[
https://issues.apache.org/jira/browse/HBASE-26487?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17461784#comment-17461784
]
Duo Zhang commented on HBASE-26487:
-----------------------------------
Update for IntegrationTestRegionReplicaReplication.
I use this command
{noformat}
./bin/hbase org.apache.hadoop.hbase.IntegrationTestRegionReplicaReplication \
-DIntegrationTestRegionReplicaReplication.num_keys_per_server=10000 \
-Dhbase.IntegrationTestRegionReplicaReplication.runtime=3600000 \
-DIntegrationTestRegionReplicaReplication.read_delay_ms=60000 \
-DIntegrationTestRegionReplicaReplication.region_replication=2 \
-DIntegrationTestRegionReplicaReplication.region_replica_id=1 \
-DIntegrationTestRegionReplicaReplication.num_read_threads=10 \
-DIntegrationTestRegionReplicaReplication.num_write_threads=10 \
-m slowDeterministic \
&>test.log </dev/null &
{noformat}
Sadly that both master and HBASE-26233 can not pass it. There are basiclly two
problems:
1. If we set IntegrationTestRegionReplicaReplication.read_delay_ms to 5000 by
default, it will be easy to find inconsistent rows, as slowDeterministic will
aggressively move regions and kill region servers, which could introduce big
replication delay.
2. The secondary replica will remain in disable read state for a very long
time. Checked the jstack of a region server, we hang in the rpc call for
trigger a flush on primary region in RegionReplicaFlushHandler. Will file
follow on issues to address this.
Anyway, at least we are not worse than master branch :)
> Run some tests to verify the new region replication framework
> -------------------------------------------------------------
>
> Key: HBASE-26487
> URL: https://issues.apache.org/jira/browse/HBASE-26487
> Project: HBase
> Issue Type: Sub-task
> Components: integration tests, test
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
>
> Make sure there is no big bugs before merging back.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
