bosh-director 1.3215.4.0 → 1.3232.0

data/lib/bosh/director/deployment_plan/update_config.rb

@@ -68,7 +68,7 @@ module Bosh::Director
           'max_in_flight' => @max_in_flight,
           'canary_watch_time' => "#{@min_canary_watch_time}-#{@max_canary_watch_time}",
           'update_watch_time' => "#{@min_update_watch_time}-#{@max_update_watch_time}",
-          'serial' => @serial
+          'serial' => serial?
         }
       end
 

data/lib/bosh/director/disk_manager.rb

@@ -7,7 +7,7 @@ module Bosh::Director
       @transactor = Transactor.new
     end
 
-    def update_persistent_disk(instance_plan, vm_recreator)
+    def update_persistent_disk(instance_plan)
       @logger.info('Updating persistent disk')
       check_persistent_disk(instance_plan)
 
@@ -18,7 +18,7 @@ module Bosh::Director
 
       disk = nil
       if instance_plan.needs_disk?
-        disk = create_and_attach_disk(instance_plan, vm_recreator)
+        disk = create_and_attach_disk(instance_plan)
         mount_and_migrate_disk(instance, disk, old_disk)
       end
 
@@ -52,19 +52,25 @@ module Bosh::Director
 
     def orphan_disk(disk)
       @transactor.retryable_transaction(Bosh::Director::Config.db) do
-        orphan_disk = Models::OrphanDisk.create(
-          disk_cid: disk.disk_cid,
-          size: disk.size,
-          availability_zone: disk.instance.availability_zone,
-          deployment_name: disk.instance.deployment.name,
-          instance_name: "#{disk.instance.job}/#{disk.instance.uuid}",
-          cloud_properties: disk.cloud_properties
-        )
-
-        orphan_snapshots(disk.snapshots, orphan_disk)
-        @logger.info("Orphaning disk: '#{disk.disk_cid}', #{disk.active ? "active" : "inactive"}")
-
-        disk.destroy
+        begin
+          parent_id = add_event('delete', disk.instance.deployment.name, "#{disk.instance.job}/#{disk.instance.uuid}", disk.disk_cid)
+          orphan_disk = Models::OrphanDisk.create(
+              disk_cid:          disk.disk_cid,
+              size:              disk.size,
+              availability_zone: disk.instance.availability_zone,
+              deployment_name:   disk.instance.deployment.name,
+              instance_name:     "#{disk.instance.job}/#{disk.instance.uuid}",
+              cloud_properties:  disk.cloud_properties
+          )
+
+          orphan_snapshots(disk.snapshots, orphan_disk)
+          @logger.info("Orphaning disk: '#{disk.disk_cid}', #{disk.active ? "active" : "inactive"}")
+          disk.destroy
+        rescue Exception => e
+          raise e
+        ensure
+          add_event('delete', orphan_disk.deployment_name, orphan_disk.instance_name, orphan_disk.disk_cid, parent_id, e)
+        end
       end
     end
 
@@ -150,7 +156,7 @@ module Bosh::Director
       rescue Bosh::Clouds::DiskNotAttached
         if disk.active
           raise CloudDiskNotAttached,
-                "`#{instance_model}' VM should have persistent disk attached " +
+                "'#{instance_model}' VM should have persistent disk attached " +
                     "but it doesn't (according to CPI)"
         end
       end
@@ -173,6 +179,22 @@ module Bosh::Director
 
     private
 
+    def add_event(action, deployment_name, instance_name, object_name = nil, parent_id = nil, error = nil)
+      event  = Config.current_job.event_manager.create_event(
+          {
+              parent_id:   parent_id,
+              user:        Config.current_job.username,
+              action:      action,
+              object_type: 'disk',
+              object_name: object_name,
+              deployment:  deployment_name,
+              instance:    instance_name,
+              task:        Config.current_job.task_id,
+              error:       error
+          })
+      event.id
+    end
+
     def orphan_mounted_persistent_disk(instance_model, disk)
       unmount_disk(instance_model, disk)
 
@@ -223,14 +245,14 @@ module Bosh::Director
         @logger.debug('Disk is already detached')
       elsif agent_disk_cid != instance.model.persistent_disk_cid
         raise AgentDiskOutOfSync,
-          "`#{instance}' has invalid disks: agent reports " +
-            "`#{agent_disk_cid}' while director record shows " +
-            "`#{instance.model.persistent_disk_cid}'"
+          "'#{instance}' has invalid disks: agent reports " +
+            "'#{agent_disk_cid}' while director record shows " +
+            "'#{instance.model.persistent_disk_cid}'"
       end
 
       instance.model.persistent_disks.each do |disk|
         unless disk.active
-          @logger.warn("`#{instance}' has inactive disk #{disk.disk_cid}")
+          @logger.warn("'#{instance}' has inactive disk #{disk.disk_cid}")
         end
       end
     end
@@ -243,28 +265,11 @@ module Bosh::Director
       AgentClient.with_vm_credentials_and_agent_id(instance_model.credentials, instance_model.agent_id)
     end
 
-    def create_and_attach_disk(instance_plan, vm_recreator)
+    def create_and_attach_disk(instance_plan)
       instance = instance_plan.instance
       disk = create_disk(instance_plan)
       @cloud.attach_disk(instance.model.vm_cid, disk.disk_cid)
-      return disk
-    rescue Bosh::Clouds::NoDiskSpace => e
-      if e.ok_to_retry
-        @logger.warn('Retrying attach disk operation after persistent disk update failed')
-        # Re-creating the vm may cause it to be re-created in a place with more storage
-        unmount_disk_for(instance_plan)
-        vm_recreator.recreate_vm(instance_plan, disk.disk_cid)
-        begin
-          @cloud.attach_disk(instance.model.vm_cid, disk.disk_cid)
-        rescue
-          orphan_disk(disk)
-          raise
-        end
-      else
-        orphan_disk(disk)
-        raise
-      end
-      return disk
+      disk
     end
 
     def mount_and_migrate_disk(instance, new_disk, old_disk)
@@ -282,7 +287,7 @@ module Bosh::Director
     def create_disk(instance_plan)
       job = instance_plan.desired_instance.job
       instance_model = instance_plan.instance.model
-
+      parent_id = add_event('create', instance_model.deployment.name, "#{instance_model.job}/#{instance_model.uuid}")
       disk_size = job.persistent_disk_type.disk_size
       cloud_properties = job.persistent_disk_type.cloud_properties
       disk_cid = @cloud.create_disk(disk_size, cloud_properties, instance_model.vm_cid)
@@ -294,6 +299,10 @@ module Bosh::Director
         size: disk_size,
         cloud_properties: cloud_properties,
       )
+    rescue Exception => e
+      raise e
+    ensure
+      add_event('create', instance_model.deployment.name, "#{instance_model.job}/#{instance_model.uuid}", disk_cid, parent_id, e)
     end
   end
 end

data/lib/bosh/director/dns/canonicalizer.rb

@@ -16,11 +16,11 @@ module Bosh::Director
     def self.validate_dns_name(string)
       if string =~ /^(\d|-)/
         raise DnsInvalidCanonicalName,
-              "Invalid DNS canonical name `#{string}', must begin with a letter"
+              "Invalid DNS canonical name '#{string}', must begin with a letter"
       end
       if string =~ /-$/
         raise DnsInvalidCanonicalName,
-              "Invalid DNS canonical name `#{string}', can't end with a hyphen"
+              "Invalid DNS canonical name '#{string}', can't end with a hyphen"
       end
       string
     end

data/lib/bosh/director/dns/dns_manager.rb

@@ -35,7 +35,7 @@ module Bosh::Director
           dns = NetAddr::CIDR.create(dns)
           unless dns.size == 1
             raise NetworkInvalidDns,
-              "Invalid DNS for network `#{network}': must be a single IP"
+              "Invalid DNS for network '#{network}': must be a single IP"
           end
 
           servers << dns.ip
@@ -160,7 +160,7 @@ module Bosh::Director
           dns = NetAddr::CIDR.create(dns)
           unless dns.size == 1
             raise NetworkInvalidDns,
-              "Invalid DNS for network `#{network}': must be a single IP"
+              "Invalid DNS for network '#{network}': must be a single IP"
           end
 
           servers << dns.ip

data/lib/bosh/director/dns/powerdns.rb

@@ -34,7 +34,7 @@ module Bosh::Director
     end
 
     def find_dns_records_by_pattern(record_pattern)
-      records = Models::Dns::Record.filter(:name.like(record_pattern))
+      records = Models::Dns::Record.grep(:name, record_pattern)
       records.filter(:domain_id => find_domain_id).all
     end
 
@@ -45,7 +45,7 @@ module Bosh::Director
       ips = []
       records.each do |record|
         ips << record.content
-        Models::Dns::Record.filter(:content.like(record.name)).each do |ptr|
+        Models::Dns::Record.grep(:content, record.name).each do |ptr|
           @logger.info("Deleting reverse DNS record: #{ptr.name} -> #{ptr.content}")
           ptr.destroy
         end

data/lib/bosh/director/download_helper.rb

@@ -13,7 +13,7 @@ module Bosh::Director
     def download_remote_file(resource, remote_file, local_file, num_redirects = 0)
       @logger.info("Downloading remote #{resource} from #{remote_file}") if @logger
       uri = URI.parse(remote_file)
-      Net::HTTP.start(uri.host, uri.port, 
+      Net::HTTP.start(uri.host, uri.port,
                       :use_ssl => uri.scheme == 'https',
                       :verify_mode => OpenSSL::SSL::VERIFY_NONE) do |http|
         http.request_get(uri.request_uri) do |response|
@@ -26,23 +26,23 @@ module Bosh::Director
               end
 
             when Net::HTTPFound
-              raise ResourceError, "Too many redirects at `#{remote_file}'." if num_redirects >= 9
+              raise ResourceError, "Too many redirects at '#{remote_file}'." if num_redirects >= 9
               location = response.header['location']
-              raise ResourceError, "No location header for redirect found at `#{remote_file}'." if location.nil?
+              raise ResourceError, "No location header for redirect found at '#{remote_file}'." if location.nil?
 
               location = URI.join(uri, location).to_s
               download_remote_file(resource, location, local_file, num_redirects + 1)
 
             when Net::HTTPNotFound
               @logger.error("Downloading remote #{resource} from #{remote_file} failed: #{response.message}") if @logger
-              raise ResourceNotFound, "No #{resource} found at `#{remote_file}'."
+              raise ResourceNotFound, "No #{resource} found at '#{remote_file}'."
 
             else
               @logger.error("Downloading remote #{resource} from #{remote_file} failed: #{response.message}") if @logger
               raise ResourceError, "Downloading remote #{resource} failed. Check task debug log for details."
           end
         end
-      end       
+      end
     rescue URI::Error, SocketError, Timeout::Error, Errno::EINVAL, Errno::ECONNRESET, Errno::ECONNREFUSED, EOFError,
            Net::HTTPBadResponse, Net::HTTPHeaderSyntaxError, Net::ProtocolError => e
       @logger.error("Downloading remote #{resource} from #{remote_file} failed: #{e.inspect}") if @logger

data/lib/bosh/director/errand/job_manager.rb

@@ -3,21 +3,20 @@ module Bosh::Director
     # @param [Bosh::Director::DeploymentPlan::Planner] deployment
     # @param [Bosh::Director::DeploymentPlan::Job] job
     # @param [Bosh::Clouds] cloud
-    # @param [Bosh::Director::EventLog::Log] event_log
     # @param [Logger] logger
-    def initialize(deployment, job, cloud, event_log, logger)
+    def initialize(deployment, job, cloud, logger)
       @deployment = deployment
       @job = job
-      @event_log = event_log
       @logger = logger
       vm_deleter = Bosh::Director::VmDeleter.new(cloud, logger)
       @disk_manager = DiskManager.new(cloud, logger)
       @job_renderer = JobRenderer.create
-      @vm_creator = Bosh::Director::VmCreator.new(cloud, logger, vm_deleter, @disk_manager, @job_renderer)
+      arp_flusher = ArpFlusher.new
+      @vm_creator = Bosh::Director::VmCreator.new(cloud, logger, vm_deleter, @disk_manager, @job_renderer, arp_flusher)
     end
 
     def create_missing_vms
-      @vm_creator.create_for_instance_plans(@job.instance_plans_with_missing_vms, @deployment.ip_provider, @event_log)
+      @vm_creator.create_for_instance_plans(@job.instance_plans_with_missing_vms, @deployment.ip_provider)
     end
 
     # Creates/updates all errand job instances
@@ -38,7 +37,7 @@ module Bosh::Director
       end
 
       @logger.info('Deleting errand instances')
-      event_log_stage = @event_log.begin_stage('Deleting errand instances', instance_plans.size, [@job.name])
+      event_log_stage = Config.event_log.begin_stage('Deleting errand instances', instance_plans.size, [@job.name])
       dns_manager = DnsManagerProvider.create
       instance_deleter = InstanceDeleter.new(@deployment.ip_provider, dns_manager, @disk_manager)
       instance_deleter.delete_instance_plans(instance_plans, event_log_stage)

data/lib/bosh/director/errand/result.rb

@@ -29,7 +29,7 @@ module Bosh::Director
     end
 
     def short_description(job_name)
-      title_prefix = "Errand `#{job_name}'"
+      title_prefix = "Errand '#{job_name}'"
       exit_code_suffix = "(exit code #{@exit_code})"
 
       if @exit_code == 0

data/lib/bosh/director/errand/runner.rb

@@ -5,11 +5,10 @@ module Bosh::Director
     # @param [Bosh::Director::Api::InstanceManager] instance_manager
     # @param [Bosh::Director::EventLog::Log] event_log
     # @param [Bosh::Director::LogsFetcher] logs_fetcher
-    def initialize(job, result_file, instance_manager, event_log, logs_fetcher)
+    def initialize(job, result_file, instance_manager, logs_fetcher)
       @job = job
       @result_file = result_file
       @instance_manager = instance_manager
-      @event_log = event_log
       @agent_task_id = nil
       @logs_fetcher = logs_fetcher
     end
@@ -22,7 +21,7 @@ module Bosh::Director
       end
 
       agent_task_result = nil
-      event_log_stage = @event_log.begin_stage('Running errand', 1)
+      event_log_stage = Config.event_log.begin_stage('Running errand', 1)
 
       begin
         event_log_stage.advance_and_track("#{@job.name}/#{instance.index}") do

data/lib/bosh/director/event_log.rb

@@ -38,16 +38,10 @@ module Bosh::Director
             layout: Logging.layouts.pattern(:pattern => '%m\n')
           ))
         end
-
-        @last_stage = Stage.new(self, 'unknown', [], 0)
       end
 
       def begin_stage(stage_name, total = nil, tags = [])
-        @last_stage = Stage.new(self, stage_name, tags, total)
-      end
-
-      def track(task_name = nil, &blk)
-        @last_stage.advance_and_track(task_name, &blk)
+        Stage.new(self, stage_name, tags, total)
       end
 
       def warn_deprecated(message)

data/lib/bosh/director/ext.rb

@@ -1,9 +1,3 @@
-module Resque
-  def redis
-    Bosh::Director::Config.redis
-  end
-end
-
 class Object
   def to_openstruct
     self

data/lib/bosh/director/instance_deleter.rb

@@ -15,9 +15,10 @@ module Bosh::Director
 
     def delete_instance_plan(instance_plan, event_log_stage)
       instance_model = instance_plan.new? ? instance_plan.instance.model : instance_plan.existing_instance
-
+      deployment_name = instance_model.deployment.name
+      instance_name = instance_model.name
+      parent_id = add_event(deployment_name, instance_name)
       event_log_stage.advance_and_track(instance_model.to_s) do
-
         error_ignorer.with_force_check do
           stop(instance_plan)
         end
@@ -46,6 +47,10 @@ module Bosh::Director
 
         instance_model.destroy
       end
+    rescue Exception => e
+      raise e
+    ensure
+      add_event(deployment_name, instance_name, parent_id, e) if parent_id
     end
 
     def delete_instance_plans(instance_plans, event_log_stage, options = {})
@@ -59,6 +64,22 @@ module Bosh::Director
 
     private
 
+    def add_event(deployment_name, instance_name, parent_id = nil, error = nil)
+      event  = Config.current_job.event_manager.create_event(
+          {
+              parent_id:   parent_id,
+              user:        Config.current_job.username,
+              action:      'delete',
+              object_type: 'instance',
+              object_name: instance_name,
+              task:        Config.current_job.task_id,
+              deployment:  deployment_name,
+              instance:    instance_name,
+              error:       error
+          })
+      event.id
+    end
+
     def stop(instance_plan)
       stopper = Stopper.new(instance_plan, 'stopped', Config, @logger)
       stopper.stop

data/lib/bosh/director/instance_updater.rb

@@ -10,7 +10,8 @@ module Bosh::Director
       vm_deleter = VmDeleter.new(cloud, logger)
       disk_manager = DiskManager.new(cloud, logger)
       job_renderer = JobRenderer.create
-      vm_creator = VmCreator.new(cloud, logger, vm_deleter, disk_manager, job_renderer)
+      arp_flusher = ArpFlusher.new
+      vm_creator = VmCreator.new(cloud, logger, vm_deleter, disk_manager, job_renderer, arp_flusher)
       vm_recreator = VmRecreator.new(vm_creator, vm_deleter)
       dns_manager = DnsManagerProvider.create
       new(
@@ -41,6 +42,8 @@ module Bosh::Director
 
     def update(instance_plan, options = {})
       instance = instance_plan.instance
+      action, context = get_action_and_context(instance_plan)
+      parent_id = add_event(instance.deployment_model.name, action, instance.model.name, context) if instance_plan.changed?
       @logger.info("Updating instance #{instance}, changes: #{instance_plan.changes.to_a.join(', ').inspect}")
 
       InstanceUpdater::InstanceState.with_instance_update(instance.model) do
@@ -72,16 +75,23 @@ module Bosh::Director
 
         recreated = false
         if needs_recreate?(instance_plan)
-          @logger.debug('Failed to update in place. Recreating VM')
-          @disk_manager.unmount_disk_for(instance_plan)
-          @vm_recreator.recreate_vm(instance_plan, nil)
-          recreated = true
+          begin
+            recreate_parent_id = add_event(instance.deployment_model.name, 'recreate', instance.model.name, nil) if action == 'update'
+            @logger.debug('Failed to update in place. Recreating VM')
+            @disk_manager.unmount_disk_for(instance_plan)
+            @vm_recreator.recreate_vm(instance_plan, nil)
+            recreated = true
+          rescue Exception => e
+            raise e
+          ensure
+            add_event(instance.deployment_model.name, 'recreate', instance.model.name, nil, recreate_parent_id, e) if recreate_parent_id
+          end
         end
 
         release_obsolete_ips(instance_plan)
 
         update_dns(instance_plan)
-        @disk_manager.update_persistent_disk(instance_plan, @vm_recreator)
+        @disk_manager.update_persistent_disk(instance_plan)
 
         unless recreated
           if instance.trusted_certs_changed?
@@ -100,10 +110,56 @@ module Bosh::Director
         )
         state_applier.apply(instance_plan.desired_instance.job.update)
       end
+    rescue Exception => e
+      raise e
+    ensure
+      add_event(instance.deployment_model.name, action, instance.model.name, nil, parent_id, e) if parent_id
     end
 
     private
 
+    def add_event(deployment_name, action, instance_name = nil, context = nil, parent_id = nil, error = nil)
+      event  = Config.current_job.event_manager.create_event(
+          {
+              parent_id:   parent_id,
+              user:        Config.current_job.username,
+              action:      action,
+              object_type: 'instance',
+              object_name: instance_name,
+              task:        Config.current_job.task_id,
+              deployment:  deployment_name,
+              instance:    instance_name,
+              error:       error,
+              context:     context ? context: {}
+          })
+      event.id
+    end
+
+    def get_action_and_context(instance_plan)
+      changes = instance_plan.changes
+      if changes.size == 1 && [:state,:recreate,:restart].include?(changes.first)
+        action = case instance_plan.instance.virtual_state
+          when 'started'
+            'start'
+          when 'stopped'
+            'stop'
+          when 'detached'
+            'stop'
+          else
+            instance_plan.instance.virtual_state
+        end
+       return action , {}
+      else
+        if instance_plan.new?
+          return 'create', {}
+        else
+          context = {changes: changes.to_a}
+          context['az'] = instance_plan.desired_az_name if instance_plan.desired_az_name
+          return 'update', context
+        end
+      end
+    end
+
     def release_obsolete_ips(instance_plan)
       instance_plan.network_plans
         .select(&:obsolete?)