Todd Nine created USERGRID-261:
----------------------------------
Summary: Review the counter code in NodeShardApproximationImpl
Key: USERGRID-261
URL: https://issues.apache.org/jira/browse/USERGRID-261
Project: Usergrid
Issue Type: Improvement
Components: Stack
Reporter: Todd Nine
Review the code in NodeShardApproximationImpl. The FlushWorker.call is taking
15% of our entire runtime. See the attached screenshot.
https://www.dropbox.com/s/v9w743uwzib5fej/Screen%20Shot%202014-12-02%20at%203.40.52%20PM.png?dl=0
Instead we should prototype changing the flush worker to work in the following
way.
1) Create a custom observable
2) As counters are incremented/decremented, publish them as an event
3) Subscribers to the event should subscribe on a buffer size or a timeout
4) The action that receives the buffered events should flush them immediately
to cassandra.
This would avoid the need to perform the current atomic long and incremental
counter classes, and would simplify the code. We should also verify this
removes some of the current performance issues in the system.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
