Akri agent fails to start in managed GKE cluster about akri HOT 3 CLOSED

I get the same error on my self-hosted k8s cluster running on Talos. As far as I cant tell crictl is not present on Talos and getting it installed on the host would be a bit of a hassle because there is no shell access to Talos. It's an API driven and immutable operating system. Basically the only way to get crictl on the host is gonna be to have a job that mounts a host path directory, downloads and extracts crictl. But also this would need to happen prior to the agent starting up, otherwise when the agent attempts to mount it (for example I set the path to /var/lib/akri/bin/crictl) it ends up creating that full path as a directory on the host. So this effectively creates a race condition for either producing a file or a directory at /var/lib/akri/bin/crictl.

I'm not really familiar with crictl. Does it NEED to live on the host? Perhaps we could have an option in the helm chart that would add an init container that downloads and extracts it to an emptyDir mount that could be mounted by the agent?

from akri.

interesting. crictl is used by Akri to query the existing containers (it is needed because the Kubernetes Device-Plugin system doesn't notify when a scheduled pod/container dies/crashes/finishes).

i'm not sure it needs to be on the host ... it probably just needs access to Kubernetes' container socket. maybe it can be installed in the agent container?

from akri.

Issue has been automatically marked as stale due to inactivity for 90 days. Update the issue to remove label, otherwise it will be automatically closed.

from akri.