feature-request

Allow deployments to prefer a node tag but fall back gracefully.

Hey all!

I've got a heterogenous cluster where some nodes are CPU heavy and others memory heavy. I'd like to deploy particular workloads to particular nodes if there's space, but I'd much prefer that deploys didn't fail if there's space on another node, just not the one I want. One way I can imagine this working is assign a list of tags [cpu-pool, pool] and have the scheduler try to match as many tags as possible, failing only if it can't match a single one. So an ANY rule in addition to your current ALL implementation, I suppose?

As it stands I'm a bit nervous to configure my preferred split because breaking deploys is a bigger downside than optimising node usage is an upside.

avatar
3
  • I said

    So an ANY rule in addition to your current ALL implementation, I suppose?

    But there's already an ANY strategy which wouldn't do what I wanted here. To keep with the theme maybe a MANY or more accurately a GREEDY ANY (match as many as possible) is probably a more accurate way to describe it.

    avatar
  • Hey Thomas, thanks for the idea here. I think it makes sense to have a way to associate tags like this. I've submitted the request to our internal tracker and will keep you up to date with any changes :)

    avatar
    platform
  • I think the best way to potentially solve this would to still use ANY but maybe allow tags with - or +.

    Having tags like +my-ideal somthing-else -last-resort

    avatar
    platform
v2024.10.16 © 2024 Petrichor Holdings, Inc.