[
https://issues.apache.org/jira/browse/BROOKLYN-356?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15548174#comment-15548174
]
ASF GitHub Bot commented on BROOKLYN-356:
-----------------------------------------
Github user aledsage commented on a diff in the pull request:
https://github.com/apache/brooklyn-server/pull/365#discussion_r81932394
--- Diff:
core/src/main/java/org/apache/brooklyn/util/core/task/BasicExecutionManager.java
---
@@ -757,9 +758,23 @@ protected void internalBeforeStart(Map<?,?> flags,
Task<?> task) {
PerThreadCurrentTaskHolder.perThreadCurrentTask.set(task);
((TaskInternal<?>)task).setStartTimeUtc(System.currentTimeMillis());
}
+
+ jitterThreadStart(task);
+
invokeCallback(flags.get("newTaskStartCallback"), task);
}
+ private void jitterThreadStart(Task<?> task) {
--- End diff --
Great idea!
I wonder if we should make it non-static so it can be set by a test, and
unset in its `tearDown` method. But let's go with static. For now, we can use
it by setting the system property when doing `mvn clean install`.
Once we've fixed (most of) the failures that it exposes, then we can think
about enabling it for some tests in jenkins - for regression testing purposes.
But we want jenkins to be deterministic!
> The sensor Transformer enricher fails to resolve its targetValue
> indeterministically.
> -------------------------------------------------------------------------------------
>
> Key: BROOKLYN-356
> URL: https://issues.apache.org/jira/browse/BROOKLYN-356
> Project: Brooklyn
> Issue Type: Bug
> Reporter: Svetoslav Neykov
>
> The sensor {{Transformer}} enricher fails to resolve its {{targetValue}}
> indeterministically, especially so on loaded systems. This is especially
> problematic if sourceSensor/triggerSensors change infrequently or do not
> change ever (when using default values).
> Bueprints using the {{Transformer}} behave perfectly fine during development
> and on test setups, but will fail on production machines (i.e. under load).
> The code [doing the value
> resolving|https://github.com/apache/brooklyn-server/blob/b59e7463a9b337c2d0e7931cd420d5bac68d8549/core/src/main/java/org/apache/brooklyn/enricher/stock/Transformer.java#L90-L94]
> tries to do so in a non-blocking fashion by spinning a thread to try to
> resolve and cancelling it after a short while without guarantees it ever got
> scheduled to run. It's more likely to fail when nesting DSLs, for example
> nesting several levels of {{$brooklyn:formatString}} and ending with a
> {{$brooklyn:attributeWhenRready}}. It's a common pattern in moderately
> complex blueprints. It needs to schedule a thread for each nesting level thus
> maximizing the chance that the value will not be resolved in the allotted
> time even if resolvabe. This is especially bad for sensors which don't get
> updated, for example {{PortAttributeSensorAndConfigKey} set only when
> initializing the entity or any config values which are always resolvable.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
