Display System partitions in Node Page

[thomasdanan]

Component:

'ui'

Why this is needed:

Currently, we can see all Persistent Volumes that were provisioned on a specific Node.
If one system partition is filling up, it may impact the node or some service running on the Node (ex: /var/lib/etcd or /var/logs, ...)

What should be done:

In the Node Page, we would need a new tab named System Partitions.
This tab should present all system partitions (except the ones mounted by containers).
The list of System partitions are displayed in a table with the following columns:

• Health
• device mount point
• Usage
• Size

The Health of a System partition is computed like we compute the Health of a Persistent Volume (same alert rules).

From a specific System Device (or from the list of System devices on one node), the Platform Administrator can click on a button to access a Grafana Dashboard when he can visualize more information about those devices and more specifically the usage over time.

@xaf-scality feel free to add more details.

PS: just put System partitions tab before the details tab.

Implementation proposal (strongly recommended):

Information can be collected as it is done in the Node Detailed Grafana dashboard

Those are the alerts reports on partitions

Test plan:

[Cuervino]

"Partitions" version of the screen:

[gdemonet]

This tab should present all system partitions (except the ones mounted by containers).

How do you plan on retrieving this @ChengYanJin?
IMO, you may start by retrieving all values of the device label for the node_filesystem_avail_bytes and node_filesystem_size_bytes metrics (filtered for the selected node), and only keep the ones which are not a status.deviceName of some Volume object on this node.

[ChengYanJin]

This tab should present all system partitions (except the ones mounted by containers).

How do you plan on retrieving this @ChengYanJin?
IMO, you may start by retrieving all values of the device label for the node_filesystem_avail_bytes and node_filesystem_size_bytes metrics (filtered for the selected node), and only keep the ones which are not a status.deviceName of some Volume object on this node.

Thanks for the metrics @gdemonet !
From the prometheus query, I actually have fstype field. I was thinking about exclude the devices with value fstype: "iso9660".

And we should exclude the devices withdevice='shm' which I think is used by containerd?
Can you confirm that?

[gdemonet]

Yes, good idea ignoring the ISO 9660 filesystems. For shared memory (device='shm'), indeed it makes sense to ignore them. Likely something we should add to node-exporter collector.diskstats configuration.
Also, I'm wondering if we should go further and only provide details for mounted partitions (which likely means we'll need to get additional info, not sure from where though)?