<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
On 06/17/2010 06:48 PM, Justin Wood (Callek) wrote:
<blockquote cite="mid:4C1AD053.9080307@gmail.com" type="cite">
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
(Client.py)>>Pull latest comm-central and update to a safe
mozilla-central revision <<<br>
<br>
*especially* if we make this default, it will depend on
instrumentation a bit. What happens if --mozilla-repo is not m-c.
Do we test for that, not support it, etc. i.e. local clones,
alternate project branches etc.<br>
</blockquote>
<br>
We could certainly ask the data source if it knows anything about
the repo URL in use, and if it does not, just print out a message
like "We don't know anything about what revisions are good on this
branch, the build may not work." and go with the tip of the default
branch for that repo. That way additional branches could be added
in the future with only a change on the back-end (or who is
submitting data to the back-end).<br>
<br>
<br>
<blockquote cite="mid:4C1AD053.9080307@gmail.com" type="cite"> Then
there is the question of what happens if there is a network
failure between client.py and the service that hosts this setup.
Do we assume m-c tip, no update, etc.<br>
</blockquote>
<br>
I would expect MoMo infrastructure failures to be highly correlated
(and rare). So from a Tinderbox perspective, this should be
super-rare, but it seems like just sticking with the current rev is
the best bet where possible. For fresh pulls, pulling from the tip
of default with a warning seems reasonable.<br>
<br>
<br>
<blockquote cite="mid:4C1AD053.9080307@gmail.com" type="cite"> Last,
do we care about potential extension alternatives
(venkman/chatzilla) etc breakages/revs?<br>
</blockquote>
<br>
I'd say no as I don't think we ever suggest enabling any extension
other than DOM inspector and I don't think it has ever been a
trouble point.<br>
<br>
<br>
<blockquote cite="mid:4C1AD053.9080307@gmail.com" type="cite">
>>When a mozilla-central changeset has all green builds, the
changeset will be counted as "good".<<<br>
<br>
Builds, Builds+Tests, Builds+Starred tests, other?<br>
<br>
IOW what are we using to determine this "good" state exactly. A
green build but thousands of busted tests is not necessarily good.
Talos can have a MAJOR (>200% regression too) do we count talos
in our infra here, if so "how"?<br>
</blockquote>
<br>
Builds + Tests. Thunderbird doesn't have Talos-type metrics right
now, and most of the performance regressions on Firefox' Talos are
minor from the perspective of Thunderbird.<br>
<br>
Andrew<br>
</body>
</html>