Sidecar braket#16
Open
vsoch wants to merge 2 commits into
Open
Conversation
In this design, when a gang is scheduled only the first works as a probe to dispatch the quantum work. A sidecar that matches the vendor can monitor the queue, and ungate the rest of the pods in the N-1 group when ready. This way, we are not consuming resources (and wasting them). Signed-off-by: vsoch <vsoch@users.noreply.github.com>
vsoch
force-pushed
the
sidecar-braket
branch
8 times, most recently
from
be0156c to
b89b7f6
Compare
this will ensure they will be scheduled quickly when ungated, since there is no real concept of a reservation. Signed-off-by: vsoch <vsoch@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This will need a lot of tweaking, but the idea is that we dispatch a probe pod for a gang that matches a backend (like Braket or qrmi) and then can use the API to determine when we are close in the queue to running. Close means we can ungate the rest of the group pods for scheduling. Given a higher priority in Kubernetes, if they are not able to be scheduled on the first pass, Kubernetes priority plugin will cancel lesser pods (pre-emption) so they can be scheduled on the next pass.
Gotta run - will pick up on this when I'm back at keyboard.