yujun777 commented on PR #68390:
URL: https://github.com/apache/doris/pull/68390#issuecomment-5795121571

   The findings from the 2026-09-23 review round are addressed in 
`6db4855adb6`, on top of the tip you reviewed.
   
   - **rename, all three findings**: the dependency re-key is dropped rather 
than patched. It cannot be made durable — the maps are rebuilt from 
`MTMV.relation`, which keeps the name the MV query spells, and a task result 
captured before the rename puts the old name back — so a rename invalidates as 
it did before. The cost is one extra COMPLETE for a rename-and-rename-back, 
which the unit tests and `test_ivm_partition_epoch_rebuild` now pin.
   - **status journal under `mvRwLock`**: `invalidateWholeMv` applies and 
submits under the lock and hands the journal write back; `alterMvProperties` 
awaits it after the unlock, the way the property record next to it already did.
   - **dirty rebuild lost on the IVM to PARTITIONS fallback**: the rebuild's 
committed snapshots are held on the task, every phase's accumulator starts from 
them, and the fallback's plan subtracts the partitions the rebuild already 
replaced.
   
   That commit also carries two related changes: the base-table query check is 
no longer IVM-only (it is `checkQueryUsable` and runs for every MV), and an MV 
whose query is no longer analyzable is invalidated with that as its reason — 
the generic record used to be written straight after it, on the same state, 
overwriting the detail.
   
   Verified with the unit tests (189/189, every new or inverted case also run 
against a build without its change) and the `mtmv_p0/ivm` regression directory. 
The two findings not taken keep their reasoning in the threads above.


-- 
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