Telegraf: internal_gather - include errors field

[voiprodrigo]

Is it possible to have gather errors tagged by input?
This would be handy.

[danielnelson]

I think this is only a documentation bug, the internal_gather measurement should have an errors field:

internal_gather,input=http,version=unknown errors=0i,gather_time_ns=0i,metrics_gathered=0i 1582242300000000000

[voiprodrigo]

Indeed. However something is not quite right. I'm having some problems with the iptables plugin, due to locking. There's some other agents on my systems that a few times a day call iptables without -n to avoid DNS reverse resolution. On systems with very large rule sets, this is an issue, because iptables -L can take minutes to finish. If I have the plugin running every 60 seconds, it will fail while that other agent is doing that. Of course that's not the plugin fault, but I would expect to see errors > 0 for the iptables input on those occasions, but no, it's always 0. However, the aggregated errors counts is updated in internal_agent.gather_errors for the affected hosts. Using 13.3.

[ssoroka]

There's some other agents on my systems that a few times a day call iptables without -n to avoid DNS reverse resolution

Don't you want to call iptables _with_ -n to avoid dns resolution? Seems like that's why it's taking minutes to complete (or did you mean they're intentionally running it without -n to gather names? I guess you can read that sentence two different ways).

In the mean time, I'll dig into the errors issue.

[voiprodrigo]

@ssoroka it’s not that I don’t want to use -n (which I do use), but that particular agent is a “blackbox” agent mandated by corporate policy.

_{Sent with GitHawk}