yuqi1129 commented on PR #13831:
URL: https://github.com/apache/gravitino/pull/13831#issuecomment-6212569015
@jerryshao thanks for the thorough review. I replied to each inline comment.
Summary:
- **Fixed:**
- The `invalidateOnKeyChange` lock contract is now documented.
- The duplicated `batchGet` branch is removed.
- The first replay sweep after startup logs at INFO instead of WARN.
All three are in 5a1312fe35.
- **Kept, with reasons in each thread:**
- The tail query during a replay drain: a single primary-key lookup,
bounded by the pages in one sweep.
- The pre-insert epoch sample: a post-insert sample reopens the
ancestor-invalidation race; the timeline is in the thread. A per-key fence will
be evaluated with hit-rate data under #13828.
- PostgreSQL transaction-start `created_at`: a documented limitation;
`clock_timestamp()` is a possible small follow-up.
On the two test gaps:
- **Mid-sweep replay page failure:** covered by
`testFailedReplayPageResumesFromSavedPageCursor` in 6a242d6b7f. It checks that
the failed page is retried from the saved `(created_at, id)` cursor with the
same ceiling, and that the sweep does not restart. The test fails if the
failure path resets the cursor.
- **Query volume of a replay-only drain:** not added, because that behavior
is intentionally unchanged (see the reply on `doPollChanges`).
--
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]
