lib/chef_metal/transport/ssh.rb in chef-metal-0.9.1 vs lib/chef_metal/transport/ssh.rb in chef-metal-0.9.2

@@ -23,38 +23,39 @@
       def execute(command, execute_options = {})
         Chef::Log.info("Executing #{options[:prefix]}#{command} on #{username}@#{host}")
         stdout = ''
         stderr = ''
         exitstatus = nil
-        channel = session.open_channel do |channel|
-          # Enable PTY unless otherwise specified, some instances require this
-          unless options[:ssh_pty_enable] == false
-            channel.request_pty do |chan, success|
-               raise "could not get pty" if !success && options[:ssh_pty_enable]
+        session # grab session outside timeout, it has its own timeout
+        with_execute_timeout(execute_options) do
+          channel = session.open_channel do |channel|
+            # Enable PTY unless otherwise specified, some instances require this
+            unless options[:ssh_pty_enable] == false
+              channel.request_pty do |chan, success|
+                 raise "could not get pty" if !success && options[:ssh_pty_enable]
+              end
             end
-          end
 
-          channel.exec("#{options[:prefix]}#{command}") do |ch, success|
-            raise "could not execute command: #{command.inspect}" unless success
+            channel.exec("#{options[:prefix]}#{command}") do |ch, success|
+              raise "could not execute command: #{command.inspect}" unless success
 
-            channel.on_data do |ch2, data|
-              stdout << data
-              stream_chunk(execute_options, data, nil)
-            end
+              channel.on_data do |ch2, data|
+                stdout << data
+                stream_chunk(execute_options, data, nil)
+              end
 
-            channel.on_extended_data do |ch2, type, data|
-              stderr << data
-              stream_chunk(execute_options, nil, data)
-            end
+              channel.on_extended_data do |ch2, type, data|
+                stderr << data
+                stream_chunk(execute_options, nil, data)
+              end
 
-            channel.on_request "exit-status" do |ch, data|
-              exitstatus = data.read_long
+              channel.on_request "exit-status" do |ch, data|
+                exitstatus = data.read_long
+              end
             end
           end
-        end
 
-        with_execute_timeout(execute_options) do
           channel.wait
         end
 
         Chef::Log.info("Completed #{command} on #{username}@#{host}: exit status #{exitstatus}")
         Chef::Log.debug("Stdout was:\n#{stdout}") if stdout != '' && !options[:stream] && !options[:stream_stdout] && Chef::Config.log_level != :debug
@@ -125,23 +126,29 @@
           @session = nil
         end
       end
 
       def available?
-        execute('pwd')
+        # If you can't pwd within 10 seconds, you can't pwd
+        execute('pwd', :timeout => 10)
         true
-      rescue Errno::ETIMEDOUT, Errno::ECONNREFUSED, Errno::ECONNRESET, Net::SSH::AuthenticationFailed, Net::SSH::Disconnect, Net::SSH::HostKeyMismatch
+      rescue Timeout::Error, Errno::ETIMEDOUT, Errno::ECONNREFUSED, Errno::ECONNRESET, Net::SSH::AuthenticationFailed, Net::SSH::Disconnect, Net::SSH::HostKeyMismatch
         Chef::Log.debug("#{username}@#{host} unavailable: could not execute 'pwd' on #{host}: #{$!.inspect}")
         false
       end
 
       protected
 
       def session
         @session ||= begin
           Chef::Log.debug("Opening SSH connection to #{username}@#{host} with options #{ssh_options.inspect}")
-          Net::SSH.start(host, username, ssh_options)
+          # Small initial connection timeout (10s) to help us fail faster when server is just dead
+          begin
+            Net::SSH.start(host, username, { :timeout => 10 }.merge(ssh_options))
+          rescue Timeout::Error
+            raise InitialConnectTimeout.new($!)
+          end
         end
       end
 
       def download(path, local_path)
         channel = Net::SCP.new(session).download(path, local_path)
@@ -180,9 +187,18 @@
             # TODO stdout/stderr is already printed at info/debug level.  Let's not print it twice, it's a lot.
             msg = "Error: command '#{command}' exited with code #{exitstatus}.\n"
             raise msg
           end
         end
+      end
+
+      class InitialConnectTimeout < Timeout::Error
+        def initialize(original_error)
+          super(original_error.message)
+          @original_error = original_error
+        end
+
+        attr_reader :original_error
       end
     end
   end
 end