[HBASE-12835] HBASE-12422 changed new HTable(Configuration) to not use managed Connections anymore

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.0.0, 1.1.0
• Hadoop Flags: Reviewed
• Assignee: Enis Soztutar
• Reporter: Enis Soztutar
• Created: 09/Jan/15 22:27
• Updated: 21/Feb/15 23:49
• Resolved: 13/Jan/15 02:53

Description

While debugging something else, I've noticed that HBASE-12422 changed it so that new HTable(Configuration, TableName) is not using managed connections anymore.

I thought the plan was to keep new HTable(Conf, TableName) to still use managed connections, but allow the new APIs to have un-managed connections. With HBASE-12422, there is a big compatibility change for used of the HTable interface that suddenly they will be launching connections per HTable instance which is a behavioral change and expensive (with all the meta cache, zk threads, thread pool, etc). I fear that this will catch unsuspecting users by surprise, that is why I think we should revert back to the deprecated behavior for 1.0. For master we do not need to make the change, but document it.

sduskis, [e-mail], ndimiduk what do you guys think?

Issue Links

• breaks: HBASE-13027 mapreduce.TableInputFormatBase should create its own Connection if needed
• is related to: HBASE-12833 [shell] table.rb leaks connections
• is related to: HBASE-12876 TestScannerWithBulkload should pass Table with unmanaged connection to LoadIncrementalHFiles#doBulkLoad()