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