services/api/app/models/container.rb

   1 require 'whitelist_update'
   2 require 'safe_json'
   3
   4 class Container < ArvadosModel
   5   include HasUuid
   6   include KindAndEtag
   7   include CommonApiTemplate
   8   include WhitelistUpdate
   9   extend CurrentApiClient
  10
  11   serialize :environment, Hash
  12   serialize :mounts, Hash
  13   serialize :runtime_constraints, Hash
  14   serialize :command, Array
  15   serialize :scheduling_parameters, Hash
  16
  17   before_validation :fill_field_defaults, :if => :new_record?
  18   before_validation :set_timestamps
  19   validates :command, :container_image, :output_path, :cwd, :priority, :presence => true
  20   validate :validate_state_change
  21   validate :validate_change
  22   validate :validate_lock
  23   validate :validate_output
  24   after_validation :assign_auth
  25   before_save :sort_serialized_attrs
  26   after_save :handle_completed
  27
  28   has_many :container_requests, :foreign_key => :container_uuid, :class_name => 'ContainerRequest', :primary_key => :uuid
  29   belongs_to :auth, :class_name => 'ApiClientAuthorization', :foreign_key => :auth_uuid, :primary_key => :uuid
  30
  31   api_accessible :user, extend: :common do |t|
  32     t.add :command
  33     t.add :container_image
  34     t.add :cwd
  35     t.add :environment
  36     t.add :exit_code
  37     t.add :finished_at
  38     t.add :locked_by_uuid
  39     t.add :log
  40     t.add :mounts
  41     t.add :output
  42     t.add :output_path
  43     t.add :priority
  44     t.add :progress
  45     t.add :runtime_constraints
  46     t.add :started_at
  47     t.add :state
  48     t.add :auth_uuid
  49     t.add :scheduling_parameters
  50   end
  51
  52   # Supported states for a container
  53   States =
  54     [
  55      (Queued = 'Queued'),
  56      (Locked = 'Locked'),
  57      (Running = 'Running'),
  58      (Complete = 'Complete'),
  59      (Cancelled = 'Cancelled')
  60     ]
  61
  62   State_transitions = {
  63     nil => [Queued],
  64     Queued => [Locked, Cancelled],
  65     Locked => [Queued, Running, Cancelled],
  66     Running => [Complete, Cancelled]
  67   }
  68
  69   def state_transitions
  70     State_transitions
  71   end
  72
  73   def update_priority!
  74     if [Queued, Locked, Running].include? self.state
  75       # Update the priority of this container to the maximum priority of any of
  76       # its committed container requests and save the record.
  77       self.priority = ContainerRequest.
  78         where(container_uuid: uuid,
  79               state: ContainerRequest::Committed).
  80         maximum('priority')
  81       self.save!
  82     end
  83   end
  84
  85   # Create a new container (or find an existing one) to satisfy the
  86   # given container request.
  87   def self.resolve(req)
  88     c_attrs = {
  89       command: req.command,
  90       cwd: req.cwd,
  91       environment: req.environment,
  92       output_path: req.output_path,
  93       container_image: resolve_container_image(req.container_image),
  94       mounts: resolve_mounts(req.mounts),
  95       runtime_constraints: resolve_runtime_constraints(req.runtime_constraints),
  96       scheduling_parameters: req.scheduling_parameters,
  97     }
  98     act_as_system_user do
  99       if req.use_existing && (reusable = find_reusable(c_attrs))
 100         reusable
 101       else
 102         Container.create!(c_attrs)
 103       end
 104     end
 105   end
 106
 107   # Return a runtime_constraints hash that complies with requested but
 108   # is suitable for saving in a container record, i.e., has specific
 109   # values instead of ranges.
 110   #
 111   # Doing this as a step separate from other resolutions, like "git
 112   # revision range to commit hash", makes sense only when there is no
 113   # opportunity to reuse an existing container (e.g., container reuse
 114   # is not implemented yet, or we have already found that no existing
 115   # containers are suitable).
 116   def self.resolve_runtime_constraints(runtime_constraints)
 117     rc = {}
 118     defaults = {
 119       'keep_cache_ram' =>
 120       Rails.configuration.container_default_keep_cache_ram,
 121     }
 122     defaults.merge(runtime_constraints).each do |k, v|
 123       if v.is_a? Array
 124         rc[k] = v[0]
 125       else
 126         rc[k] = v
 127       end
 128     end
 129     rc
 130   end
 131
 132   # Return a mounts hash suitable for a Container, i.e., with every
 133   # readonly collection UUID resolved to a PDH.
 134   def self.resolve_mounts(mounts)
 135     c_mounts = {}
 136     mounts.each do |k, mount|
 137       mount = mount.dup
 138       c_mounts[k] = mount
 139       if mount['kind'] != 'collection'
 140         next
 141       end
 142       if (uuid = mount.delete 'uuid')
 143         c = Collection.
 144           readable_by(current_user).
 145           where(uuid: uuid).
 146           select(:portable_data_hash).
 147           first
 148         if !c
 149           raise ArvadosModel::UnresolvableContainerError.new "cannot mount collection #{uuid.inspect}: not found"
 150         end
 151         if mount['portable_data_hash'].nil?
 152           # PDH not supplied by client
 153           mount['portable_data_hash'] = c.portable_data_hash
 154         elsif mount['portable_data_hash'] != c.portable_data_hash
 155           # UUID and PDH supplied by client, but they don't agree
 156           raise ArgumentError.new "cannot mount collection #{uuid.inspect}: current portable_data_hash #{c.portable_data_hash.inspect} does not match #{c['portable_data_hash'].inspect} in request"
 157         end
 158       end
 159     end
 160     return c_mounts
 161   end
 162
 163   # Return a container_image PDH suitable for a Container.
 164   def self.resolve_container_image(container_image)
 165     coll = Collection.for_latest_docker_image(container_image)
 166     if !coll
 167       raise ArvadosModel::UnresolvableContainerError.new "docker image #{container_image.inspect} not found"
 168     end
 169     coll.portable_data_hash
 170   end
 171
 172   def self.find_reusable(attrs)
 173     candidates = Container.
 174       where_serialized(:command, attrs[:command]).
 175       where('cwd = ?', attrs[:cwd]).
 176       where_serialized(:environment, attrs[:environment]).
 177       where('output_path = ?', attrs[:output_path]).
 178       where('container_image = ?', resolve_container_image(attrs[:container_image])).
 179       where_serialized(:mounts, resolve_mounts(attrs[:mounts])).
 180       where_serialized(:runtime_constraints, resolve_runtime_constraints(attrs[:runtime_constraints]))
 181
 182     # Check for Completed candidates whose output and log are both readable.
 183     select_readable_pdh = Collection.
 184       readable_by(current_user).
 185       select(:portable_data_hash).
 186       to_sql
 187     usable = candidates.
 188       where(state: Complete).
 189       where(exit_code: 0).
 190       where("log IN (#{select_readable_pdh})").
 191       where("output IN (#{select_readable_pdh})").
 192       order('finished_at ASC').
 193       limit(1).
 194       first
 195     return usable if usable
 196
 197     # Check for Running candidates and return the most likely to finish sooner.
 198     running = candidates.where(state: Running).
 199       order('progress desc, started_at asc').limit(1).first
 200     return running if not running.nil?
 201
 202     # Check for Locked or Queued ones and return the most likely to start first.
 203     locked_or_queued = candidates.where("state IN (?)", [Locked, Queued]).
 204       order('state asc, priority desc, created_at asc').limit(1).first
 205     return locked_or_queued if not locked_or_queued.nil?
 206
 207     # No suitable candidate found.
 208     nil
 209   end
 210
 211   def lock
 212     with_lock do
 213       if self.state == Locked
 214         raise AlreadyLockedError
 215       end
 216       self.state = Locked
 217       self.save!
 218     end
 219   end
 220
 221   def unlock
 222     with_lock do
 223       if self.state == Queued
 224         raise InvalidStateTransitionError
 225       end
 226       self.state = Queued
 227       self.save!
 228     end
 229   end
 230
 231   def self.readable_by(*users_list)
 232     if users_list.select { |u| u.is_admin }.any?
 233       return self
 234     end
 235     user_uuids = users_list.map { |u| u.uuid }
 236     uuid_list = user_uuids + users_list.flat_map { |u| u.groups_i_can(:read) }
 237     uuid_list.uniq!
 238     permitted = "(SELECT head_uuid FROM links WHERE link_class='permission' AND tail_uuid IN (:uuids))"
 239     joins(:container_requests).
 240       where("container_requests.uuid IN #{permitted} OR "+
 241             "container_requests.owner_uuid IN (:uuids)",
 242             uuids: uuid_list)
 243   end
 244
 245   def final?
 246     [Complete, Cancelled].include?(self.state)
 247   end
 248
 249   protected
 250
 251   def fill_field_defaults
 252     self.state ||= Queued
 253     self.environment ||= {}
 254     self.runtime_constraints ||= {}
 255     self.mounts ||= {}
 256     self.cwd ||= "."
 257     self.priority ||= 1
 258     self.scheduling_parameters ||= {}
 259   end
 260
 261   def permission_to_create
 262     current_user.andand.is_admin
 263   end
 264
 265   def permission_to_update
 266     # Override base permission check to allow auth_uuid to set progress and
 267     # output (only).  Whether it is legal to set progress and output in the current
 268     # state has already been checked in validate_change.
 269     current_user.andand.is_admin ||
 270       (!current_api_client_authorization.nil? and
 271        [self.auth_uuid, self.locked_by_uuid].include? current_api_client_authorization.uuid)
 272   end
 273
 274   def ensure_owner_uuid_is_permitted
 275     # Override base permission check to allow auth_uuid to set progress and
 276     # output (only).  Whether it is legal to set progress and output in the current
 277     # state has already been checked in validate_change.
 278     if !current_api_client_authorization.nil? and self.auth_uuid == current_api_client_authorization.uuid
 279       check_update_whitelist [:progress, :output]
 280     else
 281       super
 282     end
 283   end
 284
 285   def set_timestamps
 286     if self.state_changed? and self.state == Running
 287       self.started_at ||= db_current_time
 288     end
 289
 290     if self.state_changed? and [Complete, Cancelled].include? self.state
 291       self.finished_at ||= db_current_time
 292     end
 293   end
 294
 295   def validate_change
 296     permitted = [:state]
 297
 298     if self.new_record?
 299       permitted.push(:owner_uuid, :command, :container_image, :cwd,
 300                      :environment, :mounts, :output_path, :priority,
 301                      :runtime_constraints, :scheduling_parameters)
 302     end
 303
 304     case self.state
 305     when Queued, Locked
 306       permitted.push :priority
 307
 308     when Running
 309       permitted.push :priority, :progress, :output
 310       if self.state_changed?
 311         permitted.push :started_at
 312       end
 313
 314     when Complete
 315       if self.state_was == Running
 316         permitted.push :finished_at, :output, :log, :exit_code
 317       end
 318
 319     when Cancelled
 320       case self.state_was
 321       when Running
 322         permitted.push :finished_at, :output, :log
 323       when Queued, Locked
 324         permitted.push :finished_at
 325       end
 326
 327     else
 328       # The state_transitions check will add an error message for this
 329       return false
 330     end
 331
 332     check_update_whitelist permitted
 333   end
 334
 335   def validate_lock
 336     if [Locked, Running].include? self.state
 337       # If the Container was already locked, locked_by_uuid must not
 338       # changes. Otherwise, the current auth gets the lock.
 339       need_lock = locked_by_uuid_was || current_api_client_authorization.andand.uuid
 340     else
 341       need_lock = nil
 342     end
 343
 344     # The caller can provide a new value for locked_by_uuid, but only
 345     # if it's exactly what we expect. This allows a caller to perform
 346     # an update like {"state":"Unlocked","locked_by_uuid":null}.
 347     if self.locked_by_uuid_changed?
 348       if self.locked_by_uuid != need_lock
 349         return errors.add :locked_by_uuid, "can only change to #{need_lock}"
 350       end
 351     end
 352     self.locked_by_uuid = need_lock
 353   end
 354
 355   def validate_output
 356     # Output must exist and be readable by the current user.  This is so
 357     # that a container cannot "claim" a collection that it doesn't otherwise
 358     # have access to just by setting the output field to the collection PDH.
 359     if output_changed?
 360       c = Collection.unscoped do
 361         Collection.
 362             readable_by(current_user).
 363             where(portable_data_hash: self.output).
 364             first
 365       end
 366       if !c
 367         errors.add :output, "collection must exist and be readable by current user."
 368       end
 369     end
 370   end
 371
 372   def assign_auth
 373     if self.auth_uuid_changed?
 374       return errors.add :auth_uuid, 'is readonly'
 375     end
 376     if not [Locked, Running].include? self.state
 377       # don't need one
 378       self.auth.andand.update_attributes(expires_at: db_current_time)
 379       self.auth = nil
 380       return
 381     elsif self.auth
 382       # already have one
 383       return
 384     end
 385     cr = ContainerRequest.
 386       where('container_uuid=? and priority>0', self.uuid).
 387       order('priority desc').
 388       first
 389     if !cr
 390       return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
 391     end
 392     self.auth = ApiClientAuthorization.
 393       create!(user_id: User.find_by_uuid(cr.modified_by_user_uuid).id,
 394               api_client_id: 0)
 395   end
 396
 397   def sort_serialized_attrs
 398     if self.environment_changed?
 399       self.environment = self.class.deep_sort_hash(self.environment)
 400     end
 401     if self.mounts_changed?
 402       self.mounts = self.class.deep_sort_hash(self.mounts)
 403     end
 404     if self.runtime_constraints_changed?
 405       self.runtime_constraints = self.class.deep_sort_hash(self.runtime_constraints)
 406     end
 407     if self.scheduling_parameters_changed?
 408       self.scheduling_parameters = self.class.deep_sort_hash(self.scheduling_parameters)
 409     end
 410   end
 411
 412   def handle_completed
 413     # This container is finished so finalize any associated container requests
 414     # that are associated with this container.
 415     if self.state_changed? and self.final?
 416       act_as_system_user do
 417
 418         if self.state == Cancelled
 419           retryable_requests = ContainerRequest.where("container_uuid = ? and priority > 0 and state = 'Committed' and container_count < container_count_max", uuid)
 420         else
 421           retryable_requests = []
 422         end
 423
 424         if retryable_requests.any?
 425           c_attrs = {
 426             command: self.command,
 427             cwd: self.cwd,
 428             environment: self.environment,
 429             output_path: self.output_path,
 430             container_image: self.container_image,
 431             mounts: self.mounts,
 432             runtime_constraints: self.runtime_constraints,
 433             scheduling_parameters: self.scheduling_parameters
 434           }
 435           c = Container.create! c_attrs
 436           retryable_requests.each do |cr|
 437             cr.with_lock do
 438               # Use row locking because this increments container_count
 439               cr.container_uuid = c.uuid
 440               cr.save
 441             end
 442           end
 443         end
 444
 445         # Notify container requests associated with this container
 446         ContainerRequest.where(container_uuid: uuid,
 447                                state: ContainerRequest::Committed).each do |cr|
 448           cr.finalize!
 449         end
 450
 451         # Try to cancel any outstanding container requests made by this container.
 452         ContainerRequest.where(requesting_container_uuid: uuid,
 453                                state: ContainerRequest::Committed).each do |cr|
 454           cr.priority = 0
 455           cr.save
 456         end
 457       end
 458     end
 459   end
 460
 461 end