Judoscale::Reporter doesn't always follow Judoscale::Config.log_level

[azanar]

Expected

In config/initializers/judoscale.rb:

if defined?(Judoscale)
  Judoscale.configure do |config|
    config.log_level = :warn
  end
end

In logs: no INFO messages

Actual

In logs:

INFO -- : [Judoscale] Reporting 40 metrics

Workaround

if defined?(Judoscale)
  Judoscale.configure do |config|
    config.log_level = :warn
    Judoscale::Reporter.instance.logger.log_level = ::Logger::Severity::WARN
  end
end

Analysis

There appears to be a race condition between Judoscale::Config.instance and Judoscale::Reporter.instance.logger.

When Judoscale::Reporter.instance.logger is called, that calls this code: https://github.com/judoscale/judoscale-ruby/blob/main/judoscale-ruby/lib/judoscale/logger.rb#L9

The first time that code is called, it takes Judoscale::Config.instance.log_level at that moment, and hands it to a new instance LoggerProxy. All later calls hand back the same instance of LoggerProxy. Any later changes to Judoscale::Config.instance.log_level won't be reflected.

The Logger instance wrapped by LoggerProxy has no log level set: https://github.com/judoscale/judoscale-ruby/blob/main/judoscale-ruby/lib/judoscale/config.rb#L83

Potential Solutions

1. Assign and keep in sync Judoscale::Config.instance.log_level with Judoscale::Config.instance.logger
2. Update Judoscale::Reporter.instance.logger when Judoscale::Config.instance.log_level changes.

[adamlogic]

This is the best issue! Thanks so much for the clear repro steps, analysis, and potential solutions. 💯

I'll do some testing and create a PR for this next week. 👍