[SPARK-54173][K8S] Add support for Deployment API on K8s #52867
Closed
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
ForVic
changed the title
ForVic
changed the title
Contributor
Author
|
cc @sunchao also @holdenk @dongjoon-hyun |
resource-managers/kubernetes/core/src/main/scala/org/apache/spark/deploy/k8s/Config.scala
Outdated
Show resolved
Hide resolved
Member
|
Thanks @dongjoon-hyun for the review! |
sunchao
reviewed
Nov 6, 2025
resource-managers/kubernetes/core/src/main/scala/org/apache/spark/deploy/k8s/Config.scala
Outdated
Show resolved
Hide resolved
Member
|
Merged to master, thanks @ForVic for the contribution and @dongjoon-hyun for the review! |
zifeif2
pushed a commit
to zifeif2/spark
that referenced
this pull request
Nov 22, 2025
### What changes were proposed in this pull request? Adds support for K8s `Deployment` API to allocate pods. ### Why are the changes needed? Allocating individual pods is not ideal, and we can allocate with higher level APIs. apache#33508 helps this by adding an interface for arbitrary allocators and adds a statefulset allocator. However, dynamic allocation only works if you have implemented a PodDisruptionBudget associated with the decommission label. Since Deployment uses ReplicaSet, which supports `pod-deletion-cost` annotation, we can avoid needing to create a separate PDB resource, and allow dynamic allocation (w/ shuffle tracking) by adding a low deletion cost to executors we are scaling down. When we scale the Deployment, it will choose to scale down the pods with the low deletion cost. ### Does this PR introduce _any_ user-facing change? Yes, adds user-facing configs ``` spark.kubernetes.executor.podDeletionCost ``` ### How was this patch tested? New unit tests + passing existing unit tests + tested in a cluster with shuffle tracking and dynamic allocation enabled ### Was this patch authored or co-authored using generative AI tooling? No Closes apache#52867 from ForVic/dev/victors/deployment_allocator. Lead-authored-by: Victor Sunderland <victors@openai.com> Co-authored-by: victors-oai <victors@openai.com> Co-authored-by: Victor Sunderland <64456855+ForVic@users.noreply.github.com> Signed-off-by: Chao Sun <chao@openai.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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.
What changes were proposed in this pull request?
Adds support for K8s
DeploymentAPI to allocate pods.Why are the changes needed?
Allocating individual pods is not ideal, and we can allocate with higher level APIs. #33508 helps this by adding an interface for arbitrary allocators and adds a statefulset allocator. However, dynamic allocation only works if you have implemented a PodDisruptionBudget associated with the decommission label. Since Deployment uses ReplicaSet, which supports
pod-deletion-costannotation, we can avoid needing to create a separate PDB resource, and allow dynamic allocation (w/ shuffle tracking) by adding a low deletion cost to executors we are scaling down. When we scale the Deployment, it will choose to scale down the pods with the low deletion cost.Does this PR introduce any user-facing change?
Yes, adds user-facing configs
How was this patch tested?
New unit tests + passing existing unit tests + tested in a cluster with shuffle tracking and dynamic allocation enabled
Was this patch authored or co-authored using generative AI tooling?
No