Silencing a Rails.logger is (still) not thread-safe ... these initializers will try it's best to avoid your loggers being silenced (and more)! Must have for JRuby.

talkaholic_logger.rb

# put this under **#{Rails.root}/config/initializers** directory

# `silence` is like Chuck Norris you can't kill it, this biuty got yet again
# re-invented in Rails 4 right after Logger#silence got deprecated in 3.2 :)

begin
  require 'active_support/logger'
  require 'active_support/logger_silence'
  ActiveSupport::Logger.silencer = false # silence simply yields
  # if 'active_support/core_ext/logger' is loaded
  # the LoggerSilencer gets included into Logger
  Logger.silencer = false if Logger.respond_to?(:silencer)
rescue LoadError
  begin
    require 'active_support/buffered_logger'
    ActiveSupport::BufferedLogger.silencer = false
  rescue LoadError
  end
  # it's deprecated since 4.0 but not on 3.2 (been around since 2.3)
  require 'active_support/core_ext/logger'
  Logger.silencer = false
end

# the final frontier is ActiveRecord::Base.silence (2.3 - 3.2)
# used mostly in ActiveRecord::SessionStore, that one comes
# from "abusing" the extended ActiveSupport::Benchmarkable

if ActiveSupport.const_defined?(:Benchmarkable)
  ActiveSupport::Benchmarkable.module_eval do
    # NOTE: there's no other way but to monkey-patch :
    def silence
      # if the logger manages to handle silence "thread-safe"
      # it will work, otherwise the logger's silencer = false
      if ( logger = self.logger ).respond_to?(:silence)
        logger.silence { yield }
      else
        yield
      end
    end
  end
else
  ActiveRecord::Base.class_eval do
    # Method implemented directly on AR::Base (2.3) :
    def self.silence
      # if the logger manages to handle silence thread-safe
      # it will work, otherwise the logger's silencer = false
      if ( logger = self.logger ).respond_to?(:silence)
        logger.silence { yield }
      else
        yield
      end
    end
  end if defined? ActiveRecord::Base.silence
end

# NOTE: a "simple" alternative for all of the above might be :
#
#    logger = Rails.logger; def logger.level=(level) end
#

vociferous_reporting.rb

begin
  require 'active_support/core_ext/kernel/reporting'
rescue LoadError
end

module Kernel

  def warn(message)
    super if is_current_thread_verbose?
  end

  def silence_warnings
    verbose = is_current_thread_verbose?
    set_current_thread_verbose(false)
    yield
  ensure
    set_current_thread_verbose(verbose)
  end

  def enable_warnings
    verbose = is_current_thread_verbose?
    set_current_thread_verbose(true)
    yield
  ensure
    set_current_thread_verbose(verbose)
  end

  # used from methods such as {#quietly} :
  #   silence_stream(STDOUT) do
  #     silence_stream(STDERR) do
  #       yield
  #     end
  #   end
  # or {#silence_stderr} `silence_stream(STDERR) { yield }`
  def silence_stream(stream)
    do_silence_stream!(stream, true)
    yield
  ensure
    do_silence_stream!(stream, false)
  end

  # since AS 3.1 {#silence} is aliased as {#capture},
  # we leave `capture` as is but "re-invent" `silence`
  def silence(stream); silence_stream(stream); nil end

  private

  # a `$VERBOSE = verbose` replacement
  def set_current_thread_verbose(verbose)
    verbose # not implemented
  end

  def is_current_thread_verbose?
    $VERBOSE
  end

  # a `$VERBOSE = verbose` replacement
  def do_silence_stream!(stream, silence)
    return stream if silence || true # not implemented
  end

end if defined? Kernel.silence_stream

@harikrishnan83 I think some of it might be reproduced with MRI as well although it might be seen less - one thread enters the silencing block changing the level, than does IO so another thread might run Ruby code entering silence as well but it sees the "original" logger.level as ERROR (changed by the first one) ... etc.

currently mostly seen to happen with AR 4.0's store rails/activerecord-session_store#19

@jjb the obvious pro is it's a one liner the con is logger.level= will never really work which I personally dislike

@kares Thanks for the clarification. On Puma this issue is very frequent. When I changed Puma configuration to one Thread per Process, this problem disappeared.

So if we have to use Puma with Threading even on MRI your Gist is necessary.

I am curious about how your Gist impacts the amount logs that get generated. Will keep you posted on how it goes.