[
https://issues.apache.org/jira/browse/HBASE-12955?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14301826#comment-14301826
]
Enis Soztutar commented on HBASE-12955:
---------------------------------------
We have done that monkey in HBASE-10572 as a part of the IT for region
replicas. The test IntegrationTestTimeBoundedRequestsWithRegionReplicas kills
the servers, but expects the requests with backup RPCs (going to replicas) to
finish under some limited time (5sec, etc). Meta for this test is a single
point of failure, and will cause the test to fail (unless HBASE-11574 is used
with the test).
The serverKilling monkey is still useful I think which does not to other
actions (like snapshot, etc). Maybe we can rename serverKilling monkey to
serverKillingWithoutMeta and change the serverKilling monkey to kill meta.
> Make ITBLL less of a wuss
> -------------------------
>
> Key: HBASE-12955
> URL: https://issues.apache.org/jira/browse/HBASE-12955
> Project: HBase
> Issue Type: Task
> Reporter: stack
> Assignee: stack
>
> ITBLL avoids rolling restart and kill of the server hosting meta by default.
> {code}
> Action[] actions1 = new Action[] {
> - new RestartRandomRsExceptMetaAction(60000),
> + new RestartRandomRsAction(60000),
> new RestartActiveMasterAction(5000),
> - new RollingBatchRestartRsExceptMetaAction(5000, 1.0f, 2), //only
> allow 2 servers to be dead
> + new RollingBatchRestartRsAction(5000, 1.0f, 2), //only allow 2
> servers to be dead
> new ForceBalancerAction()
> {code}
> Enable killing of meta.
> Running tests over the w/e against 1.0, all seems to work w/ this enabled
> (splits were disabled).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
