[inputs/nstat] error on snmp6 when ipv6 is disabled

[deric]

Relevant telegraf.conf:

[[inputs.nstat]]

System info:

Debian 4.9.110-3+deb9u4 (2018-08-21), Telegraf 1.9.4

Steps to reproduce:

1. Linux kernel without IPv6 support ipv6.disable=1

Expected behavior:

Silently ignore missing /proc/net/snmp6.

Workaround

redirect snmp6 to snmp and then ignore it.

[[inputs.nstat]]
proc_net_snmp6 = "/proc/net/snmp"
namedrop = "snmp6"

Not a good idea. Incorrectly parsed snmp6 fields with enormous cardinality:

> SHOW FIELD KEY CARDINALITY FROM nstat
name: nstat
count
-----
389040

Actual behavior:

E! [inputs.nstat]: Error in plugin: open /proc/net/snmp6: no such file or directory

[danielnelson]

As a workaround you can add this option:

[[inputs.nstat]]
  proc_net_snmp6 = "/dev/null"

[deric]

@danielnelson Right, that's much better solution.

Just curious, nstat offers JSON output, why not use that?

nstat -a --json

It gives pretty much the same output, except all metrics are merged into single category (no snmp, snmp6 or nestat namespace) and you don't have to parse 3 files from disk.

Is a PR witch such change acceptable for telegraf? Or would you prefer keeping names of metrics the same way?

[danielnelson]

It is faster to read 3 files than to exec a binary, and then of course using it would add a dependency on the nstat binary. If you strace nstat you can see it opens these 3 files as well.

On naming, we try to keep everything backwards compatible so as to not break any dashboard or alerts.

[deric]

Sure, makes sense. Thanks for explanation!