services/api/app/models/job.rb

   1 class Job < ArvadosModel
   2   include HasUuid
   3   include KindAndEtag
   4   include CommonApiTemplate
   5   attr_protected :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   validate :ensure_script_version_is_commit
  13   validate :find_docker_image_locator
  14   before_validation :verify_status
  15   before_create :set_state_before_save
  16   before_save :set_state_before_save
  17
  18   has_many :commit_ancestors, :foreign_key => :descendant, :primary_key => :script_version
  19
  20   class SubmitIdReused < StandardError
  21   end
  22
  23   api_accessible :user, extend: :common do |t|
  24     t.add :submit_id
  25     t.add :priority
  26     t.add :script
  27     t.add :script_parameters
  28     t.add :script_version
  29     t.add :cancelled_at
  30     t.add :cancelled_by_client_uuid
  31     t.add :cancelled_by_user_uuid
  32     t.add :started_at
  33     t.add :finished_at
  34     t.add :output
  35     t.add :success
  36     t.add :running
  37     t.add :state
  38     t.add :is_locked_by_uuid
  39     t.add :log
  40     t.add :runtime_constraints
  41     t.add :tasks_summary
  42     t.add :dependencies
  43     t.add :nondeterministic
  44     t.add :repository
  45     t.add :supplied_script_version
  46     t.add :docker_image_locator
  47     t.add :queue_position
  48     t.add :description
  49   end
  50
  51   # Supported states for a job
  52   States = [
  53             (Queued = 'Queued'),
  54             (Running = 'Running'),
  55             (Cancelled = 'Cancelled'),
  56             (Failed = 'Failed'),
  57             (Complete = 'Complete'),
  58            ]
  59
  60   def assert_finished
  61     update_attributes(finished_at: finished_at || Time.now,
  62                       success: success.nil? ? false : success,
  63                       running: false)
  64   end
  65
  66   def self.queue
  67     self.where('started_at is ? and is_locked_by_uuid is ? and cancelled_at is ? and success is ?',
  68                nil, nil, nil, nil).
  69       order('priority desc, created_at')
  70   end
  71
  72   def queue_position
  73     i = 0
  74     Job::queue.each do |j|
  75       if j[:uuid] == self.uuid
  76         return i
  77       end
  78     end
  79     nil
  80   end
  81
  82   def self.running
  83     self.where('running = ?', true).
  84       order('priority desc, created_at')
  85   end
  86
  87   protected
  88
  89   def foreign_key_attributes
  90     super + %w(output log)
  91   end
  92
  93   def skip_uuid_read_permission_check
  94     super + %w(cancelled_by_client_uuid)
  95   end
  96
  97   def skip_uuid_existence_check
  98     super + %w(output log)
  99   end
 100
 101   def set_priority
 102     if self.priority.nil?
 103       self.priority = 0
 104     end
 105     true
 106   end
 107
 108   def ensure_script_version_is_commit
 109     if self.is_locked_by_uuid and self.started_at
 110       # Apparently client has already decided to go for it. This is
 111       # needed to run a local job using a local working directory
 112       # instead of a commit-ish.
 113       return true
 114     end
 115     if new_record? or script_version_changed?
 116       sha1 = Commit.find_commit_range(current_user, self.repository, nil, self.script_version, nil)[0] rescue nil
 117       if sha1
 118         self.supplied_script_version = self.script_version if self.supplied_script_version.nil? or self.supplied_script_version.empty?
 119         self.script_version = sha1
 120       else
 121         self.errors.add :script_version, "#{self.script_version} does not resolve to a commit"
 122         return false
 123       end
 124     end
 125   end
 126
 127   def ensure_unique_submit_id
 128     if !submit_id.nil?
 129       if Job.where('submit_id=?',self.submit_id).first
 130         raise SubmitIdReused.new
 131       end
 132     end
 133     true
 134   end
 135
 136   def find_docker_image_locator
 137     # Find the Collection that holds the Docker image specified in the
 138     # runtime constraints, and store its locator in docker_image_locator.
 139     unless runtime_constraints.is_a? Hash
 140       # We're still in validation stage, so we can't assume
 141       # runtime_constraints isn't something horrible like an array or
 142       # a string. Treat those cases as "no docker image supplied";
 143       # other validations will fail anyway.
 144       self.docker_image_locator = nil
 145       return true
 146     end
 147     image_search = runtime_constraints['docker_image']
 148     image_tag = runtime_constraints['docker_image_tag']
 149     if image_search.nil?
 150       self.docker_image_locator = nil
 151       true
 152     elsif coll = Collection.for_latest_docker_image(image_search, image_tag)
 153       self.docker_image_locator = coll.portable_data_hash
 154       true
 155     else
 156       errors.add(:docker_image_locator, "not found for #{image_search}")
 157       false
 158     end
 159   end
 160
 161   def dependencies
 162     deps = {}
 163     queue = self.script_parameters.values
 164     while not queue.empty?
 165       queue = queue.flatten.compact.collect do |v|
 166         if v.is_a? Hash
 167           v.values
 168         elsif v.is_a? String
 169           v.match(/^(([0-9a-f]{32})\b(\+[^,]+)?,?)*$/) do |locator|
 170             deps[locator.to_s] = true
 171           end
 172           nil
 173         end
 174       end
 175     end
 176     deps.keys
 177   end
 178
 179   def permission_to_update
 180     if is_locked_by_uuid_was and !(current_user and
 181                                    (current_user.uuid == is_locked_by_uuid_was or
 182                                     current_user.uuid == system_user.uuid))
 183       if script_changed? or
 184           script_parameters_changed? or
 185           script_version_changed? or
 186           (!cancelled_at_was.nil? and
 187            (cancelled_by_client_uuid_changed? or
 188             cancelled_by_user_uuid_changed? or
 189             cancelled_at_changed?)) or
 190           started_at_changed? or
 191           finished_at_changed? or
 192           running_changed? or
 193           success_changed? or
 194           output_changed? or
 195           log_changed? or
 196           tasks_summary_changed?
 197         logger.warn "User #{current_user.uuid if current_user} tried to change protected job attributes on locked #{self.class.to_s} #{uuid_was}"
 198         return false
 199       end
 200     end
 201     if !is_locked_by_uuid_changed?
 202       super
 203     else
 204       if !current_user
 205         logger.warn "Anonymous user tried to change lock on #{self.class.to_s} #{uuid_was}"
 206         false
 207       elsif is_locked_by_uuid_was and is_locked_by_uuid_was != current_user.uuid
 208         logger.warn "User #{current_user.uuid} tried to steal lock on #{self.class.to_s} #{uuid_was} from #{is_locked_by_uuid_was}"
 209         false
 210       elsif !is_locked_by_uuid.nil? and is_locked_by_uuid != current_user.uuid
 211         logger.warn "User #{current_user.uuid} tried to lock #{self.class.to_s} #{uuid_was} with uuid #{is_locked_by_uuid}"
 212         false
 213       else
 214         super
 215       end
 216     end
 217   end
 218
 219   def update_modified_by_fields
 220     if self.cancelled_at_changed?
 221       # Ensure cancelled_at cannot be set to arbitrary non-now times,
 222       # or changed once it is set.
 223       if self.cancelled_at and not self.cancelled_at_was
 224         self.cancelled_at = Time.now
 225         self.cancelled_by_user_uuid = current_user.uuid
 226         self.cancelled_by_client_uuid = current_api_client.andand.uuid
 227         @need_crunch_dispatch_trigger = true
 228       else
 229         self.cancelled_at = self.cancelled_at_was
 230         self.cancelled_by_user_uuid = self.cancelled_by_user_uuid_was
 231         self.cancelled_by_client_uuid = self.cancelled_by_client_uuid_was
 232       end
 233     end
 234     super
 235   end
 236
 237   def trigger_crunch_dispatch_if_cancelled
 238     if @need_crunch_dispatch_trigger
 239       File.open(Rails.configuration.crunch_refresh_trigger, 'wb') do
 240         # That's all, just create/touch a file for crunch-job to see.
 241       end
 242     end
 243   end
 244
 245   def verify_status
 246     changed_attributes = self.changed
 247
 248     if new_record?
 249       self.state = Queued
 250     elsif 'state'.in? changed_attributes
 251       case self.state
 252       when Queued
 253         self.running = false
 254         self.success = nil
 255       when Running
 256         if !self.started_at
 257           self.started_at = Time.now
 258         end
 259         self.running = true
 260         self.success = nil
 261       when Cancelled
 262         if !self.cancelled_at
 263           self.cancelled_at = Time.now
 264         end
 265         self.running = false
 266         self.success = false
 267       when Failed
 268         if !self.finished_at
 269           self.finished_at = Time.now
 270         end
 271         self.running = false
 272         self.success = false
 273       when Complete
 274         if !self.finished_at
 275           self.finished_at = Time.now
 276         end
 277         self.running = false
 278         self.success = true
 279       end
 280     elsif 'running'.in? changed_attributes
 281       self.state = Running
 282     elsif 'success'.in? changed_attributes
 283       if success
 284         self.state = Complete
 285       else
 286         self.state = Failed
 287       end
 288     elsif 'cancelled_at'.in? changed_attributes
 289       self.state = Cancelled
 290     end
 291   end
 292
 293   def set_state_before_save
 294     if !self.state
 295       if self.cancelled_at
 296         self.state = Cancelled
 297       elsif self.success
 298         self.state = Complete
 299       elsif (!self.success.nil? && !self.success)
 300         self.state = Failed
 301       elsif (self.running && self.success.nil? && !self.cencelled_at)
 302         self.state = Running
 303       elsif !self.started_at && !self.cancelled_at && !self.is_locked_by_uuid && self.success.nil?
 304         self.state = Queued
 305       end
 306     end
 307
 308     if self.state.in?(States)
 309       true
 310     else
 311       errors.add :state, "'#{state.inspect} must be one of: [#{States.join ', '}]"
 312       false
 313     end
 314   end
 315
 316 end