You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by "stack (JIRA)" <ji...@apache.org> on 2017/12/20 20:49:00 UTC
[jira] [Created] (HBASE-19569) Alleviate the bigbang hbase:meta
scan on MR startup by passing out target region location
stack created HBASE-19569:
-----------------------------
Summary: Alleviate the bigbang hbase:meta scan on MR startup by passing out target region location
Key: HBASE-19569
URL: https://issues.apache.org/jira/browse/HBASE-19569
Project: HBase
Issue Type: Task
Components: mapreduce
Reporter: stack
Assignee: stack
Task for myself. Idea is to see if I can pass out the TableSplit location across the cluster and have the MR table Tasks pick up the supplied location and make their first attempt at finding the region using it w/o bothering the hbase:meta. Would go to hbase:meta only if the location they had was failing.
This is general client caching issue -- populating it and getting metrics on how often we go to meta. Apparently we are profligate with our hbase:meta lookups. On a big cluster, it tends to buckle the server carrying hbase;meta. This would give some relief.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)