DNS SRV fixed in upstream Go, switch to native LookupSRV in Prometheus now?

[juliusv]

https://code.google.com/p/go/source/detail?r=77ceb1db4a1e

[juliusv]

/cc @discordianfish

[beorn7]

@fabxc Is that still valid? We use github.com/miekg/dns in retrieval/discovery/dns.go . Is that still needed/intended?

[juliusv]

We would have to re-try with the standard library. The fix back then didn't solve all problems IIRC, but it might be all good nowadays.

[grobie]

What are the reasons to switch?

[juliusv]

@grobie All things being equal, it's always nicer to use the standard library for something (if it works). Less dependencies, often simpler and almost guaranteed to be well-maintained.

[juliusv]

That said, there's no pressure here at all to change this.

[fabxc]

Having this in the back of my head for some time. Once this area is a bit
more stable I'll give it another shot.

On Thu, May 21, 2015 at 6:13 PM Julius Volz notifications@github.com
wrote:

That said, there's no pressure here at all to change this.

—
Reply to this email directly or view it on GitHub
#360 (comment)
.

[grobie]

The golang stdlib DNS resolver still doesn't support EDNS0 and there are no signs they're going to implement it anytime soon. There are DNS recursors which assume EDNS0 to be always supported. I'm in favor of closing this issue.

[brian-brazil]

Makes sense to me, this has been lying around a while with no sign of the Go library being ready for our use yet.

There are DNS recursors which assume EDNS0 to be always supported.

Such recursors are broken.

[discordianfish]

👍

[matthiasr]

👍 to closing.

Such recursors are broken.

True, but not helpful.

[beorn7]

Reopening because the head of github.com/miekg/dns apparently cannot deal with long DNS SRV requests.

Typical error message:
WARN[0684] resolving srv.eample.net failed: dns: failed to unpack truncated message source=dns.go:187

We are just staying with the vendored version we have, but looking at the commit log of github.com/miekg/dns, there seem to be some worrying bugs including race conditions.

We have to revisit the issue in that light: Investigate standard library capabilities once more on the one hand, and finding out what's wrong with newer versions of github.com/miekg/dns on the other hand.

[discordianfish]

I assume nobody reported the bug to miekg/dns yet?
@beorn7 Does this happen with all larger SRV queries or only under special circumstances?

This looks like ErrTruncated: https://godoc.org/github.com/miekg/dns#pkg-variables
And according to the miekg/dns examples, you are suppose to check for that error. So I think that the older version of the code returned err = nil when a message is truncated, now it returns err = ErrTruncated. If that's the case, this should fix it: https://github.com/prometheus/prometheus/compare/master...discordianfish:fish/fix-dns-sd?expand=1
@beorn7 Can you test this? If it works, I can also look into adding some unit tests.

[beorn7]

@discordianfish Thanks. That could very well be it. I just had no time to look into details here.

I'll have a look as soon as I'm done with all those higher-priority tasks.

[beorn7]

Addressed in #2153

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.