services/api/app/models/job.rb

   1 require 'log_reuse_info'
   2 require 'safe_json'
   3
   4 class Job < ArvadosModel
   5   include HasUuid
   6   include KindAndEtag
   7   include CommonApiTemplate
   8   extend CurrentApiClient
   9   extend LogReuseInfo
  10   serialize :components, Hash
  11   attr_protected :arvados_sdk_version, :docker_image_locator
  12   serialize :script_parameters, Hash
  13   serialize :runtime_constraints, Hash
  14   serialize :tasks_summary, Hash
  15   before_create :ensure_unique_submit_id
  16   after_commit :trigger_crunch_dispatch_if_cancelled, :on => :update
  17   before_validation :set_priority
  18   before_validation :update_state_from_old_state_attrs
  19   before_validation :update_script_parameters_digest
  20   validate :ensure_script_version_is_commit
  21   validate :find_docker_image_locator
  22   validate :find_arvados_sdk_version
  23   validate :validate_status
  24   validate :validate_state_change
  25   validate :ensure_no_collection_uuids_in_script_params
  26   before_save :tag_version_in_internal_repository
  27   before_save :update_timestamps_when_state_changes
  28
  29   has_many :commit_ancestors, :foreign_key => :descendant, :primary_key => :script_version
  30   has_many(:nodes, foreign_key: :job_uuid, primary_key: :uuid)
  31
  32   class SubmitIdReused < StandardError
  33   end
  34
  35   api_accessible :user, extend: :common do |t|
  36     t.add :submit_id
  37     t.add :priority
  38     t.add :script
  39     t.add :script_parameters
  40     t.add :script_version
  41     t.add :cancelled_at
  42     t.add :cancelled_by_client_uuid
  43     t.add :cancelled_by_user_uuid
  44     t.add :started_at
  45     t.add :finished_at
  46     t.add :output
  47     t.add :success
  48     t.add :running
  49     t.add :state
  50     t.add :is_locked_by_uuid
  51     t.add :log
  52     t.add :runtime_constraints
  53     t.add :tasks_summary
  54     t.add :nondeterministic
  55     t.add :repository
  56     t.add :supplied_script_version
  57     t.add :arvados_sdk_version
  58     t.add :docker_image_locator
  59     t.add :queue_position
  60     t.add :node_uuids
  61     t.add :description
  62     t.add :components
  63   end
  64
  65   # Supported states for a job
  66   States = [
  67             (Queued = 'Queued'),
  68             (Running = 'Running'),
  69             (Cancelled = 'Cancelled'),
  70             (Failed = 'Failed'),
  71             (Complete = 'Complete'),
  72            ]
  73
  74   after_initialize do
  75     @need_crunch_dispatch_trigger = false
  76   end
  77
  78   def self.limit_index_columns_read
  79     ["components"]
  80   end
  81
  82   def assert_finished
  83     update_attributes(finished_at: finished_at || db_current_time,
  84                       success: success.nil? ? false : success,
  85                       running: false)
  86   end
  87
  88   def node_uuids
  89     nodes.map(&:uuid)
  90   end
  91
  92   def self.queue
  93     self.where('state = ?', Queued).order('priority desc, created_at')
  94   end
  95
  96   def queue_position
  97     # We used to report this accurately, but the implementation made queue
  98     # API requests O(n**2) for the size of the queue.  See #8800.
  99     # We've soft-disabled it because it's not clear we even want this
 100     # functionality: now that we have Node Manager with support for multiple
 101     # node sizes, "queue position" tells you very little about when a job will
 102     # run.
 103     state == Queued ? 0 : nil
 104   end
 105
 106   def self.running
 107     self.where('running = ?', true).
 108       order('priority desc, created_at')
 109   end
 110
 111   def lock locked_by_uuid
 112     with_lock do
 113       unless self.state == Queued and self.is_locked_by_uuid.nil?
 114         raise AlreadyLockedError
 115       end
 116       self.state = Running
 117       self.is_locked_by_uuid = locked_by_uuid
 118       self.save!
 119     end
 120   end
 121
 122   def update_script_parameters_digest
 123     self.script_parameters_digest = self.class.sorted_hash_digest(script_parameters)
 124   end
 125
 126   def self.searchable_columns operator
 127     super - ["script_parameters_digest"]
 128   end
 129
 130   def self.full_text_searchable_columns
 131     super - ["script_parameters_digest"]
 132   end
 133
 134   def self.load_job_specific_filters attrs, orig_filters, read_users
 135     # Convert Job-specific @filters entries into general SQL filters.
 136     script_info = {"repository" => nil, "script" => nil}
 137     git_filters = Hash.new do |hash, key|
 138       hash[key] = {"max_version" => "HEAD", "exclude_versions" => []}
 139     end
 140     filters = []
 141     orig_filters.each do |attr, operator, operand|
 142       if (script_info.has_key? attr) and (operator == "=")
 143         if script_info[attr].nil?
 144           script_info[attr] = operand
 145         elsif script_info[attr] != operand
 146           raise ArgumentError.new("incompatible #{attr} filters")
 147         end
 148       end
 149       case operator
 150       when "in git"
 151         git_filters[attr]["min_version"] = operand
 152       when "not in git"
 153         git_filters[attr]["exclude_versions"] += Array.wrap(operand)
 154       when "in docker", "not in docker"
 155         image_hashes = Array.wrap(operand).flat_map do |search_term|
 156           image_search, image_tag = search_term.split(':', 2)
 157           Collection.
 158             find_all_for_docker_image(image_search, image_tag, read_users, filter_compatible_format: false).
 159             map(&:portable_data_hash)
 160         end
 161         filters << [attr, operator.sub(/ docker$/, ""), image_hashes]
 162       else
 163         filters << [attr, operator, operand]
 164       end
 165     end
 166
 167     # Build a real script_version filter from any "not? in git" filters.
 168     git_filters.each_pair do |attr, filter|
 169       case attr
 170       when "script_version"
 171         script_info.each_pair do |key, value|
 172           if value.nil?
 173             raise ArgumentError.new("script_version filter needs #{key} filter")
 174           end
 175         end
 176         filter["repository"] = script_info["repository"]
 177         if attrs[:script_version]
 178           filter["max_version"] = attrs[:script_version]
 179         else
 180           # Using HEAD, set earlier by the hash default, is fine.
 181         end
 182       when "arvados_sdk_version"
 183         filter["repository"] = "arvados"
 184       else
 185         raise ArgumentError.new("unknown attribute for git filter: #{attr}")
 186       end
 187       revisions = Commit.find_commit_range(filter["repository"],
 188                                            filter["min_version"],
 189                                            filter["max_version"],
 190                                            filter["exclude_versions"])
 191       if revisions.empty?
 192         raise ArgumentError.
 193           new("error searching #{filter['repository']} from " +
 194               "'#{filter['min_version']}' to '#{filter['max_version']}', " +
 195               "excluding #{filter['exclude_versions']}")
 196       end
 197       filters.append([attr, "in", revisions])
 198     end
 199
 200     filters
 201   end
 202
 203   def self.find_reusable attrs, params, filters, read_users
 204     if filters.empty?  # Translate older creation parameters into filters.
 205       filters =
 206         [["repository", "=", attrs[:repository]],
 207          ["script", "=", attrs[:script]],
 208          ["script_version", "not in git", params[:exclude_script_versions]],
 209         ].reject { |filter| filter.last.nil? or filter.last.empty? }
 210       if !params[:minimum_script_version].blank?
 211         filters << ["script_version", "in git",
 212                      params[:minimum_script_version]]
 213       else
 214         filters += default_git_filters("script_version", attrs[:repository],
 215                                        attrs[:script_version])
 216       end
 217       if image_search = attrs[:runtime_constraints].andand["docker_image"]
 218         if image_tag = attrs[:runtime_constraints]["docker_image_tag"]
 219           image_search += ":#{image_tag}"
 220         end
 221         image_locator = Collection.
 222           for_latest_docker_image(image_search).andand.portable_data_hash
 223       else
 224         image_locator = nil
 225       end
 226       filters << ["docker_image_locator", "=", image_locator]
 227       if sdk_version = attrs[:runtime_constraints].andand["arvados_sdk_version"]
 228         filters += default_git_filters("arvados_sdk_version", "arvados", sdk_version)
 229       end
 230       filters = load_job_specific_filters(attrs, filters, read_users)
 231     end
 232
 233     # Check specified filters for some reasonableness.
 234     filter_names = filters.map { |f| f.first }.uniq
 235     ["repository", "script"].each do |req_filter|
 236       if not filter_names.include?(req_filter)
 237         return send_error("#{req_filter} filter required")
 238       end
 239     end
 240
 241     # Search for a reusable Job, and return it if found.
 242     candidates = Job.readable_by(current_user)
 243     log_reuse_info { "starting with #{candidates.count} jobs readable by current user #{current_user.uuid}" }
 244
 245     candidates = candidates.where(
 246       'state = ? or (owner_uuid = ? and state in (?))',
 247       Job::Complete, current_user.uuid, [Job::Queued, Job::Running])
 248     log_reuse_info(candidates) { "after filtering on job state ((state=Complete) or (state=Queued/Running and (submitted by current user)))" }
 249
 250     digest = Job.sorted_hash_digest(attrs[:script_parameters])
 251     candidates = candidates.where('script_parameters_digest = ?', digest)
 252     log_reuse_info(candidates) { "after filtering on script_parameters_digest #{digest}" }
 253
 254     candidates = candidates.where('nondeterministic is distinct from ?', true)
 255     log_reuse_info(candidates) { "after filtering on !nondeterministic" }
 256
 257     # prefer Running jobs over Queued
 258     candidates = candidates.order('state desc, created_at')
 259
 260     candidates = apply_filters candidates, filters
 261     log_reuse_info(candidates) { "after filtering on repo, script, and custom filters #{filters.inspect}" }
 262
 263     chosen = nil
 264     incomplete_job = nil
 265     candidates.each do |j|
 266       if j.state != Job::Complete
 267         if !incomplete_job
 268           # We'll use this if we don't find a job that has completed
 269           log_reuse_info { "job #{j.uuid} is reusable, but unfinished; continuing search for completed jobs" }
 270           incomplete_job = j
 271         else
 272           log_reuse_info { "job #{j.uuid} is unfinished and we already have #{incomplete_job.uuid}; ignoring" }
 273         end
 274       elsif chosen == false
 275         # Ignore: we have already decided not to reuse any completed
 276         # job.
 277         log_reuse_info { "job #{j.uuid} with output #{j.output} ignored, see above" }
 278       elsif Rails.configuration.reuse_job_if_outputs_differ
 279         if Collection.readable_by(current_user).find_by_portable_data_hash(j.output)
 280           log_reuse_info { "job #{j.uuid} with output #{j.output} is reusable; decision is final." }
 281           return j
 282         else
 283           # Ignore: keep locking for an incomplete job or one whose
 284           # output is readable.
 285           log_reuse_info { "job #{j.uuid} output #{j.output} unavailable to user; continuing search" }
 286         end
 287       elsif chosen
 288         if chosen.output != j.output
 289           # If two matching jobs produced different outputs, run a new
 290           # job (or use one that's already running/queued) instead of
 291           # choosing one arbitrarily.
 292           log_reuse_info { "job #{j.uuid} output #{j.output} disagrees; forgetting about #{chosen.uuid} and ignoring any other finished jobs (see reuse_job_if_outputs_differ in application.default.yml)" }
 293           chosen = false
 294         else
 295           log_reuse_info { "job #{j.uuid} output #{j.output} agrees with chosen #{chosen.uuid}; continuing search in case other candidates have different outputs" }
 296         end
 297         # ...and that's the only thing we need to do once we've chosen
 298         # a job to reuse.
 299       elsif !Collection.readable_by(current_user).find_by_portable_data_hash(j.output)
 300         # This user cannot read the output of this job. Any other
 301         # completed job will have either the same output (making it
 302         # unusable) or a different output (making it unusable because
 303         # reuse_job_if_outputs_different is turned off). Therefore,
 304         # any further investigation of reusable jobs is futile.
 305         log_reuse_info { "job #{j.uuid} output #{j.output} is unavailable to user; this means no finished job can be reused (see reuse_job_if_outputs_differ in application.default.yml)" }
 306         chosen = false
 307       else
 308         log_reuse_info { "job #{j.uuid} with output #{j.output} can be reused; continuing search in case other candidates have different outputs" }
 309         chosen = j
 310       end
 311     end
 312     j = chosen || incomplete_job
 313     if j
 314       log_reuse_info { "done, #{j.uuid} was selected" }
 315     else
 316       log_reuse_info { "done, nothing suitable" }
 317     end
 318     return j
 319   end
 320
 321   def self.default_git_filters(attr_name, repo_name, refspec)
 322     # Add a filter to @filters for `attr_name` = the latest commit available
 323     # in `repo_name` at `refspec`.  No filter is added if refspec can't be
 324     # resolved.
 325     commits = Commit.find_commit_range(repo_name, nil, refspec, nil)
 326     if commit_hash = commits.first
 327       [[attr_name, "=", commit_hash]]
 328     else
 329       []
 330     end
 331   end
 332
 333   def cancel(cascade: false, need_transaction: true)
 334     if need_transaction
 335       ActiveRecord::Base.transaction do
 336         cancel(cascade: cascade, need_transaction: false)
 337       end
 338       return
 339     end
 340
 341     if self.state.in?([Queued, Running])
 342       self.state = Cancelled
 343       self.save!
 344     elsif self.state != Cancelled
 345       raise InvalidStateTransitionError
 346     end
 347
 348     return if !cascade
 349
 350     # cancel all children; they could be jobs or pipeline instances
 351     children = self.components.andand.collect{|_, u| u}.compact
 352
 353     return if children.empty?
 354
 355     # cancel any child jobs
 356     Job.where(uuid: children, state: [Queued, Running]).each do |job|
 357       job.cancel(cascade: cascade, need_transaction: false)
 358     end
 359
 360     # cancel any child pipelines
 361     PipelineInstance.where(uuid: children, state: [PipelineInstance::RunningOnServer, PipelineInstance::RunningOnClient]).each do |pi|
 362       pi.cancel(cascade: cascade, need_transaction: false)
 363     end
 364   end
 365
 366   protected
 367
 368   def self.sorted_hash_digest h
 369     Digest::MD5.hexdigest(Oj.dump(deep_sort_hash(h)))
 370   end
 371
 372   def foreign_key_attributes
 373     super + %w(output log)
 374   end
 375
 376   def skip_uuid_read_permission_check
 377     super + %w(cancelled_by_client_uuid)
 378   end
 379
 380   def skip_uuid_existence_check
 381     super + %w(output log)
 382   end
 383
 384   def set_priority
 385     if self.priority.nil?
 386       self.priority = 0
 387     end
 388     true
 389   end
 390
 391   def ensure_script_version_is_commit
 392     if state == Running
 393       # Apparently client has already decided to go for it. This is
 394       # needed to run a local job using a local working directory
 395       # instead of a commit-ish.
 396       return true
 397     end
 398     if new_record? or repository_changed? or script_version_changed?
 399       sha1 = Commit.find_commit_range(repository,
 400                                       nil, script_version, nil).first
 401       if not sha1
 402         errors.add :script_version, "#{script_version} does not resolve to a commit"
 403         return false
 404       end
 405       if supplied_script_version.nil? or supplied_script_version.empty?
 406         self.supplied_script_version = script_version
 407       end
 408       self.script_version = sha1
 409     end
 410     true
 411   end
 412
 413   def tag_version_in_internal_repository
 414     if state == Running
 415       # No point now. See ensure_script_version_is_commit.
 416       true
 417     elsif errors.any?
 418       # Won't be saved, and script_version might not even be valid.
 419       true
 420     elsif new_record? or repository_changed? or script_version_changed?
 421       uuid_was = uuid
 422       begin
 423         assign_uuid
 424         Commit.tag_in_internal_repository repository, script_version, uuid
 425       rescue
 426         self.uuid = uuid_was
 427         raise
 428       end
 429     end
 430   end
 431
 432   def ensure_unique_submit_id
 433     if !submit_id.nil?
 434       if Job.where('submit_id=?',self.submit_id).first
 435         raise SubmitIdReused.new
 436       end
 437     end
 438     true
 439   end
 440
 441   def resolve_runtime_constraint(key, attr_sym)
 442     if ((runtime_constraints.is_a? Hash) and
 443         (search = runtime_constraints[key]))
 444       ok, result = yield search
 445     else
 446       ok, result = true, nil
 447     end
 448     if ok
 449       send("#{attr_sym}=".to_sym, result)
 450     else
 451       errors.add(attr_sym, result)
 452     end
 453     ok
 454   end
 455
 456   def find_arvados_sdk_version
 457     resolve_runtime_constraint("arvados_sdk_version",
 458                                :arvados_sdk_version) do |git_search|
 459       commits = Commit.find_commit_range("arvados",
 460                                          nil, git_search, nil)
 461       if commits.empty?
 462         [false, "#{git_search} does not resolve to a commit"]
 463       elsif not runtime_constraints["docker_image"]
 464         [false, "cannot be specified without a Docker image constraint"]
 465       else
 466         [true, commits.first]
 467       end
 468     end
 469   end
 470
 471   def find_docker_image_locator
 472     if runtime_constraints.is_a? Hash
 473       runtime_constraints['docker_image'] ||=
 474         Rails.configuration.default_docker_image_for_jobs
 475     end
 476
 477     resolve_runtime_constraint("docker_image",
 478                                :docker_image_locator) do |image_search|
 479       image_tag = runtime_constraints['docker_image_tag']
 480       if coll = Collection.for_latest_docker_image(image_search, image_tag)
 481         [true, coll.portable_data_hash]
 482       else
 483         [false, "not found for #{image_search}"]
 484       end
 485     end
 486   end
 487
 488   def permission_to_update
 489     if is_locked_by_uuid_was and !(current_user and
 490                                    (current_user.uuid == is_locked_by_uuid_was or
 491                                     current_user.uuid == system_user.uuid))
 492       if script_changed? or
 493           script_parameters_changed? or
 494           script_version_changed? or
 495           (!cancelled_at_was.nil? and
 496            (cancelled_by_client_uuid_changed? or
 497             cancelled_by_user_uuid_changed? or
 498             cancelled_at_changed?)) or
 499           started_at_changed? or
 500           finished_at_changed? or
 501           running_changed? or
 502           success_changed? or
 503           output_changed? or
 504           log_changed? or
 505           tasks_summary_changed? or
 506           (state_changed? && state != Cancelled) or
 507           components_changed?
 508         logger.warn "User #{current_user.uuid if current_user} tried to change protected job attributes on locked #{self.class.to_s} #{uuid_was}"
 509         return false
 510       end
 511     end
 512     if !is_locked_by_uuid_changed?
 513       super
 514     else
 515       if !current_user
 516         logger.warn "Anonymous user tried to change lock on #{self.class.to_s} #{uuid_was}"
 517         false
 518       elsif is_locked_by_uuid_was and is_locked_by_uuid_was != current_user.uuid
 519         logger.warn "User #{current_user.uuid} tried to steal lock on #{self.class.to_s} #{uuid_was} from #{is_locked_by_uuid_was}"
 520         false
 521       elsif !is_locked_by_uuid.nil? and is_locked_by_uuid != current_user.uuid
 522         logger.warn "User #{current_user.uuid} tried to lock #{self.class.to_s} #{uuid_was} with uuid #{is_locked_by_uuid}"
 523         false
 524       else
 525         super
 526       end
 527     end
 528   end
 529
 530   def update_modified_by_fields
 531     if self.cancelled_at_changed?
 532       # Ensure cancelled_at cannot be set to arbitrary non-now times,
 533       # or changed once it is set.
 534       if self.cancelled_at and not self.cancelled_at_was
 535         self.cancelled_at = db_current_time
 536         self.cancelled_by_user_uuid = current_user.uuid
 537         self.cancelled_by_client_uuid = current_api_client.andand.uuid
 538         @need_crunch_dispatch_trigger = true
 539       else
 540         self.cancelled_at = self.cancelled_at_was
 541         self.cancelled_by_user_uuid = self.cancelled_by_user_uuid_was
 542         self.cancelled_by_client_uuid = self.cancelled_by_client_uuid_was
 543       end
 544     end
 545     super
 546   end
 547
 548   def trigger_crunch_dispatch_if_cancelled
 549     if @need_crunch_dispatch_trigger
 550       File.open(Rails.configuration.crunch_refresh_trigger, 'wb') do
 551         # That's all, just create/touch a file for crunch-job to see.
 552       end
 553     end
 554   end
 555
 556   def update_timestamps_when_state_changes
 557     return if not (state_changed? or new_record?)
 558
 559     case state
 560     when Running
 561       self.started_at ||= db_current_time
 562     when Failed, Complete
 563       self.finished_at ||= db_current_time
 564     when Cancelled
 565       self.cancelled_at ||= db_current_time
 566     end
 567
 568     # TODO: Remove the following case block when old "success" and
 569     # "running" attrs go away. Until then, this ensures we still
 570     # expose correct success/running flags to older clients, even if
 571     # some new clients are writing only the new state attribute.
 572     case state
 573     when Queued
 574       self.running = false
 575       self.success = nil
 576     when Running
 577       self.running = true
 578       self.success = nil
 579     when Cancelled, Failed
 580       self.running = false
 581       self.success = false
 582     when Complete
 583       self.running = false
 584       self.success = true
 585     end
 586     self.running ||= false # Default to false instead of nil.
 587
 588     @need_crunch_dispatch_trigger = true
 589
 590     true
 591   end
 592
 593   def update_state_from_old_state_attrs
 594     # If a client has touched the legacy state attrs, update the
 595     # "state" attr to agree with the updated values of the legacy
 596     # attrs.
 597     #
 598     # TODO: Remove this method when old "success" and "running" attrs
 599     # go away.
 600     if cancelled_at_changed? or
 601         success_changed? or
 602         running_changed? or
 603         state.nil?
 604       if cancelled_at
 605         self.state = Cancelled
 606       elsif success == false
 607         self.state = Failed
 608       elsif success == true
 609         self.state = Complete
 610       elsif running == true
 611         self.state = Running
 612       else
 613         self.state = Queued
 614       end
 615     end
 616     true
 617   end
 618
 619   def validate_status
 620     if self.state.in?(States)
 621       true
 622     else
 623       errors.add :state, "#{state.inspect} must be one of: #{States.inspect}"
 624       false
 625     end
 626   end
 627
 628   def validate_state_change
 629     ok = true
 630     if self.state_changed?
 631       ok = case self.state_was
 632            when nil
 633              # state isn't set yet
 634              true
 635            when Queued
 636              # Permit going from queued to any state
 637              true
 638            when Running
 639              # From running, may only transition to a finished state
 640              [Complete, Failed, Cancelled].include? self.state
 641            when Complete, Failed, Cancelled
 642              # Once in a finished state, don't permit any more state changes
 643              false
 644            else
 645              # Any other state transition is also invalid
 646              false
 647            end
 648       if not ok
 649         errors.add :state, "invalid change from #{self.state_was} to #{self.state}"
 650       end
 651     end
 652     ok
 653   end
 654
 655   def ensure_no_collection_uuids_in_script_params
 656     # Fail validation if any script_parameters field includes a string containing a
 657     # collection uuid pattern.
 658     if self.script_parameters_changed?
 659       if recursive_hash_search(self.script_parameters, Collection.uuid_regex)
 660         self.errors.add :script_parameters, "must use portable_data_hash instead of collection uuid"
 661         return false
 662       end
 663     end
 664     true
 665   end
 666
 667   # recursive_hash_search searches recursively through hashes and
 668   # arrays in 'thing' for string fields matching regular expression
 669   # 'pattern'.  Returns true if pattern is found, false otherwise.
 670   def recursive_hash_search thing, pattern
 671     if thing.is_a? Hash
 672       thing.each do |k, v|
 673         return true if recursive_hash_search v, pattern
 674       end
 675     elsif thing.is_a? Array
 676       thing.each do |k|
 677         return true if recursive_hash_search k, pattern
 678       end
 679     elsif thing.is_a? String
 680       return true if thing.match pattern
 681     end
 682     false
 683   end
 684 end