[PHOENIX-3486] RoundRobinResultIterator doesn't work correctly because of setting Scan's cache size inappropriately in PhoenixInputForamt

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 4.10.0
• Labels: HivePhoenix
• Assignee: Jeongdae Kim
• Reporter: Jeongdae Kim
• Created: 16/Nov/16 01:41
• Updated: 02/Mar/17 04:17
• Resolved: 20/Feb/17 01:40

Description

RoundRobinResultIterator uses "hbase.client.scanner.caching" to fill caches in parallel for all scans, but by setting Scan.setCaching() in PhoenixInputForrmat(phoenix-hive), RoundRobinResultIterator doesn't work correctly, because if Scan have cache size by setCaching(), HBase uses cache size from Scan.getCaching() to fill cache, not "hbase.client.scanner.caching", and RoundRobinResultIterator scans the table in parallel to fill caches every "hbase.client.scanner.caching", resulting in unintended parallel scan operation, this causes scan performance degradation.