Roberto Agostino Vitillo (:rvitillo) Reporter
	Description • 9 years ago

I spotted some sessions that seem to have missing and/or duplicated fragment numberings, e.g.:

list of subsessionCounter for fragments of session X of client Y: [1, 1, 1, 1]
list of subsessionCounter for fragments of session U of client K: [1, 2, 4, 5]

From all the sessions that have more than one fragment, about 15% have an incorrect ordering.

See http://nbviewer.ipython.org/gist/vitillo/9645b5f6849bd2051ea8.

	Roberto Agostino Vitillo (:rvitillo) Reporter
	Comment 1 • 9 years ago

George suggested to limit the data to build ids after the 5th of April to avoid some known and fixed bugs.

	Georg Fritzsche [:gfritzsche]
	Comment 2 • 9 years ago

Alessio is currently investigating this.

	Roberto Agostino Vitillo (:rvitillo) Reporter
	Comment 3 • 9 years ago

I have updated the notebook. The issue persists with submission from newer build-ids. 

As you can see in the example at the bottom, it seems that some fragments with the same subsessionCounter are not perfect dupes but just the same fragment taken at different times; once with reason="shutdown" and the other with reason="aborted-session".

	Georg Fritzsche [:gfritzsche]
	Comment 4 • 9 years ago

... i meant to assign bug 1157359.

	Georg Fritzsche [:gfritzsche]
	Comment 5 • 9 years ago

(In reply to Roberto Agostino Vitillo (:rvitillo) from comment #3)
> I have updated the notebook. The issue persists with submission from newer
> build-ids. 
> 
> As you can see in the example at the bottom, it seems that some fragments
> with the same subsessionCounter are not perfect dupes but just the same
> fragment taken at different times; once with reason="shutdown" and the other
> with reason="aborted-session".

Thanks, that is really helpful.
It looks like we are (always or sometimes) failing to remove the "aborted-session" ping after saving the "shutdown" ping.

	Roberto Agostino Vitillo (:rvitillo) Reporter
	Comment 6 • 9 years ago

In trying to understand if and how many fragments we receive out of order, I noticed that some fragments are sent multiple times, e.g. one particular fragment with reason "shutdown" has been submitted once on the 23rd and twice on the 26th for a session that ended on the 23rd.

	Georg Fritzsche [:gfritzsche]
	Comment 7 • 9 years ago

Those are identical pings being resent repeatedly?

	Roberto Agostino Vitillo (:rvitillo) Reporter
	Comment 8 • 9 years ago

Exactly, their payloads match and they have identical docids.

	Georg Fritzsche [:gfritzsche]
	Comment 9 • 9 years ago

Nightly 2015-05-06 has bug 1157359 which affects this, we should recheck from that build on.

	Georg Fritzsche [:gfritzsche]
	Comment 10 • 9 years ago

From bug 1157359 etc. we seem to be much better now - Roberto, can we close this bug over the other ones open or do you want to recheck here?

	Roberto Agostino Vitillo (:rvitillo) Reporter
	Comment 11 • 9 years ago

Let's close this.