Question: Can I configure a sampling rate?

[kop]

Hi,

Thank you for this module, looks really great.
However, I'm a little bit scary to enable this on production nginx. I was not able to find any control on amount tracing spans to create. Do I understand correctly that it will create a span for every request?

For high production traffic, it would be great to have an option to trace only, let's say, 0.5% or 1% of the requests.

[rnburn]

Sampling is usually done as a feature of the tracing libraries. I'm ok adding it to zipkin-cpp-opentracing and exposing it as an option.

I believe the Jaeger implementation already has an option for this, though I don't know if it's quite ready for use yet.

[kop]

@rnburn thank you for the answer.

Jaeger is the tracer I'm using and I'm asking because I don't know any way how to configure this in Jaeger agent. This option is available in all Jaeger client libraries and I was expecting it to be here too.

It would be great to have such option.

[rnburn]

Adding @isaachier, @yurishkuro to see if they can answer.

I think the library is still under development so it may not have all the options yet.

[isaachier]

Yes there is a way and I have a fork of this project in development now. Builds on docker but haven't tested it yet.

[yurishkuro]

while Jaeger clients provide a way to configure sampling priority, the default mode we recommend is letting the clients pull the sampling strategy from Jaeger backend (cf. jaegertracing/jaeger#365).

[rnburn]

@kop I think this is the option to set the agent host and port if that's what you're looking for.

[kop]

while Jaeger clients provide a way to configure sampling priority, the default mode we recommend is letting the clients pull the sampling strategy from Jaeger backend (cf. jaegertracing/jaeger#365).

This sounds great and should be much easier to work with, however this is something from the future. We still need a way to set sampling rate for current usage.

@kop I think this is the option to set the agent host and port if that's what you're looking for.

Thank you, this is not exactly what was my question about :)

[rnburn]

@kop I just put in some fixes to make the Jaeger module work. I did some testing and I believe writing something like this in your nginx.conf will give you what you want:

 http {
    opentracing on;
    jaeger_service_name test-nginx;
    jaeger_sampler_type probabilistic;
    jaeger_sampler_param 0.01; # Sample 10% of requests.

But note that there's a bug in Jaeger if you try to use a sample param of 100% (jaegertracing/jaeger-client-cpp#6).

@isaachier -- Could you document the different options available for jaeger_sampler_type? I had to grep through Jaeger's code to find this.

[isaachier]

Sure thing. Not sure when but soon enough.

[rnburn]

Closing. If you also want sampling for zipkin, please put in an issue to https://github.com/rnburn/zipkin-cpp-opentracing.