twobeeb commented on pull request #9589:
URL: https://github.com/apache/kafka/pull/9589#issuecomment-736079387
@ryannedolan
I have a preference for your first suggestion (the one for which there is
currently a code proposition) because it doesn't alter the behavior of MM2 in
any way for existing users.
> IMO, we'd ideally skip creating the A->B herder whenever
A->B.emit.heartbeats.enabled=false (defaults to true) and A->B.enabled=false
(defaults to false). A top-level emit.heartbeats.enabled=false would then
disable heartbeats altogether, which would trivially eliminate the extra
herders. N.B. this would just be an optimization and wouldn't required a KIP,
IMO.
That being said, I'm also fine with your latest proposition. It fits
perfectly to my personnal use case because topics will be replicated local to
central as well as central to local in the near future ; all instanciated
herders will then transmit data (and not just beats).
Could you confirm and/or elaborate on the behavior you want to see when
playing with ``emit.heartbeats.enabled`` ?
I guess that'd be good test cases :
**Simple A to B**
````
A->B.enabled=true
````
Expected : 2 herders
**A to B and A to C**
````
clusters=A, B, C
A->B.enabled=true
A->C.enabled=true
````
Expected : 4 herders (AB BA AC CA)
**Disabled heartbeats**
````
clusters=A, B, C
A->B.enabled=true
A->B.emit.heartbeats.enabled=false
B->A.emit.heartbeats.enabled=false
A->C.enabled=true
A->C.emit.heartbeats.enabled=false
C->A.emit.heartbeats.enabled=false
````
Expected : 4 herders (2 of which are unused, but still much better than
N*N-1)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
