Mihaly Toth created SOLR-10904:
----------------------------------
Summary: Unnecessary waiting during failover in case of failed
core creation
Key: SOLR-10904
URL: https://issues.apache.org/jira/browse/SOLR-10904
Project: Solr
Issue Type: Bug
Security Level: Public (Default Security Level. Issues are Public)
Affects Versions: master (7.0)
Reporter: Mihaly Toth
Background failover thread checks for bad replicas. In case one is found it
tries to create it on another node. Then it waits for the new replica to show
up in the cluster state. It waits even if the core creation (initiated by
itself) fails.
This situation does not occur on the happy path of the failover cases because
the new node was marked as alive. But in case the cluster is in an instable
state, or user is restarting the new node, or overseer is overloaded this extra
wait will result in holding up this failover thread.
Proposed solution may be
# wait for the result of the core creation
# only if previous step is successful proceed to wait for cluster state change
In code:
{code}
try {
Future<Boolean> future = updateExecutor.submit(() ->
createSolrCore(collection, createUrl, dataDir, ulogDir, coreNodeName, coreName,
shardId));
future.get(30000L, TimeUnit.MILLISECONDS);
} catch (InterruptedException | ExecutionException | TimeoutException e) {
log.error("Error creating core", e);
return false;
} finally {
MDC.remove("OverseerAutoReplicaFailoverThread.createUrl");
}
{code}
In such case we could consider moving core creation into the failover thread
from the updateExecutor.
I can post a patch with these changes if the solution seems appropriate.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
