fix(logtail): stream-publish txn tails to remove batch-wide wait (#24325)#24383
Open
ck89119 wants to merge 3 commits into
Open
fix(logtail): stream-publish txn tails to remove batch-wide wait (#24325)#24383ck89119 wants to merge 3 commits into
ck89119 wants to merge 3 commits into
Conversation
…rixorigin#24325) (matrixorigin#24326) ### Background PR matrixorigin#22475 refactored `Manager.onTxnLogTails` in `pkg/vm/engine/tae/logtail/mgr.go` so that a batch of txns are collected in parallel and then published after a batch-wide `collectWg.Wait()`: ```go for i, item := range items { mgr.collectWg.Add(1) mgr.collectPool.Submit(func() { defer mgr.collectWg.Done() txn.GetStore().WaitEvent(txnif.WalPreparing) ... state := txn.GetTxnState(true) mgr.orderedList[i] = txnTail }) } mgr.collectWg.Wait() // ← blocks on slowest txn for i := range items { mgr.generateLogtailWithTxn(...) // serial push AFTER all ready } ``` Batch size is 100. Under high-concurrency short-tx workloads (e.g. sysbench oltp_delete t=32), the batch regularly fills up and a single slow txn defers logtail publish for all others. CN observes this through `txnOperator.unlock → timestampWaiter.GetTimestamp` waiting on `NotifyLatestCommitTS`, which in RC isolation gates commit return, inflating per-txn latency. This causes issue matrixorigin#24325: sysbench oltp_delete t=32 standalone TPS drops ~28% at the exact commit that merged matrixorigin#22475, and ~17% on main HEAD vs 3.0-dev (which does not carry matrixorigin#22475). ### Fix Replace the batch-wide wait with per-slot buffered channels + a serial publisher that reads slot 0, 1, 2, ... in order: - each submit goroutine signals its own `chan *txnWithLogtails` on completion - the publisher loops through slots in index order, `<-ch` per slot - a slow txn only blocks the publisher up to its slot; it does not delay the collection of later txns nor the publishing of earlier already-ready txns All other matrixorigin#22475 changes (single `logtailQueue`, `collectPool`, `WaitEvent(WalPreparing)` event) are preserved. Monotonicity of `previousSaveTS` inside `generateLogtailWithTxn` is preserved since publish still happens in index (PrepareTS) order. ### Benchmark `sysbench oltp_delete.lua` on standalone MO, Apple Silicon arm64, GOMAXPROCS=10, 10 tables × 1M rows, threads=32, time=60s, db-ps-mode=disable, skip_trx=on, 3 independent runs (fresh prepare + cleanup + mo-service restart per run), mean of 30-60s steady window: | Version | Steady TPS | timestampWaiter per-txn | |----------------------------------|-----------:|------------------------:| | main before this PR | 7970 | 585 μs | | **main + this fix** | **9402** | **334 μs** | | 3.0-dev (no matrixorigin#22475) | 9326 | 127 μs | | 8b3700a (commit before matrixorigin#22475) | 10487 | 123 μs | TPS recovers **+17.9% over current main** and passes 3.0-dev. The remaining tw gap vs 3.0-dev (334 μs → 127 μs) is secondary and does not materially affect TPS at t=32; it can be pursued separately if needed. ### Ablation (confirming the hotspot is exactly this function) Each candidate fix applied in isolation on main HEAD, 3-run mean: | Candidate fix | Steady TPS | tw per-txn | |---------------------------------------------------------|-----------:|-----------:| | main baseline | 7970 | 585 μs | | parallelize `LogEntryWriter.Finish()` marshal | 8283 | 592 μs | | synchronous marshal in `cmdmgr.ApplyTxnRecord` | 8280 | 599 μs | | **this PR (stream-publish in `onTxnLogTails`)** | **9402** | 334 μs | Only touching `onTxnLogTails` produces the recovery; marshal-path rewrites do not, which rules out marshal deferral itself as the regression source. Approved by: @XuPeng-SH
Qodo reviews are paused for this user.Troubleshooting steps vary by plan Learn more → On a Teams plan? Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center? |
2 tasks
Contributor
|
@Mergifyio refresh |
Contributor
✅ Pull request refreshed |
Contributor
Merge Queue Status
This pull request spent 1 hour 3 minutes 34 seconds in the queue, with no time running CI. Waiting for any of
All conditions
ReasonThe merge conditions cannot be satisfied due to failing checks HintYou may have to fix your CI before adding the pull request to the queue again. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What type of PR is this?
Which issue(s) this PR fixes:
issue #24325
What this PR does / why we need it:
Cherry-pick #24326 to 4.0-dev.