1459710 - show more history of a job for physical workers

Status: RESOLVED FIXED

(Taskcluster :: UI, enhancement)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

I would like to see a longer history of what a machine runs or confirmation of a reboot as the first entry in history if that is the case.

I see something like this:
https://tools.taskcluster.net/provisioners/releng-hardware/worker-types/gecko-t-win10-64-hw/workers/mdc1/T-W1064-MS-251

but sometimes there are jobs that fail to run and the pattern is hard to see when they fail on a specific machine due to a configuration or cleanup issue.  For example look at https://bugzilla.mozilla.org/show_bug.cgi?id=1457452#c8.

I think showing up to 200 recent jobs would be a good idea.

Comment 1

[Hassan Ali (:hassan)]

We currently perform a request to get a list of the most recent tasks IDs and then for each ID, we perform an additional request to the queue service. Increasing that number to 200 means having to do 200 web requests possibly making the UI slow. To implement this feature, we would need to adjust the limit in the queue repo https://github.com/taskcluster/taskcluster-queue/blob/7ed2624a0f0b5f82f513bdc603f166f5ab7e4972/src/workerinfo.js#L9.

If we do go this direction, I would like the UI to still show 20 recent tasks show as default but let the user switch to 200 tasks if they would like to dig deeper.

Comment 2

[:Eli Perelman]

Pushing this priority of this to taskcluster-web. I would really like to not make 200 requests from the UI. :)

Comment 3

[Chris Cooper [:coop] (he/him)]

We still intend to fix this, but this will be made easier by moving to postgres.

Comment 4

[Dustin J. Mitchell [:dustin] (he/him)]

Per email chain, we still want to do this, but it's not blocking calling worker-manager complete, as for the moment we have the /provisioners/ view.

Comment 5

[Dustin J. Mitchell [:dustin] (he/him)]

Once hardware machines are managed by worker-manager, they will be tracked properly.

Comment 6

[Dustin J. Mitchell [:dustin] (he/him)]

We're no longer tracking operational issues. This is code-complete, so good enough.