Blog for running gpu based functions on Fission

[soharab-ic]

No description provided.

[netlify]

✅ Deploy Preview for fission-website ready!

Name	Link
🔨 Latest commit	d966fd3
🔍 Latest deploy log	https://app.netlify.com/sites/fission-website/deploys/66f511c567721300082d8e2a
😎 Deploy Preview	https://deploy-preview-271--fission-website.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[techmaharaj]

General:

• We can improve the intro - add some more meat around serverless functions, talk about rise in AI workloads, how serverless can help, here's how we can use Fission to deploy GPU functions.
• [Important] We should add a section on Why run GPU based functions on Fission, maybe 2-4 points with reasons, that'll be really helpful.
• Let us also mention at the beginning before pre-requisites as to what we are doing in this blog post. Currently it's not mentioned anywhere.
• We can find an opportunity to link to our GPU blog post on InfraCloud.

Specific:

• Let's be clear with pre-requisites, a system with an Nvidia card, Kubernetes cluster, fission installed (link to fission installation doc), Nvidia related stuff installed (link to respective pages) - all these should be bullet points under the sub-heading "Pre-requisites."
• Steps under the "Steps" section are also not clear.
• Can improve the sentiment example? I mean currently we're hardcoding the statement and the output in the last step just says Positive, and I was where did I give the statement? So two things can be done here,either provide a way to pass the statement as an argument OR print the statement in the output.

[sanketsudake]

Environment reference is missing here.

[soharab-ic]

Package is already associated with an environment.

[sanketsudake]

Can we provide the Kubectl edit command here?
Also, if we can do this via kubectl patch that would be easy to consume for the reader.

[sanketsudake]

Please consider patching environment spec. Not deployment.
User should not change fissions internal object.

[soharab-ic]

@sanketsudake Patching the resources is working via environment spec but nodeSelector patch is not working.
Seeing this error.

Error: error applying specs: environment apply failed: Environment.fission.io "pytorch" is invalid: spec.runtime.podspec.containers: Required value

apiVersion: fission.io/v1
kind: Environment
metadata:
  creationTimestamp: null
  name: pytorch
spec:
  builder:
    command: build
    container:
      name: ""
      resources: {}
    image: ghcr.io/fission/python-builder
  imagepullsecret: ""
  keeparchive: false
  poolsize: 1
  resources:
    requests:
      nvidia.com/gpu: "1"
    limits:
      nvidia.com/gpu: "1"
  runtime:
    container:
      name: ""
      resources: {}
    image: ghcr.io/fission/python-env
    podSpec:
      nodeSelector:
        kubernetes.io/hostname: "gpu-node03"
  version: 3

[sanketsudake]

So typically you need to have container name same environment. As we have two containers in pod

• runtime and fetcher

containers is required field https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.26/#podspec-v1-core

https://github.com/fission/fission/blob/93869d3bc8b6097143e5488366dae178488b4474/test/tests/test_specs/test_spec_merge/specs/env-nodend.yaml#L25

Probably we can skip node selector part in blog post. only keep resources requests.

[sanketsudake]

Checkout

• Change container name for runtime container to environment name in newdeploy fission#2044
• Runtime container should follow same naming convention in pool manager and new deploy fission#2043

[soharab-ic]

Skipping the node selector part.

[sanketsudake]

Yes please update blog accordingly