx_dio_mem_align-sized section once #8719
[WIP] feat(pageserver): dio-aligned buffer allocation #8720

c/storage/pageserver
pageserver: handle decompression outside vectored read_blobs #8942

c/storage/pageserver
feat(pageserver): newtype for dio-aligned buffer allocation #8730
Add aligned io buffer marker trait IoBufAligned and IoBufAlignedMut
Add runtime alignment checks
Implement perf simulation using Direct I/O to fake file
Reminder to check for fork + direct I/O https://neondb.slack.com/archives/C07BZ38E6SD/p1723657162079879
Options

Phase 2 Impl

Give feedback

Find and fix unaligned IOs for read
Find and fix unaligned IOs for write
Use performance simulator mode to get worst-case estimate for perf impact
explore sharing the direct IO merge code between InMemoryLayer and vectored read #8803
Options

Phase 3 Impl: Performance

Give feedback

Setup a manually managed testbench in EC2
Options

The text was updated successfully, but these errors were encountered:

problame · 2024-07-09T08:46:43Z

The direct IO RFC covers both read & write path. Renamed this epic to include reads and writes. (I don't think it's significantly more work).

Part of #8130, [RFC: Direct IO For Pageserver](https://github.com/neondatabase/neon/blob/problame/direct-io-rfc/docs/rfcs/034-direct-io-for-pageserver.md) ## Description Add pageserver config for evaluating/enabling direct I/O. - Disabled: current default, uses buffered io as is. - Evaluate: still uses buffered io, but could do alignment checking and perf simulation (pad latency by direct io RW to a fake file). - Enabled: uses direct io, behavior on alignment error is configurable. Signed-off-by: Yuchen Liang <yuchen@neon.tech>

yliang412 · 2024-08-12T13:26:02Z

Last week:

Drafted PS Config for Direct I/O
Setup managed testbench on EC2

This week:

Finish Simulation Mode + alignment check
Use aligned buffers on read path

problame · 2024-08-13T19:32:18Z

sync meeting:

VirtualFile wrapper
primitive for aligned buffer (can we continue to use BytesMut but with an allocation that we provide?)
vectored read planner refactor: vectored read: only read each stx_dio_offset_align/stx_dio_mem_align-sized section once #8719

yliang412 · 2024-08-19T13:41:37Z

Last week:

Worked on a couple versions of feat(pageserver): newtype for dio-aligned buffer allocation #8730.
Ext4 kernel code inspection on no extra memcpy.
Started working on vectored read: only read each stx_dio_offset_align/stx_dio_mem_align-sized section once #8719.

This week:

Finish vectored read: only read each stx_dio_offset_align/stx_dio_mem_align-sized section once #8719.

yliang412 · 2024-08-26T13:29:44Z

Last week:

Opened pageserver: do vectored read on each dio-aligned section once #8763
Benchmarked chunked vectored blob read: results

refs #6989 Problem ------- After unclean shutdown, we get restarted and read the local filesystem to make decisions on those reads. Some of the data might have not yet been fsynced when the unclean shutdown completed. Durability matters even though Pageservers are conceptually just a cache of state in S3. For example: - the cloud control plane is no control loop => pageserver responses to tenant attachmentm, etc, needs to be durable. - the storage controller does not rely on this (as much?) - we don't have layer file checksumming, so, downloaded+renamed but not fsynced layer files are technically not to be trusted - #2683 Solution -------- `syncfs` the tenants directory during startup, before we start reading from it. This is a bit overkill because we do remove some temp files (InMemoryLayer!) later during startup. Further, these temp files are particularly likely to be dirty in the kernel page cache. However, we don't want to refactor that cleanup code right now, and the dirty data on pageservers is generally not that high. Last, with [direct IO](#8130) we're going to have near-zero kernel page cache anyway quite soon.

refs #6989 Problem ------- After unclean shutdown, we get restarted, start reading the local filesystem, and make decisions based on those reads. However, some of the data might have not yet been fsynced when the unclean shutdown completed. Durability matters even though Pageservers are conceptually just a cache of state in S3. For example: - the cloud control plane is no control loop => pageserver responses to tenant attachmentm, etc, needs to be durable. - the storage controller does not rely on this (as much?) - we don't have layer file checksumming, so, downloaded+renamed but not fsynced layer files are technically not to be trusted - #2683 Solution -------- `syncfs` the tenants directory during startup, before we start reading from it. This is a bit overkill because we do remove some temp files (InMemoryLayer!) later during startup. Further, these temp files are particularly likely to be dirty in the kernel page cache. However, we don't want to refactor that cleanup code right now, and the dirty data on pageservers is generally not that high. Last, with [direct IO](#8130) we're going to have near-zero kernel page cache anyway quite soon.

Part of #8130, closes #8719. ## Problem Currently, vectored blob io only coalesce blocks if they are immediately adjacent to each other. When we switch to Direct IO, we need a way to coalesce blobs that are within the dio-aligned boundary but has gap between them. ## Summary of changes - Introduces a `VectoredReadCoalesceMode` for `VectoredReadPlanner` and `StreamingVectoredReadPlanner` which has two modes: - `AdjacentOnly` (current implementation) - `Chunked(<alignment requirement>)` - New `ChunkedVectorBuilder` that considers batching `dio-align`-sized read, the start and end of the vectored read will respect `stx_dio_offset_align` / `stx_dio_mem_align` (`vectored_read.start` and `vectored_read.blobs_at.first().start_offset` will be two different value). - Since we break the assumption that blobs within single `VectoredRead` are next to each other (implicit end offset), we start to store blob end offsets in the `VectoredRead`. - Adapted existing tests to run in both `VectoredReadCoalesceMode`. - The io alignment can also be live configured at runtime. Signed-off-by: Yuchen Liang <yuchen@neon.tech>

…lush (#8537) Part of [Epic: Bypass PageCache for user data blocks](#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](#8130) and #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. => #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 7c74112 origin/main HEAD: ef1c55c ``` <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 7c74112 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.

yliang412 · 2024-09-02T12:34:53Z

Last week:

Merged and perf validated pageserver: do vectored read on each dio-aligned section once #8763 (on preprod), decided to go with io_buffer_alignnment=512.

This week:

Polish feat(pageserver): newtype for dio-aligned buffer allocation #8730 with a WIP stacked on top to show usage
Rethink vectored_blob_io buffer interaction (where to do decompression + where to deserialize value?)
Deploy pageserver: do vectored read on each dio-aligned section once #8763 to prod region: ap-southeast-2 and ap-southeast-1
- https://github.com/neondatabase/infra/pull/1879

problame · 2024-09-04T14:45:02Z

Sync meeting:

ship 4c0a61d as a preliminary PR, to land in prod as part of next week's release
- retain .freeze() and .slice() for uncompressed buffers
- additional allocation for compressed buffers is ok
- review from @problame and @arpad-m
continue feat(pageserver): newtype for dio-aligned buffer allocation #8730
- actually do O_DIRECT (quick and dirty)
- see if it works functionally (local test bench, regression test suite run)
- some basic pagebench getpage-at-latest-lsn benchmarking
  - 100% PS pagecache hit rate, resize like prod config, or resize further and take a note
  - 100% virtualfile fd cache hit rate, resize like prod config
- fix until not failures
- => this demonstrates we can do O_DIRECT on the read path without errors, and we have a performance ballpark

Next steps after that:

InMemoryLayer direct IO for read of tail end
- develop separately
- then cherry-pick into feat(pageserver): newtype for dio-aligned buffer allocation #8730
slice up feat(pageserver): newtype for dio-aligned buffer allocation #8730 and ship it to prod in a way that allows safe rollout

jcsp added t/feature Issue type: feature, for new features or requests c/storage/pageserver Component: storage: pageserver labels Jun 21, 2024

problame mentioned this issue Jul 2, 2024

RFC: direct IO for Pageserver #8240

Draft

problame changed the title ~~pageserver: direct I/O for reads~~ pageserver: direct I/O Jul 9, 2024

yliang412 self-assigned this Jul 29, 2024

yliang412 mentioned this issue Aug 7, 2024

feat(pageserver): add direct io pageserver config #8622

Merged

5 tasks

problame mentioned this issue Aug 15, 2024

bypass PageCache for InMemoryLayer + avoid Value::deser on L0 flush #8537

Merged

yliang412 mentioned this issue Aug 20, 2024

pageserver: do vectored read on each dio-aligned section once #8763

Merged

5 tasks

problame mentioned this issue Aug 26, 2024

pageserver: startup: ensure local disk state is durable #8835

Merged

yliang412 mentioned this issue Aug 28, 2024

DO NOT MERGE: cherry pick https://github.com/neondatabase/neon/pull/8763 onto release branch for testing in pre-prod #8857

Closed

problame self-assigned this Sep 9, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

pageserver: direct I/O #8130

pageserver: direct I/O #8130

jcsp commented Jun 21, 2024 •

edited by yliang412

Loading

Tasks

Phase 1 Impl

Phase 2 Impl

Phase 3 Impl: Performance

problame commented Jul 9, 2024

yliang412 commented Aug 12, 2024

problame commented Aug 13, 2024 •

edited

Loading

yliang412 commented Aug 19, 2024

yliang412 commented Aug 26, 2024 •

edited

Loading

yliang412 commented Sep 2, 2024 •

edited

Loading

problame commented Sep 4, 2024

pageserver: direct I/O #8130

pageserver: direct I/O #8130

Comments

jcsp commented Jun 21, 2024 • edited by yliang412 Loading

Tasks

Phase 1 Impl

Phase 2 Impl

Phase 3 Impl: Performance

problame commented Jul 9, 2024

yliang412 commented Aug 12, 2024

problame commented Aug 13, 2024 • edited Loading

yliang412 commented Aug 19, 2024

yliang412 commented Aug 26, 2024 • edited Loading

yliang412 commented Sep 2, 2024 • edited Loading

problame commented Sep 4, 2024

jcsp commented Jun 21, 2024 •

edited by yliang412

Loading

problame commented Aug 13, 2024 •

edited

Loading

yliang412 commented Aug 26, 2024 •

edited

Loading

yliang412 commented Sep 2, 2024 •

edited

Loading