Print DNS errors

[2opremio]

Fixes #1577

[tomwilkie]

By default we try and resolve scope.weave.local, which will fail if people aren't running weave (most aren't), so this will pollute the logs, no?

[2opremio]

By default we try and resolve scope.weave.local, which will fail if people aren't running weave (most aren't), so this will pollute the logs, no?

True, it's similar to #1599 (comment) but I don't think the solution is to remove all errors, including legitimate ones. In fact, in the example I just mentioned we do the exact opposite.

How about only including the scope.weave.local target once weave is found?

[tomwilkie]

How about only including the scope.weave.local target once weave is found?

I don't like the hardcoding - how about we only report resolution errors once (or n times) per target and silence them after? And similarly, when the start succeeding again, we log something.

[2opremio]

I don't like the hardcoding

It's not hardcoding, I wrote scope.weave.local to continue the conversation but we will use whatever is provided as --probe.weave.hostname

It makes sense, if weave is not running, why are we even targeting that host?

how about we only report resolution errors once (or n times) per target and silence them after? And similarly, when the start succeeding again, we log something.

That's an option too, but I prefer the former because it won't display irrelevant errors

[tomwilkie]

It makes sense, if weave is not running, why are we even targeting that host?

We don't know if we are running weave or not. We assume we are, and try to use it.

I think you're suggesting we only log errors after we have successfully resolved a host? But you are also arguing we should show resolution errors? Or would you just do this as a special case for resolution of scope's weavedns hostname?

I think it too special cased, and a more general rule (like only logging the first error) would be easier to understand.

[2opremio]

We don't know if we are running weave or not. We assume we are, and try to use it.

I think you're suggesting we only log errors after we have successfully resolved a host? But you are also arguing we should show resolution errors? Or would you just do this as a special case for resolution of scope's weavedns hostname?

What I am suggesting is we only register the weave target in the probes after we have proof of weave working (e.g. after weave ps succeeds or the weave status request suceeds ...)

I think it too special cased, and a more general rule (like only logging the first error) would be easier to understand.

We can do both. Only log the first error and only add the WeaveDNS target once we know weave is successfully running.

[tomwilkie]

LGTM; I see we only get/set from the single goroutine - does it matter that we create the map on a different one?

[2opremio]

does it matter that we create the map on a different one?

Not AFAIK