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 :update_timestamps_when_state_changes
  20
  21   has_many :commit_ancestors, :foreign_key => :descendant, :primary_key => :script_version
  22   has_many(:nodes, foreign_key: :job_uuid, primary_key: :uuid)
  23
  24   class SubmitIdReused < StandardError
  25   end
  26
  27   api_accessible :user, extend: :common do |t|
  28     t.add :submit_id
  29     t.add :priority
  30     t.add :script
  31     t.add :script_parameters
  32     t.add :script_version
  33     t.add :cancelled_at
  34     t.add :cancelled_by_client_uuid
  35     t.add :cancelled_by_user_uuid
  36     t.add :started_at
  37     t.add :finished_at
  38     t.add :output
  39     t.add :success
  40     t.add :running
  41     t.add :state
  42     t.add :is_locked_by_uuid
  43     t.add :log
  44     t.add :runtime_constraints
  45     t.add :tasks_summary
  46     t.add :dependencies
  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 || Time.now,
  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 script_version_changed?
 135       sha1 = Commit.find_commit_range(current_user, self.repository, nil, self.script_version, nil)[0] rescue nil
 136       if sha1
 137         self.supplied_script_version = self.script_version if self.supplied_script_version.nil? or self.supplied_script_version.empty?
 138         self.script_version = sha1
 139       else
 140         self.errors.add :script_version, "#{self.script_version} does not resolve to a commit"
 141         return false
 142       end
 143     end
 144   end
 145
 146   def ensure_unique_submit_id
 147     if !submit_id.nil?
 148       if Job.where('submit_id=?',self.submit_id).first
 149         raise SubmitIdReused.new
 150       end
 151     end
 152     true
 153   end
 154
 155   def resolve_runtime_constraint(key, attr_sym)
 156     if ((runtime_constraints.is_a? Hash) and
 157         (search = runtime_constraints[key]))
 158       ok, result = yield search
 159     else
 160       ok, result = true, nil
 161     end
 162     if ok
 163       send("#{attr_sym}=".to_sym, result)
 164     else
 165       errors.add(attr_sym, result)
 166     end
 167     ok
 168   end
 169
 170   def find_arvados_sdk_version
 171     resolve_runtime_constraint("arvados_sdk_version",
 172                                :arvados_sdk_version) do |git_search|
 173       commits = Commit.find_commit_range(current_user, "arvados",
 174                                          nil, git_search, nil)
 175       if commits.nil? or commits.empty?
 176         [false, "#{git_search} does not resolve to a commit"]
 177       elsif not runtime_constraints["docker_image"]
 178         [false, "cannot be specified without a Docker image constraint"]
 179       else
 180         [true, commits.first]
 181       end
 182     end
 183   end
 184
 185   def find_docker_image_locator
 186     resolve_runtime_constraint("docker_image",
 187                                :docker_image_locator) do |image_search|
 188       image_tag = runtime_constraints['docker_image_tag']
 189       if coll = Collection.for_latest_docker_image(image_search, image_tag)
 190         [true, coll.portable_data_hash]
 191       else
 192         [false, "not found for #{image_search}"]
 193       end
 194     end
 195   end
 196
 197   def dependencies
 198     deps = {}
 199     queue = self.script_parameters.values
 200     while not queue.empty?
 201       queue = queue.flatten.compact.collect do |v|
 202         if v.is_a? Hash
 203           v.values
 204         elsif v.is_a? String
 205           v.match(/^(([0-9a-f]{32})\b(\+[^,]+)?,?)*$/) do |locator|
 206             deps[locator.to_s] = true
 207           end
 208           nil
 209         end
 210       end
 211     end
 212     deps.keys
 213   end
 214
 215   def permission_to_update
 216     if is_locked_by_uuid_was and !(current_user and
 217                                    (current_user.uuid == is_locked_by_uuid_was or
 218                                     current_user.uuid == system_user.uuid))
 219       if script_changed? or
 220           script_parameters_changed? or
 221           script_version_changed? or
 222           (!cancelled_at_was.nil? and
 223            (cancelled_by_client_uuid_changed? or
 224             cancelled_by_user_uuid_changed? or
 225             cancelled_at_changed?)) or
 226           started_at_changed? or
 227           finished_at_changed? or
 228           running_changed? or
 229           success_changed? or
 230           output_changed? or
 231           log_changed? or
 232           tasks_summary_changed? or
 233           state_changed?
 234         logger.warn "User #{current_user.uuid if current_user} tried to change protected job attributes on locked #{self.class.to_s} #{uuid_was}"
 235         return false
 236       end
 237     end
 238     if !is_locked_by_uuid_changed?
 239       super
 240     else
 241       if !current_user
 242         logger.warn "Anonymous user tried to change lock on #{self.class.to_s} #{uuid_was}"
 243         false
 244       elsif is_locked_by_uuid_was and is_locked_by_uuid_was != current_user.uuid
 245         logger.warn "User #{current_user.uuid} tried to steal lock on #{self.class.to_s} #{uuid_was} from #{is_locked_by_uuid_was}"
 246         false
 247       elsif !is_locked_by_uuid.nil? and is_locked_by_uuid != current_user.uuid
 248         logger.warn "User #{current_user.uuid} tried to lock #{self.class.to_s} #{uuid_was} with uuid #{is_locked_by_uuid}"
 249         false
 250       else
 251         super
 252       end
 253     end
 254   end
 255
 256   def update_modified_by_fields
 257     if self.cancelled_at_changed?
 258       # Ensure cancelled_at cannot be set to arbitrary non-now times,
 259       # or changed once it is set.
 260       if self.cancelled_at and not self.cancelled_at_was
 261         self.cancelled_at = Time.now
 262         self.cancelled_by_user_uuid = current_user.uuid
 263         self.cancelled_by_client_uuid = current_api_client.andand.uuid
 264         @need_crunch_dispatch_trigger = true
 265       else
 266         self.cancelled_at = self.cancelled_at_was
 267         self.cancelled_by_user_uuid = self.cancelled_by_user_uuid_was
 268         self.cancelled_by_client_uuid = self.cancelled_by_client_uuid_was
 269       end
 270     end
 271     super
 272   end
 273
 274   def trigger_crunch_dispatch_if_cancelled
 275     if @need_crunch_dispatch_trigger
 276       File.open(Rails.configuration.crunch_refresh_trigger, 'wb') do
 277         # That's all, just create/touch a file for crunch-job to see.
 278       end
 279     end
 280   end
 281
 282   def update_timestamps_when_state_changes
 283     return if not (state_changed? or new_record?)
 284
 285     case state
 286     when Running
 287       self.started_at ||= Time.now
 288     when Failed, Complete
 289       self.finished_at ||= Time.now
 290     when Cancelled
 291       self.cancelled_at ||= Time.now
 292     end
 293
 294     # TODO: Remove the following case block when old "success" and
 295     # "running" attrs go away. Until then, this ensures we still
 296     # expose correct success/running flags to older clients, even if
 297     # some new clients are writing only the new state attribute.
 298     case state
 299     when Queued
 300       self.running = false
 301       self.success = nil
 302     when Running
 303       self.running = true
 304       self.success = nil
 305     when Cancelled, Failed
 306       self.running = false
 307       self.success = false
 308     when Complete
 309       self.running = false
 310       self.success = true
 311     end
 312     self.running ||= false # Default to false instead of nil.
 313
 314     @need_crunch_dispatch_trigger = true
 315
 316     true
 317   end
 318
 319   def update_state_from_old_state_attrs
 320     # If a client has touched the legacy state attrs, update the
 321     # "state" attr to agree with the updated values of the legacy
 322     # attrs.
 323     #
 324     # TODO: Remove this method when old "success" and "running" attrs
 325     # go away.
 326     if cancelled_at_changed? or
 327         success_changed? or
 328         running_changed? or
 329         state.nil?
 330       if cancelled_at
 331         self.state = Cancelled
 332       elsif success == false
 333         self.state = Failed
 334       elsif success == true
 335         self.state = Complete
 336       elsif running == true
 337         self.state = Running
 338       else
 339         self.state = Queued
 340       end
 341     end
 342     true
 343   end
 344
 345   def validate_status
 346     if self.state.in?(States)
 347       true
 348     else
 349       errors.add :state, "#{state.inspect} must be one of: #{States.inspect}"
 350       false
 351     end
 352   end
 353
 354   def validate_state_change
 355     ok = true
 356     if self.state_changed?
 357       ok = case self.state_was
 358            when nil
 359              # state isn't set yet
 360              true
 361            when Queued
 362              # Permit going from queued to any state
 363              true
 364            when Running
 365              # From running, may only transition to a finished state
 366              [Complete, Failed, Cancelled].include? self.state
 367            when Complete, Failed, Cancelled
 368              # Once in a finished state, don't permit any more state changes
 369              false
 370            else
 371              # Any other state transition is also invalid
 372              false
 373            end
 374       if not ok
 375         errors.add :state, "invalid change from #{self.state_was} to #{self.state}"
 376       end
 377     end
 378     ok
 379   end
 380
 381   def ensure_no_collection_uuids_in_script_params
 382     # recursive_hash_search searches recursively through hashes and
 383     # arrays in 'thing' for string fields matching regular expression
 384     # 'pattern'.  Returns true if pattern is found, false otherwise.
 385     def recursive_hash_search thing, pattern
 386       if thing.is_a? Hash
 387         thing.each do |k, v|
 388           return true if recursive_hash_search v, pattern
 389         end
 390       elsif thing.is_a? Array
 391         thing.each do |k|
 392           return true if recursive_hash_search k, pattern
 393         end
 394       elsif thing.is_a? String
 395         return true if thing.match pattern
 396       end
 397       false
 398     end
 399
 400     # Fail validation if any script_parameters field includes a string containing a
 401     # collection uuid pattern.
 402     if self.script_parameters_changed?
 403       if recursive_hash_search(self.script_parameters, Collection.uuid_regex)
 404         self.errors.add :script_parameters, "must use portable_data_hash instead of collection uuid"
 405         return false
 406       end
 407     end
 408     true
 409   end
 410 end