[
https://issues.apache.org/jira/browse/HBASE-10572?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Enis Soztutar updated HBASE-10572:
----------------------------------
Attachment: hbase-10572_v1.patch
Here is a more complete patch. From the javadoc:
{code}
* An IntegrationTest for doing reads with a timeout, to a read-only table with
region
* replicas. ChaosMonkey is run which kills the region servers and master, but
ensures
* that meta region server is not killed, and at most 2 region servers are dead
at any point
* in time. The expected behavior is that all reads with stale mode true will
return
* before the timeout (5 sec by default). The test fails if the read requests
does not finish
* in time.
*
* <p> This test uses LoadTestTool to read and write the data from a single
client but
* multiple threads. The data is written first, then we allow the region
replicas to catch
* up. Then we start the reader threads doing get requests with stale mode
true. Chaos Monkey is
* started after some delay (20 sec by default) after the reader threads are
started so that
* there is enough time to fully cache meta.
{code}
I am using this to test HBASE-10355, and the server side patches.
> Create an IntegrationTest for region replicas
> ---------------------------------------------
>
> Key: HBASE-10572
> URL: https://issues.apache.org/jira/browse/HBASE-10572
> Project: HBase
> Issue Type: Sub-task
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Fix For: hbase-10070
>
> Attachments: hbase-10572_v0.patch, hbase-10572_v1.patch
>
>
> We should add an integration test for testing reads from replicas with chaos
> monkey.
> I am thinking of a test which:
> - Creates a table with region replication
> - Writes some data to a table with many regions per server. Ensures that the
> replicas sees the data.
> - Does time bounded gets (5 sec, etc) with timeline consistency. Failing to
> complete the reads with a smaller latency will fail the test
> - Runs Chaos Monkey to kill the region servers, master and runs balance. We
> do not want other CM actions (flush, snapshot, etc)
--
This message was sent by Atlassian JIRA
(v6.2#6252)
