add stop_signal and stop_timeout options; fix stop string command not…

… working

History.txt

@@ -1,3 +1,9 @@
+==
+  * Minor Enhancements
+    * Add stop_timeout and stop_signal options to Watch
+  * Bug Fixes
+    * Stop command string was being ignored
+
 == 0.9.0 / 2010-04-03
   * Minor Enhancements
     * Allow kqueue for OpenBSD and NetBSD

lib/god.rb

@@ -157,6 +157,8 @@ module God
   DRB_ALLOW_DEFAULT = ['127.0.0.1']
   LOG_LEVEL_DEFAULT = :info
   TERMINATE_TIMEOUT_DEFAULT = 10
+  STOP_TIMEOUT_DEFAULT = 10
+  STOP_SIGNAL_DEFAULT = 'TERM'
 
   class << self
     # user configurable

lib/god/process.rb

@@ -2,8 +2,9 @@ module God
   class Process
     WRITES_PID = [:start, :restart]
 
-    attr_accessor :name, :uid, :gid, :log, :log_cmd, :err_log, :err_log_cmd, :start, :stop, :restart,
-                  :unix_socket, :chroot, :env, :dir
+    attr_accessor :name, :uid, :gid, :log, :log_cmd, :err_log, :err_log_cmd,
+                  :start, :stop, :restart, :unix_socket, :chroot, :env, :dir,
+                  :stop_timeout, :stop_signal
 
     def initialize
       self.log = '/dev/null'
@@ -14,6 +15,8 @@ def initialize
       @pid = nil
       @unix_socket = nil
       @log_cmd = nil
+      @stop_timeout = God::STOP_TIMEOUT_DEFAULT
+      @stop_signal = God::STOP_SIGNAL_DEFAULT
     end
 
     def alive?
@@ -203,11 +206,11 @@ def call_action(action)
         command = lambda do
           applog(self, :info, "#{self.name} stop: default lambda killer")
 
-          ::Process.kill('TERM', pid) rescue nil
-          applog(self, :info, "#{self.name} sent SIGTERM")
+          ::Process.kill(@stop_signal, pid) rescue nil
+          applog(self, :info, "#{self.name} sent SIG#{@stop_signal}")
 
           # Poll to see if it's dead
-          5.times do
+          @stop_timeout.times do
             begin
               ::Process.kill(0, pid)
             rescue Errno::ESRCH
@@ -220,14 +223,14 @@ def call_action(action)
           end
 
           ::Process.kill('KILL', pid) rescue nil
-          applog(self, :info, "#{self.name} still alive; sent SIGKILL")
+          applog(self, :warn, "#{self.name} still alive after #{@stop_timeout}s; sent SIGKILL")
         end
       end
 
       if command.kind_of?(String)
         pid = nil
 
-        if @tracking_pid
+        if [:start, :restart].include?(action) && @tracking_pid
           # double fork god-daemonized processes
           # we don't want to wait for them to finish
           r, w = IO.pipe
@@ -326,13 +329,15 @@ def spawn(command)
     #
     # Returns nothing
     def ensure_stop
+      applog(self, :warn, "#{self.name} ensuring stop...")
+
       unless self.pid
         applog(self, :warn, "#{self.name} stop called but pid is uknown")
         return
       end
 
       # Poll to see if it's dead
-      10.times do
+      @stop_timeout.times do
         begin
           ::Process.kill(0, self.pid)
         rescue Errno::ESRCH
@@ -345,7 +350,7 @@ def ensure_stop
 
       # last resort
       ::Process.kill('KILL', self.pid) rescue nil
-      applog(self, :warn, "#{self.name} process still running 10 seconds after stop command returned. Force killing.")
+      applog(self, :warn, "#{self.name} still alive after #{@stop_timeout}s; sent SIGKILL")
     end
 
     private

lib/god/watch.rb

@@ -12,10 +12,13 @@ class Watch < Task
 
     extend Forwardable
     def_delegators :@process, :name, :uid, :gid, :start, :stop, :restart, :dir,
-                              :name=, :uid=, :gid=, :start=, :stop=, :restart=, :dir=,
-                              :pid_file, :pid_file=, :log, :log=, :log_cmd, :log_cmd=,
-                              :err_log, :err_log=, :err_log_cmd, :err_log_cmd=, :alive?, :pid,
-                              :unix_socket, :unix_socket=, :chroot, :chroot=, :env, :env=, :signal
+                              :name=, :uid=, :gid=, :start=, :stop=, :restart=,
+                              :dir=, :pid_file, :pid_file=, :log, :log=,
+                              :log_cmd, :log_cmd=, :err_log, :err_log=,
+                              :err_log_cmd, :err_log_cmd=, :alive?, :pid,
+                              :unix_socket, :unix_socket=, :chroot, :chroot=,
+                              :env, :env=, :signal, :stop_timeout=,
+                              :stop_signal=
     # 
     def initialize
       super

test/configs/stop_options/simple_server.rb

@@ -0,0 +1,12 @@
+#! /usr/bin/env ruby
+
+trap :USR1 do
+
+end
+
+loop do
+  STDOUT.puts('server');
+  STDOUT.flush;
+
+  sleep 10
+end

test/configs/stop_options/stop_options.god

@@ -0,0 +1,39 @@
+God.watch do |w|
+  w.name = 'stop-options'
+  w.start = File.join(GOD_ROOT, *%w[test configs stop_options simple_server.rb])
+  w.stop_signal = 'USR1'
+  w.stop_timeout = 5
+  w.interval = 5
+  w.grace = 2
+
+  w.start_if do |start|
+    start.condition(:process_running) do |c|
+      c.running = false
+    end
+  end
+
+  w.restart_if do |restart|
+    restart.condition(:cpu_usage) do |c|
+      c.above = 30.percent
+      c.times = [3, 5]
+    end
+
+    restart.condition(:memory_usage) do |c|
+      c.above = 10.megabytes
+      c.times = [3, 5]
+    end
+  end
+
+  # lifecycle
+  w.lifecycle do |on|
+    on.condition(:flapping) do |c|
+      c.to_state = [:start, :restart]
+      c.times = 3
+      c.within = 60.seconds
+      c.transition = :unmonitored
+      c.retry_in = 10.seconds
+      c.retry_times = 2
+      c.retry_within = 5.minutes
+    end
+  end
+end