SEZ9 commented on issue #10203: URL: https://github.com/apache/seatunnel/issues/10203#issuecomment-6485912659
@ryanmeowy thanks for pointing back to the "On a death inside the backfill" paragraph in your 24 Oct reply (issuecomment-6379487210) — I had missed that it already covers the mid-backfill case, and the reasoning there is what I was after: a checkpoint taken inside the backfill persists the split with both watermarks null, so `isSnapshotReadFinished()` is false and restore re-runs it as 2a; once both watermarks are checkpointed, restore treats it as 2b without replaying, independent of whether `onSplitFinished` ran. To answer your question directly: yes, please add it to the 2a/2b section of the 23 Oct table comment (issuecomment-6357660652) as well. A short version is enough — one or two sentences stating that a mid-backfill checkpoint lands in 2a because the watermark pair is still null, and that the 2a/2b boundary is purely the checkpointed pair (`SnapshotSplit.isSnapshotReadFinished()`), with `finishedUnackedSplits` rebuilt from it on restore (`IncrementalSourceReader.addSplits:152-153`). The 23 Oct comment is the one people will read as the consolidated reference, so I would rather have the 2a/2b section self-contained than rely on readers finding the 24 Oct paragraph. Once that edit is in, I consider this point closed on my side. <!-- streview-comment:2104 --> -- 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]
