On 1/19/22 3:31 AM, Modassar Ather wrote:
It creates the collection and the cores but as soon as the server is
rebooted and Solr nodes are restarted the cores get deleted.
The state.json file of the collection shows all the cores entry but the
state of the cores as down.
When Solr in cloud mode connects to a zookeeper that is empty, it will
delete all cores that it has locally, because none of them will be in
the ZK database. I filed a bug about this, but it is still open. It is
possible that when that node reboots, it is not connecting to the same
identical ZK host definition that it was connected to before the reboot,
so it is not seeing the ZK database that it was using before. If that's
happening, I do not know enough about your environment to guess why.
https://issues.apache.org/jira/browse/SOLR-13396
Or maybe on reboot, the Solr instance is getting a different IP address
than it had before the reboot, so when it registers itself in the ZK
database, it no longer sees its own cores in the collection's
state.json, and deletes them. That would probably be more likely if you
have Solr deployed in containers rather than physical machines. If that
is happening, the solution would be to override the host definition that
SolrCloud is using so it's a name rather than an IP address.
Thanks,
Shawn
