jruby_gc_stats 0.1-java

data/README.txt

@@ -0,0 +1,31 @@
+Title: jruby-gc-stats
+
+Description:
+
+This is a set of GC-monitoring methods for JRuby that mimic behavior of Ruby
+Enterprise Edition's GC methods. In our case, these are all implemented in
+Ruby, using the JVM's built-in monitoring and management APIs.
+
+Example Usage:
+
+require 'jruby/gc_stats'
+
+require 'pp'
+puts "Enabling stats..."
+GC.enable_stats
+puts "allocation size: #{GC.allocation_size}"
+puts "Running loop..."
+1_000.times {
+  ary = []
+  1_000.times {ary << 'foo' + 'bar'}
+}
+puts "collections: #{GC.collections}"
+puts "time: #{GC.time}ms"
+puts "bytes since last GC: #{GC.growth}"
+puts "size change: #{GC.allocation_size}"
+puts "Dumping..."
+GC.dump
+
+puts "Dumping caller for all threads..."
+2.times {Thread.new {sleep}}
+pp caller_for_all_threads

data/lib/jruby/gc_stats.rb

@@ -0,0 +1,138 @@
+require 'java'
+require 'jruby'
+
+module GC
+  import java.lang.management.ManagementFactory
+  import java.lang.management.MemoryType
+  GC_MBEANS = ManagementFactory.garbage_collector_mxbeans
+  POOL_MBEANS = ManagementFactory.memory_pool_mxbeans
+  
+  # no perf penalty for stats on JVM, but save current counts
+  def self.enable_stats
+    @enabled = true
+    @start_count = _collection_count
+    @start_time = _collection_time
+    @start_size = _total_pool_size
+  end
+  def self.disable_stats
+    @enabled = false
+  end
+  
+  # Number of GC runs since stat collection started.
+  # This is accumulated across all GCs in the system.
+  def self.collections
+    raise "GC stats are not enabled" unless @enabled
+    new_count = _collection_count
+    new_count - @start_count
+  end
+  
+  # Amount of time spent in GC since stat collection started
+  # This is accumulated across all GCs in the system.
+  def self.time
+    raise "GC stats are not enabled" unless @enabled
+    new_time = _collection_time
+    new_time - @start_time
+  end
+  
+  # Number of heap bytes requested since the last GC run.
+  # This includes all memory pools.
+  def self.growth
+    _usage_versus_collected
+  end
+  
+  # Dumping the basic data for each pool
+  def self.dump
+    filename = ENV['RUBY_GC_DATA_FILE']
+    begin
+      file = filename ? File.open(filename, 'w') : $stderr
+      for pool_bean in POOL_MBEANS
+        file.puts "Name: #{pool_bean.name}"
+        file.puts "  Type: #{pool_bean.type}"
+        file.puts "  Peak usage: #{pool_bean.peak_usage}"
+        file.puts "  Current usage: #{pool_bean.usage}"
+        file.puts "  Usage after last collection: #{pool_bean.collection_usage}"
+        file.puts "  Managers: #{pool_bean.memory_manager_names.to_a.join(', ')}"
+      end
+    ensure
+      file.close if filename
+    end
+  end
+  
+  # A delta in the committed (active main memory) size of all memory pools
+  def self.allocation_size
+    new_size = _total_pool_size
+    new_size - @start_size
+  end
+  
+  def self.num_allocations
+    # not sure this can be tracked on JVM; allocations happen all over
+  end
+  
+  private
+  
+  def self._collection_count
+    GC_MBEANS.inject(0) {|tot, bean| tot + bean.collection_count}
+  end
+  
+  def self._collection_time
+    GC_MBEANS.inject(0) {|tot, bean| tot + bean.collection_time}
+  end
+  
+  def self._usage_versus_collected
+    POOL_MBEANS.inject(0) do |tot, bean|
+      next tot if bean.type == MemoryType::NON_HEAP
+      tot + (bean.usage.used - bean.collection_usage.used)
+    end
+  end
+  
+  def self._total_pool_size
+    POOL_MBEANS.inject(0) do |tot, bean|
+      tot + bean.usage.committed
+    end
+  end
+end
+
+module ObjectSpace
+  def self.live_objects
+    # not sure if there's a way to get this without tooling API on JVM
+  end
+  
+  def self.allocated_objects
+    # ditto
+  end
+end
+
+class Object
+  def caller_for_all_threads
+    backtraces = {}
+    ts = JRuby.runtime.thread_service
+    ts.active_ruby_threads.each do |rthread|
+      tc = ts.get_thread_context_for_thread(rthread)
+      caller = tc.create_caller_backtrace(JRuby.runtime, 0)
+      backtraces[rthread] = caller
+    end
+    backtraces
+  end
+end
+
+if $0 == __FILE__
+  require 'pp'
+  puts "Enabling stats..."
+  GC.enable_stats
+  puts "allocation size: #{GC.allocation_size}"
+  puts "Running loop..."
+  1_000.times {
+    ary = []
+    1_000.times {ary << 'foo' + 'bar'}
+  }
+  puts "collections: #{GC.collections}"
+  puts "time: #{GC.time}ms"
+  puts "bytes since last GC: #{GC.growth}"
+  puts "size change: #{GC.allocation_size}"
+  puts "Dumping..."
+  GC.dump
+  
+  puts "Dumping caller for all threads..."
+  2.times {Thread.new {sleep}}
+  pp caller_for_all_threads
+end

metadata

@@ -0,0 +1,63 @@
+--- !ruby/object:Gem::Specification 
+name: jruby_gc_stats
+version: !ruby/object:Gem::Version 
+  prerelease: false
+  segments: 
+    - 0
+    - 1
+  version: "0.1"
+platform: java
+authors: 
+  - Charles Oliver Nutter
+autorequire: 
+bindir: bin
+cert_chain: []
+
+date: 2010-04-24 19:01:44.476000 -05:00
+default_executable: 
+dependencies: []
+
+description: A set of GC-monitoring methods for JRuby similar to those in Ruby Enterprise Edition
+email: 
+  - headius@headius.com
+executables: []
+
+extensions: []
+
+extra_rdoc_files: 
+  - README.txt
+files: 
+  - lib/jruby/gc_stats.rb
+  - README.txt
+has_rdoc: true
+homepage: http://github.com/headius/jruby_gc_stats
+licenses: []
+
+post_install_message: 
+rdoc_options: []
+
+require_paths: 
+  - lib
+required_ruby_version: !ruby/object:Gem::Requirement 
+  requirements: 
+    - - ">="
+      - !ruby/object:Gem::Version 
+        segments: 
+          - 0
+        version: "0"
+required_rubygems_version: !ruby/object:Gem::Requirement 
+  requirements: 
+    - - ">="
+      - !ruby/object:Gem::Version 
+        segments: 
+          - 0
+        version: "0"
+requirements: []
+
+rubyforge_project: 
+rubygems_version: 1.3.6
+signing_key: 
+specification_version: 3
+summary: A set of GC-monitoring methods for JRuby similar to those in Ruby Enterprise Edition
+test_files: []
+