Rangsh commented on issue #12058:
URL: https://github.com/apache/seatunnel/issues/12058#issuecomment-5610735944

   @nzw921rx thanks for the clarification — attaching the **original** 
Diagnostics HTML flame graphs (and summaries) from a baseline run on 
**`origin/dev`**, not from #12081.
   
   ### Artifacts (original HTML)
   
   Gist: https://gist.github.com/Rangsh/de401aa7d0657d3f27e6b5d537a0acbc
   
   | File | Role |
   | --- | --- |
   | `flame-wall-forward.html` / `flame-wall-reverse.html` | Primary wall-clock 
flames |
   | `flame-cpu-forward.html` / `flame-cpu-reverse.html` | CPU flames |
   | `wall-profile-summary.md` / `cpu-profile-summary.md` / 
`gc-profile-summary.md` | Diagnostics summaries |
   
   Please download the HTML and open locally in a browser (GitHub gist raw view 
is not ideal for interactive flame graphs).
   
   ### Environment (baseline)
   
   | Item | Value |
   | --- | --- |
   | Code | `origin/dev` @ `75fd4ed4b` (**not** #12081) |
   | Method | `CheckpointStorageBenchmark.checkpointOverviewIncrementalUpdate$` 
|
   | Machine | Apple M1 / 16 GB / macOS |
   | JDK | Corretto **11.0.26** |
   | Tooling | `tools/benchmarks/profile_benchmarks.sh` (`profile wall` / `cpu` 
/ `gc`) |
   
   ### Brief reading notes (still investigation, not a CV fix claim)
   
   - Wall profile is dominated by Hazelcast idle/`park` samples, so searching 
`hsync` in the full wall graph often shows ~`Matched: 0.00%` even though the 
write path is present.
   - The measured write-through path still appears in the profiles as: 
`FileMapStore.store` → `RequestFuture.get` wait, and on the WAL worker: 
`HdfsWriter.flush` → `FSDataOutputStream.hsync` → `RawLocalFileSystem` / 
`LocalFSFileOutputStream` (`file:///`).
   - GC summary from this diagnostic run: Alloc/op ≈ **289 KB/op**, GC count = 
**1**, GC time = **4 ms** (profile Score is not comparable to a normal JMH 
Score).
   
   This is **not** evidence that #12081 resolves the within-run CV. Happy to 
follow whatever next isolation step you prefer (e.g. tighter slow-vs-normal 
iteration correlation, JFR pause overlap, or another Diagnostics mode).


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to