incubator-uniffle icon indicating copy to clipboard operation
incubator-uniffle copied to clipboard

[Feature Request] Support deploy multiple shuffle servers in a single node

Open xianjingfeng opened this issue 3 years ago • 9 comments

We have 1.2T memory per host, so we need deploy multiple shuffle servers in a single node. But current partition assignment policy is not suitable to do that. Because partition maybe assign to a same host and client maybe fail if one host down

xianjingfeng avatar Jul 26 '22 10:07 xianjingfeng

As i know, partition assignment policy could be implemented by yourself.

zuston avatar Jul 26 '22 10:07 zuston

As i know, partition assignment policy could be implemented by yourself.

Yes, but not only assignment policy need to be change, such as start script, i will create a pr for this

xianjingfeng avatar Jul 26 '22 13:07 xianjingfeng

I'm also curious why we need to modify start script?

jerqi avatar Jul 27 '22 03:07 jerqi

I'm also curious why we need to modify start script?

start script will process existence

xianjingfeng avatar Jul 27 '22 12:07 xianjingfeng

If our plan is deploy on k8s, this issue should close?

xianjingfeng avatar Jul 27 '22 12:07 xianjingfeng

K8S Deployment can't solve this issue.

jerqi avatar Jul 27 '22 12:07 jerqi

I'm also curious why we need to modify start script?

start script will process existence

Yes, this is the limit with current implementation

colinmjj avatar Jul 28 '22 03:07 colinmjj

K8S Deployment can't solve this issue.

@xianjingfeng I think you can continue this.

jerqi avatar Jul 28 '22 03:07 jerqi

K8S Deployment can't solve this issue.

@xianjingfeng I think you can continue this.

ok

xianjingfeng avatar Jul 29 '22 13:07 xianjingfeng