DarvenDuan opened a new pull request, #36148:
URL: https://github.com/apache/doris/pull/36148
## Proposed changes
Issue Number: close #xxx
If a table with an existing partition of the same schema is restored, and
the version of those partitions is higher than the ones being restored, the
restored replicas' version in the FE metadata will not be updated to match the
restored partition's version. Although there will be no impact on read and
write operations, these replicas might not be dropped during repair or
rebalancing processes. This is because the TabletScheduler typically avoids
dropping replicas with the highest version among all replicas within a tablet,
potentially leading to the existence of redundant or abnormal replicas within a
tablet:
```
+-----------+-----------+---------+-------------------+------------------+---------------+------------+---------------+----------------+----------+--------+-------+------------+--------------+----------------------+------------------------------------------------+--------------------------------------------------------------+-------------------+----------------+-----------+
| ReplicaId | BackendId | Version | LstSuccessVersion | LstFailedVersion |
LstFailedTime | SchemaHash | LocalDataSize | RemoteDataSize | RowCount | State
| IsBad | IsUserDrop | VersionCount | PathHash | MetaUrl
| CompactionStatus
| CooldownReplicaId | CooldownMetaId | QueryHits |
+-----------+-----------+---------+-------------------+------------------+---------------+------------+---------------+----------------+----------+--------+-------+------------+--------------+----------------------+------------------------------------------------+--------------------------------------------------------------+-------------------+----------------+-----------+
| 16611 | 10148 | 441345 | 441345 | -1 |
NULL | 363051650 | 190695232 | 0 | 23551002 | NORMAL
| false | false | 8 | -8787066244400026431 |
http://0.0.0.0:0/api/meta/header/16608 |
http://0.0.0.0:0/api/compaction/show?tablet_id=16608 | -1
| | 0 |
| 16610 | 17020 | 441343 | 441343 | -1 |
NULL | 363051650 | 190695232 | 0 | 23551002 | NORMAL
| false | false | 8 | -7255479345457496615 |
http://xx.xx.xx.xx:8040/api/meta/header/16608 |
http://xx.xx.xx.xx:8040/api/compaction/show?tablet_id=16608 | -1
| | 0 |
| 16611 | 17021 | 441343 | 441343 | -1 |
NULL | 363051650 | 190695232 | 0 | 23551002 | NORMAL
| false | false | 8 | -8787066244400026431 |
http://xx.xx.xx.xx:8040/api/meta/header/16608 |
http://xx.xx.xx.xx:8040/api/compaction/show?tablet_id=16608 | -1
| | 0 |
| 16609 | 17022 | 441343 | 441343 | -1 |
NULL | 363051650 | 190695232 | 0 | 23551002 | NORMAL
| false | false | 8 | 2726332276095749175 |
http://xx.xx.xx.xx:8040/api/meta/header/16608 |
http://xx.xx.xx.xx:8040/api/compaction/show?tablet_id=16608 | -1
| | 0 |
+-----------+-----------+---------+-------------------+------------------+---------------+------------+---------------+----------------+----------+--------+-------+------------+--------------+----------------------+------------------------------------------------+--------------------------------------------------------------+-------------------+----------------+-----------+
```
As we can see, tablet 16608 has four replicas, but replica 16611 is
redundant and ought to be removed as its corresponding backend, 10148, has been
dropped. However, replica 16611 remains due to its higher version number,
441345, which is greater than the versions of the other replicas, 441343.
Consequently, it cannot be automatically deleted, and the FE received the
following message:
2024-06-07 16:13:58,568 INFO (tablet scheduler|40)
[TabletScheduler.deleteReplicaInternal():1090] can not delete only one replica,
tabletId =16608 replicaId =16611
2024-06-07 16:13:58,568 INFO (tablet scheduler|40)
[TabletScheduler.removeTabletCtx():1594] remove the tablet tablet id:16608,
status: REDUNDANT, state: PENDING, type: REPAIR, priority: VERY_HIGH, tablet
size: 0, visible version:441345, committed version:441345. err: the only one
latest replia can not be dropped, tabletId =16608, replicaId =16611. because:
the only one latest replia can not be dropped, tabletId =16608, replicaId =16611
<!--Describe your changes.-->
optimize:
update replica version after restore
--
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]
