[
https://issues.apache.org/jira/browse/APEXCORE-380?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15198496#comment-15198496
]
ASF GitHub Bot commented on APEXCORE-380:
-----------------------------------------
Github user PramodSSImmaneni commented on a diff in the pull request:
https://github.com/apache/incubator-apex-core/pull/271#discussion_r56440964
--- Diff:
engine/src/main/java/com/datatorrent/stram/engine/GenericNode.java ---
@@ -608,9 +610,9 @@ else if (tracker.ports[trackerIndex] == null) {
if (need2sleep) {
if (handleIdleTime && insideWindow) {
((IdleTimeHandler) operator).handleIdleTime();
- }
- else {
+ } else {
Thread.sleep(spinMillis);
+ spinMillis = Math.min(maxSpinMillis, spinMillis + 1);
--- End diff --
I had looked at this before in a difference scenario. Can we do a better
heuristic. We can have another timeout setting (which is less than streaming
window size) where we continue to do busy way till that timeout and if there is
no data till then, then initiate the stepwise sleep.
> Idle time sleep time should increase from 0 to a configurable max value
> -----------------------------------------------------------------------
>
> Key: APEXCORE-380
> URL: https://issues.apache.org/jira/browse/APEXCORE-380
> Project: Apache Apex Core
> Issue Type: Improvement
> Affects Versions: 3.4.0
> Reporter: Vlad Rozov
> Assignee: Vlad Rozov
> Fix For: 3.4.0
>
>
> When all active input ports of an operator are empty, the operator either
> calls idle time handler or goes into a wait. In the later case, the wait time
> should start with 0 and increase to the configured/default idle spin time to
> handle cases when there is trickle and pour stream of tuples.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
