helm: add charts for memtierd and memory-qos

[fmuyassarov]

This commit adds two new Helm charts for installing memory-qos(#115) and memtierd(#117) NRI plugins via Helm.

[fmuyassarov]

@kad @askervin PR is now updated. But I have not added #121 changes, which I think we need. But since it is under active reviews, perhaps we can do it once the patch has landed.

Both plugins are now outside the resource management dir, and the corresponding images are called
ghcr.io/containers/nri-plugins/nri-memtierd & ghcr.io/containers/nri-plugins/nri-memory-qos.

Same with the labels and DaemonSet object naming, nri-memtierd & nri-memory-qos.

[askervin]

Cool, thanks @fmuyassarov! This works fine already. The last wishes from me:

• Would you add the track-working-set-size class to deployments/helm/memtierd/templates/configmap.yaml
• Is it possible to make this volume mount a configurable option in helm installation?

After adding the track-working-set-size class and helm install it should be possible to deploy test/e2e/files/nri-memtierd-test-pod.yaml without errors.

Deploying test/e2e/files/nri-memory-qos-test-pod.yaml already works fine after helm installing nri-memory-qos. Great!

[fmuyassarov]

Cool, thanks @fmuyassarov! This works fine already. The last wishes from me:

* Would you add the [track-working-set-size](https://github.com/askervin/nri-plugins/blob/5RA_memtierd_host_root/deployment/overlays/memtierd/sample-configmap.yaml#L31) class to deployments/helm/memtierd/templates/configmap.yaml

* Is it possible to make [this](https://github.com/askervin/nri-plugins/blob/5RA_memtierd_host_root/deployment/overlays/memtierd/daemonset.yaml#L46) volume mount a configurable option in helm installation?

After adding the track-working-set-size class and helm install it should be possible to deploy test/e2e/files/nri-memtierd-test-pod.yaml without errors.

Deploying test/e2e/files/nri-memory-qos-test-pod.yaml already works fine after helm installing nri-memory-qos. Great!

@askervin done. PTAL.

[askervin]

@fmuyassarov, sorry, I overlooked a couple of things... there are also other cases where deployments/helm/memtierd/templates/daemonset.yaml has diverged from deployment/overlays/memtierd/daemonset.yaml.

We do not want to pass the --host-root parameter, nor do we want to mount the whole host root filesystem. Also the --run-dir /run-dir is an essential argument in order to get memtierd.output files into the directory that user gives with outputDir to helm.

Setting outputDir seems to work here nicely.

[fmuyassarov]

@askervin I think we also need - --cgroups-dir - /sys/fs/cgroup ?

[askervin]

@askervin I think we also need - --cgroups-dir - /sys/fs/cgroup ?

Yes we do.

It's good to check that helm template --set outputDir=/var/run/nri-memtierd nri-memtierd deployment/helm/memtierd output includes the daemonset that will be created wtih kustomize, too, but it actually has the outputDir mounted that is just commented in the kustomize daemonset.

[marquiz]

Just a mental note that the documentation is a total mess. Installation instructions are under "Resource policy plugins" but the actual plugin documentation under "Memory plugins" 🤪 I don't think should fix the mess in this PR

[fmuyassarov]

ow right. I will submit a patch for that and also for selectorLabel for TA and balloons plugins. Thanks for noticing.

[marquiz]

Thanks @fmuyassarov for working on this. I don't have anything to add for the initial version. We should carry over some of the things (like labels vs selectorLabels) to the other charts as well.
LGTM

[fmuyassarov]

Thanks @fmuyassarov for working on this. I don't have anything to add for the initial version. We should carry over some of the things (like labels vs selectorLabels) to the other charts as well. LGTM

Submitted in #124. Thanks for the reviews @marquiz