zipkin timestamps and NTP

[Arens]

Hello guys,

Before today, every services I wanted to log using Zipkin were running on my localhost and everything seemed to work fine. Now I have communications between my localhost and a distant server.

What happens is I still have a Parent-Child relation between my spans inside my trace but I see in the Zipkin UI that I have parents spans which start *after* children spans.

So I was wondering how does Zipkin collector handle spans coming from two different servers but which belong to the same trace.

Is it mandatory to setup a NTP server or did I miss something on how the collector works ? 

I did not find any documentation on the subject so maybe the problem is on my side but it would be great for me to have a better understanding of how all this work anyway :).

Thanks,

-- 

Hugo

[Radu Brumariu]

Look at the json representation of the trace ( http://.../api/get/<traceid> ) and see what the timestamps are for the annotations.

Also check your clock on the two machines. I suspect clock skew can do this to the traces. It is, generally, good practice to have your machines in sync with a ntp server ( you can use one of the public ones, no need to run your own ).

There are some clock skew time adjusters in the code that should run at query time, but I haven't dig too deep into that side of the code. Perhaps someone with more understanding of it can comment on how that works.

HTH,

Radu

--

---
You received this message because you are subscribed to the Google Groups "zipkin-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to zipkin-user...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.