661543 - Missing data for test runs done later

Status: RESOLVED INCOMPLETE

(Webtools Graveyard :: Graph Server, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

The following graph: http://graphs-new.mozilla.org/graph.html#tests=[[73,1,13]]&sel=1306339720701.0562,1306343141411.617&displayrange=7&datatype=running is missing the data for:

http://tbpl.mozilla.org/?rev=4ae5181b22f7

Which lies between:
acb4e51fa8a6 and 32b009addadd

http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=da44483257cf&tochange=acb4e51fa8a6

This test run was explicitly run after acb4e51fa8a6 and 32b009addadd

Comment 1

[Robert Helmer [:rhelmer]]

(In reply to comment #0)
> The following graph:
> http://graphs-new.mozilla.org/graph.html#tests=[[73,1,13]]&sel=1306339720701.
> 0562,1306343141411.617&displayrange=7&datatype=running is missing the data
> for:
> 
> http://tbpl.mozilla.org/?rev=4ae5181b22f7
> 
> Which lies between:
> acb4e51fa8a6 and 32b009addadd
> 
> http://hg.mozilla.org/mozilla-central/
> pushloghtml?fromchange=da44483257cf&tochange=acb4e51fa8a6
> 
> This test run was explicitly run after acb4e51fa8a6 and 32b009addadd

Found it:
http://graphs-new.mozilla.org/graph.html#tests=[[73,1,13]]&sel=1306333171029,1306339890092&displayrange=365&datatype=running

Here's the problem. The x-axis is ordered by the timestamp that the graphserver has for each checkin (which I believe is the time that the report came in, not sure if it's set by the server or provided by talos, will check).

Looking at the timestamps in the graphserver JSON for these revisions you can see that they are out of order:

"da44483257cf": 20110525045325
"4ae5181b22f7": 20110525072732
"32b009addadd": 20110525064816

Which makes sense since they were run later as you say in the summary.

Bug 659299 might be a way to deal with this, but maybe the default should be the time that checkins happened rather than the time the test was run...

Comment 2

[Robert Helmer [:rhelmer]]

The data submitted by talos to graphserver (via HTTP POST) is:

[
 "machine_name",
 "test_name",
 "branch_name",
 "ref_changeset",
 "ref_build_id",
 "date_run"
]

So I think we'd need to change this so the time of the push was reported as well, since it'd be far too slow to query hgweb for this info from the UI.