hazelcast - Declaring a Member dead when it stops responding to actual requests -
i experienced multiple outages our hazelcast setup (hazelcast embedded within payara) in situation, non-hazelcast workload of process caused gc overhead errors.
during time hazelcast responded heartbeats, however, did not respond actual requests fetched data. caused system stall, if backup of data available.
therefore want ask if there way kick off member out of cluster if stops responding other kinds of requests heartbeat.
you might want consider using hazelcast in "lite" mode on payara server instances. if turn on "lite" mode on instances might face ecessive gc, these instances should not affect hazelcast performance, nodes not store data hazelcast cluster.
however, necessary complement solution additional full instances, keep data. these additional instances should not run application of yours , function separate hazelcast node. section focused on throughput , performance in blog post describes scenarios how use "lite" payara cluster members (backed lite hazelcast members).
to summarize - if turn nodes "lite" , every "lite" node add additional "full" node runs itself, no deployed applications, separate memory used applications , hazelcast separate jvms. gc on jvm "lite" node , contains no data hazelcast cluster, should not affect hazelcast performance.
Comments
Post a Comment