Binlog: Improve ZstdInMemoryDecompressorMaxSize management

[mattlord]

Description

Vitess supports MySQL's binlog transaction compression. That support lives primarily in a single file: https://github.com/vitessio/vitess/blob/main/go/mysql/binlog_event_compression.go

There's an important variable which controls HOW that work is done. That variable is currently a const:

vitess/go/mysql/binlog_event_compression.go

Lines 62 to 65 in f6067e0

	// At what size should we switch from the in-memory buffer
	// decoding to streaming mode which is much slower, but does
	// not require everything be done in memory.
	zstdInMemoryDecompressorMaxSize = 128 << (10 * 2) // 128MiB

That is currently hardcoded at 128MiB, which was somewhat arbitrary. The thinking was that all transactions will be compressed and you want to process them as fast as possible, while still being able to support payloads of virtually any size — the compressed transaction payload is limited by MySQL's max_allowed_packet size, but the size of the uncompressed payload is not strictly limited.
The in-memory buffer based decoding is fast but memory intensive, so the 128MiB threshold can be too high for environments that are memory constrained (e.g. 512MiB of memory or less). This PR makes this setting configurable via a vttablet flag — --binlog-in-memory-decompressor-max-size — so that it can be configured at the vttablet level based on the details of the execution environment for the process.

Note

This PR also changes how we handle the larger payloads. In local testing done for this PR I realized that the MaxMemory and/or MaxWindowSize options (same for streaming) that were added in v21+ via #16328 means that we CANNOT process compressed payloads which have an uncompressed size larger than the given size (example here). So this PR moves to this for the streaming method:
zstd.NewReader(nil, zstd.WithDecoderLowmem(true), zstd.WithDecoderConcurrency(1))
Which allows us to process the large payload (> ZstdInMemoryDecompressorMaxSize), no matter the size, but using the least amount of memory possible as it instructs the reader to limit memory allocations and limit it to 1 in flight window or block.

It's for this reason that, while this isn't the kind of thing we would normally backport (a new flag), the new flag and the noted change above are critical for those using MySQL with --binlog_transaction_compression. And the zstd.WithDecoderMaxMemory usage is new in v21 via #16328 so I think we should backport this to v21. You can see the failure users could encounter w/o it on main here: https://gist.github.com/mattlord/17c7dbf7985b8805bb6db3efcbaf2218

Related Issue(s)

• Fixes: Feature Request: Introduce VTTablet flag to control zstdInMemoryDecompressorMaxSize #17219

Checklist

• "Backport to:" labels have been added if this change should be back-ported to release branches
• If this change is to be back-ported to previous releases, a justification is included in the PR description
• Tests were added or are not required
• Did the new or modified tests pass consistently locally and on CI?
• Documentation: Document flag added in https://github.com/vitessio/vitess/pull/17220 website#1889

[vitess-bot]

Review Checklist

Hello reviewers! 👋 Please follow this checklist when reviewing this Pull Request.

General

• Ensure that the Pull Request has a descriptive title.
• Ensure there is a link to an issue (except for internal cleanup and flaky test fixes), new features should have an RFC that documents use cases and test cases.

Tests

• Bug fixes should have at least one unit or end-to-end test, enhancement and new features should have a sufficient number of tests.

Documentation

• Apply the release notes (needs details) label if users need to know about this change.
• New features should be documented.
• There should be some code comments as to why things are implemented the way they are.
• There should be a comment at the top of each new or modified test to explain what the test does.

New flags

• Is this flag really necessary?
• Flag names must be clear and intuitive, use dashes (-), and have a clear help text.

If a workflow is added or modified:

• Each item in Jobs should be named in order to mark it as required.
• If the workflow needs to be marked as required, the maintainer team must be notified.

Backward compatibility

• Protobuf changes should be wire-compatible.
• Changes to _vt tables and RPCs need to be backward compatible.
• RPC changes should be compatible with vitess-operator
• If a flag is removed, then it should also be removed from vitess-operator and arewefastyet, if used there.
• vtctl command output order should be stable and awk-able.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 67.41%. Comparing base (f6067e0) to head (623813d).
Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #17220      +/-   ##
==========================================
+ Coverage   67.40%   67.41%   +0.01%     
==========================================
  Files        1570     1570              
  Lines      252903   252917      +14     
==========================================
+ Hits       170460   170501      +41     
+ Misses      82443    82416      -27     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[shlomi-noach]

Let's add validation to the entered value - there have to be some bounds I guess? And first read it into a local variable before assigning it into mysql.ZstdInMemoryDecompressorMaxSize.

[mattlord]

The bounds are set by the type. If it’s 0 then streaming mode will always be used. If it’s the max then the in-memory buffers will always be used.

Why would we read it into a local variable first?

[shlomi-noach]

Why would we read it into a local variable first?

We'd need to if there was need for boundary checking. But per your comment boundary checking is not requried, so a local variable is not needed.