Ivan Bessonov created IGNITE-20754:
--------------------------------------
Summary: Fix log prefix truncation in logit storage
Key: IGNITE-20754
URL: https://issues.apache.org/jira/browse/IGNITE-20754
Project: Ignite
Issue Type: Improvement
Reporter: Ivan Bessonov
{code:java}
Utils.runInThread(executor, () -> {
this.indexDB.truncatePrefix(firstIndexKept);
this.segmentLogDB.truncatePrefix(firstIndexKept);
this.confDB.truncatePrefix(firstIndexKept);
}); {code}
Currently, this code doesn't work if we truncate log after loading raft
snapshot from the leader. The problem is in the gap in index file, that
prevents new log entries to be written.
Possible solution - differentiate between "safe" log truncation and "unsafe".
Here I mean that if index N is present in the log, then we can truncate log to
it.
Otherwise, truncation is equivalent to entire log invalidation, and it should
be performed differently. Either synchronously, or somehow guaranteeing that
same files won't be used for writes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
