Gabriel39 commented on PR #65851:
URL: https://github.com/apache/doris/pull/65851#issuecomment-5154845233

   Current-head integration blockers and remaining memory coverage 
(`a83d213137`)
   
   The two previously reported P1 issues have substantive code fixes, but the 
latest head is still not merge-ready:
   
   1. **P1 — A supported projection from a table containing VARIANT crashes in 
FE.** External Regression runs `SELECT id FROM test_variant_repro`; `id` is 
supported and only the unprojected sibling is VARIANT, but planning fails with 
`Cannot invoke "java.util.Collection.isEmpty()" because "elements" is null`. 
This violates the intended graceful-degradation contract that one unsupported 
column must not make supported columns unreadable. Please collect the FE stack, 
fix the null path, and keep this existing regression green.
   2. **P2 — Row-lineage INSERT validation changed its error contract.** Two 
regressions expect `Cannot specify invisible column _row_id`, but the pinned 
write schema excludes the reserved invisible field early and now reports 
`column not found` / `Unknown column`. The operation is still rejected, but the 
established validation behavior and regression contract need to be preserved or 
deliberately updated with justification.
   3. **The large-split test does not close all FE OOM paths.** It forces 
`num_files_in_batch_mode=1` on format v2, which proves laziness only after the 
streaming gate is entered. The production default is 1024, so fewer large files 
can still yield many byte splits on the synchronous path; format v3 also 
unconditionally returns `-1` from `streamingSplitEstimate()`. `planScan()` then 
retains every range in an `ArrayList`, leaving memory proportional to final 
split count. Please add coverage for the default threshold, a few huge files, 
and a large v3 scan, or add a bounded path for those cases.
   4. **FE UT is red.** 
`IcebergRowLevelDmlTransformTest.synthesizeDeleteOnPluginTableBuildsSinkTargetingIt`
 passes a null `ConnectContext`, while synthesis now dereferences its statement 
context. This appears to be a test-fixture regression rather than a production 
caller, but it still blocks the current head.
   
   The confirmed explicit-DEFAULT, duplicate historical-name carrier, and 
omitted nullable MERGE-column correctness issues are filed inline. Latest 
checks: https://github.com/apache/doris/pull/65851/checks


-- 
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]

Reply via email to