Add tapestats to collect tape devices statistics

[luizluca]

It is heavily based on diskstats to allow metrics reuse by simply
s/disk/tape/ the query.

Signed-off-by: Luiz Angelo Daros de Luca luizluca@gmail.com

[SuperQ]

Thanks for the new collector. We typically implement access to proc/sys files via the Prometheus procfs library rather than directly in the node_exporter.

If you wouldn't mind implementing the file parser in the library first, it would be great to add this.

(Yes, there are some legacy collectors that still use proc/sys directly)

[luizluca]

Thanks for the new collector. We typically implement access to proc/sys files via the Prometheus procfs library rather than directly in the node_exporter.

If you wouldn't mind implementing the file parser in the library first, it would be great to add this.

No problem. The collector is quite simple. It might not be a big deal.

(Yes, there are some legacy collectors that still use proc/sys directly)

As a suggestion, it would be nice to have a comment on top of every "legacy" usage with a "// TODO: migrate to procfs library". As programming is mostly a bunch of ctrl+c,ctrl+v, without a warning those codes tend to replicate.

[luizluca]

Done. It requires prometheus/procfs#382

[luizluca]

Rebased and solved CHANGELOG.md conflicts.
procfs side is already merged.

[discordianfish]

Nice LGTM!

[SuperQ]

CC @bluecmd

[luizluca]

All changes applied. Rebased on master to use procfs v0.7.0

[SuperQ]

LGTM.

[SuperQ]

Actually, should we name this scsi_tape instead of tapestats to match the procfs files?

[luizluca]

I though collectors abstracts the underlying OS. If someone implements tapestats for windows, FreeBSD, it would not make sense to mention a Linux specific name.

Anyway, I can rename it if needed.

[SuperQ]

Most of the naming has been based on the Linux names for things, since we typically start with Linux implementations.

tapestats is fine with me.

@bluecmd Any opinions?

[bluecmd]

Should this really be multiplied?

[SuperQ]

Seems like a copy-n-paste bug.

[luizluca]

Seems like a copy-n-paste bug.

Yes, it is... I fixed that. Thanks!

[bluecmd]

I think this looks fine (save for the one comment above), I have been using the following script for a few days to test these metrics out, and it has worked fine.

#!/bin/bash

for stp in /sys/class/scsi_tape/st[0-999]
do
  drv="$(basename "${stp}")"
  echo "node_tape_io_now{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/in_flight")"
  echo "node_tape_io_time_nanoseconds_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/io_ns")"
  echo "node_tape_io_others_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/other_cnt")"
  echo "node_tape_read_bytes_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/read_byte_cnt")"
  echo "node_tape_reads_completed_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/read_cnt")"
  echo "node_tape_read_time_nanoseconds_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/read_ns")"
  echo "node_tape_residual_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/resid_cnt")"
  echo "node_tape_written_bytes_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/write_byte_cnt")"
  echo "node_tape_writes_completed_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/write_cnt")"
  echo "node_tape_write_time_nanoseconds_total{device=\""${drv}"\"} $(<"/sys/class/scsi_tape/${drv}/stats/write_ns")"
done

Looking forward to having these metrics straight in node_exporter!

[bluecmd]

@SuperQ can this be merged?

[SuperQ]

Done, thanks!