-
Notifications
You must be signed in to change notification settings - Fork 203
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
WIP: docs: Add RFC for multi-node consolidation partitioning #1547
base: main
Are you sure you want to change the base?
WIP: docs: Add RFC for multi-node consolidation partitioning #1547
Conversation
Signed-off-by: Cameron McAvoy <[email protected]>
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: cnmcavoy The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
Pull Request Test Coverage Report for Build 10887856678Details
💛 - Coveralls |
This PR has been inactive for 14 days. StaleBot will close this stale PR after 14 more days of inactivity. |
|
||
Multi-node consolidation should be partitioned into multiple consolidation actions, each of which is responsible for consolidating nodes that are homogenous in terms of cpu architecture and nodepool. This will allow for more successful consolidation actions, and will reduce the number of nodes in the cluster more effectively. | ||
|
||
A naive, simple approach would be to have Karpenter create M bins for multi-node consolidation, based on cpu architecture and the nodepool of the node. Multi-node consolidation would sort the bins by the number of nodes contained, and attempt multi-node consolidation on the nodes within each bin, starting with the largest, descending. After any solution is found, return the solution. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think the issue is more around how we order our nodes in consolidation, right?
We order our nodes by the estimated cost to disrupt the pods on that node, and then find some contiguous group of nodes from 0 that are compatible. If the set of candidates are ordered in such a way that each alternating node has a different architecture, you'll never be able to get multi-node consolidation. How were you thinking this might impact that complication today?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Does the estimated disruption cost ordering need to be changed? The disruption cost is independent of whether the node is compatible to be consolidated with another node.
What I am proposing is to bin the nodes together with nodes they are compatible. This part is the hand-wavey, and a bit arbitrary; Karpenter has to decide where the boundaries for whether candidates should be consolidated. If Karpenter picks bad boundaries (current situation) then there are two many or two few bins, and consolidation fails. This is also where I am most interested in input... what are the boundaries for each bin that is roughly compatible. Just CPU architecture? CPU architecture and node pool? Maybe the node's operating system should be a boundary.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
What I am proposing is to bin the nodes together with nodes they are compatible.
The problem is more about compatibility in between pods, right? I can see what you're saying that right now, every node is in its own bucket (no pre-defined ordering). We can either create arbitrary buckets/groupings and use that as a heuristic for grouping nodes in our consolidation ordering. The closest generalization for partitioning would be if we could group all nodes from the same nodepool together, but I think even then you're still hitting the same multi-arch issue on your NodePool, so you actually probably want to make the grouping separate from the NodePool.
If we just make it on CPU architecture, you probably solve most of the issues here, but I need to think more about how this might work in practice.
This PR has been inactive for 14 days. StaleBot will close this stale PR after 14 more days of inactivity. |
This PR has been inactive for 14 days. StaleBot will close this stale PR after 14 more days of inactivity. |
Fixes #853
Description
Add RFC for multi-node consolidation partitioning
How was this change tested?
N/A
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.