[
https://issues.apache.org/jira/browse/CASSANDRA-6620?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13882330#comment-13882330
]
Edward Capriolo commented on CASSANDRA-6620:
--------------------------------------------
Dtests may not be revealing the problem for two reasons.
1) You doing 100 operations not 10000
for n in xrange(100, 200):
insert_c1c2(cursor1, n, "QUORUM")
query_c1c2(cursor2, n, "QUORUM")
2) Every test in the suite has wait_other_notice=True
# shutdown another node and test we get unavailabe exception
node2.stop(wait_other_notice=True)
My test is not waiting for others to notice.
> Collateral damage from killing a node
> -------------------------------------
>
> Key: CASSANDRA-6620
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6620
> Project: Cassandra
> Issue Type: Bug
> Reporter: Edward Capriolo
> Attachments: nodelogs.txt
>
>
> I have designed a new scenario with farsandra:
> 3 nodes with Replication factor = 2 a Counter column family. I perform 10,000
> inserts to node 1. I kill off node 2, do 10000 more inserts. restart node 2.
> Sometimes I made it completely though this test. However sometimes I do not.
> I have seen the client throw time out exceptions. it seems like the death of
> node 2 greatly upsets node 1 and it times out a request. Since the default is
> ready 1 should this be happening?
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
