Can logger level set per API request?

Hi all, can Logger level set per API request / process?

Suppose there is a buggy API call, and you want to debug what happen during these API call. could you set Logger level to debug for that API request only? maybe via http header / query parameter / other http payload.

I have been looking around Logger module on elixir and :logger application on erlang, but there it doesn’t seem there’s any way to set level on per request process.

Well, unfortunately not really, at least not without some hacking. There is PR in OTP that would provide such functionality, but it is stalled in discussion:

Currently the only possible approach is to increase default verbosity and then do manual filtering in front of the filter pipeline, but this can be a little bit problematic approach as it may cause unneeded performance degradation and in case of filter failures, it may DoS the service with logs volume.


I think you can set some Logger options at runtime, e.g.

Logger.configure(level: :debug)
# Or on a specific module:
Logger.put_module_level(SomeModule, :debug)

I’ve done this in cases like the one you mentioned where you just need to see what’s going on… something like firing a “tracer round”, e.g. triggered by a query parameter or request header.

However, there are caveats to this, e.g. the logger settings applied only affect the current process, so if your response dispatches execution to other processes (e.g. a database lookup), you may not be enabling the visibility that you think you are.

There are some other subtle and complex caveats (and I’ll defer to the expertise of others here). For instance, if you have configured your app in such a way that logging macros cause the logging statements to be removed from the compiled artifact entirely (e.g. via :compile_time_purge_matching), then no amount of run-time tweaks will make any difference because the statements are no longer present.

Yes, you could set Logger options during runtime. However setted Logger options will affect either whole application, or a particular module. both are still global (it affect other request also), as opposed to local/process based. Link given on @hauleth explain a lot on why logger doesn’t have local/process based logging. It also shine light on other options such as tracing debugger / erlang trace module.