Built-in health check for GCS-FUSE-CSI-DRIVER

[tred77]

Hi all,

Some background:
I am facing this issue GoogleCloudPlatform/gcsfuse#1726
Which is fixed in the newer version of the driver but since I am using stable release channel of GKE, it is not available yet.

Meanwhile for this case and might be relevant also for some other potential issues, I was thinking if I can develop a health check for my running container, using kubernetees livenessProbe/readinessProbe to let kubernetees restart the unhealthy container. To do so, the simplest action would be the ls command, if it fails, then the mount is not live anymore. But since ls command is translated to the list operation which causes the original problem, adding a periodic list operation would make my system more flaky.

I was wondering if there is better way of checking the health of the mount?

[songjiaxun]

It will be challenging for the CSI driver to implement such a monitoring mechanism. Probably we can leverage the CSI Volume Health Monitoring Feature, but since this feature is still alpha, GKE cannot use this feature for now.

We are working to expose GCSFuse metrics via CSI. And maybe in the future, applications can rely on these metrics to decide if the GCSFuse volume is healthy.