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