[ https://issues.apache.org/jira/browse/CASSANDRA-13562?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16074128#comment-16074128 ]
ZhaoYang edited comment on CASSANDRA-13562 at 7/5/17 12:32 AM: --------------------------------------------------------------- you may consider checking `phi_convict_threshold` in cassandra.yaml and turning on the logging for phi value in gossip. then you will get a better idea why c* node thinks another as down node. was (Author: jasonstack): you may consider checking `phi_convict_threshold` in cassandra.yaml and turning the logging for phi value in gossip. then you will get a better idea why c* node thinks another as down node. > nodes in cluster gets into split-brain mode > ------------------------------------------- > > Key: CASSANDRA-13562 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13562 > Project: Cassandra > Issue Type: Bug > Components: Core > Reporter: Jaydeepkumar Chovatia > Fix For: 3.0.x > > > We have seen nodes in Cassandra (3.0.11) ring gets into split-brain somehow. > We don't know exact reproducible steps but here is our observation: > Let's assume we have 5 node cluster n1,n2,n3,n4,n5. In this bug when do > nodetool status on each node then each one has different view of DN node > e.g. > n1 sees n3 as DN and other nodes are UN > n3 sees n4 as DN and other nodes are UN > n4 sees n5 as DN and other nodes are UN and so on... > One thing we have observed is once n/w link is broken and restored then > sometimes nodes go into this split-brain mode but we still don't have exact > reproducible steps. > Please let us know if I am missing anything specific here. -- This message was sent by Atlassian JIRA (v6.4.14#64029) --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org