Composing metrics with client_python

[Thomas Schmidt]

Hello all, 

I have the following use-case: I'm instrumenting a python codebase where backing services (eg. Kafka, S3, HDFS, HTTP client) are injected into a service that provides the business logic. 

In some cases there may be multiple instance of the same backing service per service, which creates either: 

* Issue with the python client, which justifiably refuses to register metrics that already exist

* Issue with the exposition that exposes multiple instance of each metrics (unsure if duplication exist in that case, but it seems to)

What I would like is to have a base CollectorRegistry that can be use for collecting lower-level metrics, but if a metric already exist, instead of raising an exception, simply pass over it, so that multiple instances of a backing service would report as a single set of metrics. 

Is there a simple way to achieve that using the python client currently, and if not what are possible alternatives?

Thank you all in advance, 

– Thomas 

[Brian Brazil]

If it's the same backing service, why aren't they all using the same metric objects? Metrics are meant to be defined at the file level, which shouldn't have these problems.

 

--

Brian Brazil

www.robustperception.io