Nest-Binding: Suppress occasionally failing update echo cancellations. #2930
Conversation
|
I've been running the patch for ~4hs now, and there's no trace of issue in the logs. Prior to that, I'd get: every ~10-15 minutes. In my environment, I only ever use sendCommand in my rules/scripts, and these are infrequently done (during Alarm arming/disarming events for the house). As a result, I'm not really testing the update codepath for anything other than suppressing echos. |
|
Thanks @mrguessed! Please check again tomorrow if you can, just to make sure this resolves it for good. Google Group discussion here. |
|
Your patch is correct, and avoids the thread-safety issue... in openHAB Core. As it turns it out, the issue is in DateTimeType.java, specifically the use of a shared/static It's an oldie, but a goodie, I remember running into that problem in the late 90's:
Your new changes avoid the issue since you're no longer using |
|
Criminy, what a nuisance to have a non-threadsafe toString() method. My patch should also run faster and create less garbage, so I'll merge it. Thanks very much, Mark. |
Nest-Binding: Suppress occasionally failing update echo cancellations.
|
Yeah, it's not limited to it's I suspect if there's enough "DateTime" concurrency going on in an openHAB deployment that there will be little corruptions going on all over. eg. multiple Bindings parsing, or logging, I think it can be fixed by changing the const's into methods, of the same name, and having them return a new object for each call. Unfortunately, the const declarations in this class were made |
This code change attempts to suppress rare cases where an attempt to cancel an update echo fails. An update echo is when the
execute()method polls the Nest API and sends updates on the event bus for bound items. The binding'sinternalReceiveUpdatemethod, which is intended to handle updates from all other sources, consults a dynamic list of events to ignore and upon finding and removing a matching event, does not forward the update back to the Nest API. In rare cases and as yet unexplained, the echo cancellation fails and the binding sends the update back to the Nest API.This code change employs two approaches to suppress this behaviour:
itemNameand string version of the newStateas previously, it placesUpdateobjects on the list, which are comprised of theitemNameand thestate. TheUpdateclass'sequalsmethod is used by the list'sremovemethod to find and remove the first matching update. PreviouslyString.equalswas used, but appeared to be occasionally failing for as yet unknown reasons.Note: This PR only attempts to work around the issue of unsuppressed echoes. My working theory, lacking other insight, is some kind of JVM bug. Please test this PR extensively before I merge it (and I will continue to test it), but hopefully some kind of fix or workaround can make it into openHAB 1.7.1.