bosh_agent 1.2411.0 → 1.2416.0

data/lib/bosh_agent/handler.rb

@@ -1,10 +1,9 @@
-# Copyright (c) 2009-2012 VMware, Inc.
-
 module Bosh::Agent
 
   class Handler
     include Bosh::Exec
 
+    attr_accessor :current_long_running_task
     attr_accessor :nats
     attr_reader :processors
 
@@ -17,7 +16,7 @@ module Bosh::Agent
 
     # Seconds  until we kill the agent so it can be restarted:
     KILL_AGENT_THREAD_TIMEOUT_ON_ERRORS = 15 # When there's an unexpected error
-    KILL_AGENT_THREAD_TIMEOUT_ON_RESTART = 1 # When we force a restart 
+    KILL_AGENT_THREAD_TIMEOUT_ON_RESTART = 1 # When we force a restart
 
     def initialize
       @agent_id  = Config.agent_id
@@ -36,7 +35,7 @@ module Bosh::Agent
       @lock = Mutex.new
 
       @results = []
-      @long_running_agent_task = []
+      self.current_long_running_task = {}
       @restarting_agent = false
 
       @nats_fail_count = 0
@@ -55,7 +54,7 @@ module Bosh::Agent
         klazz = Bosh::Agent::Message.const_get(c)
         if klazz.respond_to?(:process)
           # CamelCase -> under_score -> downcased
-          processor_key = c.to_s.gsub(/(.)([A-Z])/,'\1_\2').downcase
+          processor_key = c.to_s.gsub(/(.)([A-Z])/, '\1_\2').downcase
           @processors[processor_key] = klazz
         end
       end
@@ -66,12 +65,13 @@ module Bosh::Agent
       @processors[method]
     end
 
+    # rubocop:disable MethodLength
     def start
-      ['TERM', 'INT', 'QUIT'].each { |s| trap(s) { shutdown } }
+      %w(TERM INT QUIT).each { |s| trap(s) { shutdown } }
 
       EM.run do
         begin
-          @nats = NATS.connect(:uri => @nats_uri, :autostart => false) { on_connect }
+          @nats = NATS.connect(uri: @nats_uri, autostart: false) { on_connect }
           Config.nats = @nats
         rescue Errno::ENETUNREACH, Timeout::Error => e
           @logger.info("Unable to talk to nats - retry (#{e.inspect})")
@@ -82,12 +82,12 @@ module Bosh::Agent
         setup_heartbeats
 
         if @process_alerts
-          if (@smtp_port.nil? || @smtp_user.nil? || @smtp_password.nil?)
-            @logger.error "Cannot start alert processor without having SMTP port, user and password configured"
-            @logger.error "Agent will be running but alerts will NOT be properly processed"
+          if @smtp_port.nil? || @smtp_user.nil? || @smtp_password.nil?
+            @logger.error 'Cannot start alert processor without having SMTP port, user and password configured'
+            @logger.error 'Agent will be running but alerts will NOT be properly processed'
           else
             @logger.debug("SMTP: #{@smtp_password}")
-            @processor = Bosh::Agent::AlertProcessor.start("127.0.0.1", @smtp_port, @smtp_user, @smtp_password)
+            @processor = Bosh::Agent::AlertProcessor.start('127.0.0.1', @smtp_port, @smtp_user, @smtp_password)
             setup_syslog_monitor
           end
         end
@@ -100,10 +100,14 @@ module Bosh::Agent
       retry if @nats_fail_count < MAX_NATS_RETRIES
       @logger.fatal("Unable to reconnect to NATS after #{MAX_NATS_RETRIES} retries, exiting...")
     end
+    # rubocop:enable MethodLength
 
     def shutdown
-      @logger.info("Exit")
-      NATS.stop { EM.stop; exit }
+      @logger.info('Exit')
+      NATS.stop do
+        EM.stop
+        exit
+      end
     end
 
     def on_connect
@@ -118,7 +122,7 @@ module Bosh::Agent
         @hbp.enable(interval)
         @logger.info("Heartbeats are enabled and will be sent every #{interval} seconds")
       else
-        @logger.warn("Heartbeats are disabled")
+        @logger.warn('Heartbeats are disabled')
       end
     end
 
@@ -126,10 +130,11 @@ module Bosh::Agent
       Bosh::Agent::SyslogMonitor.start(@nats, @agent_id)
     end
 
+    # rubocop:disable MethodLength
     def handle_message(json)
       msg = Yajl::Parser.new.parse(json)
 
-      unless msg["reply_to"]
+      unless msg['reply_to']
         @logger.info("Missing reply_to in: #{msg}")
         return
       end
@@ -145,8 +150,8 @@ module Bosh::Agent
       method = msg['method']
       args = msg['arguments']
 
-      if method == "get_state"
-        method = "state"
+      if method == 'get_state'
+        method = 'state'
       end
 
       processor = lookup(method)
@@ -154,9 +159,11 @@ module Bosh::Agent
         EM.defer do
           process_in_thread(processor, reply_to, method, args)
         end
-      elsif method == "get_task"
+      elsif method == 'cancel_task'
+        handle_cancel_task(reply_to, args.first)
+      elsif method == 'get_task'
         handle_get_task(reply_to, args.first)
-      elsif method == "shutdown"
+      elsif method == 'shutdown'
         handle_shutdown(reply_to)
       else
         re = RemoteException.new("unknown message #{msg.inspect}")
@@ -165,19 +172,21 @@ module Bosh::Agent
     rescue Yajl::ParseError => e
       @logger.info("Failed to parse message: #{json}: #{e.inspect}: #{e.backtrace}")
     end
+    # rubocop:enable MethodLength
 
+    # rubocop:disable MethodLength
     def process_in_thread(processor, reply_to, method, args)
       if processor.respond_to?(:long_running?)
         if @restarting_agent
-          exception = RemoteException.new("restarting agent")
+          exception = RemoteException.new('restarting agent')
           publish(reply_to, exception.to_hash)
         else
           @lock.synchronize do
-            if @long_running_agent_task.empty?
-              process_long_running(reply_to, processor, args)
-            else
-              exception = RemoteException.new("already running long running task")
+            if current_long_running_task[:task_id]
+              exception = RemoteException.new('already running long running task')
               publish(reply_to, exception.to_hash)
+            else
+              process_long_running(reply_to, processor, args)
             end
           end
         end
@@ -198,17 +207,32 @@ module Bosh::Agent
       # log an error as this would otherwise be lost
       @logger.error("#{processor.to_s}: #{e.message}\n#{e.backtrace.join("\n")}")
     end
+    # rubocop:enable MethodLength
+
+    def handle_cancel_task(reply_to, agent_task_id)
+      if current_long_running_task?(agent_task_id)
+        if current_long_running_task[:processor].respond_to?(:cancel)
+          current_long_running_task[:processor].cancel
+          publish(reply_to, { 'value' => 'canceled' })
+          self.current_long_running_task = {}
+        else
+          publish(reply_to, { 'exception' => "could not cancel task #{agent_task_id}" })
+        end
+      else
+        publish(reply_to, { 'exception' => 'unknown agent_task_id' })
+      end
+    end
 
     def handle_get_task(reply_to, agent_task_id)
-      if @long_running_agent_task == [agent_task_id]
-        publish(reply_to, {"value" => {"state" => "running", "agent_task_id" => agent_task_id}})
+      if current_long_running_task?(agent_task_id)
+        publish(reply_to, { 'value' => { 'state' => 'running', 'agent_task_id' => agent_task_id } })
       else
         rs = @results.find { |time, task_id, result| task_id == agent_task_id }
         if rs
-          time, task_id, result = rs
+          _, _, result = rs
           publish(reply_to, result)
         else
-          publish(reply_to, {"exception" => "unknown agent_task_id" })
+          publish(reply_to, { 'exception' => 'unknown agent_task_id' })
         end
       end
     end
@@ -230,7 +254,7 @@ module Bosh::Agent
           @nats.publish(reply_to, json, &blk)
         end
       else
-        msg = "message > NATS_MAX_PAYLOAD, stored in blobstore"
+        msg = 'message > NATS_MAX_PAYLOAD, stored in blobstore'
         exception = RemoteException.new(msg, nil, unencrypted)
         @logger.fatal(msg)
         EM.next_tick do
@@ -242,14 +266,14 @@ module Bosh::Agent
     def process_long_running(reply_to, processor, args)
       agent_task_id = generate_agent_task_id
 
-      @long_running_agent_task = [agent_task_id]
+      self.current_long_running_task = { task_id: agent_task_id, processor: processor }
 
-      payload = {:value => {:state => "running", :agent_task_id => agent_task_id}}
+      payload = { value: { state: 'running', agent_task_id: agent_task_id } }
       publish(reply_to, payload)
 
       result = process(processor, args)
       @results << [Time.now.to_i, agent_task_id, result]
-      @long_running_agent_task = []
+      self.current_long_running_task = {}
     end
 
     def kill_main_thread_in(seconds)
@@ -261,17 +285,17 @@ module Bosh::Agent
     end
 
     def process(processor, args)
-      begin
-        result = processor.process(args)
-        return {:value => result}
-      rescue Bosh::Agent::Error => e
-        @logger.info("#{e.inspect}: #{e.backtrace}")
-        return RemoteException.from(e).to_hash
-      rescue Exception => e
-        kill_main_thread_in(KILL_AGENT_THREAD_TIMEOUT_ON_ERRORS)
-        @logger.error("#{e.inspect}: #{e.backtrace}")
-        return {:exception => "#{e.inspect}: #{e.backtrace}"}
-      end
+      result = processor.process(args)
+      return { value: result }
+    rescue Bosh::Agent::Error => e
+      @logger.info("#{e.inspect}: #{e.backtrace}")
+      return RemoteException.from(e).to_hash
+    # rubocop:disable RescueException
+    rescue Exception => e
+    # rubocop:enable RescueException
+      kill_main_thread_in(KILL_AGENT_THREAD_TIMEOUT_ON_ERRORS)
+      @logger.error("#{e.inspect}: #{e.backtrace}")
+      return { exception: "#{e.inspect}: #{e.backtrace}" }
     end
 
     def generate_agent_task_id
@@ -279,33 +303,34 @@ module Bosh::Agent
     end
 
     ##
-    # When there's a network change on an existing vm, director sends a prepare_network_change message to the vm      
+    # When there's a network change on an existing vm, director sends a prepare_network_change message to the vm
     # agent. After agent replies to director with a `true` message, the post_prepare_network_change method is called
     # (via EM callback).
     #
-    # The post_prepare_network_change  method will delete the udev network persistent rules, delete the agent settings 
-    # and then it should restart the agent to get the new agent settings (set by director-cpi). For a simple network 
-    # change (i.e. dns changes) this is enough, as when the agent is restarted it will apply the new network settings. 
-    # But for other network changes (i.e. IP change), the CPI will be responsible to reboot or recreate the vm if needed.    
+    # The post_prepare_network_change  method will delete the udev network persistent rules, delete the agent settings
+    # and then it should restart the agent to get the new agent settings (set by director-cpi). For a simple network
+    # change (i.e. dns changes) this is enough, as when the agent is restarted it will apply the new network settings.
+    # But for other network changes (i.e. IP change), the CPI will be responsible to reboot or recreate the vm if
+    # needed.
     def post_prepare_network_change
       if Bosh::Agent::Config.configure
         udev_file = '/etc/udev/rules.d/70-persistent-net.rules'
         if File.exist?(udev_file)
-          @logger.info("deleting 70-persistent-net.rules - again")
+          @logger.info('deleting 70-persistent-net.rules - again')
           File.delete(udev_file)
         end
-        @logger.info("Removing settings.json")
+        @logger.info('Removing settings.json')
         settings_file = Bosh::Agent::Config.settings_file
         File.delete(settings_file)
       end
 
-      @logger.info("Restarting agent to prepare for a network change")
+      @logger.info('Restarting agent to prepare for a network change')
       kill_main_thread_in(KILL_AGENT_THREAD_TIMEOUT_ON_RESTART)
     end
 
     def handle_shutdown(reply_to)
       @logger.info("Shutting down #{URI.parse(Config.mbus).scheme.upcase} connection")
-      payload = {:value => "shutdown"}
+      payload = { value: 'shutdown' }
 
       if Bosh::Agent::Config.configure
         # We should never come back up again
@@ -330,30 +355,30 @@ module Bosh::Agent
     end
 
     def decrypt(msg)
-      [ "session_id", "encrypted_data" ].each do |key|
+      %w(session_id encrypted_data).each do |key|
         unless msg.key?(key)
           @logger.info("Missing #{key} in #{msg}")
           return
         end
       end
 
-      message_session_id = msg["session_id"]
-      reply_to = msg["reply_to"]
+      message_session_id = msg['session_id']
+      reply_to = msg['reply_to']
 
-      encryption_handler = lookup_encryption_handler(:session_id => message_session_id)
+      encryption_handler = lookup_encryption_handler(session_id: message_session_id)
 
       # save message handler for the reply
       @session_reply_map[reply_to] = encryption_handler
 
       # Log exceptions from the EncryptionHandler, but stay quiet on the wire.
       begin
-        msg = encryption_handler.decrypt(msg["encrypted_data"])
+        msg = encryption_handler.decrypt(msg['encrypted_data'])
       rescue Bosh::Core::EncryptionHandler::CryptError => e
         log_encryption_error(e)
         return
       end
 
-      msg["reply_to"] = reply_to
+      msg['reply_to'] = reply_to
 
       @logger.info("Decrypted Message: #{msg}")
       msg
@@ -364,17 +389,21 @@ module Bosh::Agent
     end
 
     def encrypt(reply_to, payload)
-      encryption_handler = lookup_encryption_handler(:reply_to => reply_to)
+      encryption_handler = lookup_encryption_handler(reply_to: reply_to)
       session_id = encryption_handler.session_id
 
       payload = {
-        "session_id" => session_id,
-        "encrypted_data" => encryption_handler.encrypt(payload)
+        'session_id' => session_id,
+        'encrypted_data' => encryption_handler.encrypt(payload)
       }
 
       payload
     end
 
+    def current_long_running_task?(agent_task_id)
+      current_long_running_task[:task_id] == agent_task_id
+    end
+
   end
 
   # Built-in message handlers
@@ -382,13 +411,13 @@ module Bosh::Agent
 
     class Ping
       def self.process(args)
-        "pong"
+        'pong'
       end
     end
 
     class Noop
       def self.process(args)
-        "nope"
+        'nope'
       end
     end
 
@@ -399,7 +428,7 @@ module Bosh::Agent
           Bosh::Agent::Monit.start_services
         end
 
-        "started"
+        'started'
 
       rescue => e
         raise Bosh::Agent::MessageHandlerError, "Cannot start job: #{e}"
@@ -418,7 +447,7 @@ module Bosh::Agent
           Bosh::Agent::Monit.stop_services
         end
 
-        "stopped"
+        'stopped'
 
       rescue => e
         # Monit retry logic should make it really hard to get here but if it happens we should yell.

data/lib/bosh_agent/message/run_errand.rb

@@ -3,10 +3,23 @@ require 'open3'
 module Bosh::Agent
   module Message
     class RunErrand
+      CANCEL_GRACE_PERIOD_SECONDS = 30
+
       def self.process(args)
         self.new(args).start
       end
 
+      def self.cancel
+        pid = running_errand_pid
+        Process.kill('-TERM', pid) if errand_running?
+        CANCEL_GRACE_PERIOD_SECONDS.times do
+          break unless errand_running?
+          sleep 1
+        end
+        Process.kill('-KILL', pid) if errand_running?
+      rescue Errno::ESRCH
+      end
+
       def self.long_running?
         true
       end
@@ -27,9 +40,9 @@ module Bosh::Agent
 
         job_template_name = job_templates.first.fetch('name')
 
-        env  = { 'PATH' => '/usr/sbin:/usr/bin:/sbin:/bin' }
-        cmd  = "#{@base_dir}/jobs/#{job_template_name}/bin/run"
-        opts = { :unsetenv_others => true }
+        env = { 'PATH' => '/usr/sbin:/usr/bin:/sbin:/bin', 'TMPDIR' => ENV['TMPDIR'] }
+        cmd = "#{@base_dir}/jobs/#{job_template_name}/bin/run"
+        opts = { unsetenv_others: true, pgroup: true }
 
         unless File.executable?(cmd)
           raise Bosh::Agent::MessageHandlerError,
@@ -37,9 +50,20 @@ module Bosh::Agent
         end
 
         begin
-          stdout, stderr, status = Open3.capture3(env, cmd, opts)
+          stdout, stderr, status = Open3.popen3(env, cmd, opts) { |i, o, e, t|
+            self.class.running_errand_pid = t.pid
+
+            out_reader = Thread.new { o.read }
+            err_reader = Thread.new { e.read }
+
+            i.close
+
+            [out_reader.value, err_reader.value, t.value]
+          }
+          self.class.running_errand_pid = nil
+
           {
-            'exit_code' => status.exitstatus,
+            'exit_code' => extract_status_code(status),
             'stdout' => stdout,
             'stderr' => stderr,
           }
@@ -48,6 +72,22 @@ module Bosh::Agent
           raise Bosh::Agent::MessageHandlerError, e.inspect
         end
       end
+
+      def extract_status_code(status)
+        status.exitstatus || (status.termsig + 128)
+      end
+
+      class << self
+        attr_accessor :running_errand_pid
+      end
+
+      def self.errand_running?
+        return false unless running_errand_pid
+        Process.kill(0, running_errand_pid)
+        true
+      rescue Errno::ESRCH
+        false
+      end
     end
   end
 end

data/lib/bosh_agent/version.rb

@@ -2,7 +2,7 @@
 
 module Bosh
   module Agent
-    VERSION = '1.2411.0'
+    VERSION = '1.2416.0'
     BOSH_PROTOCOL = "1"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: bosh_agent
 version: !ruby/object:Gem::Version
-  version: 1.2411.0
+  version: 1.2416.0
   prerelease: 
 platform: ruby
 authors:
@@ -9,7 +9,7 @@ authors:
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2014-04-16 00:00:00.000000000 Z
+date: 2014-04-17 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: netaddr
@@ -162,7 +162,7 @@ dependencies:
     requirements:
     - - ~>
       - !ruby/object:Gem::Version
-        version: 1.2411.0
+        version: 1.2416.0
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
@@ -170,7 +170,7 @@ dependencies:
     requirements:
     - - ~>
       - !ruby/object:Gem::Version
-        version: 1.2411.0
+        version: 1.2416.0
 - !ruby/object:Gem::Dependency
   name: bosh_common
   requirement: !ruby/object:Gem::Requirement
@@ -178,7 +178,7 @@ dependencies:
     requirements:
     - - ~>
       - !ruby/object:Gem::Version
-        version: 1.2411.0
+        version: 1.2416.0
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
@@ -186,7 +186,7 @@ dependencies:
     requirements:
     - - ~>
       - !ruby/object:Gem::Version
-        version: 1.2411.0
+        version: 1.2416.0
 - !ruby/object:Gem::Dependency
   name: blobstore_client
   requirement: !ruby/object:Gem::Requirement
@@ -194,7 +194,7 @@ dependencies:
     requirements:
     - - ~>
       - !ruby/object:Gem::Version
-        version: 1.2411.0
+        version: 1.2416.0
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
@@ -202,7 +202,7 @@ dependencies:
     requirements:
     - - ~>
       - !ruby/object:Gem::Version
-        version: 1.2411.0
+        version: 1.2416.0
 - !ruby/object:Gem::Dependency
   name: rspec
   requirement: !ruby/object:Gem::Requirement
@@ -238,7 +238,7 @@ dependencies:
 description: ! 'This agent listens for instructions from the bosh director on each
   server that bosh manages.
 
-  a126f7'
+  7ca225'
 email: support@cloudfoundry.com
 executables:
 - bosh_agent
@@ -344,7 +344,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
       segments:
       - 0
-      hash: 19318725305148884
+      hash: 3410278131475347007
 requirements: []
 rubyforge_project: 
 rubygems_version: 1.8.23