-
Notifications
You must be signed in to change notification settings - Fork 1.4k
Kafka Partitions Per Pod Scaling #2581
Copy link
Copy link
Closed
kedacore/keda-docs
#1539Labels
feature-requestAll issues for new features that have not been committed toAll issues for new features that have not been committed toneeds-discussionstale-bot-ignoreAll issues that should not be automatically closed by our stale botAll issues that should not be automatically closed by our stale bot
Metadata
Metadata
Assignees
Labels
feature-requestAll issues for new features that have not been committed toAll issues for new features that have not been committed toneeds-discussionstale-bot-ignoreAll issues that should not be automatically closed by our stale botAll issues that should not be automatically closed by our stale bot
Type
Projects
Status
Ready To Ship
Proposal
Enhance the Kafka Scaler or create a new KEDA Scaler to support scaling Kafka by partitions per pod as opposed to simple increase in pod count.
Use-Case
Say I have 12 partitions for my topic.
At 3 pods I will have four partitions per pod:
4 + 4 + 4 = 12At 4 pods I will have three partitions per pod:
3 + 3 + 3 + 3 = 12If I naively scale to 5 pods, I will end up with 3 pods with 2 partitions each and 2 pods with still 3 partitions each:
2 + 2 + 2 + 3 + 3 = 12To scale correctly I need to jump from 4 pods directly to 6 pods:
2 + 2 + 2 + 2 + 2 + 2 = 12And instead of scaling to 7, 8, 9, 10, or 11, it should scale directly from 6 to 12.
1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 = 12As such the list of valid scale targets is not a simple sequence from 0 to 12. It is rather a sequence of the divisors of the partition count:
0, 1, 2, 3, 4, 6, 12Anything else?
No response