Aled Sage created BROOKLYN-322:
----------------------------------
Summary: On rebind, entity temporarily (incorrectly) reports
serviceUp=false
Key: BROOKLYN-322
URL: https://issues.apache.org/jira/browse/BROOKLYN-322
Project: Brooklyn
Issue Type: Bug
Reporter: Aled Sage
With Brooklyn 0.10.0-SNAPSHOT (but also with older versions, I presume)...
When rebinding to many entities, some software processes temporarily show
serviceUp=false before going back to showing the correct serviceUp=true.
The problem is with the polling of their feeds. It starts polling immediately,
rather than waiting for the entity to be managed. This can cause some feeds to
throw an exception (e.g. when {{SoftwareProcessImpl.connectServiceUpIsRunning}}
calls {{driver.isRunning()}}, and that tries to execute an ssh command). The
exception causes the feed to set it to whatever value the {{onException}} tells
it to use.
It is more likely to happen when there are many entities, because the entities
will be unmanaged for longer.
The solution is to not poll when the entity is not managed.
This is a simple fix in ((Poller}}.
It might cause problems for those trying to use a Poller to poll just once: it
might skip that single poll because the entity is unmanaged. I think we can
ignore that. I don't know of any use-cases doing that. Also, I suspect that it
might be a good thing! On rebind we don't necessarily want to do that single
poll again, and when the feed is first registered we'd hope that it is done
when the entity is managed (if done in {{SoftwareProcess.connectSensors()}},
but not if done in the entity's init method).
We could revisit this in the future, so that if the entity is unmanaged (but e
think it will soon be managed), then we reschedule it for when the entity is
actually managed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
