[ https://issues.apache.org/jira/browse/YARN-1514?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14027224#comment-14027224 ]
Tsuyoshi OZAWA commented on YARN-1514: -------------------------------------- Updated a patch. This patch includes following changes: * Adding dependency on zookeeper-tests' jar. * Made number of applications, number of application attempts , and ZK connection configurable. * Added YarnTestDriver to launch this perf tool from bin/hadoop command. Example of running jar is as follows: {code} bin/hadoop jar ../hadoop-common/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/target/hadoop-yarn-server-resourcemanager-3.0.0-SNAPSHOT-tests.jar TestZKRMStateStorePerf -appSize 100 -appattemptsize 100 -hostPort localhost:2181 {code} One problem is too much logging. If we run the command, the following logs are generated: {quote} ... 14/06/10 23:25:55 INFO recovery.RMStateStore: Storing info for app: application_1352994193343_0099 14/06/10 23:26:31 INFO recovery.TestZKRMStateStore: ZKRMStateStore takes 2791 msec to loadState. .... ZKRMStateStore takes 2791 msec to loadState. ... 14/06/10 23:26:31 INFO recovery.ZKRMStateStore: Watcher event type: NodeDeleted with state:SyncConnected for path:/Test/ZKRMStateRoot/RMAppRoot/application_1352994193343_0000 for Service org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore in state org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore: STARTED {quote} This is because of ZKRMStateStore generates log via {{LOG.info()}}. [~kkambatl], what do you think about this? Should we improve it on this JIRA? > Utility to benchmark ZKRMStateStore#loadState for ResourceManager-HA > -------------------------------------------------------------------- > > Key: YARN-1514 > URL: https://issues.apache.org/jira/browse/YARN-1514 > Project: Hadoop YARN > Issue Type: Sub-task > Reporter: Tsuyoshi OZAWA > Assignee: Tsuyoshi OZAWA > Fix For: 2.5.0 > > Attachments: YARN-1514.1.patch, YARN-1514.wip-2.patch, > YARN-1514.wip.patch > > > ZKRMStateStore is very sensitive to ZNode-related operations as discussed in > YARN-1307, YARN-1378 and so on. Especially, ZKRMStateStore#loadState is > called when RM-HA cluster does failover. Therefore, its execution time > impacts failover time of RM-HA. > We need utility to benchmark time execution time of ZKRMStateStore#loadStore > as development tool. -- This message was sent by Atlassian JIRA (v6.2#6252)