[naily astute] Add remove_nodes methods.

astute/lib/astute/orchestrator.rb

@@ -30,6 +30,12 @@ module Astute
       return
     end
 
+    def remove_nodes(reporter, task_id, nodes)
+      context = Context.new(task_id, reporter)
+      result = simple_remove_nodes(context, nodes)
+      return result
+    end
+
     def verify_networks(reporter, task_id, nodes, networks)
       context = Context.new(task_id, reporter)
       result = @check_network.call(context, nodes, networks)
@@ -40,6 +46,41 @@ module Astute
     end
 
     private
+    def simple_remove_nodes(ctx, nodes)
+      if nodes.empty?
+        Astute.logger.info "#{ctx.task_id}: Nodes to remove are not provided. Do nothing."
+        return {'nodes' => nodes}
+      end
+      uids = nodes.map {|n| n['uid']}
+      Astute.logger.info "#{ctx.task_id}: Starting removing of nodes: #{uids.inspect}"
+      remover = MClient.new(ctx, "erase_node", uids, check_result=false)
+      result = remover.erase_node(:reboot => true)
+      Astute.logger.debug "#{ctx.task_id}: Data resieved from nodes: #{result.inspect}"
+      inaccessible_uids = uids - result.map {|n| n.results[:sender]}
+      error_nodes = []
+      erased_nodes = []
+      result.each do |n|
+        if n.results[:statuscode] != 0
+          error_nodes << {'uid' => n.results[:sender],
+                         'error' => "RPC agent 'erase_node' failed. Result: #{n.results.inspect}"}
+        elsif not n.results[:data][:rebooted]
+          error_nodes << {'uid' => n.results[:sender],
+                         'error' => "RPC method 'erase_node' failed with message: #{n.results[:data][:error_msg]}"}
+        else
+          erased_nodes << {'uid' => n.results[:sender]}
+        end
+      end
+      error_nodes.concat(inaccessible_uids.map {|n| {'uid' => n, 'error' => "Node not answered by RPC."}})
+      if error_nodes.empty?
+        answer = {'nodes' => erased_nodes}
+      else
+        answer = {'status' => 'error', 'nodes' => erased_nodes, 'error_nodes' => error_nodes}
+        Astute.logger.error "#{ctx.task_id}: Removing of nodes #{uids.inspect} ends with errors: #{error_nodes.inspect}"
+      end
+      Astute.logger.info "#{ctx.task_id}: Finished removing of nodes: #{uids.inspect}"
+      return answer
+    end
+
     def deploy_piece(ctx, nodes)
       nodes_roles = nodes.map { |n| { n['uid'] => n['role'] } }
       Astute.logger.info "#{ctx.task_id}: Starting deployment of nodes => roles: #{nodes_roles.inspect}"

mcagent/erase_node.ddl

@@ -6,7 +6,7 @@ metadata    :name        => "Erase node bootloader",
             :url         => "http://mirantis.com",
             :timeout     => 40
 
-action "erase_node", :description => "Zeroing of raw disk" do
+action "erase_node", :description => "Zeroing of boot device" do
     display :always
 end
 

mcagent/erase_node.rb

@@ -5,36 +5,7 @@ module MCollective
   module Agent
     class Erase_node<RPC::Agent
       action "erase_node" do
-        validate :file, String
-        if File.exist?(request[:file])
-          if request.data.key?(:get_data) and request.data[:get_data] != ''
-            header = get_data(request[:file], 512)
-            reply[:erased_data] = Base64.encode64(header)
-          end
-          request_reboot = request.data.key?(:reboot) and request.data[:reboot] != ''
-          if request.data.key?(:dry_run) and request.data[:dry_run] != ''
-            reply[:status] = true
-            reply[:dry_run] = true
-            if request_reboot
-              reply[:reboot] = true
-            end
-          else
-            begin
-              erase_data(request[:file], 512)
-              reply[:status] = true
-              if request_reboot
-                reboot
-                reply[:reboot] = true
-              end
-            rescue Exception => e
-              reply[:status] = false
-              reply[:error_msg] = e.message
-            end
-          end
-        else
-          reply[:status] = false
-          reply[:error_msg] = "File \"#{request[:file]}\" not exist."
-        end
+        erase_node
       end
 
       action "reboot_node" do
@@ -43,8 +14,69 @@ module MCollective
 
       private
 
+      def erase_node
+        request_reboot = request.data[:reboot]
+        dry_run = request.data[:dry_run]
+        error_msg = []
+        reply[:status] = 0  # Shell exitcode behaviour
+
+        begin
+          boot_device = get_boot_device
+          erase_data(boot_device, 512) unless dry_run
+          reply[:erased] = true
+        rescue Exception => e
+          reply[:erased] = false
+          reply[:status] += 1
+          msg = "MBR can't be erased. Reason: #{e.message};"
+          Log.error(msg)
+          error_msg << msg
+        end
+
+        begin
+          reboot if not dry_run and request_reboot
+          reply[:rebooted] = true
+        rescue Exception => e
+          reply[:rebooted] = false
+          reply[:status] += 1
+          msg = "Can't reboot node. Reason: #{e.message};"
+          Log.error(msg)
+          error_msg << "Can't reboot node. Reason: #{e.message};"
+        end
+
+        unless error_msg.empty?
+          reply[:error_msg] = error_msg.join(' ')
+        end
+      end
+
+      def get_boot_device
+        dev_map = '/boot/grub/device.map'
+        grub_conf = '/boot/grub/grub.conf'
+        # Look boot device at GRUB device.map file
+        if File.file?(dev_map) and File.readable?(dev_map)
+          open(dev_map).readlines.each do |l|
+            line = l.strip
+            unless line.start_with?('#')
+              grub_dev, kernel_dev = line.split(%r{[ |\t]+})
+              return kernel_dev if grub_dev == '(hd0)'
+            end
+          end
+        end
+        # Look boot device at GRUB config autogenerated file
+        if File.file?(grub_conf) and File.readable?(grub_conf)
+          open(grub_conf).readlines.each do |l|
+            line = l.strip
+            if line.start_with?('#boot=')
+              grub_dev, kernel_dev = line.split('=')
+              return kernel_dev
+            end
+          end
+        end
+        # If nothing found
+        raise 'Boot device not found.'
+      end
+
       def reboot
-        cmd = "/bin/sleep 5; /sbin/shutdown -r"
+        cmd = "/bin/sleep 5; /sbin/shutdown -r now"
         pid = fork { system(cmd) }
         Process.detach(pid)
       end

naily/lib/naily/dispatcher.rb

@@ -45,6 +45,13 @@ module Naily
       report_result(result, reporter)
     end
 
+    def remove_nodes(data)
+      reporter = Naily::Reporter.new(@producer, data['respond_to'], data['args']['task_uuid'])
+      nodes = data['args']['nodes']
+      result = @orchestrator.remove_nodes(reporter, data['args']['task_uuid'], nodes)
+      report_result(result, reporter)
+    end
+
     private
     def report_result(result, reporter)
       result = {} unless result.instance_of?(Hash)

naily/lib/naily/server.rb

@@ -38,7 +38,7 @@ module Naily
       unless @delegate.respond_to?(data['method'])
         Naily.logger.error "Unsupported RPC call #{data['method']}"
         if data['respond_to']
-          reporter = Naily::Reporter.new(@publisher, data['respond_to'], data['args']['task_uuid'])
+          reporter = Naily::Reporter.new(@producer, data['respond_to'], data['args']['task_uuid'])
           reporter.report({'status' => 'error', 'error' => "Unsupported method '#{data['method']}' called."})
         end
         return
@@ -51,7 +51,7 @@ module Naily
       rescue Exception => e
         Naily.logger.error "Error running RPC method #{data['method']}: #{e.message}, trace: #{e.backtrace.inspect}"
         if data['respond_to']
-          reporter = Naily::Reporter.new(@publisher, data['respond_to'], data['args']['task_uuid'])
+          reporter = Naily::Reporter.new(@producer, data['respond_to'], data['args']['task_uuid'])
           reporter.report({'status' => 'error', 'error' => "Error occured while running method '#{data['method']}'. See logs of Naily for details."})
         end
         return