Support Autoscalers actively querying for queued jobs and provisioning instances accordingly #5798
Assignees
Comments
ZainRizvi
added a commit
that referenced
this issue
Dec 10, 2024
… scale up code to enable larger scale ups (#6015) Part of #5798 ### Changes included - Fix a big bug with how we provisioned instances for ephemeral instances. In the old code, the fewer ephemeral runners we had available, the _less_ likely we were to overprovision them, which sounds like the opposite of what we actually want to do. Changed that logic to more explicitly give ephemeral runners a 50% chance of overprovisioning the incoming request by two instances, which is what I think [this PR](#5501) was intending to do (we can tweak these numbers as desired). This fix might handle most of our ephemeral runner woes by itself... - Extend the allRunnersBusy function to enable it to handle scale up requests of more than a single instance (which will be leveraged in a future PR) - Renames allRUnnersBusy to something more accurate name
|
Fixed a different bug with autoscalers fixed queuing for ephemeral jobs |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Goal is to reduce queueing by ensuring we provision instances even when their scale up request got dropped somehow
The text was updated successfully, but these errors were encountered: