Zhen Zhang created HELIX-540:
--------------------------------
Summary: Full-auto rebalancer not able to map all partitions
Key: HELIX-540
URL: https://issues.apache.org/jira/browse/HELIX-540
Project: Apache Helix
Issue Type: Bug
Reporter: Zhen Zhang
We discover a bug where full-auto rebalancer not be able to mapping all
partitions even when capacity is enough.
Assuming we have 2 partition and 2 node, state model is LeaderStandby and
replica is 1. For some reason, the current state mapping becomes:
{noformat}
partition_0:
node_0: LEADER
node_1: STANDBY
{noformat}
The extra STANDBY mapping prevents the full-auto rebalancer
(AutoRebalanceStrategy#computePartitionAssignment) to assign LEADER to
partition_1, so partition_1 will remain unmapped even though node capacity is
enough.
For workaround, we can disable the partition in STANDBY state on the node, thus
"frees" the quota and the unmapped partition will get assigned LEADER. We can
then re-enable the partition on the node.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
