David Smiley created SOLR-17287:
-----------------------------------
Summary: RESTORECORE should reset/clear the UpdateLog
Key: SOLR-17287
URL: https://issues.apache.org/jira/browse/SOLR-17287
Project: Solr
Issue Type: Bug
Security Level: Public (Default Security Level. Issues are Public)
Reporter: David Smiley
I think the semantics of a RESTORECORE (the core level API, not SolrCloud)
should be wholistic to the core and thus consider the UpdateLog. As such, I
think the UpdateLog should be left in an empty state and be ACTIVE. Recently,
in SOLR-16924, it was enhanced to call UpdateLog.applyBufferedUpdates with the
goal of transitioning from state BUFFERING to ACTIVE but that doesn't do
anything if it's not in a buffering state to begin with (it'll be BUFFERING in
SolrCloud always; isn't obvious).
To prove there is a problem, I modified TestRestoreCore.testSimpleRestore (a
good test!) to have a configured UpdateLog and I used RTG after the restore to
see if I could get a document that was added *after* the backup was performed.
I could. It doesn't matter if someone doesn't use RTG, it's just a means of
demonstrating the state is dirty; it should be empty. Thus if a node crashed
after a restore, the buffer would be replayed on startup for stuff added prior
to the RESTORECORE, which isn't what we want.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
