I'm currently running Ignite as a Hadoop accelerator via IGFS backed by a
secondary HDFS store. The cluster currently functions as expected, but I'm
seeing some anomalies in the metrics reported by Ignite. In particular, on a
fresh cluster, when a node restarts (exits and rejoins), the igfs-meta TX
commit rate jumps up from ~0/sec to ~4/sec for all the nodes in the cluster.
This rate increases as additional nodes leave/join. From the looks of it, it
seems to be correlated to the TX rollback rate. Is there a reason why this
might be the case?

<http://apache-ignite-users.70518.x6.nabble.com/file/n13155/rollback.png> 

In the screenshot, I've restarted all the nodes in the cluster a little
after 14:30, and restarted one of the nodes around 16:30.



--
View this message in context: 
http://apache-ignite-users.70518.x6.nabble.com/igfs-meta-behavior-when-node-restarts-tp13155.html
Sent from the Apache Ignite Users mailing list archive at Nabble.com.

Reply via email to