Deadlock on Offline Shadow Update

[kc9zyz]

Preconditions:

1. Two Device shadows are associated (as in a gateway scenerio), but no update has been sent so the update/accepted and update/rejected topics are not yet subscribed to.

Process to Re-Create:

1. Remove connectivity from the library
2. Attempt to update one of the shadows, receive a subscribeTimoutException
3. Attempt to update the second shadow

Result

The call to update the second shadow never returns. The offending call is here:

def basicShadowSubscribe(self, srcShadowName, srcShadowAction, srcCallback):
        self._shadowSubUnsubOperationLock.acquire()
        currentShadowAction = _shadowAction(srcShadowName, srcShadowAction)
        if currentShadowAction.isDelta:
            self._mqttCoreHandler.subscribe(currentShadowAction.getTopicDelta(), 0, srcCallback)
        else:
            self._mqttCoreHandler.subscribe(currentShadowAction.getTopicAccept(), 0, srcCallback)
            self._mqttCoreHandler.subscribe(currentShadowAction.getTopicReject(), 0, srcCallback)
        time.sleep(2)
        self._shadowSubUnsubOperationLock.release()

On line 68 a lock is acquired. Notice that on lines 71, 73, and 74, a subscribeTimeoutException may be raised (as happened in the second step to recreate), jumping past the lock release call on line 76. Any cases of locks being acquired/released should be using context managers or exception handling to ensure that the lock is released if an operation fails.

[liuszeng]

Hi @kc9zyz ,

Thank you very much for pointing this out! We will incorporate the bug fix into the next release.

Thanks,
Liusu

[liuszeng]

Addressed in release v1.2.0.

[kc9zyz]

@liuszeng Thanks for the update.