Comments on: Profiling Buildbot http://atlee.ca/blog/2009/07/28/profiling-buildbot/ programming, photography, media, and anything else that strikes my fancy Fri, 27 Jan 2012 19:08:24 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Robert Kaiser http://atlee.ca/blog/2009/07/28/profiling-buildbot/comment-page-1/#comment-55602 Robert Kaiser Thu, 30 Jul 2009 01:22:36 +0000 http://atlee.ca/blog/?p=450#comment-55602 Sending less data from the slaves doesn't sound like a good idea, all the log data is really helpful when looking at problems (and I tend to have a number of those where I need to look at those logs) Sending less data from the slaves doesn’t sound like a good idea, all the log data is really helpful when looking at problems (and I tend to have a number of those where I need to look at those logs)

]]> By: Axel Hecht http://atlee.ca/blog/2009/07/28/profiling-buildbot/comment-page-1/#comment-55577 Axel Hecht Wed, 29 Jul 2009 07:31:55 +0000 http://atlee.ca/blog/?p=450#comment-55577 The sync might actually be that high because we have tons of atomic steps that are either not really part of the build, or work arounds. TinderboxPrint is a prominent example, I guess that the plethora of repo clobbers for tools are, too. I guess the l10n builds are the most drastic here, we're having 36 steps. Of those, 12 are required to do the job, to my counting. Which is generous, as I've let building config and nsprpub pass. Not sure we ever need the latter. 7 steps are just TinderboxPrint and echo's. 5 are slave maintaince (buildtools and compare-locales repos). There are a few other steps which may or may not deserve a step, and thus a separate log file. Killing tinderbox would help our load (the bz2 part, too). And a more thorough review of what we do in the builders and why. And we should use a better tool for slave maintainance than regular builders, too. The sync might actually be that high because we have tons of atomic steps that are either not really part of the build, or work arounds.

TinderboxPrint is a prominent example, I guess that the plethora of repo clobbers for tools are, too.

I guess the l10n builds are the most drastic here, we’re having 36 steps. Of those, 12 are required to do the job, to my counting. Which is generous, as I’ve let building config and nsprpub pass. Not sure we ever need the latter. 7 steps are just TinderboxPrint and echo’s. 5 are slave maintaince (buildtools and compare-locales repos). There are a few other steps which may or may not deserve a step, and thus a separate log file.

Killing tinderbox would help our load (the bz2 part, too). And a more thorough review of what we do in the builders and why.

And we should use a better tool for slave maintainance than regular builders, too.

]]> By: Anonymous http://atlee.ca/blog/2009/07/28/profiling-buildbot/comment-page-1/#comment-55571 Anonymous Wed, 29 Jul 2009 03:03:54 +0000 http://atlee.ca/blog/?p=450#comment-55571 also take a look at http://code.google.com/p/unladen-swallow/ also take a look at http://code.google.com/p/unladen-swallow/

]]>