VTTablet: Transmit raw MySQL packets for Fields and Rows over wire with grpc buffer

[harshit-gangal]

Description

In this PR, VTTablet sends raw MySQL packets to VTGate avoiding two conversion

1. MySQL to sqltypes.Result
2. sqltypes.Result to Proto

In this approach, the raw packets received from the MySQL is prefixed with the proto header and using mem.BufferSlice are send as-is over GRPC transport layer avoiding the double cost of conversions.

VTGate on the receiving side will convert the raw packets to sqltypes.Result, earlier the conversion was from proto to sqltypes.Result.

Related Issue(s)

• [wip] Raw MySQL packets #15522
• Closes Use Raw MySQL packets when passing results back from VTTablet to VTGate #17080
• Resolves VTTablet allocation reduction #17172

Checklist

• Tests were added or are not required
• Did the new or modified tests pass consistently locally and on CI?
• Documentation was added or is not required

[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.

[vitess-bot]

Hello! 👋

This Pull Request is now handled by arewefastyet. The current HEAD and future commits will be benchmarked.

You can find the performance comparison on the arewefastyet website.

[codecov]

Codecov Report

Attention: Patch coverage is 36.00917% with 279 lines in your changes missing coverage. Please review.

Project coverage is 67.31%. Comparing base (216fd70) to head (a300c8a).
Report is 10 commits behind head on main.

Files with missing lines	Patch %	Lines
go/mysql/conn.go	8.73%	115 Missing ⚠️
go/mysql/query.go	23.64%	113 Missing ⚠️
go/mysql/result.go	60.27%	29 Missing ⚠️
go/vt/vttablet/grpctabletconn/conn.go	41.17%	10 Missing ⚠️
go/vt/servenv/grpc_codec.go	78.26%	5 Missing ⚠️
go/vt/vttablet/grpcqueryservice/server.go	16.66%	5 Missing ⚠️
go/vt/vttablet/tabletserver/query_executor.go	90.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #17135      +/-   ##
==========================================
- Coverage   67.39%   67.31%   -0.08%     
==========================================
  Files        1570     1574       +4     
  Lines      252917   253484     +567     
==========================================
+ Hits       170446   170632     +186     
- Misses      82471    82852     +381     

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

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests
• JS Bundle Analysis - Avoid shipping oversized bundles

[vmg]

Alright, after a lot of tweaking this week I've got a branch going in our benchmarks which appears to behave properly. The results are, sadly, not good:

As you can see, we have managed to reduce allocation costs in the vttablet by almost 40%. That's an extremely significant amount which unfortunately doesn't appear to translate into savings anywhere else. There's no increase in QPS and no reduction on latency.

Why is this happening? Well, most of the performance wins to be had with pooling were already covered by this PR which I shipped a few weeks ago, enabling the v2 codec for GRPC. The idea here was that the CPU costs of parsing MySQL packets could be amortized further by moving the packet parsing from the tablets to the gates, and hence reducing GC costs on the former, but this doesn't appear to materialize.

I've also checked the possibility that the constant allocations we optimized away could be acting as a memory ballast and forcing GC to be triggering more often, but that doesn't appear to be the case because these appear to be all ephemeral.

Overall, I'm still not 100% convinced this is a dead end because I feel like we're missing something in this performance analysis. But I think we've already spent enough time on it (particularly @harshit-gangal who very kindly backported this patch and made it work again on newer Vitess) that we should stick it on the back-burner. I'll keep thinking about this however.