Multi-ICMP probe

[SuperQ]

In order to provide a more smokeping-like experience, it would be nice to have an ICMP probe mode that sends out multiple packets and buckets the probe time in a histogram and counts lost packets.

[brian-brazil]

I'm not sure about this, why would it be limited to just ICMP?

Multiple scrapes from Prometheus could provide this for all modules.

[SuperQ]

My thinking is that ICMP is lighter-weight than most of the TCP-based probes.

[brian-brazil]

If we support it for one, we should support it for all.

Smokeping is the best smokeping.

[SuperQ]

Yea, thinking about this more, I think we're already doing better than smokeping with the right PromQL.

[RichiH]

Smokeping is an island of data and you end up doing checks twice, though.

Some guidance for @qybl as to what's needed would be appreciated as he can implement this on company time.

[SuperQ]

I had some ideas today about a better way to do smokeping-like stats.

Running Prometheus with a high scrape interval and using recording rules is OK, but I think there are better options.

What I'm thinking is not a Blackbox exporter, where the scrapes are scheduled by Prometheus, but a network prober that has its own scheduling and exports more like a normal target.

[RichiH]

Is there an inherent reason why this needs to be done in an extra system? I am not saying I am opposed, just wondering.

Can you re-open this ticket for now?

[brian-brazil]

I'd consider this out of scope for the blackbox exporter. A blackbox probe is on or off. Smokeping is gathering metrics across multiple probes.

[RichiH]

And smokeping looks at the Blackbox state of a system. Either way, if you wouldn't merge this, doing it elsewhere is better. Please give final no if that's the case. Richard Sent by mobile; excuse my brevity and the wall of text Gmail appends by default.

…

On Feb 27, 2017 14:07, "Brian Brazil" ***@***.***> wrote: I'd consider this out of scope for the blackbox exporter. A blackbox probe is on or off. Smokeping is gathering metrics across multiple probes. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#115 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAuEI_pt2PS26ZCJEb1MAS4r9HgYN3gFks5rgsqmgaJpZM4MLPNf> .

[brian-brazil]

I don't see a way to have this make sense in the blackbox exporter, and I'm not even sure it works well within Prometheus. It's a specialised use case.

[SuperQ]

I think this is not a good fit for the blackbox exporter. But I think something similar has some more useful potential for network probing. I'd be keen on chatting about it.

[RichiH]

To create a public log: @SuperQ and me keep mulling things on Hangouts and we are coming up with ideas for a network latency exporter.

[SuperQ]

FYI: I have a working prototype histogram ping prober.