feature #457: retries is now an option for IMs. Also the number of retries defaults to 0 for all drivers

share/etc/oned.conf

@@ -116,15 +116,19 @@ DEFAULT_DEVICE_PREFIX = "hd"
 
 #-------------------------------------------------------------------------------
 #  KVM Information Driver Manager Configuration
+#    -r number of retries when monitoring a host
+#    -t number of threads, i.e. number of hosts monitored at the same time
 #-------------------------------------------------------------------------------
 IM_MAD = [
       name       = "im_kvm",
       executable = "one_im_ssh",
-      arguments  = "kvm" ]
+      arguments  = "-r 0 -t 15 kvm" ]
 #-------------------------------------------------------------------------------
 
 #-------------------------------------------------------------------------------
 #  XEN Information Driver Manager Configuration
+#    -r number of retries when monitoring a host
+#    -t number of threads, i.e. number of hosts monitored at the same time
 #-------------------------------------------------------------------------------
 #IM_MAD = [
 #    name       = "im_xen",

src/im_mad/im_ssh/one_im_ssh.rb

@@ -32,6 +32,7 @@ $: << RUBY_LIB_LOCATION
 
 require 'OpenNebulaDriver'
 require 'CommandManager'
+require 'getoptlong'
 
 #-------------------------------------------------------------------------------
 # The SSH Information Manager Driver
@@ -41,13 +42,14 @@ class InformationManager < OpenNebulaDriver
     #---------------------------------------------------------------------------
     # Init the driver
     #---------------------------------------------------------------------------
-    def initialize(hypervisor, num)
-        super(num, true)
+    def initialize(hypervisor, threads, retries)
+        super(threads, true)
 
         @config = read_configuration
 
         @hypervisor = hypervisor
         @remote_dir = @config['SCRIPTS_REMOTE_DIR']
+        @retries    = retries
 
         # register actions
         register_action(:MONITOR, method("action_monitor"))
@@ -72,8 +74,11 @@ class InformationManager < OpenNebulaDriver
 
         cmd_string = "#{@remote_dir}/im/run_probes #{@hypervisor} #{host}"      
 
-        cmd = RemotesCommand.run(cmd_string, host, @remote_dir, log_lambda, 2)
-    
+        cmd = RemotesCommand.run(cmd_string, 
+                                 host, 
+                                 @remote_dir, 
+                                 log_lambda, 
+                                 @retries)
         if cmd.code == 0
             send_message("MONITOR", RESULT[:success], number, cmd.stdout)
         else
@@ -90,6 +95,33 @@ end
 #-------------------------------------------------------------------------------
 #-------------------------------------------------------------------------------
 
-hypervisor = ARGV[0]||''
-im = InformationManager.new(hypervisor, 15)
+opts = GetoptLong.new(
+    [ '--retries',    '-r', GetoptLong::OPTIONAL_ARGUMENT ],
+    [ '--threads',    '-t', GetoptLong::OPTIONAL_ARGUMENT ]
+)
+
+hypervisor = ''
+retries    = 0
+threads    = 15
+
+begin
+    opts.each do |opt, arg|
+        case opt
+            when '--retries'
+                retries = arg.to_i
+            when '--threads'
+                threads = arg.to_i
+        end
+    end
+rescue Exception => e
+    exit(-1)
+end 
+
+if ARGV.length >= 1 
+    hypervisor = ARGV.shift
+end
+
+puts retries, threads, hypervisor
+
+im = InformationManager.new(hypervisor, threads, retries)
 im.start_driver

src/mad/ruby/CommandManager.rb

@@ -150,7 +150,7 @@ class RemotesCommand < SSHCommand
     MAGIC_RC = 42
 
     # Creates a command and runs it
-    def self.run(command, host, remote_dir, logger=nil, stdin=nil, retries=1)
+    def self.run(command, host, remote_dir, logger=nil, stdin=nil, retries=0)
         cmd_file = command.split(' ')[0]
 
         cmd_string = "'if [ -x \"#{cmd_file}\" ]; then #{command}; else\

src/mad/ruby/VirtualMachineDriver.rb

@@ -65,10 +65,11 @@ class VirtualMachineDriver < OpenNebulaDriver
     # -------------------------------------------------------------------------
     # Register default actions for the protocol
     # -------------------------------------------------------------------------
-    def initialize(concurrency=10, threaded=true)
+    def initialize(concurrency=10, threaded=true, retries=0)
         super(concurrency,threaded)
 
-        @hosts = Array.new
+        @hosts   = Array.new
+        @retries = retries
 
         register_action(ACTION[:deploy].to_sym, method("deploy"))
         register_action(ACTION[:shutdown].to_sym, method("shutdown"))
@@ -107,14 +108,14 @@ class VirtualMachineDriver < OpenNebulaDriver
     # -------------------------------------------------------------------------
     # Execute a command associated to an action and id in a remote host.
     # -------------------------------------------------------------------------
-    def remotes_action(command, id, host, action, remote_dir, std_in=nil, retries=1)
+    def remotes_action(command, id, host, action, remote_dir, std_in=nil)
+
         command_exe = RemotesCommand.run(command, 
                                          host, 
                                          remote_dir, 
                                          log_method(id),
                                          std_in, 
-                                         retries)
-
+                                         @retries)
         if command_exe.code == 0
             result = :success
             info   = command_exe.stdout