Include initContainers when calculating pod overhead #3572
Merged
Conversation
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
2i2c-org#3569 changed the cryptnono daemonset to have different resource requests for the init containers as well as the container. While working on 2i2c-org#3566, I noticed this was generating wrong choices - the overhead was calculated wrong (too small). We were intentionally ignoring init containers while calculating overhead, and turns out the scheduler and the autoscaler both do take it into consideration. The effective resource request for a pod is the higher of the resource requests for the containers *or* the init containers - this ensures that a pod with higher requests for init containers than containers (like our cryptnono pod!) will actually run. This is documented at https://kubernetes.io/docs/concepts/workloads/pods/init-containers/#resource-sharing-within-containers, and implemented in Kubernetes itself at https://github.com/kubernetes/kubernetes/blob/9bd0ef5f173de3cc2d1d629a4aee499d53690aee/pkg/api/v1/resource/helpers.go#L50 (this is the library code that the cluster autoscaler uses). This PR updates the two places we currently have that calculate effective resource requests (I assume eventually these will be merged into one - I haven't kept up with the team's work last quarter here). I've updated the node-capacity-info.json file, which is what seems to be used by the generator script right now.
yuvipanda
added a commit
to yuvipanda/pilot-hubs
that referenced
this pull request
Jan 4, 2024
Brings in the new memory / cpu limits set up for 2i2c-org#3572 - without that, the largest set up size doesn't actually spawn on r5.4xlarge due to insufficient memory.
Member
Author
|
Put up the adjusted memory limits in #3573 - I'm not sure why an update for r5.4xlarge was needed (seems unrelated to the fixes here, as the fix should be a no-op). |
Contributor
|
@yuvipanda this looks great, I added a commit where I re-executed the generate command for all clusters based on a comment in the daemonset-requests.yaml file: # This file isn't updated by automation, but can easily be updated by manually
# running a command once for each cluster:
#
# ls config/clusters | xargs -I {} deployer generate resource-allocation daemonset-requests {}
# |
Member
Author
|
Ah, thank you @consideRatio! I'll do that next time I change any code that affects these generated. |
consideRatio
approved these changes
Jan 5, 2024
Contributor
consideRatio
left a comment
consideRatio
left a comment
There was a problem hiding this comment.
Thank you for working this Yuvi!!
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
#3569 changed the cryptnono daemonset to have different resource requests for the init containers as well as the container. While working on #3566, I noticed this was generating wrong choices - the overhead was calculated wrong (too small).
We were intentionally ignoring init containers while calculating overhead, and turns out the scheduler and the autoscaler both do take it into consideration. The effective resource request for a pod is the higher of the resource requests for the containers or the init containers - this ensures that a pod with higher requests for init containers than containers (like our cryptnono pod!) will actually run. This is documented at
https://kubernetes.io/docs/concepts/workloads/pods/init-containers/#resource-sharing-within-containers, and implemented in Kubernetes itself at
https://github.com/kubernetes/kubernetes/blob/9bd0ef5f173de3cc2d1d629a4aee499d53690aee/pkg/api/v1/resource/helpers.go#L50 (this is the library code that the cluster autoscaler uses).
This PR updates the two places we currently have that calculate effective resource requests (I assume eventually these will be merged into one - I haven't kept up with the team's work last quarter here).
I've updated the node-capacity-info.json file, which is what seems to be used by the generator script right now.