Node reconnects to Zookeeper in event of GC

classic Classic list List threaded Threaded
4 messages Options
swattal swattal
Reply | Threaded
Open this post in threaded view
|

Node reconnects to Zookeeper in event of GC

Hello,

We are evaluating Apache Ignite as a caching solution and i was wondering if
the server/client nodes reconnect to the ZK after a major GC pause is over?
If the reconnect doesn't happen automatically then what is the recommended
way of exposing the node back to the ZK and the rest of the cluster?

Would really appreciate if i could any pointers in this regard.

Thanks,
Sumit



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
ezhuravlev ezhuravlev
Reply | Threaded
Open this post in threaded view
|

Re: Node reconnects to Zookeeper in event of GC

Hi,

If server node was dropped out of the cluster after GC pause which were longer than failureDetectionTImeout(https://apacheignite.readme.io/docs/tcpip-discovery#section-failure-detection-timeout), it will be stopped and won't try to reconnect to the cluster again till it will be restarted. You can configure SegmentaionPolicy for this purpose: https://ignite.apache.org/releases/latest/javadoc/org/apache/ignite/plugin/segmentation/SegmentationPolicy.html. Also, client will try to reconnect to the cluster automatically, even after lt was dropped from the cluster.

Evgenii

пт, 3 янв. 2020 г. в 12:43, swattal <[hidden email]>:
Hello,

We are evaluating Apache Ignite as a caching solution and i was wondering if
the server/client nodes reconnect to the ZK after a major GC pause is over?
If the reconnect doesn't happen automatically then what is the recommended
way of exposing the node back to the ZK and the rest of the cluster?

Would really appreciate if i could any pointers in this regard.

Thanks,
Sumit



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
swattal swattal
Reply | Threaded
Open this post in threaded view
|

Re: Node reconnects to Zookeeper in event of GC

Thanks for the response Evgenii.

If i choose to use NOOP SegmentationPolicy will this cause the Split brains
issue in the cluster? We are not using ignite.sh script to start the servers
but using the Java API to start the cluster. Is there a possibility to
re-register the server node back to the ZK using a watcher?

Thanks,
Sumit



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
ilya.kasnacheev ilya.kasnacheev
Reply | Threaded
Open this post in threaded view
|

Re: Node reconnects to Zookeeper in event of GC

Hello!

I don't think a node should re-register. Node should either observe failureDetectionTimeout (and corresponding Zk timeouts) or segment from the cluster and shutdown.

You can restart node later, manually or automatically.

Regards,
--
Ilya Kasnacheev


вт, 7 янв. 2020 г. в 03:16, swattal <[hidden email]>:
Thanks for the response Evgenii.

If i choose to use NOOP SegmentationPolicy will this cause the Split brains
issue in the cluster? We are not using ignite.sh script to start the servers
but using the Java API to start the cluster. Is there a possibility to
re-register the server node back to the ZK using a watcher?

Thanks,
Sumit



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/