[
https://issues.apache.org/jira/browse/DERBY-3208?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12543017
]
Knut Anders Hatlen commented on DERBY-3208:
-------------------------------------------
Yes, at one extreme one could imagine an implementation which queued up all
calls to serviceNow() and guaranteed that performWork() would be called exactly
the same number of times as the subscriber had called serviceNow(). Or one
could imagine an implementation that queued up multiple requests from the same
subscriber, but occasionally lost some of them. Since the original description
was so vague, I thought it was best not to restrict the aspects that are
unimportant for the correct behaviour of the current code.
> Callers of DaemonService.serviceNow() assume behaviour not guaranteed by the
> interface
> --------------------------------------------------------------------------------------
>
> Key: DERBY-3208
> URL: https://issues.apache.org/jira/browse/DERBY-3208
> Project: Derby
> Issue Type: Bug
> Components: Services, Store
> Affects Versions: 10.4.0.0
> Reporter: Knut Anders Hatlen
> Priority: Minor
>
> There are two callers of DaemonService.serviceNow() in the code:
> - Clock.rotateClock()
> - LogToFile.flush()
> They implicitly assume that if they are not currently waiting to be serviced
> by the daemon service, a call to serviceNow() is guaranteed to result in
> their performWork() method being invoked at some point in the future. They
> maintain flags (Clock.cleanerRunning and LogToFile.checkpointDaemonCalled)
> which tell whether serviceNow() has been called, and if they are true,
> serviceNow() is not called again. When performWork() is called, they reset
> the flag, allowing others to call serviceNow() again.
> However, DaemonService.serviceNow() does not guarantee that performWork()
> always gets called in these cases. Its javadoc says the following:
> /**
> Service this subscription ASAP. Does not guarantee that the daemon
> will actually do anything about it.
> @param clientNumber the number that uniquely identify the client
> */
> If one of the calls to serviceNow() does not result in a call to
> performWork(), the flags which prevent new calls to serviceNow() are not
> reset. Since both Clock and LogToFile are subscribed to the daemon service in
> on-demand-only mode, performWork() is only called if the subscriber calls
> serviceNow(). That means there is no way to reset the flags, and
> Clock/LogToFile are never allowed to call serviceNow() again.
> Although BasicDaemon's implementation of the DaemonService interface does in
> fact give the necessary guarantees (without explicitly stating them, though)
> to make this protocol work, Clock and LogToFile should be coded against the
> specification of the interface rather than against the undocumented behaviour
> of an implementation of the interface.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
