Serban Teodorescu created CASSANDRA-20558:
---------------------------------------------
Summary: CassandraDataLayer uses configuration list of IPs instead
of the full ring/datacenter
Key: CASSANDRA-20558
URL: https://issues.apache.org/jira/browse/CASSANDRA-20558
Project: Apache Cassandra
Issue Type: Bug
Components: Analytics Library
Reporter: Serban Teodorescu
Assignee: Serban Teodorescu
In
[https://github.com/apache/cassandra-analytics/blob/trunk/cassandra-analytics-core/src/main/java/org/apache/cassandra/spark/data/CassandraDataLayer.java#L418-L435]
instanceMap is built based on clusterConfig:
{code:java}
instanceMap =
clusterConfig.stream().collect(Collectors.toMap(SidecarInstance::hostname,
Function.identity()));
{code}
If the sidecarContactPoints has a list of all Cassandra IPs there will be no
impact. Probably it will still work if the existing ones can ensure quorum. But
if not, it will fail with the following error:
{code}
25/04/10 16:58:03 WARN TaskSetManager: Lost task 36.0 in stage 0.0 (TID 36)
(ip-10-218-174-202.ec2.internal executor 4):
org.apache.cassandra.spark.data.partitioner.NotEnoughReplicasException:
Required 2 replicas but only 0 responded
at
org.apache.cassandra.spark.data.partitioner.MultipleReplicas.openAll(MultipleReplicas.java:101)
{code}
I no longer have the executor error and stacktrace, but the instanceMap is used
in listInstance(); MultipleReplicas.openReplicaOrRetry is indirectly using this.
Snapshot creation at createSnapshot() is not using this, but the RingResponse
information:
{code:java}
ring.stream().filter(...)
{code}
So the snapshot is created on all nodes. On the other hand, clearSnapshot()
uses clusterConfig, so we end up with snapshots that are not deleted, so that
needs to be fixed as well.
I'm working on a fix, I'll add a PR this week.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
