Ability to skip tracing for certain URL patterns

Question

Ability to skip tracing for certain URL patterns

pavolloffay opened this issue 7 years ago · comments

This should be helpful for use cases like: checking application health on openshift, disable favicon, swagger docs..

Pavol Loffay · Answer 1 · Fri Apr 07 2017 21:33:18 GMT+0800 (China Standard Time)

I am thinking about implementation, I'm wondering whether setting sampling.priority=0 on newly started server span would be the cleanest solution here.

cc @bhs @yurishkuro @adriancole

Ben Sigelman · Answer 2 · Sat Apr 08 2017 00:38:30 GMT+0800 (China Standard Time)

@pavolloffay IMO it would make more sense to just omit the Span entirely... of course this is a judgement call, but if the goal is to ignore something it would be better in my mind to skip the cost of allocating Spans just to discard them shortly thereafter.

Adrian Cole · Answer 3 · Sun Apr 09 2017 09:41:12 GMT+0800 (China Standard Time)

I agree with @bhs: instrumentation policy is different than sampling policy. Simpler to not instrument always uninteresting paths (by affecting which paths the trace filter applies to). If that's not possible or practical, then move to sampling policy

Pavol Loffay · Answer 4 · Mon Apr 10 2017 21:26:43 GMT+0800 (China Standard Time)

Thanks, I was curious what opinion do you have on this.

I asked because this will require to pass something in servlet attributes to let know higher layers (spring MVC, jaxrs) know that request is not traced. Which is not so clean but would perform better.

Adrian Cole · Answer 5 · Mon Apr 10 2017 22:46:09 GMT+0800 (China Standard Time)

hmm.. this is an interesting problem. Usually spring MVC and jaxrs do their own instrumentation. Are you saying that these will do instrumentation while there's also a servlet-level filter in place?

Pavol Loffay · Answer 6 · Mon Apr 10 2017 22:50:36 GMT+0800 (China Standard Time)

At the moment this filter is reused by spring-web. I plan to use it also in jax-rs to trace unauthenticated/no URL mapping requests etc..

Adrian Cole · Answer 7 · Mon Apr 10 2017 22:56:43 GMT+0800 (China Standard Time)

Is it code re-use (like one layer calls library code here) or are you sharing spans between layers? Ex one instrumentation depends on span created by this filter? curious

Pavol Loffay · Answer 8 · Mon Apr 10 2017 23:02:31 GMT+0800 (China Standard Time)

in spring-web yes, I run into various issues and this seemed as the best for spring-web. In jax-rs preferred solution would be to create a new span I guess.

Adrian Cole · Answer 9 · Tue Apr 11 2017 10:50:19 GMT+0800 (China Standard Time)

Ok. If makes sense to you, maybe a test dep on one or the other side of this (or emulate the pattern mentioned in a test). Because that would help make sure you can do what you want even if someone else is maintaining. Theres often an assumption of a single span per RPC (at layer where http method might be the span name) though you are right that if different people are writing instrumentation and applying it without coordnation you might end up with double or triply implemented spans. On 10 Apr 2017 11:02 pm, "Pavol Loffay" <notifications@github.com> wrote: in spring-web yes, I run into various issues and this seemed as the best for spring-web. In jax-rs preferred solution would be to create a new span I guess. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#15 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAD612VsDLre3tIgp90Yb5Dp57nTwr47ks5rukSHgaJpZM4MzSmA> .