Linux mdadm error parsing mdraids - Invalid syntax since version 1.10.0

[toffiebotha]

After upgrading to the latest version of node-exporter (from version 1.9.1), we're seeing frequent errors as below

level=ERROR source=collector.go:168 msg="collector failed" name=mdadm duration_seconds=0.000213012 err="error parsing mdraids: strconv.ParseUint: parsing": invalid syntax"

On our commercial Azure Kubernetes Service, this error does not occur.
But in Azure US Government AKS clusters, the error appears.

This is the only major difference I know of.
As it is a managed service, we don't directly control any of the RAID or other underlying disks the nodes run from, so not sure what could be causing this.

We've also never recorded, configured or used any RAID metrics.
Any suggestions on what more information to supply to get this resolved would be appreciated

[TheMeier]

This indicates an error here https://github.com/prometheus/procfs/blob/master/sysfs/mdraid.go#L52

Maybe you could paste the contents of the files being parsed there if they don't containt sensitive information.

[toffiebotha]

@TheMeier Thanks for the input - Here's all the files and paths and contents from what I could gather in the code:
Please let me know if I've missed anything

In /sys/block/md0/md (no other md* path found)

• level
  • File is empty
• array_size
  • default
• array_state
  • clear
• chunk_size
  • 0
• metadata_version
  • none
• raid_disks
  • File is empty
• uuid
  • File does not exist
• dev-* folders
  • None exist
• degraded
  • File does not exist
• sync_*
  • Files do not exist

[toffiebotha]

I believe I found the problem in the code - The raid_disks file is empty and ReadUintFromFile ends up failing to parse it to uInt

     if val, err := util.ReadUintFromFile(filepath.Join(path, "raid_disks")); err == nil {
			md.Disks = val
		} else {
			return mdraids, err
		}

Not being a fulltime developer and not having all the context of how this should work and behave in this scenario, I'm hesitant to raise a PR.

[TheMeier]

That is exactly what I was suspecting too. I don't have any mdraid myself though...

[NeonSludge]

Root cause: prometheus/procfs#771
There is a PR that will fix this: prometheus/procfs#777