check_gpu_bots script

This is a script I hacked up during my wrangling shift. Hopefully this is only a temporary measure, and will be deprecated by the awesome work being done on the hackability CY and the sheriff-o-matic.


While wrangling, I found it cumbersome and time consuming to keep track of various waterfalls in a browser window with 10+ tabs. This script is meant to be a quick-and-dirty way to avoid having to stare at the waterfalls quite so much during a shift.

How to get it

Here is an older version of the script available for direct download for convenience, but there will be little continued support for it. The canonical version is now checked into the Chromium repository at src/gpu/tools/


The script, when run directly with no args, simply polls the waterfall's JSON API, gathers some information about (hopefully all of) the relevant GPU bots. It then prints out a summary of the offline or failing bots, and exits.

It can also be set up to repeat itself periodically, sort of like an auto-refresh.

The feature that made it most useful to me is that it can also optionally email you when there's something that needs to be looked into. By default, "needs to be looked into" is roughly defined as
  • there is at least one bot whose most recent build failed, and whose most recent build had passed on the previous execution of the script, OR
  • there is at least one bot which has been in the "offline" state for over an hour, and hadn't been the last time the script executed
As described above, it only sends an email once per bot, the first time something goes wrong. I did that to prevent getting spammed. Currently this can't be overridden with command-line arguments, but I intend to add the ability to override this.

NOTE that currently only gmail addresses are supported as the source email address, since the script hard-codes gmail's smtp server address. In addition, if you use a weak password, gmail may require you to explicitly allow "less secure apps".

Running the script on repeat, and having it set up to email you in case of a problem, takes away a good deal of staring at waterfalls. I recommend that you take a look at the waterfall from time to time, just because this script is not time-tested, and bugs aren't outside the realm of possibility.


  • ./
    • This runs once, and prints the results to stdout.
  • ./ --help
    • This prints basic usage information, including the list of possible command-line arguments.
  • ./ --email-from <source email> --email-to <dest email> --send-email-on-error --repeat-delay 10
    • '--repeat-delay 10' sets the script to poll the waterfall every 10 minutes. '--send-email-on-error' tells the script to email you in case of error/exception (for example, an unreachable server), as well as the normal cases. The rest tells the script to send an email from <source email> to <dest email>. You do have to be in control of <source email> and will need to supply the password, either by saving it in a file on disk, or supplying it to the script's prompt. If you opt to save the password on disk, it is very strongly recommended that you use a dummy account for the source email address. Also note that the script can accept multiple destination email addresses.


Please report bugs to Feedback, features requests, and suggestions are also welcome.
Glenn Hartmann,
Sep 22, 2014, 12:20 PM