Change balancer sloppyness from 0.1 to 0.3
------------------------------------------
Key: HBASE-2241
URL: https://issues.apache.org/jira/browse/HBASE-2241
Project: Hadoop HBase
Issue Type: Bug
Reporter: stack
Fix For: 0.20.4, 0.21.0
This is a quick workaround until we do a better balancer.
Taking a region off line when cluster is under load is bad news. Latency goes
up as we wait on regions to come up in new locations.
The load balancer should only cut in if the cluster is way out of alignment.
I'd argue that 10% deviance from the avg. is not good enough reason moving
regions around when cluster is under load.
Balancer already has a knack for cutting in at most inopportune moments: during
cluster startup, when new node is added to a small cluster, or moving a region
just after its been opened on a node. We'll need to do a better balancer but
meantime lets just allow that region loading can be sloppier, say 20% or 30%
off the average before balancer cuts in.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
