Jason Gerlowski created SOLR-17965:
--------------------------------------
Summary: Reload-collection shouldn't ignore "recovering" replicas
Key: SOLR-17965
URL: https://issues.apache.org/jira/browse/SOLR-17965
Project: Solr
Issue Type: Bug
Components: SolrCloud
Affects Versions: 9.9, main (10.0)
Reporter: Jason Gerlowski
I was surprised to discover recently that Solr's "reload-collection" operation
only triggers core-reloads on cores that are currently "active".
Maybe this was an intentional decision at the time (need to do some
history-spelunking and confirm), but in practice it produces some pretty
undesirable outcomes.
For instance many users run "reload-collection" operations after a configset
update, as a way to make all cores aware of the new configset. If this is done
and a core happens to be in recovery, it won't be reloaded and will (quietly)
still use the older configset version. If the older configset is missing a
definition for a field that starts appearing in docs after the "reload", the
out-of-sync core will find itself continually "down" or "recovering" as it
struggles to process docs forwarded by the leader.
A naive fix here would be to have reload-collection impact all cores,
regardless of their current status. If that ends up not being possible, at a
minimum the reload-collection API should more clearly signal "partial success"
in cases where it could only reload a subset of a collection's replicas.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
