services/api/app/models/container.rb

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: AGPL-3.0
   4
   5 require 'log_reuse_info'
   6 require 'whitelist_update'
   7 require 'safe_json'
   8 require 'update_priority'
   9
  10 class Container < ArvadosModel
  11   include ArvadosModelUpdates
  12   include HasUuid
  13   include KindAndEtag
  14   include CommonApiTemplate
  15   include WhitelistUpdate
  16   extend CurrentApiClient
  17   extend DbCurrentTime
  18   extend LogReuseInfo
  19
  20   # Posgresql JSONB columns should NOT be declared as serialized, Rails 5
  21   # already know how to properly treat them.
  22   attribute :secret_mounts, :jsonbHash, default: {}
  23   attribute :runtime_status, :jsonbHash, default: {}
  24   attribute :runtime_auth_scopes, :jsonbArray, default: []
  25   attribute :output_storage_classes, :jsonbArray, default: lambda { Rails.configuration.DefaultStorageClasses }
  26   attribute :output_properties, :jsonbHash, default: {}
  27
  28   serialize :environment, Hash
  29   serialize :mounts, Hash
  30   serialize :runtime_constraints, Hash
  31   serialize :command, Array
  32   serialize :scheduling_parameters, Hash
  33
  34   after_find :fill_container_defaults_after_find
  35   before_validation :fill_field_defaults, :if => :new_record?
  36   before_validation :set_timestamps
  37   before_validation :check_lock
  38   before_validation :check_unlock
  39   validates :command, :container_image, :output_path, :cwd, :priority, { presence: true }
  40   validates :priority, numericality: { only_integer: true, greater_than_or_equal_to: 0 }
  41   validate :validate_runtime_status
  42   validate :validate_state_change
  43   validate :validate_change
  44   validate :validate_lock
  45   validate :validate_output
  46   after_validation :assign_auth
  47   before_save :sort_serialized_attrs
  48   before_save :update_secret_mounts_md5
  49   before_save :scrub_secrets
  50   before_save :clear_runtime_status_when_queued
  51   after_save :update_cr_logs
  52   after_save :handle_completed
  53   after_save :propagate_priority
  54   after_commit { UpdatePriority.run_update_thread }
  55
  56   has_many :container_requests, :foreign_key => :container_uuid, :class_name => 'ContainerRequest', :primary_key => :uuid
  57   belongs_to :auth, :class_name => 'ApiClientAuthorization', :foreign_key => :auth_uuid, :primary_key => :uuid
  58
  59   api_accessible :user, extend: :common do |t|
  60     t.add :command
  61     t.add :container_image
  62     t.add :cwd
  63     t.add :environment
  64     t.add :exit_code
  65     t.add :finished_at
  66     t.add :locked_by_uuid
  67     t.add :log
  68     t.add :mounts
  69     t.add :output
  70     t.add :output_path
  71     t.add :priority
  72     t.add :progress
  73     t.add :runtime_constraints
  74     t.add :runtime_status
  75     t.add :started_at
  76     t.add :state
  77     t.add :auth_uuid
  78     t.add :scheduling_parameters
  79     t.add :runtime_user_uuid
  80     t.add :runtime_auth_scopes
  81     t.add :lock_count
  82     t.add :gateway_address
  83     t.add :interactive_session_started
  84     t.add :output_storage_classes
  85     t.add :output_properties
  86   end
  87
  88   # Supported states for a container
  89   States =
  90     [
  91      (Queued = 'Queued'),
  92      (Locked = 'Locked'),
  93      (Running = 'Running'),
  94      (Complete = 'Complete'),
  95      (Cancelled = 'Cancelled')
  96     ]
  97
  98   State_transitions = {
  99     nil => [Queued],
 100     Queued => [Locked, Cancelled],
 101     Locked => [Queued, Running, Cancelled],
 102     Running => [Complete, Cancelled],
 103     Complete => [Cancelled]
 104   }
 105
 106   def self.limit_index_columns_read
 107     ["mounts"]
 108   end
 109
 110   def self.full_text_searchable_columns
 111     super - ["secret_mounts", "secret_mounts_md5", "runtime_token", "gateway_address", "output_storage_classes"]
 112   end
 113
 114   def self.searchable_columns *args
 115     super - ["secret_mounts_md5", "runtime_token", "gateway_address", "output_storage_classes"]
 116   end
 117
 118   def logged_attributes
 119     super.except('secret_mounts', 'runtime_token')
 120   end
 121
 122   def state_transitions
 123     State_transitions
 124   end
 125
 126   # Container priority is the highest "computed priority" of any
 127   # matching request. The computed priority of a container-submitted
 128   # request is the priority of the submitting container. The computed
 129   # priority of a user-submitted request is a function of
 130   # user-assigned priority and request creation time.
 131   def update_priority!
 132     return if ![Queued, Locked, Running].include?(state)
 133     p = ContainerRequest.
 134         where('container_uuid=? and priority>0', uuid).
 135         includes(:requesting_container).
 136         lock(true).
 137         map do |cr|
 138       if cr.requesting_container
 139         cr.requesting_container.priority
 140       else
 141         (cr.priority << 50) - (cr.created_at.to_time.to_f * 1000).to_i
 142       end
 143     end.max || 0
 144     update_attributes!(priority: p)
 145   end
 146
 147   def propagate_priority
 148     return true unless saved_change_to_priority?
 149     act_as_system_user do
 150       # Update the priority of child container requests to match new
 151       # priority of the parent container (ignoring requests with no
 152       # container assigned, because their priority doesn't matter).
 153       ContainerRequest.
 154         where(requesting_container_uuid: self.uuid,
 155               state: ContainerRequest::Committed).
 156         where('container_uuid is not null').
 157         includes(:container).
 158         map(&:container).
 159         map(&:update_priority!)
 160     end
 161   end
 162
 163   # Create a new container (or find an existing one) to satisfy the
 164   # given container request.
 165   def self.resolve(req)
 166     if req.runtime_token.nil?
 167       runtime_user = if req.modified_by_user_uuid.nil?
 168                        current_user
 169                      else
 170                        User.find_by_uuid(req.modified_by_user_uuid)
 171                      end
 172       runtime_auth_scopes = ["all"]
 173     else
 174       auth = ApiClientAuthorization.validate(token: req.runtime_token)
 175       if auth.nil?
 176         raise ArgumentError.new "Invalid runtime token"
 177       end
 178       runtime_user = User.find_by_id(auth.user_id)
 179       runtime_auth_scopes = auth.scopes
 180     end
 181     c_attrs = act_as_user runtime_user do
 182       {
 183         command: req.command,
 184         cwd: req.cwd,
 185         environment: req.environment,
 186         output_path: req.output_path,
 187         container_image: resolve_container_image(req.container_image),
 188         mounts: resolve_mounts(req.mounts),
 189         runtime_constraints: resolve_runtime_constraints(req.runtime_constraints),
 190         scheduling_parameters: req.scheduling_parameters,
 191         secret_mounts: req.secret_mounts,
 192         runtime_token: req.runtime_token,
 193         runtime_user_uuid: runtime_user.uuid,
 194         runtime_auth_scopes: runtime_auth_scopes,
 195         output_storage_classes: req.output_storage_classes,
 196       }
 197     end
 198     act_as_system_user do
 199       if req.use_existing && (reusable = find_reusable(c_attrs))
 200         reusable
 201       else
 202         Container.create!(c_attrs)
 203       end
 204     end
 205   end
 206
 207   # Return a runtime_constraints hash that complies with requested but
 208   # is suitable for saving in a container record, i.e., has specific
 209   # values instead of ranges.
 210   #
 211   # Doing this as a step separate from other resolutions, like "git
 212   # revision range to commit hash", makes sense only when there is no
 213   # opportunity to reuse an existing container (e.g., container reuse
 214   # is not implemented yet, or we have already found that no existing
 215   # containers are suitable).
 216   def self.resolve_runtime_constraints(runtime_constraints)
 217     rc = {}
 218     runtime_constraints.each do |k, v|
 219       if v.is_a? Array
 220         rc[k] = v[0]
 221       else
 222         rc[k] = v
 223       end
 224     end
 225     if rc['keep_cache_ram'] == 0
 226       rc['keep_cache_ram'] = Rails.configuration.Containers.DefaultKeepCacheRAM
 227     end
 228     rc
 229   end
 230
 231   # Return a mounts hash suitable for a Container, i.e., with every
 232   # readonly collection UUID resolved to a PDH.
 233   def self.resolve_mounts(mounts)
 234     c_mounts = {}
 235     mounts.each do |k, mount|
 236       mount = mount.dup
 237       c_mounts[k] = mount
 238       if mount['kind'] != 'collection'
 239         next
 240       end
 241
 242       uuid = mount.delete 'uuid'
 243
 244       if mount['portable_data_hash'].nil? and !uuid.nil?
 245         # PDH not supplied, try by UUID
 246         c = Collection.
 247           readable_by(current_user).
 248           where(uuid: uuid).
 249           select(:portable_data_hash).
 250           first
 251         if !c
 252           raise ArvadosModel::UnresolvableContainerError.new "cannot mount collection #{uuid.inspect}: not found"
 253         end
 254         mount['portable_data_hash'] = c.portable_data_hash
 255       end
 256     end
 257     return c_mounts
 258   end
 259
 260   # Return a container_image PDH suitable for a Container.
 261   def self.resolve_container_image(container_image)
 262     coll = Collection.for_latest_docker_image(container_image)
 263     if !coll
 264       raise ArvadosModel::UnresolvableContainerError.new "docker image #{container_image.inspect} not found"
 265     end
 266     coll.portable_data_hash
 267   end
 268
 269   def self.find_reusable(attrs)
 270     log_reuse_info { "starting with #{Container.all.count} container records in database" }
 271     candidates = Container.where_serialized(:command, attrs[:command], md5: true)
 272     log_reuse_info(candidates) { "after filtering on command #{attrs[:command].inspect}" }
 273
 274     candidates = candidates.where('cwd = ?', attrs[:cwd])
 275     log_reuse_info(candidates) { "after filtering on cwd #{attrs[:cwd].inspect}" }
 276
 277     candidates = candidates.where_serialized(:environment, attrs[:environment], md5: true)
 278     log_reuse_info(candidates) { "after filtering on environment #{attrs[:environment].inspect}" }
 279
 280     candidates = candidates.where('output_path = ?', attrs[:output_path])
 281     log_reuse_info(candidates) { "after filtering on output_path #{attrs[:output_path].inspect}" }
 282
 283     image = resolve_container_image(attrs[:container_image])
 284     candidates = candidates.where('container_image = ?', image)
 285     log_reuse_info(candidates) { "after filtering on container_image #{image.inspect} (resolved from #{attrs[:container_image].inspect})" }
 286
 287     candidates = candidates.where_serialized(:mounts, resolve_mounts(attrs[:mounts]), md5: true)
 288     log_reuse_info(candidates) { "after filtering on mounts #{attrs[:mounts].inspect}" }
 289
 290     secret_mounts_md5 = Digest::MD5.hexdigest(SafeJSON.dump(self.deep_sort_hash(attrs[:secret_mounts])))
 291     candidates = candidates.where('secret_mounts_md5 = ?', secret_mounts_md5)
 292     log_reuse_info(candidates) { "after filtering on secret_mounts_md5 #{secret_mounts_md5.inspect}" }
 293
 294     if attrs[:runtime_constraints]['cuda'].nil?
 295       attrs[:runtime_constraints]['cuda'] = {
 296         'device_count' => 0,
 297         'driver_version' => '',
 298         'hardware_capability' => '',
 299       }
 300     end
 301     resolved_runtime_constraints = [resolve_runtime_constraints(attrs[:runtime_constraints])]
 302     if resolved_runtime_constraints[0]['cuda']['device_count'] == 0
 303       # If no CUDA requested, extend search to include older container
 304       # records that don't have a 'cuda' section in runtime_constraints
 305       resolved_runtime_constraints << resolved_runtime_constraints[0].except('cuda')
 306     end
 307
 308     candidates = candidates.where_serialized(:runtime_constraints, resolved_runtime_constraints, md5: true, multivalue: true)
 309     log_reuse_info(candidates) { "after filtering on runtime_constraints #{attrs[:runtime_constraints].inspect}" }
 310
 311     log_reuse_info { "checking for state=Complete with readable output and log..." }
 312
 313     select_readable_pdh = Collection.
 314       readable_by(current_user).
 315       select(:portable_data_hash).
 316       to_sql
 317
 318     usable = candidates.where(state: Complete, exit_code: 0)
 319     log_reuse_info(usable) { "with state=Complete, exit_code=0" }
 320
 321     usable = usable.where("log IN (#{select_readable_pdh})")
 322     log_reuse_info(usable) { "with readable log" }
 323
 324     usable = usable.where("output IN (#{select_readable_pdh})")
 325     log_reuse_info(usable) { "with readable output" }
 326
 327     usable = usable.order('finished_at ASC').limit(1).first
 328     if usable
 329       log_reuse_info { "done, reusing container #{usable.uuid} with state=Complete" }
 330       return usable
 331     end
 332
 333     # Check for non-failing Running candidates and return the most likely to finish sooner.
 334     log_reuse_info { "checking for state=Running..." }
 335     running = candidates.where(state: Running).
 336               where("(runtime_status->'error') is null").
 337               order('progress desc, started_at asc').
 338               limit(1).first
 339     if running
 340       log_reuse_info { "done, reusing container #{running.uuid} with state=Running" }
 341       return running
 342     else
 343       log_reuse_info { "have no containers in Running state" }
 344     end
 345
 346     # Check for Locked or Queued ones and return the most likely to start first.
 347     locked_or_queued = candidates.
 348                        where("state IN (?)", [Locked, Queued]).
 349                        order('state asc, priority desc, created_at asc').
 350                        limit(1).first
 351     if locked_or_queued
 352       log_reuse_info { "done, reusing container #{locked_or_queued.uuid} with state=#{locked_or_queued.state}" }
 353       return locked_or_queued
 354     else
 355       log_reuse_info { "have no containers in Locked or Queued state" }
 356     end
 357
 358     log_reuse_info { "done, no reusable container found" }
 359     nil
 360   end
 361
 362   def lock
 363     self.with_lock do
 364       if self.state != Queued
 365         raise LockFailedError.new("cannot lock when #{self.state}")
 366       end
 367       self.update_attributes!(state: Locked)
 368     end
 369   end
 370
 371   def check_lock
 372     if state_was == Queued and state == Locked
 373       if self.priority <= 0
 374         raise LockFailedError.new("cannot lock when priority<=0")
 375       end
 376       self.lock_count = self.lock_count+1
 377     end
 378   end
 379
 380   def unlock
 381     self.with_lock do
 382       if self.state != Locked
 383         raise InvalidStateTransitionError.new("cannot unlock when #{self.state}")
 384       end
 385       self.update_attributes!(state: Queued)
 386     end
 387   end
 388
 389   def check_unlock
 390     if state_was == Locked and state == Queued
 391       if self.locked_by_uuid != current_api_client_authorization.uuid
 392         raise ArvadosModel::PermissionDeniedError.new("locked by a different token")
 393       end
 394       if self.lock_count >= Rails.configuration.Containers.MaxDispatchAttempts
 395         self.state = Cancelled
 396         self.runtime_status = {error: "Failed to start container.  Cancelled after exceeding 'Containers.MaxDispatchAttempts' (lock_count=#{self.lock_count})"}
 397       end
 398     end
 399   end
 400
 401   def self.readable_by(*users_list)
 402     # Load optional keyword arguments, if they exist.
 403     if users_list.last.is_a? Hash
 404       kwargs = users_list.pop
 405     else
 406       kwargs = {}
 407     end
 408     if users_list.select { |u| u.is_admin }.any?
 409       return super
 410     end
 411     Container.where(ContainerRequest.readable_by(*users_list).where("containers.uuid = container_requests.container_uuid").arel.exists)
 412   end
 413
 414   def final?
 415     [Complete, Cancelled].include?(self.state)
 416   end
 417
 418   def self.for_current_token
 419     return if !current_api_client_authorization
 420     _, _, _, container_uuid = Thread.current[:token].split('/')
 421     if container_uuid.nil?
 422       Container.where(auth_uuid: current_api_client_authorization.uuid).first
 423     else
 424       Container.where('auth_uuid=? or (uuid=? and runtime_token=?)',
 425                       current_api_client_authorization.uuid,
 426                       container_uuid,
 427                       current_api_client_authorization.token).first
 428     end
 429   end
 430
 431   protected
 432
 433   def fill_field_defaults
 434     self.state ||= Queued
 435     self.environment ||= {}
 436     self.runtime_constraints ||= {}
 437     self.mounts ||= {}
 438     self.cwd ||= "."
 439     self.priority ||= 0
 440     self.scheduling_parameters ||= {}
 441   end
 442
 443   def permission_to_create
 444     current_user.andand.is_admin
 445   end
 446
 447   def permission_to_destroy
 448     current_user.andand.is_admin
 449   end
 450
 451   def ensure_owner_uuid_is_permitted
 452     # validate_change ensures owner_uuid can't be changed at all --
 453     # except during create, which requires admin privileges. Checking
 454     # permission here would be superfluous.
 455     true
 456   end
 457
 458   def set_timestamps
 459     if self.state_changed? and self.state == Running
 460       self.started_at ||= db_current_time
 461     end
 462
 463     if self.state_changed? and [Complete, Cancelled].include? self.state
 464       self.finished_at ||= db_current_time
 465     end
 466   end
 467
 468   # Check that well-known runtime status keys have desired data types
 469   def validate_runtime_status
 470     [
 471       'error', 'errorDetail', 'warning', 'warningDetail', 'activity'
 472     ].each do |k|
 473       if self.runtime_status.andand.include?(k) && !self.runtime_status[k].is_a?(String)
 474         errors.add(:runtime_status, "'#{k}' value must be a string")
 475       end
 476     end
 477   end
 478
 479   def validate_change
 480     permitted = [:state]
 481     progress_attrs = [:progress, :runtime_status, :log, :output, :output_properties]
 482     final_attrs = [:exit_code, :finished_at]
 483
 484     if self.new_record?
 485       permitted.push(:owner_uuid, :command, :container_image, :cwd,
 486                      :environment, :mounts, :output_path, :priority,
 487                      :runtime_constraints, :scheduling_parameters,
 488                      :secret_mounts, :runtime_token,
 489                      :runtime_user_uuid, :runtime_auth_scopes,
 490                      :output_storage_classes)
 491     end
 492
 493     case self.state
 494     when Locked
 495       permitted.push :priority, :runtime_status, :log, :lock_count
 496
 497     when Queued
 498       permitted.push :priority
 499
 500     when Running
 501       permitted.push :priority, :output_properties, *progress_attrs
 502       if self.state_changed?
 503         permitted.push :started_at, :gateway_address
 504       end
 505       if !self.interactive_session_started_was
 506         permitted.push :interactive_session_started
 507       end
 508
 509     when Complete
 510       if self.state_was == Running
 511         permitted.push *final_attrs, *progress_attrs
 512       end
 513
 514     when Cancelled
 515       case self.state_was
 516       when Running
 517         permitted.push :finished_at, *progress_attrs
 518       when Queued, Locked
 519         permitted.push :finished_at, :log, :runtime_status
 520       end
 521
 522     else
 523       # The state_transitions check will add an error message for this
 524       return false
 525     end
 526
 527     if self.state_was == Running &&
 528        !current_api_client_authorization.nil? &&
 529        (current_api_client_authorization.uuid == self.auth_uuid ||
 530         current_api_client_authorization.token == self.runtime_token)
 531       # The contained process itself can write final attrs but can't
 532       # change priority or log.
 533       permitted.push *final_attrs
 534       permitted = permitted - [:log, :priority]
 535     elsif !current_user.andand.is_admin
 536       raise PermissionDeniedError
 537     elsif self.locked_by_uuid && self.locked_by_uuid != current_api_client_authorization.andand.uuid
 538       # When locked, progress fields cannot be updated by the wrong
 539       # dispatcher, even though it has admin privileges.
 540       permitted = permitted - progress_attrs
 541     end
 542     check_update_whitelist permitted
 543   end
 544
 545   def validate_lock
 546     if [Locked, Running].include? self.state
 547       # If the Container was already locked, locked_by_uuid must not
 548       # changes. Otherwise, the current auth gets the lock.
 549       need_lock = locked_by_uuid_was || current_api_client_authorization.andand.uuid
 550     else
 551       need_lock = nil
 552     end
 553
 554     # The caller can provide a new value for locked_by_uuid, but only
 555     # if it's exactly what we expect. This allows a caller to perform
 556     # an update like {"state":"Unlocked","locked_by_uuid":null}.
 557     if self.locked_by_uuid_changed?
 558       if self.locked_by_uuid != need_lock
 559         return errors.add :locked_by_uuid, "can only change to #{need_lock}"
 560       end
 561     end
 562     self.locked_by_uuid = need_lock
 563   end
 564
 565   def validate_output
 566     # Output must exist and be readable by the current user.  This is so
 567     # that a container cannot "claim" a collection that it doesn't otherwise
 568     # have access to just by setting the output field to the collection PDH.
 569     if output_changed?
 570       c = Collection.
 571             readable_by(current_user, {include_trash: true}).
 572             where(portable_data_hash: self.output).
 573             first
 574       if !c
 575         errors.add :output, "collection must exist and be readable by current user."
 576       end
 577     end
 578   end
 579
 580   def update_cr_logs
 581     # If self.final?, this update is superfluous: the final log/output
 582     # update will be done when handle_completed calls finalize! on
 583     # each requesting CR.
 584     return if self.final? || !saved_change_to_log?
 585     leave_modified_by_user_alone do
 586       ContainerRequest.where(container_uuid: self.uuid).each do |cr|
 587         cr.update_collections(container: self, collections: ['log'])
 588         cr.save!
 589       end
 590     end
 591   end
 592
 593   def assign_auth
 594     if self.auth_uuid_changed?
 595          return errors.add :auth_uuid, 'is readonly'
 596     end
 597     if not [Locked, Running].include? self.state
 598       # Don't need one. If auth already exists, expire it.
 599       #
 600       # We use db_transaction_time here (not db_current_time) to
 601       # ensure the token doesn't validate later in the same
 602       # transaction (e.g., in a test case) by satisfying expires_at >
 603       # transaction timestamp.
 604       self.auth.andand.update_attributes(expires_at: db_transaction_time)
 605       self.auth = nil
 606       return
 607     elsif self.auth
 608       # already have one
 609       return
 610     end
 611     if self.runtime_token.nil?
 612       if self.runtime_user_uuid.nil?
 613         # legacy behavior, we don't have a runtime_user_uuid so get
 614         # the user from the highest priority container request, needed
 615         # when performing an upgrade and there are queued containers,
 616         # and some tests.
 617         cr = ContainerRequest.
 618                where('container_uuid=? and priority>0', self.uuid).
 619                order('priority desc').
 620                first
 621         if !cr
 622           return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
 623         end
 624         self.runtime_user_uuid = cr.modified_by_user_uuid
 625         self.runtime_auth_scopes = ["all"]
 626       end
 627
 628       # Generate a new token. This runs with admin credentials as it's done by a
 629       # dispatcher user, so expires_at isn't enforced by API.MaxTokenLifetime.
 630       self.auth = ApiClientAuthorization.
 631                     create!(user_id: User.find_by_uuid(self.runtime_user_uuid).id,
 632                             api_client_id: 0,
 633                             scopes: self.runtime_auth_scopes)
 634     end
 635   end
 636
 637   def sort_serialized_attrs
 638     if self.environment_changed?
 639       self.environment = self.class.deep_sort_hash(self.environment)
 640     end
 641     if self.mounts_changed?
 642       self.mounts = self.class.deep_sort_hash(self.mounts)
 643     end
 644     if self.runtime_constraints_changed?
 645       self.runtime_constraints = self.class.deep_sort_hash(self.runtime_constraints)
 646     end
 647     if self.scheduling_parameters_changed?
 648       self.scheduling_parameters = self.class.deep_sort_hash(self.scheduling_parameters)
 649     end
 650     if self.runtime_auth_scopes_changed?
 651       self.runtime_auth_scopes = self.runtime_auth_scopes.sort
 652     end
 653   end
 654
 655   def update_secret_mounts_md5
 656     if self.secret_mounts_changed?
 657       self.secret_mounts_md5 = Digest::MD5.hexdigest(
 658         SafeJSON.dump(self.class.deep_sort_hash(self.secret_mounts)))
 659     end
 660   end
 661
 662   def scrub_secrets
 663     # this runs after update_secret_mounts_md5, so the
 664     # secret_mounts_md5 will still reflect the secrets that are being
 665     # scrubbed here.
 666     if self.state_changed? && self.final?
 667       self.secret_mounts = {}
 668       self.runtime_token = nil
 669     end
 670   end
 671
 672   def clear_runtime_status_when_queued
 673     # Avoid leaking status messages between different dispatch attempts
 674     if self.state_was == Locked && self.state == Queued
 675       self.runtime_status = {}
 676     end
 677   end
 678
 679   def handle_completed
 680     # This container is finished so finalize any associated container requests
 681     # that are associated with this container.
 682     if saved_change_to_state? and self.final?
 683       # These get wiped out by with_lock (which reloads the record),
 684       # so record them now in case we need to schedule a retry.
 685       prev_secret_mounts = secret_mounts_before_last_save
 686       prev_runtime_token = runtime_token_before_last_save
 687
 688       # Need to take a lock on the container to ensure that any
 689       # concurrent container requests that might try to reuse this
 690       # container will block until the container completion
 691       # transaction finishes.  This ensure that concurrent container
 692       # requests that try to reuse this container are finalized (on
 693       # Complete) or don't reuse it (on Cancelled).
 694       self.with_lock do
 695         act_as_system_user do
 696           if self.state == Cancelled
 697             retryable_requests = ContainerRequest.where("container_uuid = ? and priority > 0 and state = 'Committed' and container_count < container_count_max", uuid)
 698           else
 699             retryable_requests = []
 700           end
 701
 702           if retryable_requests.any?
 703             c_attrs = {
 704               command: self.command,
 705               cwd: self.cwd,
 706               environment: self.environment,
 707               output_path: self.output_path,
 708               container_image: self.container_image,
 709               mounts: self.mounts,
 710               runtime_constraints: self.runtime_constraints,
 711               scheduling_parameters: self.scheduling_parameters,
 712               secret_mounts: prev_secret_mounts,
 713               runtime_token: prev_runtime_token,
 714               runtime_user_uuid: self.runtime_user_uuid,
 715               runtime_auth_scopes: self.runtime_auth_scopes
 716             }
 717             c = Container.create! c_attrs
 718             retryable_requests.each do |cr|
 719               cr.with_lock do
 720                 leave_modified_by_user_alone do
 721                   # Use row locking because this increments container_count
 722                   cr.container_uuid = c.uuid
 723                   cr.save!
 724                 end
 725               end
 726             end
 727           end
 728
 729           # Notify container requests associated with this container
 730           ContainerRequest.where(container_uuid: uuid,
 731                                  state: ContainerRequest::Committed).each do |cr|
 732             leave_modified_by_user_alone do
 733               cr.finalize!
 734             end
 735           end
 736
 737           # Cancel outstanding container requests made by this container.
 738           ContainerRequest.
 739             includes(:container).
 740             where(requesting_container_uuid: uuid,
 741                   state: ContainerRequest::Committed).each do |cr|
 742             leave_modified_by_user_alone do
 743               cr.update_attributes!(priority: 0)
 744               cr.container.reload
 745               if cr.container.state == Container::Queued || cr.container.state == Container::Locked
 746                 # If the child container hasn't started yet, finalize the
 747                 # child CR now instead of leaving it "on hold", i.e.,
 748                 # Queued with priority 0.  (OTOH, if the child is already
 749                 # running, leave it alone so it can get cancelled the
 750                 # usual way, get a copy of the log collection, etc.)
 751                 cr.update_attributes!(state: ContainerRequest::Final)
 752               end
 753             end
 754           end
 755         end
 756       end
 757     end
 758   end
 759 end