mrhhsg commented on PR #69284: URL: https://github.com/apache/doris/pull/69284#issuecomment-6265096468
<!-- autopilot-response --> Reviewed the current review state of this PR (head `d781e14f59ee4647b0ea0cf4c76a3fec7515b9c7`). - `IC_kwDOBfSlxs8AAAABdWwwFA` (`/review`): this is the review trigger comment. The `Code Review` workflow for this head is still queued, so no review findings exist yet. Nothing to fix from this comment. - `IC_kwDOBfSlxs8AAAABdWwwLA` (contribution checklist prompt): the PR description already covers all five requested points. Problem and root cause: one `TDigest` per GROUP BY group eagerly reserved `2 * compression` processed centroids and `8 * compression + 1` unprocessed centroids, i.e. 800008 bytes per group at the default compression 10000, exhausting the ASAN secondary allocator chunk table at roughly 524288 live groups. Fix: let both vectors grow on demand while keeping `_max_processed` / `_max_unprocessed` thresholds unchanged, and stop the exact-size `reserve(total)` in `_merge_unprocessed` so repeated merges keep amortized geometric growth. Behavior impact: numerical results and the serialized format are unchanged; only allocation behavior changes. No new feature was added. The refactor/optimization rationale for `_merge_unprocessed` is explained in the PR body. There are no unresolved inline review threads on this PR. No code changes were made in this round and the branch head is unchanged. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
