yujun777 commented on code in PR #68170:
URL: https://github.com/apache/doris/pull/68170#discussion_r4057013903
##########
fe/fe-core/src/main/java/org/apache/doris/mtmv/ivm/IvmFailureReason.java:
##########
@@ -38,6 +38,7 @@ public enum IvmFailureReason {
public boolean requiresCompleteRefresh() {
return this == BINLOG_BROKEN
+ || this == STREAM_UNSUPPORTED
Review Comment:
Addressed in 0044712b546, by deciding the attempt list up front rather than
by routing a failure after the fact.
`buildAttempts` now asks whether the MV has a base table without a usable
stream. When the request allows fallback, the attempts become `[COMPLETE]`: the
incremental rewrite reads the stream of every base table and a partition
refresh reads those of the PCT tables it realigns as well
(`collectPctResetPartitionIds` feeds `IvmFullRefreshMTMV` RESET reads), so no
other attempt can succeed, and COMPLETE is the only one that reconciles. That
also skips the partition attempt's partial barrier altogether. `[COMPLETE]` --
rather than `[IVM, COMPLETE]` -- because `MTMV.validateIvmRefreshStart` rejects
an IVM attempt while a baseline barrier is pending. The pending-baseline
pre-step then bows out through its existing "covered by the pending COMPLETE
attempt" branch, and a request that does not allow fallback still fails as
before: COMPLETE would refresh more than it asked for.
The stream check is the one the reconcile already uses (extracted into
`usableIvmStream`), and it iterates the same `relation.getBaseTables()` the
reconcile provisions streams for, so there is no separate predicate to keep in
sync.
Verified end to end: with the MV's streams dropped by hand, `REFRESH
MATERIALIZED VIEW mv PARTITIONS FALLBACK` now succeeds with
`RefreshMode=COMPLETE` and `IvmFallbackReason=STREAM_UNSUPPORTED`, and the
COMPLETE attempt recreates the streams; strict `REFRESH ... PARTITIONS` still
fails with `IVM stream not found`. Added
`regression-test/suites/mtmv_p0/ivm/test_ivm_partitions_fallback_stream_unusable`
for exactly that pair, including a task-level assertion.
One boundary worth stating: a stream that becomes unusable between this
check and the read still fails that refresh. The barrier then survives, and the
next refresh starts from the same check, sees the stream is gone and rebuilds
through COMPLETE -- so it recovers on the following run instead of needing an
explicit REFRESH COMPLETE.
--
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]
