metrics: free inode script code presumes a single mount

[grahamwhaley]

Testing a PR using a kind (Kubernetes IN Docker) cluster for testing, I hit this error:

INFO: And grab some stats
INFO: idle [94.38] free [23529112] launch [0] node [kind-worker2] inodes_free [30909596
30909596
30909596]
./k8s_scale_nc.sh: line 124: 30909596
30909596
30909596: syntax error in expression (error token is "30909596
30909596")

Tracking that down to the code:

local inode_free=$(kubectl exec -ti $name -- sh -c "df -i | awk '/^overlay/ {print \$4}'" | sed 's/\r//')
...
local inode_used=$((node_baseinode[$node]-inode_free))

The root problem looking like the code presumes there will only be one overlay mount point, but under kind it seems there are three.... from an exec sh into the stats pod, we can see:

/ # df
Filesystem           1K-blocks      Used Available Use% Mounted on
overlay              490818480  23955044 441861504   5% /
tmpfs                    65536         0     65536   0% /dev
tmpfs                 16415596         0  16415596   0% /sys/fs/cgroup
overlay              490818480  23955044 441861504   5% /etc/hosts
overlay              490818480  23955044 441861504   5% /dev/termination-log
/dev/root            490818480  23955044 441861504   5% /etc/hostname
/dev/root            490818480  23955044 441861504   5% /etc/resolv.conf
shm                      65536         0     65536   0% /dev/shm
tmpfs                 16415596        12  16415584   0% /var/run/secrets/kubernetes.io/serviceaccount

Not sure on the best solution here. Looking for input @dklyle @askervin - can we just take the first instance, or should we sum all instances, or should we treat each instance separately?

I'm also wondering if moving from the stats pod to the collectd data would fix this or make the solution better?