1157408 - fhr v4 previousSubsessionId recording broken

Status: RESOLVED FIXED

(Toolkit :: Telemetry, defect)

Description

[brendan c]

In the assumed most common case of an FF client with no profile copying and/or branching, the graph with--
* nodes representing sessions, and
* edges representing a pointer from one session to the previousSubsessionId recorded for that session
-- should be a simple chain of nodes, each of which points to exactly one previousSubsessionId, and each of which is pointed at by exactly one subsequent subsession (except, obviously, for the first and last session). The graph should have no branching, there should not be multiple independent subgraphs, and there should be no isolated nodes.

In practice, based on the "bug1149666.20150413" data extract created by Mark, we see all three of the pathological topologies, and almost no records have the correct topology (not a single one in 50 has the correct topology for longer than 4 subsessions)

This noteboook shows the subsession graph for 50 randomly selected clientIds. Note the abundance of star like topologies.

http://nbviewer.ipython.org/urls/gist.githubusercontent.com/bcolloran/26a515c93ad6b8245bce/raw/gistfile1.txt

This seems likely to be closely related to https://bugzilla.mozilla.org/show_bug.cgi?id=1157359

Comment 1

[brendan c]

(In reply to brendan c from comment #0)
> In the assumed most common case of an FF client with no profile copying
> and/or branching, the graph with--
> * nodes representing sessions, and
> * edges representing a pointer from one session to the previousSubsessionId
> recorded for that session

oops, the above should have been
* nodes representing SUBsessions, and
* edges representing a pointer from one subsession to the previousSubsessionId recorded for that subsession

Comment 2

[Georg Fritzsche [:gfritzsche]]

Bug 1157359 found that the persisting the previousSubsessionId used was broken (and missing basic test-coverage).
That will part will be adressed there.

Comment 3

[Georg Fritzsche [:gfritzsche]]

Brendan, should we also have previousSessionId?
There might always be cases where we miss one subsession (bugs, eviction, ...) and can't construct a chain.
Or would the clientId + ordering in time be enough?

Comment 4

[brendan c]

Sure, that could be useful as well.

Comment 5

[Georg Fritzsche [:gfritzsche]]

Ok, bug 1157359 should fix the issues seen here (or at least the majority).
We need to recheck with the data from this.

Comment 6

[Georg Fritzsche [:gfritzsche]]

Brendan, the situation definitely improved here from the other bugs.
Can we close this bug out over the newer ones files or are there specific issues here not covered by those?

Comment 7

[brendan c]

Sure, if there are issues we can file a new bug or reopen.