Skipping CI for Draft Pull Request.
If you want CI signal for your change, please convert it to an actual PR.
You can still manually trigger a test run with /test all

sequenceDiagram
  rect rgba(173, 216, 230, 0.5)
    Note over scatterRegionWithType,ReadPoolPressureFilter: Internal scatter leader candidate filtering
    scatterRegionWithType->>splitScatterReadCPUByStore: StoreKindLoads + recentMaxProvider
    splitScatterReadCPUByStore-->>scatterRegionWithType: readCPUByStore map
    scatterRegionWithType->>filterAllowedLeaderCandidateStores: candidates + readCPUByStore
    filterAllowedLeaderCandidateStores->>ReadPoolPressureFilter: Target(store) for each candidate
    ReadPoolPressureFilter-->>filterAllowedLeaderCandidateStores: statusStoreBusy or statusOK
    filterAllowedLeaderCandidateStores-->>scatterRegionWithType: filtered allowed candidates
  end

sequenceDiagram
  rect rgba(144, 238, 144, 0.5)
    Note over recordBatch,dispatch: Split-scatter epoch-version guard lifecycle
    recordBatch->>pending map: store pending item with regionWaitVersion
    collectTopPending->>pending map: read pending under lock
    collectTopPending->>collectTopPending: compare live region version vs regionWaitVersion
    collectTopPending->>collectTopPending: skip if not yet at minimum version
    dispatch->>pending map: recheck live region against regionWaitVersion
    dispatch->>dispatch: verify live region version >= regionWaitVersion
    dispatch->>pending map: skip dispatch if epoch requirement not met
  end

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands.}

Codecov Report

❌ Patch coverage is 73.83966% with 62 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.12%. Comparing base (6e3d72f) to head (01702d5).
⚠️ Report is 14 commits behind head on master.

❌ Your patch check has failed because the patch coverage (73.83%) is below the target coverage (74.00%). You can increase the patch coverage or adjust the target coverage.

@@            Coverage Diff             @@
##           master   #10856      +/-   ##
==========================================
- Coverage   79.14%   79.12%   -0.02%     
==========================================
  Files         536      541       +5     
  Lines       74033    75044    +1011     
==========================================
+ Hits        58594    59380     +786     
- Misses      11311    11464     +153     
- Partials     4128     4200      +72     

/retest

Case1

After introducing split scatter, for test cases where the read pool is already under significant and relatively uniform pressure, unnecessary scattering will cause performance degradation.

Case2

Therefore, we introduced a target CPU filter mechanism. If the CPU usage of a target exceeds 70% of the unified read pool CPU, we will no longer select it as a target. This restores the QPS to normal, but the tail latency will still be high due to task backlog.

Strangely, after further slowing down the speed, it was found that even after the scatter operator stopped, the subsequent wait duration still increased after a period of scatter operation.

Case3

Therefore, we introduced another mechanism: if the CPU usage difference between the source and target is less than 10%, we will no longer perform scheduling. This aligns with the design philosophy, as the goal of split scatter is to distribute the results of load base splitting, and the ultimate goal of this distribution is to prevent CPU overruns.

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: bufferflies
Once this PR has been reviewed and has the lgtm label, please assign niubell for approval. For more information see the Code Review Process.
Please ensure that each of them provides their approval before proceeding.

The full list of commands accepted by this bot can be found here.

[LGTM Timeline notifier]

Timeline:

• 2026-06-24 06:51:03.810404288 +0000 UTC m=+103147.473629801: ☑️ agreed by bufferflies.

/retest

@lhy1024: The following test failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Full PR test history. Your PR dashboard.