I may be going on a tangent from what's in your mind, but i think a means to visualize this stuff would be really useful. I can see value in a few different distinct, but complementary things:
1. state flow of instances in a job. This would cover transitions between states like PENDING and RUNNING, and what happens after termination (based on whether the job is a service, cron, etc) 2. *plan* for Processes in a task, as followed by thermos. This would be handy to see that parallel/sequential invocations are as you expect. 3. plan *execution* of processes in a task, as executed thus far by thermos. Sounds like this is what you described above, so this would be a yet-to-finish view of (2). As for your approach thus far, the apache mail server seems to eat attachments, so you'll probably be better off posting a diff to our ReviewBoard, or pushing a branch to github for folks to peek at. -=Bill On Thu, Apr 9, 2015 at 8:49 AM, Michael Schenck <[email protected]> wrote: > Hi all, > > I need to be able to plot state transitions from a Job-level at bare > minimum. I've put together a patch to thermos-observer (attached: > thermos-statsd.patch). > > Now, I understand that the Observer is likely to be going away, so I was > thinking about creating a similar patch for the executor, possibly for > possibly in ThermosTaskRunner.compute_status > <https://github.com/mschenck/aurora/blob/master/src/main/python/apache/aurora/executor/thermos_task_runner.py#L183-L205>. > My last thought would be to include a Process that *tails* the checkpoint > log similarly to the Observer. > > What is the preferred approach to accomplishing this goal? > > Best regards, > Michael Schenck >
