We've discovered a case where the node-exporter NodeFilesystemAlmostOutOfSpace alert is firing for a customer use case that could be excluded from the checks as a whole.  

The customer is using the [IBM Cloud Object Storage Plugin](https://github.com/IBM/ibmcloud-object-storage-plugin), which creates a tmpfs mountpoint on the cluster node to store the API key or AccessKey + SecretKey for AWS S3.  However, it creates a 4k filesystem, just large enough to store a single file, holding the key.

This triggers the node-exporter to think the filesystem is out or almost out of disk space, and raise an alert, which is paging OpenShift Dedicated SREs on-call.

It might be possible to exclude these on OSD clusters with the `--collector.filesystem.mount-points-exclude`, but doing so for any plugin or 3rd party component that might do this would be difficult to scale.

It's been suggested that the node-exporter could possibly handle this, perhaps with a 1M minimum filesystem size requirement to be included in the monitored filesystems (or something along those lines)?

I realize this isn't a bug, per se, but we're hoping the node-exporter can be more efficient than manual SRE intervention.

N/A

100%

1. Install and setup the IBM Cloud Object Storage plugin S3 component.
2. Observe the nodeFileSystemFillingUp alert triggered by the stored passwd file

nodeFileSystemFillingUp alert triggered

Hopefully nodeFileSystemFillingUp alert is suppressed.

The code in question is here: https://github.com/IBM/ibmcloud-object-storage-plugin/blob/157a391b710bb1a89da609d45c48eb5d2ba3e1d8/driver/driver.go#L259