github-actions[bot] commented on code in PR #69554:
URL: https://github.com/apache/doris/pull/69554#discussion_r4367374261
##########
fe/fe-core/src/main/java/org/apache/doris/planner/OlapTableSink.java:
##########
@@ -1246,13 +1246,13 @@ private TOlapTableLocationParam createLocation(long
dbId, OlapTable table) throw
tablet.getId(), baseTabletId,
partition.getId());
bePathsMap =
getChangelogColocatedReplicaBackendPathMap(baseTablet, tablet);
} else {
- bePathsMap =
tablet.getNormalReplicaBackendPathMap();
+ bePathsMap = tablet.getLoadReplicaBackendPathMap();
Review Comment:
[P1] Account for filtered replicas in write-failure handling
This initial-plan path also loses rows for existing partitions, beyond the
dynamic RPC instance in thread 4366121929. Use replication_num=3,
min_load_replica_num=1 and the legacy writer
(enable_memtable_on_sink_node=false), with one-bucket partitions A on
BEs{1,2,3} and B on{1,4,5}. Mark alive BEs2/3 disable_load: this call reduces A
to{1}, which passes the following required=1 check. If BE1's tablet-writer-open
RPC fails after planning, IndexChannel counts one failed node against
total3-required1=2; excluded BEs2/3 never consume that allowance. Repeated
write/close failures add the same node to a set, so the writer can return OK
with no A commit infos while B succeeds through4/5.
DatabaseTransactionMgr.checkCommitStatus() infers involved partitions solely
from successful commit infos and therefore checks B but never A. The load can
report success with every A row missing; previously alive BEs2/3 remained
destinations and could receive those rows.
Count filtered replicas in failure handling, or otherwise require successful
participation of every touched partition before success. Checking only nonempty
or size>=loadRequiredReplicaNum is insufficient here. Add a mixed-partition
regression with a real selected-target failure; the new destination-list tests
do not cover it.
--
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]
