<gchristensen>
what if build failures on x86_64-linux were always X's
<gchristensen>
("tier-1 platform")
<gchristensen>
timeouts gray blobs
<gchristensen>
"not supported" gray blobs
NinjaTrappeur has quit [Ping timeout: 252 seconds]
<MichaelRaskin>
Well, there is staging
NinjaTrappeur has joined #nixos-borg
<MichaelRaskin>
I would say if we want to maintain «red is bad», we cannot go beyond «base branch builds OK, but merging PR breaks the build», and on staging there will be impact-evaluation build requests that fail and are consiously deferred.
<gchristensen>
LnL: I'm going to be trying out the versioned messages shortly
<gchristensen>
(back soon. lunch.)
<LnL>
lunch?
<gchristensen>
lunch!
<gchristensen>
I'm at ZuriHac today
<gchristensen>
oh dear I need to add support for versioned BuildJobs I think
<LnL>
ah that explains the timezone error :)
<LnL>
versioning the jobs makes the queue much more complicated
orivej has joined #nixos-borg
jtojnar has joined #nixos-borg
<gchristensen>
oh yeah that is right
tilpner has joined #nixos-borg
<LnL>
unless you do something like splitting the queue per version, similar to the platforms
<LnL>
but that seems a bit awkward
<gchristensen>
I actually wanted this versioning thing support for build log messages, so maybe I should just make a enum for that message and not version the build job msg
<LnL>
if it's from the builder to the server you can version it
<LnL>
in the other direction you can only add optional data unless you know what to dispatch where
<LnL>
a serialisation error + the rabbitmq retry might make stuff route correctly eventually
<LnL>
but it's probably not a good idea to depend on that
jtojnar has quit [Read error: Connection reset by peer]