github-actions[bot] commented on code in PR #65359:
URL: https://github.com/apache/doris/pull/65359#discussion_r3543299657
##########
regression-test/suites/external_table_p0/iceberg/test_iceberg_deletion_vector.groovy:
##########
@@ -531,7 +531,7 @@ s3.path-style-access=true
run {
sql """
/* ${splitCacheProfileTag} */
- SELECT count(*), sum(id)
+ SELECT /*+ SET_VAR(parallel_pipeline_task_num=1) */ count(*),
sum(id)
Review Comment:
The new hint still leaves this profile assertion dependent on how many BEs
receive splits for the one Iceberg data file. `parallel_pipeline_task_num=1`
flows into the per-BE pipeline instance count, but the scan ranges are still
generated from `file_split_size=1` and assigned by
`FederationBackendPolicy.computeScanRangeAssignment()`, which round-robins
remote splits across the available backends. On the BE side the DV cache used
by `IcebergReaderMixin::read_deletion_vector()` is the `_kv_cache` created in
each `FileScanLocalState`, so each backend that scans any split of this same
file can load the same deletion vector once and contribute another `2048` to
the merged `NumDeleteRows` total. In a multi-BE external regression run this
can still report `2048 * backend_count`, so the flaky exact `2048` assertion is
not fully fixed. Please either force this profiled scan to a single
backend/local scan state, or make the assertion account for backend-local cache
scope instead of assuming
one global load.
--
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]
