Advertise the readiness of the driver on each node.

[dysosmus]

This PR set the label gcs.csi.ofek.dev/driver-ready={true,false} on each node running csi-gcs to reflect the state of readiness of the driver.

This allows pods to avoid nodes where csi-gcs is still starting up.

[ofek]

Thanks! I'll let @maennchen give the final approval.

Can you please briefly explain/show how this would be used?

[dysosmus]

Hi, indeed I forgot to provide a use case.

When pods get scheduled on freshly created nodes and the driver just started many pods will temporarily fail with MountVolume.MountDevice failed for volume xxxx usually because the driver is still registering, for example. It only persists for a few seconds. However, we noticed at the company where I work, it has a tendency to make our users worried.

Having the driver readiness signalled at node level with a label allow us to inject proper affinity/node selector in pods mounting csi-gcs volume and ensure pods only get scheduled after the driver is ready to mount volumes. This results in a smoother user experience.
It's of course a trade-off of longer scheduling time vs temporary warnings at pod startup.

[ofek]

Great! Can you please rebase/merge master and add such info to https://github.com/ofek/csi-gcs/blob/master/docs/troubleshooting.md#todo

(and rename header)

[dysosmus]

I updated the documentation. If I have the time, and if you are interested, I can contribute the mutating webhook too in a separate PR.

[ofek]

Yes please!