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     t.add :cost
  87     t.add :subrequests_cost
  88   end
  89
  90   # Supported states for a container
  91   States =
  92     [
  93      (Queued = 'Queued'),
  94      (Locked = 'Locked'),
  95      (Running = 'Running'),
  96      (Complete = 'Complete'),
  97      (Cancelled = 'Cancelled')
  98     ]
  99
 100   State_transitions = {
 101     nil => [Queued],
 102     Queued => [Locked, Cancelled],
 103     Locked => [Queued, Running, Cancelled],
 104     Running => [Complete, Cancelled],
 105     Complete => [Cancelled]
 106   }
 107
 108   def self.limit_index_columns_read
 109     ["mounts"]
 110   end
 111
 112   def self.full_text_searchable_columns
 113     super - ["secret_mounts", "secret_mounts_md5", "runtime_token", "gateway_address", "output_storage_classes"]
 114   end
 115
 116   def self.searchable_columns *args
 117     super - ["secret_mounts_md5", "runtime_token", "gateway_address", "output_storage_classes"]
 118   end
 119
 120   def logged_attributes
 121     super.except('secret_mounts', 'runtime_token')
 122   end
 123
 124   def state_transitions
 125     State_transitions
 126   end
 127
 128   # Container priority is the highest "computed priority" of any
 129   # matching request. The computed priority of a container-submitted
 130   # request is the priority of the submitting container. The computed
 131   # priority of a user-submitted request is a function of
 132   # user-assigned priority and request creation time.
 133   def update_priority!
 134     return if ![Queued, Locked, Running].include?(state)
 135     p = ContainerRequest.
 136         where('container_uuid=? and priority>0', uuid).
 137         includes(:requesting_container).
 138         lock(true).
 139         map do |cr|
 140       if cr.requesting_container
 141         cr.requesting_container.priority
 142       else
 143         (cr.priority << 50) - (cr.created_at.to_time.to_f * 1000).to_i
 144       end
 145     end.max || 0
 146     update_attributes!(priority: p)
 147   end
 148
 149   def propagate_priority
 150     return true unless saved_change_to_priority?
 151     act_as_system_user do
 152       # Update the priority of child container requests to match new
 153       # priority of the parent container (ignoring requests with no
 154       # container assigned, because their priority doesn't matter).
 155       ContainerRequest.
 156         where(requesting_container_uuid: self.uuid,
 157               state: ContainerRequest::Committed).
 158         where('container_uuid is not null').
 159         includes(:container).
 160         map(&:container).
 161         map(&:update_priority!)
 162     end
 163   end
 164
 165   # Create a new container (or find an existing one) to satisfy the
 166   # given container request.
 167   def self.resolve(req)
 168     if req.runtime_token.nil?
 169       runtime_user = if req.modified_by_user_uuid.nil?
 170                        current_user
 171                      else
 172                        User.find_by_uuid(req.modified_by_user_uuid)
 173                      end
 174       runtime_auth_scopes = ["all"]
 175     else
 176       auth = ApiClientAuthorization.validate(token: req.runtime_token)
 177       if auth.nil?
 178         raise ArgumentError.new "Invalid runtime token"
 179       end
 180       runtime_user = User.find_by_id(auth.user_id)
 181       runtime_auth_scopes = auth.scopes
 182     end
 183     c_attrs = act_as_user runtime_user do
 184       {
 185         command: req.command,
 186         cwd: req.cwd,
 187         environment: req.environment,
 188         output_path: req.output_path,
 189         container_image: resolve_container_image(req.container_image),
 190         mounts: resolve_mounts(req.mounts),
 191         runtime_constraints: resolve_runtime_constraints(req.runtime_constraints),
 192         scheduling_parameters: req.scheduling_parameters,
 193         secret_mounts: req.secret_mounts,
 194         runtime_token: req.runtime_token,
 195         runtime_user_uuid: runtime_user.uuid,
 196         runtime_auth_scopes: runtime_auth_scopes,
 197         output_storage_classes: req.output_storage_classes,
 198       }
 199     end
 200     act_as_system_user do
 201       if req.use_existing && (reusable = find_reusable(c_attrs))
 202         reusable
 203       else
 204         Container.create!(c_attrs)
 205       end
 206     end
 207   end
 208
 209   # Return a runtime_constraints hash that complies with requested but
 210   # is suitable for saving in a container record, i.e., has specific
 211   # values instead of ranges.
 212   #
 213   # Doing this as a step separate from other resolutions, like "git
 214   # revision range to commit hash", makes sense only when there is no
 215   # opportunity to reuse an existing container (e.g., container reuse
 216   # is not implemented yet, or we have already found that no existing
 217   # containers are suitable).
 218   def self.resolve_runtime_constraints(runtime_constraints)
 219     rc = {}
 220     runtime_constraints.each do |k, v|
 221       if v.is_a? Array
 222         rc[k] = v[0]
 223       else
 224         rc[k] = v
 225       end
 226     end
 227     if rc['keep_cache_ram'] == 0
 228       rc['keep_cache_ram'] = Rails.configuration.Containers.DefaultKeepCacheRAM
 229     end
 230     if rc['keep_cache_disk'] == 0 and rc['keep_cache_ram'] == 0
 231       # If neither ram nor disk cache was specified and
 232       # DefaultKeepCacheRAM==0, default to disk cache with size equal
 233       # to RAM constraint (but at least 2 GiB and at most 32 GiB).
 234       rc['keep_cache_disk'] = [[rc['ram'] || 0, 2 << 30].max, 32 << 30].min
 235     end
 236     rc
 237   end
 238
 239   # Return a mounts hash suitable for a Container, i.e., with every
 240   # readonly collection UUID resolved to a PDH.
 241   def self.resolve_mounts(mounts)
 242     c_mounts = {}
 243     mounts.each do |k, mount|
 244       mount = mount.dup
 245       c_mounts[k] = mount
 246       if mount['kind'] != 'collection'
 247         next
 248       end
 249
 250       uuid = mount.delete 'uuid'
 251
 252       if mount['portable_data_hash'].nil? and !uuid.nil?
 253         # PDH not supplied, try by UUID
 254         c = Collection.
 255           readable_by(current_user).
 256           where(uuid: uuid).
 257           select(:portable_data_hash).
 258           first
 259         if !c
 260           raise ArvadosModel::UnresolvableContainerError.new "cannot mount collection #{uuid.inspect}: not found"
 261         end
 262         mount['portable_data_hash'] = c.portable_data_hash
 263       end
 264     end
 265     return c_mounts
 266   end
 267
 268   # Return a container_image PDH suitable for a Container.
 269   def self.resolve_container_image(container_image)
 270     coll = Collection.for_latest_docker_image(container_image)
 271     if !coll
 272       raise ArvadosModel::UnresolvableContainerError.new "docker image #{container_image.inspect} not found"
 273     end
 274     coll.portable_data_hash
 275   end
 276
 277   def self.find_reusable(attrs)
 278     log_reuse_info { "starting with #{Container.all.count} container records in database" }
 279     candidates = Container.where_serialized(:command, attrs[:command], md5: true)
 280     log_reuse_info(candidates) { "after filtering on command #{attrs[:command].inspect}" }
 281
 282     candidates = candidates.where('cwd = ?', attrs[:cwd])
 283     log_reuse_info(candidates) { "after filtering on cwd #{attrs[:cwd].inspect}" }
 284
 285     candidates = candidates.where_serialized(:environment, attrs[:environment], md5: true)
 286     log_reuse_info(candidates) { "after filtering on environment #{attrs[:environment].inspect}" }
 287
 288     candidates = candidates.where('output_path = ?', attrs[:output_path])
 289     log_reuse_info(candidates) { "after filtering on output_path #{attrs[:output_path].inspect}" }
 290
 291     image = resolve_container_image(attrs[:container_image])
 292     candidates = candidates.where('container_image = ?', image)
 293     log_reuse_info(candidates) { "after filtering on container_image #{image.inspect} (resolved from #{attrs[:container_image].inspect})" }
 294
 295     candidates = candidates.where_serialized(:mounts, resolve_mounts(attrs[:mounts]), md5: true)
 296     log_reuse_info(candidates) { "after filtering on mounts #{attrs[:mounts].inspect}" }
 297
 298     secret_mounts_md5 = Digest::MD5.hexdigest(SafeJSON.dump(self.deep_sort_hash(attrs[:secret_mounts])))
 299     candidates = candidates.where('secret_mounts_md5 = ?', secret_mounts_md5)
 300     log_reuse_info(candidates) { "after filtering on secret_mounts_md5 #{secret_mounts_md5.inspect}" }
 301
 302     if attrs[:runtime_constraints]['cuda'].nil?
 303       attrs[:runtime_constraints]['cuda'] = {
 304         'device_count' => 0,
 305         'driver_version' => '',
 306         'hardware_capability' => '',
 307       }
 308     end
 309     resolved_runtime_constraints = [resolve_runtime_constraints(attrs[:runtime_constraints])]
 310     if resolved_runtime_constraints[0]['cuda']['device_count'] == 0
 311       # If no CUDA requested, extend search to include older container
 312       # records that don't have a 'cuda' section in runtime_constraints
 313       resolved_runtime_constraints << resolved_runtime_constraints[0].except('cuda')
 314     end
 315     if resolved_runtime_constraints[0]['keep_cache_disk'] == 0
 316       # If no disk cache requested, extend search to include older container
 317       # records that don't have a 'keep_cache_disk' field in runtime_constraints
 318       if resolved_runtime_constraints.length == 2
 319         # exclude the one that also excludes CUDA
 320         resolved_runtime_constraints << resolved_runtime_constraints[1].except('keep_cache_disk')
 321       end
 322       resolved_runtime_constraints << resolved_runtime_constraints[0].except('keep_cache_disk')
 323     end
 324
 325     candidates = candidates.where_serialized(:runtime_constraints, resolved_runtime_constraints, md5: true, multivalue: true)
 326     log_reuse_info(candidates) { "after filtering on runtime_constraints #{attrs[:runtime_constraints].inspect}" }
 327
 328     log_reuse_info { "checking for state=Complete with readable output and log..." }
 329
 330     select_readable_pdh = Collection.
 331       readable_by(current_user).
 332       select(:portable_data_hash).
 333       to_sql
 334
 335     usable = candidates.where(state: Complete, exit_code: 0)
 336     log_reuse_info(usable) { "with state=Complete, exit_code=0" }
 337
 338     usable = usable.where("log IN (#{select_readable_pdh})")
 339     log_reuse_info(usable) { "with readable log" }
 340
 341     usable = usable.where("output IN (#{select_readable_pdh})")
 342     log_reuse_info(usable) { "with readable output" }
 343
 344     usable = usable.order('finished_at ASC').limit(1).first
 345     if usable
 346       log_reuse_info { "done, reusing container #{usable.uuid} with state=Complete" }
 347       return usable
 348     end
 349
 350     # Check for non-failing Running candidates and return the most likely to finish sooner.
 351     log_reuse_info { "checking for state=Running..." }
 352     running = candidates.where(state: Running).
 353               where("(runtime_status->'error') is null").
 354               order('progress desc, started_at asc').
 355               limit(1).first
 356     if running
 357       log_reuse_info { "done, reusing container #{running.uuid} with state=Running" }
 358       return running
 359     else
 360       log_reuse_info { "have no containers in Running state" }
 361     end
 362
 363     # Check for Locked or Queued ones and return the most likely to start first.
 364     locked_or_queued = candidates.
 365                        where("state IN (?)", [Locked, Queued]).
 366                        order('state asc, priority desc, created_at asc').
 367                        limit(1).first
 368     if locked_or_queued
 369       log_reuse_info { "done, reusing container #{locked_or_queued.uuid} with state=#{locked_or_queued.state}" }
 370       return locked_or_queued
 371     else
 372       log_reuse_info { "have no containers in Locked or Queued state" }
 373     end
 374
 375     log_reuse_info { "done, no reusable container found" }
 376     nil
 377   end
 378
 379   def lock
 380     self.with_lock do
 381       if self.state != Queued
 382         raise LockFailedError.new("cannot lock when #{self.state}")
 383       end
 384       self.update_attributes!(state: Locked)
 385     end
 386   end
 387
 388   def check_lock
 389     if state_was == Queued and state == Locked
 390       if self.priority <= 0
 391         raise LockFailedError.new("cannot lock when priority<=0")
 392       end
 393       self.lock_count = self.lock_count+1
 394     end
 395   end
 396
 397   def unlock
 398     self.with_lock do
 399       if self.state != Locked
 400         raise InvalidStateTransitionError.new("cannot unlock when #{self.state}")
 401       end
 402       self.update_attributes!(state: Queued)
 403     end
 404   end
 405
 406   def check_unlock
 407     if state_was == Locked and state == Queued
 408       if self.locked_by_uuid != current_api_client_authorization.uuid
 409         raise ArvadosModel::PermissionDeniedError.new("locked by a different token")
 410       end
 411       if self.lock_count >= Rails.configuration.Containers.MaxDispatchAttempts
 412         self.state = Cancelled
 413         self.runtime_status = {error: "Failed to start container.  Cancelled after exceeding 'Containers.MaxDispatchAttempts' (lock_count=#{self.lock_count})"}
 414       end
 415     end
 416   end
 417
 418   def self.readable_by(*users_list)
 419     # Load optional keyword arguments, if they exist.
 420     if users_list.last.is_a? Hash
 421       kwargs = users_list.pop
 422     else
 423       kwargs = {}
 424     end
 425     if users_list.select { |u| u.is_admin }.any?
 426       return super
 427     end
 428     Container.where(ContainerRequest.readable_by(*users_list).where("containers.uuid = container_requests.container_uuid").arel.exists)
 429   end
 430
 431   def final?
 432     [Complete, Cancelled].include?(self.state)
 433   end
 434
 435   def self.for_current_token
 436     return if !current_api_client_authorization
 437     _, _, _, container_uuid = Thread.current[:token].split('/')
 438     if container_uuid.nil?
 439       Container.where(auth_uuid: current_api_client_authorization.uuid).first
 440     else
 441       Container.where('auth_uuid=? or (uuid=? and runtime_token=?)',
 442                       current_api_client_authorization.uuid,
 443                       container_uuid,
 444                       current_api_client_authorization.token).first
 445     end
 446   end
 447
 448   protected
 449
 450   def fill_field_defaults
 451     self.state ||= Queued
 452     self.environment ||= {}
 453     self.runtime_constraints ||= {}
 454     self.mounts ||= {}
 455     self.cwd ||= "."
 456     self.priority ||= 0
 457     self.scheduling_parameters ||= {}
 458   end
 459
 460   def permission_to_create
 461     current_user.andand.is_admin
 462   end
 463
 464   def permission_to_destroy
 465     current_user.andand.is_admin
 466   end
 467
 468   def ensure_owner_uuid_is_permitted
 469     # validate_change ensures owner_uuid can't be changed at all --
 470     # except during create, which requires admin privileges. Checking
 471     # permission here would be superfluous.
 472     true
 473   end
 474
 475   def set_timestamps
 476     if self.state_changed? and self.state == Running
 477       self.started_at ||= db_current_time
 478     end
 479
 480     if self.state_changed? and [Complete, Cancelled].include? self.state
 481       self.finished_at ||= db_current_time
 482     end
 483   end
 484
 485   # Check that well-known runtime status keys have desired data types
 486   def validate_runtime_status
 487     [
 488       'error', 'errorDetail', 'warning', 'warningDetail', 'activity'
 489     ].each do |k|
 490       if self.runtime_status.andand.include?(k) && !self.runtime_status[k].is_a?(String)
 491         errors.add(:runtime_status, "'#{k}' value must be a string")
 492       end
 493     end
 494   end
 495
 496   def validate_change
 497     permitted = [:state]
 498     final_attrs = [:finished_at]
 499     progress_attrs = [:progress, :runtime_status, :subrequests_cost, :cost,
 500                       :log, :output, :output_properties, :exit_code]
 501
 502     if self.new_record?
 503       permitted.push(:owner_uuid, :command, :container_image, :cwd,
 504                      :environment, :mounts, :output_path, :priority,
 505                      :runtime_constraints, :scheduling_parameters,
 506                      :secret_mounts, :runtime_token,
 507                      :runtime_user_uuid, :runtime_auth_scopes,
 508                      :output_storage_classes)
 509     end
 510
 511     case self.state
 512     when Locked
 513       permitted.push :priority, :runtime_status, :log, :lock_count
 514
 515     when Queued
 516       permitted.push :priority
 517
 518     when Running
 519       permitted.push :priority, :output_properties, :gateway_address, *progress_attrs
 520       if self.state_changed?
 521         permitted.push :started_at
 522       end
 523       if !self.interactive_session_started_was
 524         permitted.push :interactive_session_started
 525       end
 526
 527     when Complete
 528       if self.state_was == Running
 529         permitted.push *final_attrs, *progress_attrs
 530       end
 531
 532     when Cancelled
 533       case self.state_was
 534       when Running
 535         permitted.push :finished_at, *progress_attrs
 536       when Queued, Locked
 537         permitted.push :finished_at, :log, :runtime_status, :cost
 538       end
 539
 540     else
 541       # The state_transitions check will add an error message for this
 542       return false
 543     end
 544
 545     if self.state_was == Running &&
 546        !current_api_client_authorization.nil? &&
 547        (current_api_client_authorization.uuid == self.auth_uuid ||
 548         current_api_client_authorization.token == self.runtime_token)
 549       # The contained process itself can write final attrs but can't
 550       # change priority or log.
 551       permitted.push *final_attrs
 552       permitted = permitted - [:log, :priority]
 553     elsif !current_user.andand.is_admin
 554       raise PermissionDeniedError
 555     elsif self.locked_by_uuid && self.locked_by_uuid != current_api_client_authorization.andand.uuid
 556       # When locked, progress fields cannot be updated by the wrong
 557       # dispatcher, even though it has admin privileges.
 558       permitted = permitted - progress_attrs
 559     end
 560     check_update_whitelist permitted
 561   end
 562
 563   def validate_lock
 564     if [Locked, Running].include? self.state
 565       # If the Container was already locked, locked_by_uuid must not
 566       # changes. Otherwise, the current auth gets the lock.
 567       need_lock = locked_by_uuid_was || current_api_client_authorization.andand.uuid
 568     else
 569       need_lock = nil
 570     end
 571
 572     # The caller can provide a new value for locked_by_uuid, but only
 573     # if it's exactly what we expect. This allows a caller to perform
 574     # an update like {"state":"Unlocked","locked_by_uuid":null}.
 575     if self.locked_by_uuid_changed?
 576       if self.locked_by_uuid != need_lock
 577         return errors.add :locked_by_uuid, "can only change to #{need_lock}"
 578       end
 579     end
 580     self.locked_by_uuid = need_lock
 581   end
 582
 583   def validate_output
 584     # Output must exist and be readable by the current user.  This is so
 585     # that a container cannot "claim" a collection that it doesn't otherwise
 586     # have access to just by setting the output field to the collection PDH.
 587     if output_changed?
 588       c = Collection.
 589             readable_by(current_user, {include_trash: true}).
 590             where(portable_data_hash: self.output).
 591             first
 592       if !c
 593         errors.add :output, "collection must exist and be readable by current user."
 594       end
 595     end
 596   end
 597
 598   def update_cr_logs
 599     # If self.final?, this update is superfluous: the final log/output
 600     # update will be done when handle_completed calls finalize! on
 601     # each requesting CR.
 602     return if self.final? || !saved_change_to_log?
 603     leave_modified_by_user_alone do
 604       ContainerRequest.where(container_uuid: self.uuid).each do |cr|
 605         cr.update_collections(container: self, collections: ['log'])
 606         cr.save!
 607       end
 608     end
 609   end
 610
 611   def assign_auth
 612     if self.auth_uuid_changed?
 613          return errors.add :auth_uuid, 'is readonly'
 614     end
 615     if not [Locked, Running].include? self.state
 616       # Don't need one. If auth already exists, expire it.
 617       #
 618       # We use db_transaction_time here (not db_current_time) to
 619       # ensure the token doesn't validate later in the same
 620       # transaction (e.g., in a test case) by satisfying expires_at >
 621       # transaction timestamp.
 622       self.auth.andand.update_attributes(expires_at: db_transaction_time)
 623       self.auth = nil
 624       return
 625     elsif self.auth
 626       # already have one
 627       return
 628     end
 629     if self.runtime_token.nil?
 630       if self.runtime_user_uuid.nil?
 631         # legacy behavior, we don't have a runtime_user_uuid so get
 632         # the user from the highest priority container request, needed
 633         # when performing an upgrade and there are queued containers,
 634         # and some tests.
 635         cr = ContainerRequest.
 636                where('container_uuid=? and priority>0', self.uuid).
 637                order('priority desc').
 638                first
 639         if !cr
 640           return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
 641         end
 642         self.runtime_user_uuid = cr.modified_by_user_uuid
 643         self.runtime_auth_scopes = ["all"]
 644       end
 645
 646       # Generate a new token. This runs with admin credentials as it's done by a
 647       # dispatcher user, so expires_at isn't enforced by API.MaxTokenLifetime.
 648       self.auth = ApiClientAuthorization.
 649                     create!(user_id: User.find_by_uuid(self.runtime_user_uuid).id,
 650                             api_client_id: 0,
 651                             scopes: self.runtime_auth_scopes)
 652     end
 653   end
 654
 655   def sort_serialized_attrs
 656     if self.environment_changed?
 657       self.environment = self.class.deep_sort_hash(self.environment)
 658     end
 659     if self.mounts_changed?
 660       self.mounts = self.class.deep_sort_hash(self.mounts)
 661     end
 662     if self.runtime_constraints_changed?
 663       self.runtime_constraints = self.class.deep_sort_hash(self.runtime_constraints)
 664     end
 665     if self.scheduling_parameters_changed?
 666       self.scheduling_parameters = self.class.deep_sort_hash(self.scheduling_parameters)
 667     end
 668     if self.runtime_auth_scopes_changed?
 669       self.runtime_auth_scopes = self.runtime_auth_scopes.sort
 670     end
 671   end
 672
 673   def update_secret_mounts_md5
 674     if self.secret_mounts_changed?
 675       self.secret_mounts_md5 = Digest::MD5.hexdigest(
 676         SafeJSON.dump(self.class.deep_sort_hash(self.secret_mounts)))
 677     end
 678   end
 679
 680   def scrub_secrets
 681     # this runs after update_secret_mounts_md5, so the
 682     # secret_mounts_md5 will still reflect the secrets that are being
 683     # scrubbed here.
 684     if self.state_changed? && self.final?
 685       self.secret_mounts = {}
 686       self.runtime_token = nil
 687     end
 688   end
 689
 690   def clear_runtime_status_when_queued
 691     # Avoid leaking status messages between different dispatch attempts
 692     if self.state_was == Locked && self.state == Queued
 693       self.runtime_status = {}
 694     end
 695   end
 696
 697   def handle_completed
 698     # This container is finished so finalize any associated container requests
 699     # that are associated with this container.
 700     if saved_change_to_state? and self.final?
 701       # These get wiped out by with_lock (which reloads the record),
 702       # so record them now in case we need to schedule a retry.
 703       prev_secret_mounts = secret_mounts_before_last_save
 704       prev_runtime_token = runtime_token_before_last_save
 705
 706       # Need to take a lock on the container to ensure that any
 707       # concurrent container requests that might try to reuse this
 708       # container will block until the container completion
 709       # transaction finishes.  This ensure that concurrent container
 710       # requests that try to reuse this container are finalized (on
 711       # Complete) or don't reuse it (on Cancelled).
 712       self.with_lock do
 713         act_as_system_user do
 714           if self.state == Cancelled
 715             retryable_requests = ContainerRequest.where("container_uuid = ? and priority > 0 and state = 'Committed' and container_count < container_count_max", uuid)
 716           else
 717             retryable_requests = []
 718           end
 719
 720           if retryable_requests.any?
 721             scheduling_parameters = {
 722               # partitions: empty if any are empty, else the union of all parameters
 723               "partitions": retryable_requests
 724                               .map { |req| req.scheduling_parameters["partitions"] || [] }
 725                               .reduce { |cur, new| (cur.empty? or new.empty?) ? [] : (cur | new) },
 726
 727               # preemptible: true if all are true, else false
 728               "preemptible": retryable_requests
 729                                .map { |req| req.scheduling_parameters["preemptible"] }
 730                                .all?,
 731
 732               # max_run_time: 0 if any are 0 (unlimited), else the maximum
 733               "max_run_time": retryable_requests
 734                                 .map { |req| req.scheduling_parameters["max_run_time"] || 0 }
 735                                 .reduce do |cur, new|
 736                 if cur == 0 or new == 0
 737                   0
 738                 elsif new > cur
 739                   new
 740                 else
 741                   cur
 742                 end
 743               end,
 744             }
 745
 746             c_attrs = {
 747               command: self.command,
 748               cwd: self.cwd,
 749               environment: self.environment,
 750               output_path: self.output_path,
 751               container_image: self.container_image,
 752               mounts: self.mounts,
 753               runtime_constraints: self.runtime_constraints,
 754               scheduling_parameters: scheduling_parameters,
 755               secret_mounts: prev_secret_mounts,
 756               runtime_token: prev_runtime_token,
 757               runtime_user_uuid: self.runtime_user_uuid,
 758               runtime_auth_scopes: self.runtime_auth_scopes
 759             }
 760             c = Container.create! c_attrs
 761             retryable_requests.each do |cr|
 762               cr.with_lock do
 763                 leave_modified_by_user_alone do
 764                   # Use row locking because this increments container_count
 765                   cr.cumulative_cost += self.cost + self.subrequests_cost
 766                   cr.container_uuid = c.uuid
 767                   cr.save!
 768                 end
 769               end
 770             end
 771           end
 772
 773           # Notify container requests associated with this container
 774           ContainerRequest.where(container_uuid: uuid,
 775                                  state: ContainerRequest::Committed).each do |cr|
 776             leave_modified_by_user_alone do
 777               cr.finalize!
 778             end
 779           end
 780
 781           # Cancel outstanding container requests made by this container.
 782           ContainerRequest.
 783             includes(:container).
 784             where(requesting_container_uuid: uuid,
 785                   state: ContainerRequest::Committed).each do |cr|
 786             leave_modified_by_user_alone do
 787               cr.update_attributes!(priority: 0)
 788               cr.container.reload
 789               if cr.container.state == Container::Queued || cr.container.state == Container::Locked
 790                 # If the child container hasn't started yet, finalize the
 791                 # child CR now instead of leaving it "on hold", i.e.,
 792                 # Queued with priority 0.  (OTOH, if the child is already
 793                 # running, leave it alone so it can get cancelled the
 794                 # usual way, get a copy of the log collection, etc.)
 795                 cr.update_attributes!(state: ContainerRequest::Final)
 796               end
 797             end
 798           end
 799         end
 800       end
 801     end
 802   end
 803 end