Intelligent Worker Cycling RFC

[petemoore]

An RFC to ensure that outdated workers are refreshed or removed from worker pools when worker pool definitions are updated.

Originated in https://github.com/taskcluster/taskcluster/issues/2891.

[tomprince]

Rendered

[petemoore]

Rendered

I didn't understand this comment, sorry! Ah - is the issue that the doc name is 0161-deploymentid.md? If that is what you meant, indeed I can change that to 0161-intelligent-worker-cycling.md.

[petemoore]

I think this looks great. Thanks!

The one minor concern I have is whether the reregister call occurs often enough to refresh configuration in a timely fashion, or if there should be a more active way to push new config.

My thinking is that we can set the frequency of worker manager reregister calls to the minimum of how often we wish to check that the worker configuration is current, and how often we want credentials to expire. If the calling frequency matches the minimum of the two, both requirements should be met with a minimum total number of API calls required (since a single API call covers both refreshes). This may mean we can further reduce the validity period of temporary credentials returned from the workermanager.reregisterWorker call.

[tomprince]

Rendered

I didn't understand this comment, sorry! Ah - is the issue that the doc name is 0161-deploymentid.md? If that is what you meant, indeed I can change that to 0161-intelligent-worker-cycling.md.

No, it is a convenient link to the rendered version of the RFC for easy of reading. I would've added to the description if I had permission.

[petemoore]

CC @SimonSapin @lissyx @davehouse @dividehex @escapewindow

[petemoore]

Closing for now as I'm not actively working on it.