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
   9 class Container < ArvadosModel
  10   include ArvadosModelUpdates
  11   include HasUuid
  12   include KindAndEtag
  13   include CommonApiTemplate
  14   include WhitelistUpdate
  15   extend CurrentApiClient
  16   extend DbCurrentTime
  17   extend LogReuseInfo
  18
  19   # Posgresql JSONB columns should NOT be declared as serialized, Rails 5
  20   # already know how to properly treat them.
  21   attribute :secret_mounts, :jsonbHash, default: {}
  22   attribute :runtime_status, :jsonbHash, default: {}
  23   attribute :runtime_auth_scopes, :jsonbArray, default: []
  24   attribute :output_storage_classes, :jsonbArray, default: lambda { Rails.configuration.DefaultStorageClasses }
  25   attribute :output_properties, :jsonbHash, default: {}
  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
  54   has_many :container_requests, :foreign_key => :container_uuid, :class_name => 'ContainerRequest', :primary_key => :uuid
  55   belongs_to :auth, :class_name => 'ApiClientAuthorization', :foreign_key => :auth_uuid, :primary_key => :uuid
  56
  57   api_accessible :user, extend: :common do |t|
  58     t.add :command
  59     t.add :container_image
  60     t.add :cwd
  61     t.add :environment
  62     t.add :exit_code
  63     t.add :finished_at
  64     t.add :locked_by_uuid
  65     t.add :log
  66     t.add :mounts
  67     t.add :output
  68     t.add :output_path
  69     t.add :priority
  70     t.add :progress
  71     t.add :runtime_constraints
  72     t.add :runtime_status
  73     t.add :started_at
  74     t.add :state
  75     t.add :auth_uuid
  76     t.add :scheduling_parameters
  77     t.add :runtime_user_uuid
  78     t.add :runtime_auth_scopes
  79     t.add :lock_count
  80     t.add :gateway_address
  81     t.add :interactive_session_started
  82     t.add :output_storage_classes
  83     t.add :output_properties
  84     t.add :cost
  85     t.add :subrequests_cost
  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           select("priority, requesting_container_uuid, created_at").
 136           lock(true).
 137           map do |cr|
 138       if cr.requesting_container_uuid
 139         Container.where(uuid: cr.requesting_container_uuid).pluck(:priority).first
 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 = ? and state = ? and container_uuid is not null',
 155               self.uuid, ContainerRequest::Committed).
 156         pluck(:container_uuid).each do |container_uuid|
 157         Container.find_by_uuid(container_uuid).update_priority!
 158       end
 159     end
 160   end
 161
 162   # Create a new container (or find an existing one) to satisfy the
 163   # given container request.
 164   def self.resolve(req)
 165     if req.runtime_token.nil?
 166       runtime_user = if req.modified_by_user_uuid.nil?
 167                        current_user
 168                      else
 169                        User.find_by_uuid(req.modified_by_user_uuid)
 170                      end
 171       runtime_auth_scopes = ["all"]
 172     else
 173       auth = ApiClientAuthorization.validate(token: req.runtime_token)
 174       if auth.nil?
 175         raise ArgumentError.new "Invalid runtime token"
 176       end
 177       runtime_user = User.find_by_id(auth.user_id)
 178       runtime_auth_scopes = auth.scopes
 179     end
 180     c_attrs = act_as_user runtime_user do
 181       {
 182         command: req.command,
 183         cwd: req.cwd,
 184         environment: req.environment,
 185         output_path: req.output_path,
 186         container_image: resolve_container_image(req.container_image),
 187         mounts: resolve_mounts(req.mounts),
 188         runtime_constraints: resolve_runtime_constraints(req.runtime_constraints),
 189         scheduling_parameters: req.scheduling_parameters,
 190         secret_mounts: req.secret_mounts,
 191         runtime_token: req.runtime_token,
 192         runtime_user_uuid: runtime_user.uuid,
 193         runtime_auth_scopes: runtime_auth_scopes,
 194         output_storage_classes: req.output_storage_classes,
 195       }
 196     end
 197     act_as_system_user do
 198       if req.use_existing && (reusable = find_reusable(c_attrs))
 199         reusable
 200       else
 201         Container.create!(c_attrs)
 202       end
 203     end
 204   end
 205
 206   # Return a runtime_constraints hash that complies with requested but
 207   # is suitable for saving in a container record, i.e., has specific
 208   # values instead of ranges.
 209   #
 210   # Doing this as a step separate from other resolutions, like "git
 211   # revision range to commit hash", makes sense only when there is no
 212   # opportunity to reuse an existing container (e.g., container reuse
 213   # is not implemented yet, or we have already found that no existing
 214   # containers are suitable).
 215   def self.resolve_runtime_constraints(runtime_constraints)
 216     rc = {}
 217     runtime_constraints.each do |k, v|
 218       if v.is_a? Array
 219         rc[k] = v[0]
 220       else
 221         rc[k] = v
 222       end
 223     end
 224     if rc['keep_cache_ram'] == 0
 225       rc['keep_cache_ram'] = Rails.configuration.Containers.DefaultKeepCacheRAM
 226     end
 227     if rc['keep_cache_disk'] == 0 and rc['keep_cache_ram'] == 0
 228       rc['keep_cache_disk'] = bound_keep_cache_disk(rc['ram'])
 229     end
 230     rc
 231   end
 232
 233   # Return a mounts hash suitable for a Container, i.e., with every
 234   # readonly collection UUID resolved to a PDH.
 235   def self.resolve_mounts(mounts)
 236     c_mounts = {}
 237     mounts.each do |k, mount|
 238       mount = mount.dup
 239       c_mounts[k] = mount
 240       if mount['kind'] != 'collection'
 241         next
 242       end
 243
 244       uuid = mount.delete 'uuid'
 245
 246       if mount['portable_data_hash'].nil? and !uuid.nil?
 247         # PDH not supplied, try by UUID
 248         c = Collection.
 249           readable_by(current_user).
 250           where(uuid: uuid).
 251           select(:portable_data_hash).
 252           first
 253         if !c
 254           raise ArvadosModel::UnresolvableContainerError.new "cannot mount collection #{uuid.inspect}: not found"
 255         end
 256         mount['portable_data_hash'] = c.portable_data_hash
 257       end
 258     end
 259     return c_mounts
 260   end
 261
 262   # Return a container_image PDH suitable for a Container.
 263   def self.resolve_container_image(container_image)
 264     coll = Collection.for_latest_docker_image(container_image)
 265     if !coll
 266       raise ArvadosModel::UnresolvableContainerError.new "docker image #{container_image.inspect} not found"
 267     end
 268     coll.portable_data_hash
 269   end
 270
 271   def self.find_reusable(attrs)
 272     log_reuse_info { "starting with #{Container.all.count} container records in database" }
 273     candidates = Container.where_serialized(:command, attrs[:command], md5: true)
 274     log_reuse_info(candidates) { "after filtering on command #{attrs[:command].inspect}" }
 275
 276     candidates = candidates.where('cwd = ?', attrs[:cwd])
 277     log_reuse_info(candidates) { "after filtering on cwd #{attrs[:cwd].inspect}" }
 278
 279     candidates = candidates.where_serialized(:environment, attrs[:environment], md5: true)
 280     log_reuse_info(candidates) { "after filtering on environment #{attrs[:environment].inspect}" }
 281
 282     candidates = candidates.where('output_path = ?', attrs[:output_path])
 283     log_reuse_info(candidates) { "after filtering on output_path #{attrs[:output_path].inspect}" }
 284
 285     image = resolve_container_image(attrs[:container_image])
 286     candidates = candidates.where('container_image = ?', image)
 287     log_reuse_info(candidates) { "after filtering on container_image #{image.inspect} (resolved from #{attrs[:container_image].inspect})" }
 288
 289     candidates = candidates.where_serialized(:mounts, resolve_mounts(attrs[:mounts]), md5: true)
 290     log_reuse_info(candidates) { "after filtering on mounts #{attrs[:mounts].inspect}" }
 291
 292     secret_mounts_md5 = Digest::MD5.hexdigest(SafeJSON.dump(self.deep_sort_hash(attrs[:secret_mounts])))
 293     candidates = candidates.where('secret_mounts_md5 = ?', secret_mounts_md5)
 294     log_reuse_info(candidates) { "after filtering on secret_mounts_md5 #{secret_mounts_md5.inspect}" }
 295
 296     resolved_runtime_constraints = resolve_runtime_constraints(attrs[:runtime_constraints])
 297     # Ideally we would completely ignore Keep cache constraints when making
 298     # reuse considerations, but our database structure makes that impractical.
 299     # The best we can do is generate a search that matches on all likely values.
 300     runtime_constraint_variations = {
 301       keep_cache_disk: [
 302         # Check for constraints without keep_cache_disk
 303         # (containers that predate the constraint)
 304         nil,
 305         # Containers that use keep_cache_ram instead
 306         0,
 307         # The default value
 308         bound_keep_cache_disk(resolved_runtime_constraints['ram']),
 309         # The minimum default bound
 310         bound_keep_cache_disk(0),
 311         # The maximum default bound (presumably)
 312         bound_keep_cache_disk(1 << 60),
 313         # The requested value
 314         resolved_runtime_constraints.delete('keep_cache_disk'),
 315       ].uniq,
 316       keep_cache_ram: [
 317         # Containers that use keep_cache_disk instead
 318         0,
 319         # The default value
 320         Rails.configuration.Containers.DefaultKeepCacheRAM,
 321         # The requested value
 322         resolved_runtime_constraints.delete('keep_cache_ram'),
 323       ].uniq,
 324     }
 325     resolved_cuda = resolved_runtime_constraints['cuda']
 326     if resolved_cuda.nil? or resolved_cuda['device_count'] == 0
 327       runtime_constraint_variations[:cuda] = [
 328         # Check for constraints without cuda
 329         # (containers that predate the constraint)
 330         nil,
 331         # The default "don't need CUDA" value
 332         {
 333           'device_count' => 0,
 334           'driver_version' => '',
 335           'hardware_capability' => '',
 336         },
 337         # The requested value
 338         resolved_runtime_constraints.delete('cuda')
 339       ].uniq
 340     end
 341     reusable_runtime_constraints = hash_product(runtime_constraint_variations)
 342                                      .map { |v| resolved_runtime_constraints.merge(v) }
 343
 344     candidates = candidates.where_serialized(:runtime_constraints, reusable_runtime_constraints, md5: true, multivalue: true)
 345     log_reuse_info(candidates) { "after filtering on runtime_constraints #{attrs[:runtime_constraints].inspect}" }
 346
 347     log_reuse_info { "checking for state=Complete with readable output and log..." }
 348
 349     select_readable_pdh = Collection.
 350       readable_by(current_user).
 351       select(:portable_data_hash).
 352       to_sql
 353
 354     usable = candidates.where(state: Complete, exit_code: 0)
 355     log_reuse_info(usable) { "with state=Complete, exit_code=0" }
 356
 357     usable = usable.where("log IN (#{select_readable_pdh})")
 358     log_reuse_info(usable) { "with readable log" }
 359
 360     usable = usable.where("output IN (#{select_readable_pdh})")
 361     log_reuse_info(usable) { "with readable output" }
 362
 363     usable = usable.order('finished_at ASC').limit(1).first
 364     if usable
 365       log_reuse_info { "done, reusing container #{usable.uuid} with state=Complete" }
 366       return usable
 367     end
 368
 369     # Check for non-failing Running candidates and return the most likely to finish sooner.
 370     log_reuse_info { "checking for state=Running..." }
 371     running = candidates.where(state: Running).
 372               where("(runtime_status->'error') is null").
 373               order('progress desc, started_at asc').
 374               limit(1).first
 375     if running
 376       log_reuse_info { "done, reusing container #{running.uuid} with state=Running" }
 377       return running
 378     else
 379       log_reuse_info { "have no containers in Running state" }
 380     end
 381
 382     # Check for Locked or Queued ones and return the most likely to start first.
 383     locked_or_queued = candidates.
 384                        where("state IN (?)", [Locked, Queued]).
 385                        order('state asc, priority desc, created_at asc').
 386                        limit(1).first
 387     if locked_or_queued
 388       log_reuse_info { "done, reusing container #{locked_or_queued.uuid} with state=#{locked_or_queued.state}" }
 389       return locked_or_queued
 390     else
 391       log_reuse_info { "have no containers in Locked or Queued state" }
 392     end
 393
 394     log_reuse_info { "done, no reusable container found" }
 395     nil
 396   end
 397
 398   def lock
 399     self.with_lock do
 400       if self.state != Queued
 401         raise LockFailedError.new("cannot lock when #{self.state}")
 402       end
 403       self.update_attributes!(state: Locked)
 404     end
 405   end
 406
 407   def check_lock
 408     if state_was == Queued and state == Locked
 409       if self.priority <= 0
 410         raise LockFailedError.new("cannot lock when priority<=0")
 411       end
 412       self.lock_count = self.lock_count+1
 413     end
 414   end
 415
 416   def unlock
 417     self.with_lock do
 418       if self.state != Locked
 419         raise InvalidStateTransitionError.new("cannot unlock when #{self.state}")
 420       end
 421       self.update_attributes!(state: Queued)
 422     end
 423   end
 424
 425   def check_unlock
 426     if state_was == Locked and state == Queued
 427       if self.locked_by_uuid != current_api_client_authorization.uuid
 428         raise ArvadosModel::PermissionDeniedError.new("locked by a different token")
 429       end
 430       if self.lock_count >= Rails.configuration.Containers.MaxDispatchAttempts
 431         self.state = Cancelled
 432         self.runtime_status = {error: "Failed to start container.  Cancelled after exceeding 'Containers.MaxDispatchAttempts' (lock_count=#{self.lock_count})"}
 433       end
 434     end
 435   end
 436
 437   def self.readable_by(*users_list)
 438     # Load optional keyword arguments, if they exist.
 439     if users_list.last.is_a? Hash
 440       kwargs = users_list.pop
 441     else
 442       kwargs = {}
 443     end
 444     if users_list.select { |u| u.is_admin }.any?
 445       return super
 446     end
 447     Container.where(ContainerRequest.readable_by(*users_list).where("containers.uuid = container_requests.container_uuid").arel.exists)
 448   end
 449
 450   def final?
 451     [Complete, Cancelled].include?(self.state)
 452   end
 453
 454   def self.for_current_token
 455     return if !current_api_client_authorization
 456     _, _, _, container_uuid = Thread.current[:token].split('/')
 457     if container_uuid.nil?
 458       Container.where(auth_uuid: current_api_client_authorization.uuid).first
 459     else
 460       Container.where('auth_uuid=? or (uuid=? and runtime_token=?)',
 461                       current_api_client_authorization.uuid,
 462                       container_uuid,
 463                       current_api_client_authorization.token).first
 464     end
 465   end
 466
 467   protected
 468
 469   def self.bound_keep_cache_disk(value)
 470     value ||= 0
 471     min_value = 2 << 30
 472     max_value = 32 << 30
 473     if value < min_value
 474       min_value
 475     elsif value > max_value
 476       max_value
 477     else
 478       value
 479     end
 480   end
 481
 482   def self.hash_product(**kwargs)
 483     # kwargs is a hash that maps parameters to an array of values.
 484     # This function enumerates every possible hash where each key has one of
 485     # the values from its array.
 486     # The output keys are strings since that's what container hash attributes
 487     # want.
 488     # A nil value yields a hash without that key.
 489     [[:_, nil]].product(
 490       *kwargs.map { |(key, values)| [key.to_s].product(values) },
 491     ).map { |param_pairs| Hash[param_pairs].compact }
 492   end
 493
 494   def fill_field_defaults
 495     self.state ||= Queued
 496     self.environment ||= {}
 497     self.runtime_constraints ||= {}
 498     self.mounts ||= {}
 499     self.cwd ||= "."
 500     self.priority ||= 0
 501     self.scheduling_parameters ||= {}
 502   end
 503
 504   def permission_to_create
 505     current_user.andand.is_admin
 506   end
 507
 508   def permission_to_destroy
 509     current_user.andand.is_admin
 510   end
 511
 512   def ensure_owner_uuid_is_permitted
 513     # validate_change ensures owner_uuid can't be changed at all --
 514     # except during create, which requires admin privileges. Checking
 515     # permission here would be superfluous.
 516     true
 517   end
 518
 519   def set_timestamps
 520     if self.state_changed? and self.state == Running
 521       self.started_at ||= db_current_time
 522     end
 523
 524     if self.state_changed? and [Complete, Cancelled].include? self.state
 525       self.finished_at ||= db_current_time
 526     end
 527   end
 528
 529   # Check that well-known runtime status keys have desired data types
 530   def validate_runtime_status
 531     [
 532       'error', 'errorDetail', 'warning', 'warningDetail', 'activity'
 533     ].each do |k|
 534       if self.runtime_status.andand.include?(k) && !self.runtime_status[k].is_a?(String)
 535         errors.add(:runtime_status, "'#{k}' value must be a string")
 536       end
 537     end
 538   end
 539
 540   def validate_change
 541     permitted = [:state]
 542     final_attrs = [:finished_at]
 543     progress_attrs = [:progress, :runtime_status, :subrequests_cost, :cost,
 544                       :log, :output, :output_properties, :exit_code]
 545
 546     if self.new_record?
 547       permitted.push(:owner_uuid, :command, :container_image, :cwd,
 548                      :environment, :mounts, :output_path, :priority,
 549                      :runtime_constraints, :scheduling_parameters,
 550                      :secret_mounts, :runtime_token,
 551                      :runtime_user_uuid, :runtime_auth_scopes,
 552                      :output_storage_classes)
 553     end
 554
 555     case self.state
 556     when Locked
 557       permitted.push :priority, :runtime_status, :log, :lock_count
 558
 559     when Queued
 560       permitted.push :priority
 561
 562     when Running
 563       permitted.push :priority, :output_properties, :gateway_address, *progress_attrs
 564       if self.state_changed?
 565         permitted.push :started_at
 566       end
 567       if !self.interactive_session_started_was
 568         permitted.push :interactive_session_started
 569       end
 570
 571     when Complete
 572       if self.state_was == Running
 573         permitted.push *final_attrs, *progress_attrs
 574       end
 575
 576     when Cancelled
 577       case self.state_was
 578       when Running
 579         permitted.push :finished_at, *progress_attrs
 580       when Queued, Locked
 581         permitted.push :finished_at, :log, :runtime_status, :cost
 582       end
 583
 584     else
 585       # The state_transitions check will add an error message for this
 586       return false
 587     end
 588
 589     if self.state_was == Running &&
 590        !current_api_client_authorization.nil? &&
 591        (current_api_client_authorization.uuid == self.auth_uuid ||
 592         current_api_client_authorization.token == self.runtime_token)
 593       # The contained process itself can write final attrs but can't
 594       # change priority or log.
 595       permitted.push *final_attrs
 596       permitted = permitted - [:log, :priority]
 597     elsif !current_user.andand.is_admin
 598       raise PermissionDeniedError
 599     elsif self.locked_by_uuid && self.locked_by_uuid != current_api_client_authorization.andand.uuid
 600       # When locked, progress fields cannot be updated by the wrong
 601       # dispatcher, even though it has admin privileges.
 602       permitted = permitted - progress_attrs
 603     end
 604     check_update_whitelist permitted
 605   end
 606
 607   def validate_lock
 608     if [Locked, Running].include? self.state
 609       # If the Container was already locked, locked_by_uuid must not
 610       # changes. Otherwise, the current auth gets the lock.
 611       need_lock = locked_by_uuid_was || current_api_client_authorization.andand.uuid
 612     else
 613       need_lock = nil
 614     end
 615
 616     # The caller can provide a new value for locked_by_uuid, but only
 617     # if it's exactly what we expect. This allows a caller to perform
 618     # an update like {"state":"Unlocked","locked_by_uuid":null}.
 619     if self.locked_by_uuid_changed?
 620       if self.locked_by_uuid != need_lock
 621         return errors.add :locked_by_uuid, "can only change to #{need_lock}"
 622       end
 623     end
 624     self.locked_by_uuid = need_lock
 625   end
 626
 627   def validate_output
 628     # Output must exist and be readable by the current user.  This is so
 629     # that a container cannot "claim" a collection that it doesn't otherwise
 630     # have access to just by setting the output field to the collection PDH.
 631     if output_changed?
 632       c = Collection.
 633             readable_by(current_user, {include_trash: true}).
 634             where(portable_data_hash: self.output).
 635             first
 636       if !c
 637         errors.add :output, "collection must exist and be readable by current user."
 638       end
 639     end
 640   end
 641
 642   def update_cr_logs
 643     # If self.final?, this update is superfluous: the final log/output
 644     # update will be done when handle_completed calls finalize! on
 645     # each requesting CR.
 646     return if self.final? || !saved_change_to_log?
 647     leave_modified_by_user_alone do
 648       ContainerRequest.where(container_uuid: self.uuid).each do |cr|
 649         cr.update_collections(container: self, collections: ['log'])
 650         cr.save!
 651       end
 652     end
 653   end
 654
 655   def assign_auth
 656     if self.auth_uuid_changed?
 657          return errors.add :auth_uuid, 'is readonly'
 658     end
 659     if not [Locked, Running].include? self.state
 660       # Don't need one. If auth already exists, expire it.
 661       #
 662       # We use db_transaction_time here (not db_current_time) to
 663       # ensure the token doesn't validate later in the same
 664       # transaction (e.g., in a test case) by satisfying expires_at >
 665       # transaction timestamp.
 666       self.auth.andand.update_attributes(expires_at: db_transaction_time)
 667       self.auth = nil
 668       return
 669     elsif self.auth
 670       # already have one
 671       return
 672     end
 673     if self.runtime_token.nil?
 674       if self.runtime_user_uuid.nil?
 675         # legacy behavior, we don't have a runtime_user_uuid so get
 676         # the user from the highest priority container request, needed
 677         # when performing an upgrade and there are queued containers,
 678         # and some tests.
 679         cr = ContainerRequest.
 680                where('container_uuid=? and priority>0', self.uuid).
 681                order('priority desc').
 682                first
 683         if !cr
 684           return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
 685         end
 686         self.runtime_user_uuid = cr.modified_by_user_uuid
 687         self.runtime_auth_scopes = ["all"]
 688       end
 689
 690       # Generate a new token. This runs with admin credentials as it's done by a
 691       # dispatcher user, so expires_at isn't enforced by API.MaxTokenLifetime.
 692       self.auth = ApiClientAuthorization.
 693                     create!(user_id: User.find_by_uuid(self.runtime_user_uuid).id,
 694                             api_client_id: 0,
 695                             scopes: self.runtime_auth_scopes)
 696     end
 697   end
 698
 699   def sort_serialized_attrs
 700     if self.environment_changed?
 701       self.environment = self.class.deep_sort_hash(self.environment)
 702     end
 703     if self.mounts_changed?
 704       self.mounts = self.class.deep_sort_hash(self.mounts)
 705     end
 706     if self.runtime_constraints_changed?
 707       self.runtime_constraints = self.class.deep_sort_hash(self.runtime_constraints)
 708     end
 709     if self.scheduling_parameters_changed?
 710       self.scheduling_parameters = self.class.deep_sort_hash(self.scheduling_parameters)
 711     end
 712     if self.runtime_auth_scopes_changed?
 713       self.runtime_auth_scopes = self.runtime_auth_scopes.sort
 714     end
 715   end
 716
 717   def update_secret_mounts_md5
 718     if self.secret_mounts_changed?
 719       self.secret_mounts_md5 = Digest::MD5.hexdigest(
 720         SafeJSON.dump(self.class.deep_sort_hash(self.secret_mounts)))
 721     end
 722   end
 723
 724   def scrub_secrets
 725     # this runs after update_secret_mounts_md5, so the
 726     # secret_mounts_md5 will still reflect the secrets that are being
 727     # scrubbed here.
 728     if self.state_changed? && self.final?
 729       self.secret_mounts = {}
 730       self.runtime_token = nil
 731     end
 732   end
 733
 734   def clear_runtime_status_when_queued
 735     # Avoid leaking status messages between different dispatch attempts
 736     if self.state_was == Locked && self.state == Queued
 737       self.runtime_status = {}
 738     end
 739   end
 740
 741   def handle_completed
 742     # This container is finished so finalize any associated container requests
 743     # that are associated with this container.
 744     if saved_change_to_state? and self.final?
 745       # These get wiped out by with_lock (which reloads the record),
 746       # so record them now in case we need to schedule a retry.
 747       prev_secret_mounts = secret_mounts_before_last_save
 748       prev_runtime_token = runtime_token_before_last_save
 749
 750       # Need to take a lock on the container to ensure that any
 751       # concurrent container requests that might try to reuse this
 752       # container will block until the container completion
 753       # transaction finishes.  This ensure that concurrent container
 754       # requests that try to reuse this container are finalized (on
 755       # Complete) or don't reuse it (on Cancelled).
 756       self.with_lock do
 757         act_as_system_user do
 758           if self.state == Cancelled
 759             retryable_requests = ContainerRequest.where("container_uuid = ? and priority > 0 and state = 'Committed' and container_count < container_count_max", uuid)
 760           else
 761             retryable_requests = []
 762           end
 763
 764           if retryable_requests.any?
 765             scheduling_parameters = {
 766               # partitions: empty if any are empty, else the union of all parameters
 767               "partitions": retryable_requests
 768                               .map { |req| req.scheduling_parameters["partitions"] || [] }
 769                               .reduce { |cur, new| (cur.empty? or new.empty?) ? [] : (cur | new) },
 770
 771               # preemptible: true if all are true, else false
 772               "preemptible": retryable_requests
 773                                .map { |req| req.scheduling_parameters["preemptible"] }
 774                                .all?,
 775
 776               # supervisor: true if all any true, else false
 777               "supervisor": retryable_requests
 778                                .map { |req| req.scheduling_parameters["supervisor"] }
 779                                .any?,
 780
 781               # max_run_time: 0 if any are 0 (unlimited), else the maximum
 782               "max_run_time": retryable_requests
 783                                 .map { |req| req.scheduling_parameters["max_run_time"] || 0 }
 784                                 .reduce do |cur, new|
 785                 if cur == 0 or new == 0
 786                   0
 787                 elsif new > cur
 788                   new
 789                 else
 790                   cur
 791                 end
 792               end,
 793             }
 794
 795             c_attrs = {
 796               command: self.command,
 797               cwd: self.cwd,
 798               environment: self.environment,
 799               output_path: self.output_path,
 800               container_image: self.container_image,
 801               mounts: self.mounts,
 802               runtime_constraints: self.runtime_constraints,
 803               scheduling_parameters: scheduling_parameters,
 804               secret_mounts: prev_secret_mounts,
 805               runtime_token: prev_runtime_token,
 806               runtime_user_uuid: self.runtime_user_uuid,
 807               runtime_auth_scopes: self.runtime_auth_scopes
 808             }
 809             c = Container.create! c_attrs
 810             retryable_requests.each do |cr|
 811               cr.with_lock do
 812                 leave_modified_by_user_alone do
 813                   # Use row locking because this increments container_count
 814                   cr.cumulative_cost += self.cost + self.subrequests_cost
 815                   cr.container_uuid = c.uuid
 816                   cr.save!
 817                 end
 818               end
 819             end
 820           end
 821
 822           # Notify container requests associated with this container
 823           ContainerRequest.where(container_uuid: uuid,
 824                                  state: ContainerRequest::Committed).each do |cr|
 825             leave_modified_by_user_alone do
 826               cr.finalize!
 827             end
 828           end
 829
 830           # Cancel outstanding container requests made by this container.
 831           ContainerRequest.
 832             where(requesting_container_uuid: uuid,
 833                   state: ContainerRequest::Committed).
 834             in_batches(of: 15).each_record do |cr|
 835             leave_modified_by_user_alone do
 836               cr.set_priority_zero
 837               container_state = Container.where(uuid: cr.container_uuid).pluck(:state).first
 838               if container_state == Container::Queued || container_state == Container::Locked
 839                 # If the child container hasn't started yet, finalize the
 840                 # child CR now instead of leaving it "on hold", i.e.,
 841                 # Queued with priority 0.  (OTOH, if the child is already
 842                 # running, leave it alone so it can get cancelled the
 843                 # usual way, get a copy of the log collection, etc.)
 844                 cr.update_attributes!(state: ContainerRequest::Final)
 845               end
 846             end
 847           end
 848         end
 849       end
 850     end
 851   end
 852 end