morningman opened a new pull request, #64525:
URL: https://github.com/apache/doris/pull/64525
## What problem does this PR solve?
Stabilizes the `branch-4.0` P0 regression pipeline by fixing test failures
observed in
regression build #198126 (12 failed). This PR addresses **7 of them**.
### Backports of apache/master fixes
- **#64439** `[fix](load) fix empty statistics for forwarded INSERT` — real
product fix: a
forwarded `INSERT ... SELECT` reported empty `JobDetails`
(`ScannedRows=0`) because the
non-Nereids-distribute fallback `Coordinator` was built without the
`jobId`. Fixes
`load_p0/insert/test_insert_statistic`.
- **#63859** `[test](regression) Add debug point ANN index-only scan test` —
replaces the
fragile `ScanBytes`-profile-based ANN suites (an unreliable index-only
proxy) with
debug-point suites. Fixes `ann_index_p0/ann_index_only_scan_distance_expr`
and
`ann_index_p0/ann_index_only_scan_metric_direction` (both were new
failures).
- **#64402** `[fix](case) fix insert_group_commit_into_max_filter_ratio` —
cherry-pick of the
(currently open) master PR; aligns the case with the
`enable_insert_strict` /
`enable_strict_cast` decoupling (#63794). Fixes
`insert_p0/insert_group_commit_into_max_filter_ratio`.
- **Partial of #64238**: only the `scanner_profile` `actualRows` regex fix
(`PhysicalOlapScan[id]`
is rendered with the numeric nereids id, not the table name). The BE
CPU-counter rename in
#64238 is intentionally NOT backported to a release branch. Fixes
`query_profile/scanner_profile`.
### Test robustness / isolation fixes (the same defect also exists on
apache/master)
- `temp_table_p0/test_temp_table`: the replica-count detection was broken —
the regex `/:(\d+)/`
missed the space in the canonical `tag.location.default: 3`, and
`force_olap_table_replication_num`
was never consulted, so on a 3-replica cluster `SHOW TABLETS` returned 9
rows while the test
expected 3. Detection is now whitespace-tolerant, sums the allocation
per-tag counts (mirroring
`PropertyAnalyzer.analyzeReplicaAllocation` precedence) and falls back to
`force_olap_table_replication_num`, while still asserting the real replica
count.
- `backup_restore/test_backup_restore_reset_index_id` (+
`test_backup_restore_inverted_idx` and
`restore_p0/test_validate_restore_inverted_idx`): cross-suite race on the
process-global
`restore_reset_index_id` FE config — a concurrent suite flipped it to
`false` between this
suite's BACKUP and RESTORE. The three suites that mutate this global are
now `nonConcurrent`
so they run serially. The FE reset logic itself is correct; this is purely
test isolation.
## Release note
None
## Check List (For Committer)
- [x] Test <!-- regression tests -->
## Check List (For Reviewer who provide ASF feature)
- [ ] Performance regression
--
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]
