Get the log_messages counter to work properly

[bboreham]

There is supposed to be a metric exposed called log_messages so you can alert on too many errors, warnings, etc. But it doesn't work. [UPDATE: I think right now it is giving all zeros]

If there is another library in your program trying to use a counter named log_messages, it's unpredictable which one wins.

Maybe some kind of namespacing in the name should be added, but from the Prometheus docs it seems that the AlreadyRegisteredError is designed for the case above.

[bboreham]

Right now it looks more like we don't set up logrus so any logging that goes that way from weaveworks/common or the like is not getting counted and not affected by -log-level

[bboreham]

The only path I can see that goes via promrus is weaveworks/common/ServerLoggingInterceptor - if that took a logger as a parameter we'd be set.

[saurav-malani]

I tried running the program and curled the metrics, and I am getting non-zero numbers. So, I think this is no longer an issue.

[bboreham]

Thanks for looking into this @saurav-malani.

I poked around a bit further and came to the conclusion the reason it's not an issue right now is the version of promrus that Cortex is vendoring has the metric log_messages whereas Cortex' own code is using log_messages_total. So the names don't clash and we don't get the issue.

If we updated vendor to the current promrus it quite likely would be a problem again.

[gouthamve]

We don't have this metric anymore. Closing :)