✨ Priorityqueue: Optionally return items within a priority fairly #3261
+476
−39
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
k8s-ci-robot
added
the
cncf-cla: yes
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: alvaroaleman The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
k8s-ci-robot
added
approved
|
/hold |
k8s-ci-robot
added
the
do-not-merge/hold
alvaroaleman
changed the title
This change makes it possible for the priorityqueue to return items fairly across one or more fairness dimensions. These fairness dimensions are extracted from items using a configurable extractor func. Fairness is only ensured within a given priority, as the purpose of priorities is to return higher-priority items before lower-priority items. An example use-case is to use the namespace as fairness dimension, which then ensures that one very busy namespace can not starve off other namespaces. An example use-case for more than one dimension are multi-cluster controllers, where fairness should first be achieved across clusters and then across namespaces within each cluster. By default, no extractor func is configured so this doesn't change any behavior. We can revisit this in the future. Naively, the whole problem looks like it could be solved by sorting the items in the queue appropriately, but that doesn't really work because: * Most (all?) sorting algorithm assume that the order of two elements can be determined just by looking at those two elements, but for fairness we have to consider the items in front of the current elements as well as the items that we already handed out * We can not hand out locked items and have to ignore them for the purpose of fairness. This in turn means that whenever an item gets unlocked, we'd have to redo our sorting This change solves the problem by introducing a `fairQueue` structure which holds all items that are ready to be handed out and in the highest priority bracket that has ready items. The priorityqueue then hands out items from its internal `fairQueue` and manages its content. The `fairQueue` works by using a slice of all the fairness dinensions it has ever observed and a map of fairness dimension to item slice. Whenever it hands out an item, it iterates over this slice and checks the corresponding map for items until it found an item to hand out.
|
PR needs rebase. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. |
k8s-ci-robot
added
the
needs-rebase
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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 change makes it possible for the priorityqueue to return items fairly across one or more fairness dimensions. These fairness dimensions are extracted from items using a configurable extractor func. Fairness is only ensured within a given priority, as the purpose of priorities is to return higher-priority items before lower-priority items.
An example use-case is to use the namespace as fairness dimension, which then ensures that one very busy namespace can not starve off other namespaces.
An example use-case for more than one dimension are multi-cluster controllers, where fairness should first be achieved across clusters and then across namespaces within each cluster.
By default, no extractor func is configured so this doesn't change any behavior. We can revisit this in the future.
Naively, the whole problem looks like it could be solved by sorting the items in the queue appropriately, but that doesn't really work because:
This change solves the problem by introducing a
fairQueuestructure which holds all items that are ready to be handed out and in the highest priority bracket that has ready items. The priorityqueue then hands out items from its internalfairQueueand manages its content.The
fairQueueworks by using a slice of all the fairness dinensions it has ever observed and a map of fairness dimension to item slice. Whenever it hands out an item, it iterates over this slice and checks the corresponding map for items until it found an item to hand out.