huaxiangsun commented on a change in pull request #3052:
URL: https://github.com/apache/hbase/pull/3052#discussion_r595548965
##########
File path:
hbase-server/src/main/java/org/apache/hadoop/hbase/master/MasterMetaBootstrap.java
##########
@@ -61,25 +60,25 @@ void assignMetaReplicas()
throw new IllegalStateException("hbase:meta must be initialized first
before we can " +
"assign out its replicas");
}
- ServerName metaServername =
MetaTableLocator.getMetaRegionLocation(this.master.getZooKeeper());
+
for (int i = 1; i < numReplicas; i++) {
// Get current meta state for replica from zk.
- RegionState metaState =
MetaTableLocator.getMetaRegionState(master.getZooKeeper(), i);
RegionInfo hri = RegionReplicaUtil.getRegionInfoForReplica(
- RegionInfoBuilder.FIRST_META_REGIONINFO, i);
- LOG.debug(hri.getRegionNameAsString() + " replica region state from
zookeeper=" + metaState);
- if (metaServername.equals(metaState.getServerName())) {
- metaState = null;
- LOG.info(hri.getRegionNameAsString() +
- " old location is same as current hbase:meta location; setting
location as null...");
- }
Review comment:
I think it is a bad optimization. What it tries to do is to avoid
assigning meta replica to the same region server where primary region is
located. This part of old logic is untested as there is no unassign procedure
for the meta replica region. It is ok to keep the original assignment and let
balancer to move them apart.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
