evented_bluepill 0.0.46

data/lib/bluepill/process.rb

@@ -0,0 +1,422 @@
+require "state_machine"
+require "daemons"
+
+module Bluepill
+  class Process
+    CONFIGURABLE_ATTRIBUTES = [
+      :start_command, 
+      :stop_command, 
+      :restart_command, 
+      
+      :stdout,
+      :stderr,
+      :stdin,
+      
+      :daemonize, 
+      :pid_file, 
+      :working_dir,
+      :environment,
+      
+      :start_grace_time, 
+      :stop_grace_time, 
+      :restart_grace_time,
+      
+      :uid,
+      :gid,
+      
+      :monitor_children,
+      :child_process_template
+    ]
+    
+    attr_accessor :name, :watches, :triggers, :logger, :skip_ticks_until
+    attr_accessor *CONFIGURABLE_ATTRIBUTES
+    attr_reader :children, :statistics
+    
+    state_machine :initial => :unmonitored do
+      # These are the idle states, i.e. only an event (either external or internal) will trigger a transition.
+      # The distinction between down and unmonitored is that down
+      # means we know it is not running and unmonitored is that we don't care if it's running.
+      state :unmonitored, :up, :down
+
+      # These are transitionary states, we expect the process to change state after a certain period of time.
+      state :starting, :stopping, :restarting
+
+      event :tick do
+        transition :starting => :up, :if => :process_running?
+        transition :starting => :down, :unless => :process_running?
+
+        transition :up => :up, :if => :process_running?
+        transition :up => :down, :unless => :process_running?
+        
+        # The process failed to die after entering the stopping state. Change the state to reflect
+        # reality.
+        transition :stopping => :up, :if => :process_running? 
+        transition :stopping => :down, :unless => :process_running?
+        
+        transition :down => :up, :if => :process_running?
+        transition :down => :starting, :unless => :process_running?
+        
+        transition :restarting => :up, :if => :process_running?
+        transition :restarting => :down, :unless => :process_running?
+      end
+
+      event :start do
+        transition [:unmonitored, :down] => :starting
+      end
+
+      event :stop do
+        transition :up => :stopping
+      end
+
+      event :unmonitor do
+        transition any => :unmonitored
+      end
+
+      event :restart do
+        transition [:up, :down] => :restarting
+      end
+
+      before_transition any => any, :do => :notify_triggers
+
+      after_transition any => :starting, :do => :start_process
+      after_transition any => :stopping, :do => :stop_process
+      after_transition any => :restarting, :do => :restart_process
+
+      after_transition any => any, :do => :record_transition
+    end
+    
+    def initialize(process_name, options = {})      
+      @name = process_name
+      @event_mutex = Monitor.new
+      @transition_history = Util::RotationalArray.new(10)
+      @watches = []
+      @triggers = []
+      @children = []
+      @statistics = ProcessStatistics.new
+      
+      # These defaults are overriden below if it's configured to be something else.
+      @monitor_children =  false
+      @start_grace_time = @stop_grace_time = @restart_grace_time = 3
+      @environment = {}
+      
+      CONFIGURABLE_ATTRIBUTES.each do |attribute_name|
+        self.send("#{attribute_name}=", options[attribute_name]) if options.has_key?(attribute_name)
+      end
+            
+      # Let state_machine do its initialization stuff
+      super() # no arguments intentional
+    end
+
+    def tick
+      return if self.skipping_ticks?
+      self.skip_ticks_until = nil
+
+      # clear the memoization per tick
+      @process_running = nil
+
+      # run state machine transitions
+      super
+
+      if self.up?
+        self.run_watches
+        
+        if self.monitor_children?
+          refresh_children!
+          children.each {|child| child.tick}
+        end
+      end
+    end
+    
+    def logger=(logger)
+      @logger = logger
+      self.watches.each {|w| w.logger = logger }
+      self.triggers.each {|t| t.logger = logger }
+    end
+    
+    # State machine methods
+    def dispatch!(event, reason = nil)
+      @event_mutex.synchronize do
+        @statistics.record_event(event, reason)
+        self.send("#{event}")
+      end
+    end
+    
+    def record_transition(transition)
+      unless transition.loopback?
+        @transitioned = true
+        
+        # When a process changes state, we should clear the memory of all the watches
+        self.watches.each { |w| w.clear_history! }
+        
+        # Also, when a process changes state, we should re-populate its child list
+        if self.monitor_children?
+          self.logger.warning "Clearing child list"
+          self.children.clear
+        end
+        logger.info "Going from #{transition.from_name} => #{transition.to_name}"
+      end
+    end
+    
+    def notify_triggers(transition)
+      self.triggers.each {|trigger| trigger.notify(transition)}
+    end
+    
+    # Watch related methods
+    def add_watch(name, options = {})
+      self.watches << ConditionWatch.new(name, options.merge(:logger => self.logger))
+    end
+    
+    def add_trigger(name, options = {})
+      self.triggers << Trigger[name].new(self, options.merge(:logger => self.logger))
+    end
+
+    def run_watches
+      now = Time.now.to_i
+
+      threads = self.watches.collect do |watch|
+        [watch, Thread.new { Thread.current[:events] = watch.run(self.actual_pid, now) }]
+      end
+      
+      @transitioned = false
+      
+      threads.inject([]) do |events, (watch, thread)|
+        thread.join
+        if thread[:events].size > 0
+          logger.info "#{watch.name} dispatched: #{thread[:events].join(',')}"
+          thread[:events].each do |event|
+            events << [event, watch.to_s]
+          end
+        end
+        events
+      end.each do |(event, reason)|
+        break if @transitioned
+        self.dispatch!(event, reason)
+      end
+    end
+    
+    def determine_initial_state
+      if self.process_running?(true)
+        self.state = 'up'
+      else
+        # TODO: or "unmonitored" if bluepill was started in no auto-start mode.
+        self.state = 'down'
+      end
+    end
+    
+    def handle_user_command(cmd)
+      case cmd
+      when "start"
+        if self.process_running?(true)
+          logger.warning("Refusing to re-run start command on an already running process.")
+        else
+          dispatch!(:start, "user initiated")
+        end
+      when "stop"
+        stop_process
+        dispatch!(:unmonitor, "user initiated")
+      when "restart"
+        restart_process
+      when "unmonitor"
+        # When the user issues an unmonitor cmd, reset any triggers so that
+        # scheduled events gets cleared
+        triggers.each {|t| t.reset! }
+        dispatch!(:unmonitor, "user initiated")
+      end
+    end
+    
+    # System Process Methods
+    def process_running?(force = false)
+      @process_running = nil if force # clear existing state if forced
+      
+      @process_running ||= signal_process(0)
+      # the process isn't running, so we should clear the PID
+      self.clear_pid unless @process_running
+      @process_running
+    end
+    
+    def start_process
+      logger.warning "Executing start command: #{start_command}"
+      
+      if self.daemonize?
+        System.daemonize(start_command, self.system_command_options)
+        
+      else
+        # This is a self-daemonizing process
+        with_timeout(start_grace_time) do
+          result = System.execute_blocking(start_command, self.system_command_options)
+          
+          unless result[:exit_code].zero?
+            logger.warning "Start command execution returned non-zero exit code:"
+            logger.warning result.inspect
+          end          
+        end
+      end
+            
+      self.skip_ticks_for(start_grace_time)
+    end
+    
+    def stop_process      
+      if stop_command
+        cmd = self.prepare_command(stop_command)
+        logger.warning "Executing stop command: #{cmd}"
+        
+        with_timeout(stop_grace_time) do
+          result = System.execute_blocking(cmd, self.system_command_options)
+          
+          unless result[:exit_code].zero?
+            logger.warning "Stop command execution returned non-zero exit code:"
+            logger.warning result.inspect
+          end
+        end
+                
+      else
+        logger.warning "Executing default stop command. Sending TERM signal to #{actual_pid}"
+        signal_process("TERM")
+      end
+      self.unlink_pid # TODO: we only write the pid file if we daemonize, should we only unlink it if we daemonize?
+      
+      self.skip_ticks_for(stop_grace_time)
+    end
+    
+    def restart_process
+      if restart_command
+        cmd = self.prepare_command(restart_command)
+        
+        logger.warning "Executing restart command: #{cmd}"
+        
+        with_timeout(restart_grace_time) do
+          result = System.execute_blocking(cmd, self.system_command_options)
+
+          unless result[:exit_code].zero?
+            logger.warning "Restart command execution returned non-zero exit code:"
+            logger.warning result.inspect
+          end
+        end
+        
+        self.skip_ticks_for(restart_grace_time)
+      else
+        logger.warning "No restart_command specified. Must stop and start to restart"
+        self.stop_process
+        # the tick will bring it back.
+      end
+    end
+    
+    def daemonize?
+      !!self.daemonize
+    end
+    
+    def monitor_children?
+      !!self.monitor_children
+    end
+    
+    def signal_process(code)
+      ::Process.kill(code, actual_pid)
+      true
+    rescue
+      false
+    end
+    
+    def actual_pid
+      @actual_pid ||= begin
+        if pid_file
+          if File.exists?(pid_file)
+            str = File.read(pid_file)
+            str.to_i if str.size > 0
+          else
+            logger.warning("pid_file #{pid_file} does not exist or cannot be read")
+            nil
+          end
+        end
+      end
+    end
+    
+    def actual_pid=(pid)
+      @actual_pid = pid
+    end
+    
+    def clear_pid
+      @actual_pid = nil
+    end
+    
+    def unlink_pid
+      File.unlink(pid_file) if pid_file && File.exists?(pid_file)
+    end
+    
+     # Internal State Methods
+    def skip_ticks_for(seconds)
+      # TODO: should this be addative or longest wins?
+      #       i.e. if two calls for skip_ticks_for come in for 5 and 10, should it skip for 10 or 15?
+      self.skip_ticks_until = (self.skip_ticks_until || Time.now.to_i) + seconds.to_i
+    end
+       
+    def skipping_ticks?
+      self.skip_ticks_until && self.skip_ticks_until > Time.now.to_i
+    end
+    
+    def refresh_children!
+      # First prune the list of dead children
+      @children.delete_if {|child| !child.process_running?(true) }
+      
+      # Add new found children to the list
+      new_children_pids = System.get_children(self.actual_pid) - @children.map {|child| child.actual_pid}
+ 
+      unless new_children_pids.empty?
+        logger.info "Existing children: #{@children.collect{|c| c.actual_pid}.join(",")}. Got new children: #{new_children_pids.inspect} for #{actual_pid}"
+      end
+      
+      # Construct a new process wrapper for each new found children
+      new_children_pids.each do |child_pid|
+        child = self.child_process_template.deep_copy
+        
+        child.name = "<child(pid:#{child_pid})>"
+        child.actual_pid = child_pid
+        child.logger = self.logger.prefix_with(child.name)
+        
+        child.initialize_state_machines
+        child.state = "up"
+        
+        @children << child
+      end
+    end
+
+    def deep_copy
+      # TODO: This is a kludge. Ideally, process templates 
+      # would be facotries, and not a template object.
+      mutex, triggers, @event_mutex, @triggers = @event_mutex, @triggers, nil, nil
+      clone = Marshal.load(Marshal.dump(self))
+      clone.instance_variable_set("@event_mutex", Monitor.new)
+      clone.instance_variable_set("@triggers", triggers.collect{ |t| t.deep_copy })
+      @event_mutex = mutex
+      @triggers = triggers
+      clone
+    end
+
+    def prepare_command(command)
+      command.to_s.gsub("{{PID}}", actual_pid.to_s)
+    end
+    
+    def system_command_options
+      {
+        :uid         => self.uid, 
+        :gid         => self.gid, 
+        :working_dir => self.working_dir,
+        :environment => self.environment,
+        :pid_file    => self.pid_file,
+        :logger      => self.logger,
+        :stdin       => self.stdin,
+        :stdout      => self.stdout,
+        :stderr      => self.stderr
+      }
+    end
+    
+    def with_timeout(secs, &blk)
+      Timeout.timeout(secs.to_f, &blk)
+      
+    rescue Timeout::Error
+      logger.err "Execution is taking longer than expected. Unmonitoring."
+      logger.err "Did you forget to tell bluepill to daemonize this process?"
+      self.dispatch!("unmonitor")
+    end
+  end
+end
+ 

data/lib/bluepill/process_conditions/always_true.rb

@@ -0,0 +1,17 @@
+module Bluepill
+  module ProcessConditions
+    class AlwaysTrue < ProcessCondition
+      def initialize(options = {})
+        @below = options[:below]
+      end
+      
+      def run(pid)
+        1
+      end
+      
+      def check(value)
+        true
+      end
+    end
+  end
+end

data/lib/bluepill/process_conditions/cpu_usage.rb

@@ -0,0 +1,18 @@
+module Bluepill
+  module ProcessConditions
+    class CpuUsage < ProcessCondition
+      def initialize(options = {})
+        @below = options[:below]
+      end
+      
+      def run(pid)
+        # third col in the ps axu output
+        System.cpu_usage(pid).to_f
+      end
+      
+      def check(value)
+        value < @below
+      end
+    end
+  end
+end

data/lib/bluepill/process_conditions/http.rb

@@ -0,0 +1,52 @@
+require 'net/http'
+require 'uri'
+
+module Bluepill
+  module ProcessConditions
+    class Http < ProcessCondition
+      def initialize(options = {})
+        @uri = URI.parse(options[:url])
+        @kind = case options[:kind]
+                  when Fixnum then Net::HTTPResponse::CODE_TO_OBJ[options[:kind].to_s]
+                  when String, Symbol then Net.const_get("HTTP#{options[:kind].to_s.camelize}")
+                else
+                  Net::HTTPSuccess
+                end
+        @pattern = options[:pattern] || nil
+        @open_timeout = (options[:open_timeout] || options[:timeout] || 5).to_i
+        @read_timeout = (options[:read_timeout] || options[:timeout] || 5).to_i
+      end
+
+      def run(pid)
+        session = Net::HTTP.new(@uri.host, @uri.port)
+        session.open_timeout = @open_timeout
+        session.read_timeout = @read_timeout
+        hide_net_http_bug do
+          session.start do |http|
+            http.get(@uri.path)
+          end
+        end
+      rescue
+        $!
+      end
+
+      def check(value)
+        return false unless value.kind_of?(@kind)
+        return true  unless @pattern
+        return false unless value.class.body_permitted?
+        @pattern === value.body
+      end
+
+    private
+      def hide_net_http_bug
+        yield
+      rescue NoMethodError => e
+        if e.to_s =~ /#{Regexp.escape(%q|undefined method `closed?' for nil:NilClass|)}/
+          raise Errno::ECONNREFUSED, "Connection refused attempting to contact #{@uri.scheme}://#{@uri.host}:#{@uri.port}"
+        else
+          raise
+        end
+      end
+    end
+  end
+end

data/lib/bluepill/process_conditions/mem_usage.rb

@@ -0,0 +1,31 @@
+module Bluepill
+  module ProcessConditions
+    class MemUsage < ProcessCondition
+      MB = 1024 ** 2
+      FORMAT_STR = "%d%s"
+      MB_LABEL = "MB"
+      KB_LABEL = "KB"
+      
+      def initialize(options = {})
+        @below = options[:below]
+      end
+      
+      def run(pid)
+        # rss is on the 5th col
+        System.memory_usage(pid).to_f
+      end
+      
+      def check(value)
+        value.kilobytes < @below
+      end
+      
+      def format_value(value)
+        if value.kilobytes >= MB
+          FORMAT_STR % [(value / 1024).round, MB_LABEL]
+        else
+          FORMAT_STR % [value, KB_LABEL]
+        end
+      end
+    end
+  end
+end

data/lib/bluepill/process_conditions/process_condition.rb

@@ -0,0 +1,21 @@
+module Bluepill
+  module ProcessConditions
+    class ProcessCondition  
+      def initialize(options = {})
+        @options = options
+      end
+      
+      def run(pid)
+        raise "Implement in subclass!"
+      end
+      
+      def check(value)
+        raise "Implement in subclass!"
+      end
+      
+      def format_value(value)
+        value
+      end
+    end
+  end
+end

data/lib/bluepill/process_conditions.rb

@@ -0,0 +1,13 @@
+module Bluepill
+  module ProcessConditions    
+    def self.[](name)
+      const_get(name.to_s.camelcase)
+    end
+  end
+end
+
+require "bluepill/process_conditions/process_condition"
+Dir["#{File.dirname(__FILE__)}/process_conditions/*.rb"].each do |pc|
+  require pc
+end
+

data/lib/bluepill/process_statistics.rb

@@ -0,0 +1,24 @@
+module Bluepill
+  class ProcessStatistics    
+    STRFTIME = "%m/%d/%Y %H:%I:%S"
+    # possibly persist this data.
+    def initialize
+      @events = Util::RotationalArray.new(10)
+    end
+
+    def record_event(event, reason)
+      @events.push([event, reason, Time.now])
+    end
+
+    def to_s
+      str = []
+      @events.each do |(event, reason, time)|
+        str << "  #{event} at #{time.strftime(STRFTIME)} - #{reason || "unspecified"}"
+      end
+      if str.size > 0
+        str << "event history:"
+      end
+      str.reverse.join("\n")
+    end
+  end
+end

data/lib/bluepill/socket.rb

@@ -0,0 +1,47 @@
+require 'socket'
+
+module Bluepill
+  module Socket
+    TIMEOUT = 10
+
+    extend self
+
+    def client(base_dir, name, &b)
+      UNIXSocket.open(socket_path(base_dir, name), &b)
+    end
+    
+    def client_command(base_dir, name, command)
+      client(base_dir, name) do |socket|
+        Timeout.timeout(TIMEOUT) do
+          socket.puts command
+          Marshal.load(socket)
+        end
+      end
+    rescue EOFError, Timeout::Error
+      abort("Socket Timeout: Server may not be responding")
+    end
+    
+    def server(base_dir, name)
+      socket_path = self.socket_path(base_dir, name)
+      begin
+        UNIXServer.open(socket_path)
+      rescue Errno::EADDRINUSE
+        # if sock file has been created.  test to see if there is a server
+        begin
+          UNIXSocket.open(socket_path)
+        rescue Errno::ECONNREFUSED
+          File.delete(socket_path)
+          return UNIXServer.open(socket_path)
+        else
+          logger.err("Server is already running!")
+          exit(7)
+        end
+      end
+    end
+    
+    def socket_path(base_dir, name)
+      File.join(base_dir, 'socks', name + ".sock") 
+    end
+  end
+end
+