mrhhsg opened a new pull request, #69507:
URL: https://github.com/apache/doris/pull/69507
### What problem does this PR solve?
Issue Number: None
Related PR: None
Problem Summary: A projected scanner can retain rows from an earlier small
block in its private padding buffer while returning a later block that reaches
the soft batch target. The shared SQL LIMIT counts only returned rows, so peer
scanners can continue reading selective tails even when those buffered rows
would satisfy the LIMIT. When a shared LIMIT is present, merge pending padding
with the current block in source order before projection and charge the counter
only for rows actually returned. Scans without a shared LIMIT and scans that
cannot merge padding retain their existing behavior.
### Release note
Projected scans with a shared LIMIT may return a block larger than the soft
batch target to avoid unnecessary reads by peer scanners.
### Check List (For Author)
- Test:
- Unit Test: 73 focused BE tests passed; the ThreadPool interleaving
test passed in three additional runs. An unrelated baseline test was
temporarily excluded from the test build and restored afterward; the full BE
test suite was not validated.
- Build: `BUILD_TYPE=ASAN ./build.sh --be` passed.
- Static checks: BE formatting, build hygiene, and changed-line
clang-tidy passed.
- Regression test: Not run. Deterministic BE tests cover the
cross-scanner interleaving and source-read counts.
- Performance benchmark: Not run; no latency improvement is claimed.
- Behavior changed: Yes (bounded projected scans flush mergeable pending
rows before returning).
- Does this need documentation: No
--
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]
