Scale Apps CPU Pinning

[stavros-k]

I don't know if Devs are planning to add this feature but i haven't see it mentioned here or in jira.
I know in docker CPU limiting/pinning is supported, i don't know if that's the case in k8s or helm.
If someone could post a feature request in Jira, i could vote it!
I'm afraid of doing so, because i can't provide enough info, and don't want to spam with tickets without the necessary info and get ignored in the end because of that!
Thanks

 

[morganL]

I assume you just want to limit the CPU resources consumed by an application..... if so, explaining you situation and which apps consume too many resources is useful.

Pinning is a much more complex process to manage for many applications.

 

[stavros-k]

I assume you just want to limit the CPU resources consumed by an application..... if so, explaining you situation and which apps consume too many resources is useful.

Pinning is a much more complex process to manage for many applications.

For some dumb reason i wrote my reply on the report button (i might need some sleep lol),

For example i use handbrake (in a docker) which runs most of the day, and in my current setup(unraid) i have limited the cpu to 8 cores (TR2920x) so i got 4 more for the rest of the system to be happy and functional.

If i leave handbrake with unlimited cpu resources i believe it will slow down the system

 

[peter.m]

What you're asking for is container requests and limits, and sure, those are supported in k8s :)

Resource Management for Pods and Containers

When you specify a Pod, you can optionally specify how much of each resource a container needs. The most common resources to specify are CPU and memory (RAM); there are others. When you specify the resource request for containers in a Pod, the kube-scheduler uses this information to decide which...

kubernetes.io

Cpu pinning is also supported, but that is something completely different - it allows you to pin the container to a specific core/cores in the CPU

 

[stavros-k]

What you're asking for is container requests and limits, and sure, those are supported in k8s :) As is cpu pinning, but that is something completely different

Good to hear k8s support it as well!

Just to clarify because i didn't mention it in my first post, I'm asking for GUI implementation!

 

[peter.m]

It will be up to the helm chart for a given application to ask the question - looking at ix-charts they don't, but it would be just a couple of lines in questions.yaml to do so (here's a peek https://github.com/truenas/charts/blob/master/charts/plex/1.0.0/questions.yaml). Oh and i just remembered, k3s in scale doesn't enable with metrics-server ootb, and it is required for this to work.

 

[stavros-k]

It will be up to the helm chart for a given application to ask the question - looking at ix-charts they don't, but it would be just a couple of lines in questions.yaml to do so (here's a peek https://github.com/truenas/charts/blob/master/charts/plex/1.0.0/questions.yaml)

Good to hear that it could be easy as that for an implementation! I'll wait a little to see if it lands on the next updates!

 

[marshalleq]

Just to add an extra vote for this (also coming in from Unraid). I've been using CPU pinning for isolating cpu intensive tasks from those that require responsiveness e.g. web sites don't share the same cores as an encoding cluster I run (across multiple machines). Also, for video surprisingly you get a noticeably better result if you limit the threads that are encoding to under 8. So to this end on my threadripper box I assign 4x8 threads per encoding binary (each in a separate docker). I have in the past also done this with VM's and dockers limiting each to the specific use case of the workload within each. I do similar on my xeon box. Further I tend to leave the first core free as recommended by unraid for general system processing, though I have no idea if that applies to TrueNAS and that did always seem odd. Many thanks.

 

[marshalleq]

I left truenas, but ive Migrated to it again today, I haven't seen this feature anywhere. - anyone know if it was implemented or if there was a feature request done?

 

[morganL]

I don't believe there is a NAS ticket or any SCALE user stories indicating that the feature is required.

 

[truecharts]

I left truenas, but ive Migrated to it again today, I haven't seen this feature anywhere. - anyone know if it was implemented or if there was a feature request done?

We support CPU limiting by default and even set it by default as well, to prevent a single app from completely slowing down the whole system.

 

[morganL]

We support CPU limiting by default and even set it by default as well, to prevent a single app from completely slowing down the whole system.

Is that limiting working on SCALE 23.10 or have you had issues??

 

[truecharts]

Is that limiting working on SCALE 23.10 or have you had issues??

That is using Kubernetes-Native resource limiting and, as far as we're aware, working perfectly fine.
If it's not we've at least not gotten any reports about it (yet) in years.