Adding helm charts for the snapshot-agent - #19
Conversation
| initContainers: | ||
| - name: install-cuda-checkpoint | ||
| image: alpine:latest | ||
| command: ["sh", "-c", "apk add --no-cache wget && wget -qO /opt/bin/cuda-checkpoint https://raw.githubusercontent.com/NVIDIA/cuda-checkpoint/main/bin/x86_64_Linux/cuda-checkpoint && chmod +x /opt/bin/cuda-checkpoint && echo 'cuda-checkpoint installed'"] |
There was a problem hiding this comment.
Not a blocker for this PR. I wonder if it would be better to fork cuda-checkpoint into the time-slicing repo and build it into the snapshot-agent image at build time rather than pulling the main branch at run time? It would help avoid instabilities/incompatibilities caused by upstream changes to cuda-checkpoint.
There was a problem hiding this comment.
Not sure what the usual pattern is, but don't think a fork into this repo is the best move. Can we pin it to a commit install that version instead?
| port: 9001 | ||
|
|
||
| securityContext: | ||
| privileged: true |
There was a problem hiding this comment.
Not a blocker for this PR. We should maybe file a follow-up issue in our backlog to scope the permissions down to exactly what we need rather than broad full host access - for e.g. SYS_PTRACE, SYS_ADMIN, CHECKPOINT_RESTORE capabilities might suffice. This requires some testing, so let's not block this PR on this.
|
/approve |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
|
/approve |
2 participants
What does this PR do?
Why is this change needed?
Adds the helm charts needed to deploy the snapshot daemon.
Sample command:
How was this tested?
Checklist
git commit -s) per DCOmake test)make lint)Related Issues