[
https://issues.apache.org/jira/browse/HBASE-2468?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mingjie Lai updated HBASE-2468:
-------------------------------
Attachment: HBASE-2468-trunk.patch
Basic ideas:
1. At o.a.h.h.client.HConnectionManager.locateRegionInMeta(), prefetch certain
number of table regions in global region loc cache when scanning META, instead
of only caching the desired queried table+row region.
2. For a fresh table load, *all* region locations of the table will be
prefetched/cached. While after initialization, each miss could fetch ahead some
number of rows from META.
Patched files:
- o.a.h.h.client.HConnectionManager: added a new method to perform META table
prewarm.
- o.a.h.h.client.MetaScanner: overloaded MetaScanner.metaScan, to support
scanning meta table within certain steps:
public static void metaScan(HBaseConfiguration configuration,
MetaScannerVisitor visitor, byte[] tableName, int rowLimit)
- o.a.h.h.client.TestFromClientSide: added a test case for region cache
prewarm.
Potential issues:
- o.a.h.h.client.HTable constructor triggers the cache prewarm right now, by
calling HConnection.locateRegion(). I was suggested to defer calling
locateRegion() from HTable constructor, but some of existing unit test cases
failed because they implicitly rely on the constructor to do something, i.e.,
waiting for a table to be fully created after a table creation.
- For 2, ``each miss could fetch ahead some number of rows'': right now, this
number is hard-coded as 10. Is it necessary to make it configurable?
> Improvements to prewarm META cache on clients
> ---------------------------------------------
>
> Key: HBASE-2468
> URL: https://issues.apache.org/jira/browse/HBASE-2468
> Project: Hadoop HBase
> Issue Type: Improvement
> Components: client
> Reporter: Todd Lipcon
> Assignee: Mingjie Lai
> Fix For: 0.21.0
>
> Attachments: HBASE-2468-trunk.patch
>
>
> A couple different use cases cause storms of reads to META during startup.
> For example, a large MR job will cause each map task to hit meta since it
> starts with an empty cache.
> A couple possible improvements have been proposed:
> - MR jobs could ship a copy of META for the table in the DistributedCache
> - Clients could prewarm cache by doing a large scan of all the meta for the
> table instead of random reads for each miss
> - Each miss could fetch ahead some number of rows in META
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
