Cadvisor create container metrics in diffirent ways for containerd and cadvisor for "name" filed

[xiaoping8385]

Hi there, we recently notice that when using cadvisor to get metrics from the clusters, our docker clusters have a human-readable value for the name label, but the containerd clusters have a long hex value,checking the source code we found: cadvisor generates the Aliases field in different ways for docker and containerd.

dockerd: Aliases = [name, id]: https://github.com/google/cadvisor/blob/aa0afbee6da04e8d8a214333bf2747f0ae27f7a0/container/docker/handler.go#L215 containerd: Aliases = [id, name]: https://github.com/google/cadvisor/blob/2dd0cd908dca0c7e3ca255f601ab80a0ef2ddbc1/container/containerd/handler.go#L119

As far as the prometheus reference is hard coded as “container.Aliases[0]“,https://github.com/google/cadvisor/blob/2b6fbacac7598e0140b5bc8428e3bdd7d86cf5b9/metrics/prometheus.go#L1849 it will leads to two different results if using prometheus interface to obtain the metrics.

So is there any reason we design like this? or it is a cadvisor bug?

[xiaoping8385]

@abhi I notice you submitted the change at the very beginning, could you please take a look at this, thanks!

[bobbypage]

/cc @qiutongs

[pknight37]

Hi do you have any update on this issue?

[qiutongs]

I don't know the history but I also noticed this difference. The docker has k8s_<container-name>_<pod-name>_<namespace>_<pod-uid>_<restart-count> format.

I would consider this is a cadvisor bug as it doesn't provide consistent naming format. A few months ago, I merged this patch https://github.com/google/cadvisor/commit/56f032e06ddf2c7a700fcce51cb9e009d809f5b0 to a separate branch https://github.com/google/cadvisor/tree/containerd-cri. I couldn't merge it to master because it integrates with CRI package and will incur a circular dependency issue when cadvisor is vendored back into k8s.

[ystkfujii]

Hi do you have any update on this issue?