Caizhi Weng created FLINK-28605:
-----------------------------------
Summary: Throw exception intentionally when new snapshots are
committed during restore
Key: FLINK-28605
URL: https://issues.apache.org/jira/browse/FLINK-28605
Project: Flink
Issue Type: Improvement
Components: Table Store
Affects Versions: table-store-0.2.0
Reporter: Caizhi Weng
Fix For: table-store-0.2.0
Currently snapshots are committed in {{notifyCheckpointComplete}}. If the job
fails between a successful checkpoint and the call of
{{notifyCheckpointComplete}}, these snapshots will be committed after job
restarts.
However when the writer starts they also need to read from the latest snapshot
(to build the latest structure of LSM tree). These two steps may happen
concurrently and what the writers see may not be the latest snapshot.
To fix this problem, we can throw exception intentionally after new snapshots
are committed during restore. In this way the job will be forcefully restarted
and it is very likely that the writers can see the latest snapshot.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
