services/api/app/models/job.rb

   1 class Job < ArvadosModel
   2   include HasUuid
   3   include KindAndEtag
   4   include CommonApiTemplate
   5   attr_protected :arvados_sdk_version, :docker_image_locator
   6   serialize :script_parameters, Hash
   7   serialize :runtime_constraints, Hash
   8   serialize :tasks_summary, Hash
   9   before_create :ensure_unique_submit_id
  10   after_commit :trigger_crunch_dispatch_if_cancelled, :on => :update
  11   before_validation :set_priority
  12   before_validation :update_state_from_old_state_attrs
  13   validate :ensure_script_version_is_commit
  14   validate :find_arvados_sdk_version
  15   validate :find_docker_image_locator
  16   validate :validate_status
  17   validate :validate_state_change
  18   validate :ensure_no_collection_uuids_in_script_params
  19   before_save :tag_version_in_internal_repository
  20   before_save :update_timestamps_when_state_changes
  21
  22   has_many :commit_ancestors, :foreign_key => :descendant, :primary_key => :script_version
  23   has_many(:nodes, foreign_key: :job_uuid, primary_key: :uuid)
  24
  25   class SubmitIdReused < StandardError
  26   end
  27
  28   api_accessible :user, extend: :common do |t|
  29     t.add :submit_id
  30     t.add :priority
  31     t.add :script
  32     t.add :script_parameters
  33     t.add :script_version
  34     t.add :cancelled_at
  35     t.add :cancelled_by_client_uuid
  36     t.add :cancelled_by_user_uuid
  37     t.add :started_at
  38     t.add :finished_at
  39     t.add :output
  40     t.add :success
  41     t.add :running
  42     t.add :state
  43     t.add :is_locked_by_uuid
  44     t.add :log
  45     t.add :runtime_constraints
  46     t.add :tasks_summary
  47     t.add :nondeterministic
  48     t.add :repository
  49     t.add :supplied_script_version
  50     t.add :arvados_sdk_version
  51     t.add :docker_image_locator
  52     t.add :queue_position
  53     t.add :node_uuids
  54     t.add :description
  55   end
  56
  57   # Supported states for a job
  58   States = [
  59             (Queued = 'Queued'),
  60             (Running = 'Running'),
  61             (Cancelled = 'Cancelled'),
  62             (Failed = 'Failed'),
  63             (Complete = 'Complete'),
  64            ]
  65
  66   def assert_finished
  67     update_attributes(finished_at: finished_at || db_current_time,
  68                       success: success.nil? ? false : success,
  69                       running: false)
  70   end
  71
  72   def node_uuids
  73     nodes.map(&:uuid)
  74   end
  75
  76   def self.queue
  77     self.where('state = ?', Queued).order('priority desc, created_at')
  78   end
  79
  80   def queue_position
  81     Job::queue.each_with_index do |job, index|
  82       if job[:uuid] == self.uuid
  83         return index
  84       end
  85     end
  86     nil
  87   end
  88
  89   def self.running
  90     self.where('running = ?', true).
  91       order('priority desc, created_at')
  92   end
  93
  94   def lock locked_by_uuid
  95     transaction do
  96       self.reload
  97       unless self.state == Queued and self.is_locked_by_uuid.nil?
  98         raise AlreadyLockedError
  99       end
 100       self.state = Running
 101       self.is_locked_by_uuid = locked_by_uuid
 102       self.save!
 103     end
 104   end
 105
 106   protected
 107
 108   def foreign_key_attributes
 109     super + %w(output log)
 110   end
 111
 112   def skip_uuid_read_permission_check
 113     super + %w(cancelled_by_client_uuid)
 114   end
 115
 116   def skip_uuid_existence_check
 117     super + %w(output log)
 118   end
 119
 120   def set_priority
 121     if self.priority.nil?
 122       self.priority = 0
 123     end
 124     true
 125   end
 126
 127   def ensure_script_version_is_commit
 128     if self.state == Running
 129       # Apparently client has already decided to go for it. This is
 130       # needed to run a local job using a local working directory
 131       # instead of a commit-ish.
 132       return true
 133     end
 134     if new_record? or repository_changed? or script_version_changed?
 135       sha1 = Commit.find_commit_range(current_user, repository,
 136                                       nil, script_version, nil).first
 137       if not sha1
 138         errors.add :script_version, "#{script_version} does not resolve to a commit"
 139         return false
 140       end
 141       if supplied_script_version.nil? or supplied_script_version.empty?
 142         self.supplied_script_version = script_version
 143       end
 144       self.script_version = sha1
 145     end
 146     true
 147   end
 148
 149   def tag_version_in_internal_repository
 150     if self.state == Running
 151       # No point now. See ensure_script_version_is_commit.
 152       true
 153     elsif new_record? or repository_changed? or script_version_changed?
 154       uuid_was = uuid
 155       begin
 156         assign_uuid
 157         Commit.tag_in_internal_repository repository, script_version, uuid
 158       rescue
 159         uuid = uuid_was
 160         raise
 161       end
 162     end
 163   end
 164
 165   def ensure_unique_submit_id
 166     if !submit_id.nil?
 167       if Job.where('submit_id=?',self.submit_id).first
 168         raise SubmitIdReused.new
 169       end
 170     end
 171     true
 172   end
 173
 174   def resolve_runtime_constraint(key, attr_sym)
 175     if ((runtime_constraints.is_a? Hash) and
 176         (search = runtime_constraints[key]))
 177       ok, result = yield search
 178     else
 179       ok, result = true, nil
 180     end
 181     if ok
 182       send("#{attr_sym}=".to_sym, result)
 183     else
 184       errors.add(attr_sym, result)
 185     end
 186     ok
 187   end
 188
 189   def find_arvados_sdk_version
 190     resolve_runtime_constraint("arvados_sdk_version",
 191                                :arvados_sdk_version) do |git_search|
 192       commits = Commit.find_commit_range(current_user, "arvados",
 193                                          nil, git_search, nil)
 194       if commits.empty?
 195         [false, "#{git_search} does not resolve to a commit"]
 196       elsif not runtime_constraints["docker_image"]
 197         [false, "cannot be specified without a Docker image constraint"]
 198       else
 199         [true, commits.first]
 200       end
 201     end
 202   end
 203
 204   def find_docker_image_locator
 205     resolve_runtime_constraint("docker_image",
 206                                :docker_image_locator) do |image_search|
 207       image_tag = runtime_constraints['docker_image_tag']
 208       if coll = Collection.for_latest_docker_image(image_search, image_tag)
 209         [true, coll.portable_data_hash]
 210       else
 211         [false, "not found for #{image_search}"]
 212       end
 213     end
 214   end
 215
 216   def permission_to_update
 217     if is_locked_by_uuid_was and !(current_user and
 218                                    (current_user.uuid == is_locked_by_uuid_was or
 219                                     current_user.uuid == system_user.uuid))
 220       if script_changed? or
 221           script_parameters_changed? or
 222           script_version_changed? or
 223           (!cancelled_at_was.nil? and
 224            (cancelled_by_client_uuid_changed? or
 225             cancelled_by_user_uuid_changed? or
 226             cancelled_at_changed?)) or
 227           started_at_changed? or
 228           finished_at_changed? or
 229           running_changed? or
 230           success_changed? or
 231           output_changed? or
 232           log_changed? or
 233           tasks_summary_changed? or
 234           state_changed?
 235         logger.warn "User #{current_user.uuid if current_user} tried to change protected job attributes on locked #{self.class.to_s} #{uuid_was}"
 236         return false
 237       end
 238     end
 239     if !is_locked_by_uuid_changed?
 240       super
 241     else
 242       if !current_user
 243         logger.warn "Anonymous user tried to change lock on #{self.class.to_s} #{uuid_was}"
 244         false
 245       elsif is_locked_by_uuid_was and is_locked_by_uuid_was != current_user.uuid
 246         logger.warn "User #{current_user.uuid} tried to steal lock on #{self.class.to_s} #{uuid_was} from #{is_locked_by_uuid_was}"
 247         false
 248       elsif !is_locked_by_uuid.nil? and is_locked_by_uuid != current_user.uuid
 249         logger.warn "User #{current_user.uuid} tried to lock #{self.class.to_s} #{uuid_was} with uuid #{is_locked_by_uuid}"
 250         false
 251       else
 252         super
 253       end
 254     end
 255   end
 256
 257   def update_modified_by_fields
 258     if self.cancelled_at_changed?
 259       # Ensure cancelled_at cannot be set to arbitrary non-now times,
 260       # or changed once it is set.
 261       if self.cancelled_at and not self.cancelled_at_was
 262         self.cancelled_at = db_current_time
 263         self.cancelled_by_user_uuid = current_user.uuid
 264         self.cancelled_by_client_uuid = current_api_client.andand.uuid
 265         @need_crunch_dispatch_trigger = true
 266       else
 267         self.cancelled_at = self.cancelled_at_was
 268         self.cancelled_by_user_uuid = self.cancelled_by_user_uuid_was
 269         self.cancelled_by_client_uuid = self.cancelled_by_client_uuid_was
 270       end
 271     end
 272     super
 273   end
 274
 275   def trigger_crunch_dispatch_if_cancelled
 276     if @need_crunch_dispatch_trigger
 277       File.open(Rails.configuration.crunch_refresh_trigger, 'wb') do
 278         # That's all, just create/touch a file for crunch-job to see.
 279       end
 280     end
 281   end
 282
 283   def update_timestamps_when_state_changes
 284     return if not (state_changed? or new_record?)
 285
 286     case state
 287     when Running
 288       self.started_at ||= db_current_time
 289     when Failed, Complete
 290       self.finished_at ||= db_current_time
 291     when Cancelled
 292       self.cancelled_at ||= db_current_time
 293     end
 294
 295     # TODO: Remove the following case block when old "success" and
 296     # "running" attrs go away. Until then, this ensures we still
 297     # expose correct success/running flags to older clients, even if
 298     # some new clients are writing only the new state attribute.
 299     case state
 300     when Queued
 301       self.running = false
 302       self.success = nil
 303     when Running
 304       self.running = true
 305       self.success = nil
 306     when Cancelled, Failed
 307       self.running = false
 308       self.success = false
 309     when Complete
 310       self.running = false
 311       self.success = true
 312     end
 313     self.running ||= false # Default to false instead of nil.
 314
 315     @need_crunch_dispatch_trigger = true
 316
 317     true
 318   end
 319
 320   def update_state_from_old_state_attrs
 321     # If a client has touched the legacy state attrs, update the
 322     # "state" attr to agree with the updated values of the legacy
 323     # attrs.
 324     #
 325     # TODO: Remove this method when old "success" and "running" attrs
 326     # go away.
 327     if cancelled_at_changed? or
 328         success_changed? or
 329         running_changed? or
 330         state.nil?
 331       if cancelled_at
 332         self.state = Cancelled
 333       elsif success == false
 334         self.state = Failed
 335       elsif success == true
 336         self.state = Complete
 337       elsif running == true
 338         self.state = Running
 339       else
 340         self.state = Queued
 341       end
 342     end
 343     true
 344   end
 345
 346   def validate_status
 347     if self.state.in?(States)
 348       true
 349     else
 350       errors.add :state, "#{state.inspect} must be one of: #{States.inspect}"
 351       false
 352     end
 353   end
 354
 355   def validate_state_change
 356     ok = true
 357     if self.state_changed?
 358       ok = case self.state_was
 359            when nil
 360              # state isn't set yet
 361              true
 362            when Queued
 363              # Permit going from queued to any state
 364              true
 365            when Running
 366              # From running, may only transition to a finished state
 367              [Complete, Failed, Cancelled].include? self.state
 368            when Complete, Failed, Cancelled
 369              # Once in a finished state, don't permit any more state changes
 370              false
 371            else
 372              # Any other state transition is also invalid
 373              false
 374            end
 375       if not ok
 376         errors.add :state, "invalid change from #{self.state_was} to #{self.state}"
 377       end
 378     end
 379     ok
 380   end
 381
 382   def ensure_no_collection_uuids_in_script_params
 383     # recursive_hash_search searches recursively through hashes and
 384     # arrays in 'thing' for string fields matching regular expression
 385     # 'pattern'.  Returns true if pattern is found, false otherwise.
 386     def recursive_hash_search thing, pattern
 387       if thing.is_a? Hash
 388         thing.each do |k, v|
 389           return true if recursive_hash_search v, pattern
 390         end
 391       elsif thing.is_a? Array
 392         thing.each do |k|
 393           return true if recursive_hash_search k, pattern
 394         end
 395       elsif thing.is_a? String
 396         return true if thing.match pattern
 397       end
 398       false
 399     end
 400
 401     # Fail validation if any script_parameters field includes a string containing a
 402     # collection uuid pattern.
 403     if self.script_parameters_changed?
 404       if recursive_hash_search(self.script_parameters, Collection.uuid_regex)
 405         self.errors.add :script_parameters, "must use portable_data_hash instead of collection uuid"
 406         return false
 407       end
 408     end
 409     true
 410   end
 411 end