englefly commented on PR #68314:
URL: https://github.com/apache/doris/pull/68314#issuecomment-5808210797
Pushed `9df0ec6fa0a` with fixes for the pipeline failures.
**Reproduced and fixed locally**
* **P0 + cloud_p0: `datatype_p0/complex_types/test_pruned_columns`** —
`profile is not complete for lazy_pruned_column_recovery_..., state: null`. The
check I added read a `Profile Completion State` column from
`/rest/v1/query_profile`, but those rows only expose `Task State` / `Profile
ID` / `Sql Statement`, so the state was always `null` and the assertion could
never pass (it failed at the previous head too). It now uses the regression
framework's `ProfileAction#getProfileBySql()`, which polls until the profile
text is complete and contains `LazyReadPrunedTime`; the non-zero
`LazyReadPrunedTime` assertion is kept. Local: `datatype_p0/complex_types` → 8
suites, 0 failed.
* **P0: `shape_check/tpcds_sf1000/dphyper/query{16,94,95}`** — `Expect
-----hashAgg[DISTINCT_GLOBAL] But real ------hashAgg[GLOBAL]`. These
expectations are stale on the current base, not related to this PR (tpcds
tables are flat scalars; nothing here goes through nested-column pruning).
Reproduced locally: `shape_check/tpcds_sf1000` → 308 suites, exactly those 3
failed, and the actual outputs match what CI reports. Refreshed the three
`.out` files from that run; the branch-4.2 rebase of #56802 needs the same
refresh and its refreshed files differ from these only in the alias text of
`PhysicalOlapScan`. Local after refresh: `shape_check/tpcds_sf1000/dphyper` →
99 suites, 0 failed.
* **Clang Formatter** — the single violation is in
`be/test/exec/scan/vgeneric_iterators_test.cpp`, which this PR does not touch:
it came in with #68324 and fails the formatter check for every PR on the
current branch-4.2 (verified against `branch-4.2` itself). Applied the exact
clang-format 16 change CI asks for. A full scan of the CI's source set (`be/src
be/test cloud/src cloud/test`, honoring `.clang-format-ignore`) is now 0
violations over 4350 files.
**Looked at, not caused by this PR**
* **FE UT** — the 4 CI failures are
`FrontendServiceImplTest.testCreatePartitionRange`,
`...testListTableStatusUsesCombinedStatsForExpensiveColumns`,
`...testListTableStatusPrunesAllOptionalColumnsWhenRequiredColumnsIsEmpty` (all
"available backend num is 0") and
`HudiBatchFsViewOwnerTest.synchronousListingCancellationReturnsBeforeBlockedTaskAndRetainsLease`
(Mockito `WantedButNotInvoked`). All four are timing/load sensitive around the
UT mock-backend heartbeat window: they reproduced locally only while the
machine was busy and pass on a quiet run — `PruneNestedColumnTest`,
`VariantPruningLogicTest`, `TopNRuntimeFilterTest`, `IcebergScanNodeTest`,
`MaterializeProbeVisitorTest`, `FrontendServiceImplTest` → **220 tests, 0
failures**. `FrontendServiceImplTest` and `UtFrameUtils` are byte-identical
between the pre-rebase and current base and this PR does not touch that path,
so a re-run of the FE UT job should be green.
* **External Regression** — the failing test is
`job_p0.streaming_job.cdc.tvf.test_streaming_job_cdc_stream_mysql_alter_offset`
(a CDC streaming job, no nested-column pruning involved). #68324 fails External
Regression in the same way, so this looks like a known flaky/base issue.
* **cloud_p0: `cloud_p0/test_read_cluster_var_property`** — cloud-only suite
that scrapes a BE `/brpc_metrics` endpoint; it cannot be run locally. The
`check_coverage_fe` failure is the coverage gate on top of the FE UT job, so it
should clear together with it.
**Verification on the pushed commit**
* FE UT (6 access-path/related classes incl. `FrontendServiceImplTest`) →
220 tests, 0 failures; `mvn checkstyle:check -pl fe-core` → 0 violations.
* Regression on a local single-node cluster:
`nereids_rules_p0/column_pruning` → 9 suites 0 failed,
`datatype_p0/complex_types` → 8 suites 0 failed,
`shape_check/tpcds_sf1000/dphyper` → 99 suites 0 failed.
* BE UT was green in the previous round (`ColumnReaderTest.*`,
`SegmentIteratorLazyPrunedTest.*`, `SlotDescriptorTest.*` → 61 tests) and no BE
source changed since; only the clang-format line wrap in a BE test file.
Worth re-running `run buildall` on `9df0ec6fa0a`.
--
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]
