1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: AGPL-3.0
5 require 'log_reuse_info'
8 class Job < ArvadosModel
11 include CommonApiTemplate
12 extend CurrentApiClient
14 serialize :components, Hash
15 attr_protected :arvados_sdk_version, :docker_image_locator
16 serialize :script_parameters, Hash
17 serialize :runtime_constraints, Hash
18 serialize :tasks_summary, Hash
19 before_create :ensure_unique_submit_id
20 after_commit :trigger_crunch_dispatch_if_cancelled, :on => :update
21 before_validation :set_priority
22 before_validation :update_state_from_old_state_attrs
23 before_validation :update_script_parameters_digest
24 validate :ensure_script_version_is_commit
25 validate :find_docker_image_locator
26 validate :find_arvados_sdk_version
27 validate :validate_status
28 validate :validate_state_change
29 validate :ensure_no_collection_uuids_in_script_params
30 before_save :tag_version_in_internal_repository
31 before_save :update_timestamps_when_state_changes
33 has_many :commit_ancestors, :foreign_key => :descendant, :primary_key => :script_version
34 has_many(:nodes, foreign_key: :job_uuid, primary_key: :uuid)
36 class SubmitIdReused < RequestError
39 api_accessible :user, extend: :common do |t|
43 t.add :script_parameters
46 t.add :cancelled_by_client_uuid
47 t.add :cancelled_by_user_uuid
54 t.add :is_locked_by_uuid
56 t.add :runtime_constraints
58 t.add :nondeterministic
60 t.add :supplied_script_version
61 t.add :arvados_sdk_version
62 t.add :docker_image_locator
69 # Supported states for a job
72 (Running = 'Running'),
73 (Cancelled = 'Cancelled'),
75 (Complete = 'Complete'),
79 @need_crunch_dispatch_trigger = false
82 def self.limit_index_columns_read
87 update_attributes(finished_at: finished_at || db_current_time,
88 success: success.nil? ? false : success,
97 self.where('state = ?', Queued).order('priority desc, created_at')
101 # We used to report this accurately, but the implementation made queue
102 # API requests O(n**2) for the size of the queue. See #8800.
103 # We've soft-disabled it because it's not clear we even want this
104 # functionality: now that we have Node Manager with support for multiple
105 # node sizes, "queue position" tells you very little about when a job will
107 state == Queued ? 0 : nil
111 self.where('running = ?', true).
112 order('priority desc, created_at')
115 def lock locked_by_uuid
117 unless self.state == Queued and self.is_locked_by_uuid.nil?
118 raise AlreadyLockedError
121 self.is_locked_by_uuid = locked_by_uuid
126 def update_script_parameters_digest
127 self.script_parameters_digest = self.class.sorted_hash_digest(script_parameters)
130 def self.searchable_columns operator
131 super - ["script_parameters_digest"]
134 def self.full_text_searchable_columns
135 super - ["script_parameters_digest"]
138 def self.load_job_specific_filters attrs, orig_filters, read_users
139 # Convert Job-specific @filters entries into general SQL filters.
140 script_info = {"repository" => nil, "script" => nil}
141 git_filters = Hash.new do |hash, key|
142 hash[key] = {"max_version" => "HEAD", "exclude_versions" => []}
145 orig_filters.each do |attr, operator, operand|
146 if (script_info.has_key? attr) and (operator == "=")
147 if script_info[attr].nil?
148 script_info[attr] = operand
149 elsif script_info[attr] != operand
150 raise ArgumentError.new("incompatible #{attr} filters")
155 git_filters[attr]["min_version"] = operand
157 git_filters[attr]["exclude_versions"] += Array.wrap(operand)
158 when "in docker", "not in docker"
159 image_hashes = Array.wrap(operand).flat_map do |search_term|
160 image_search, image_tag = search_term.split(':', 2)
162 find_all_for_docker_image(image_search, image_tag, read_users, filter_compatible_format: false).
163 map(&:portable_data_hash)
165 filters << [attr, operator.sub(/ docker$/, ""), image_hashes]
167 filters << [attr, operator, operand]
171 # Build a real script_version filter from any "not? in git" filters.
172 git_filters.each_pair do |attr, filter|
174 when "script_version"
175 script_info.each_pair do |key, value|
177 raise ArgumentError.new("script_version filter needs #{key} filter")
180 filter["repository"] = script_info["repository"]
181 if attrs[:script_version]
182 filter["max_version"] = attrs[:script_version]
184 # Using HEAD, set earlier by the hash default, is fine.
186 when "arvados_sdk_version"
187 filter["repository"] = "arvados"
189 raise ArgumentError.new("unknown attribute for git filter: #{attr}")
191 revisions = Commit.find_commit_range(filter["repository"],
192 filter["min_version"],
193 filter["max_version"],
194 filter["exclude_versions"])
197 new("error searching #{filter['repository']} from " +
198 "'#{filter['min_version']}' to '#{filter['max_version']}', " +
199 "excluding #{filter['exclude_versions']}")
201 filters.append([attr, "in", revisions])
207 def self.find_reusable attrs, params, filters, read_users
208 if filters.empty? # Translate older creation parameters into filters.
210 [["repository", "=", attrs[:repository]],
211 ["script", "=", attrs[:script]],
212 ["script_version", "not in git", params[:exclude_script_versions]],
213 ].reject { |filter| filter.last.nil? or filter.last.empty? }
214 if !params[:minimum_script_version].blank?
215 filters << ["script_version", "in git",
216 params[:minimum_script_version]]
218 filters += default_git_filters("script_version", attrs[:repository],
219 attrs[:script_version])
221 if image_search = attrs[:runtime_constraints].andand["docker_image"]
222 if image_tag = attrs[:runtime_constraints]["docker_image_tag"]
223 image_search += ":#{image_tag}"
225 image_locator = Collection.
226 for_latest_docker_image(image_search).andand.portable_data_hash
230 filters << ["docker_image_locator", "=", image_locator]
231 if sdk_version = attrs[:runtime_constraints].andand["arvados_sdk_version"]
232 filters += default_git_filters("arvados_sdk_version", "arvados", sdk_version)
234 filters = load_job_specific_filters(attrs, filters, read_users)
237 # Check specified filters for some reasonableness.
238 filter_names = filters.map { |f| f.first }.uniq
239 ["repository", "script"].each do |req_filter|
240 if not filter_names.include?(req_filter)
241 return send_error("#{req_filter} filter required")
245 # Search for a reusable Job, and return it if found.
246 candidates = Job.readable_by(current_user)
247 log_reuse_info { "starting with #{candidates.count} jobs readable by current user #{current_user.uuid}" }
249 candidates = candidates.where(
250 'state = ? or (owner_uuid = ? and state in (?))',
251 Job::Complete, current_user.uuid, [Job::Queued, Job::Running])
252 log_reuse_info(candidates) { "after filtering on job state ((state=Complete) or (state=Queued/Running and (submitted by current user)))" }
254 digest = Job.sorted_hash_digest(attrs[:script_parameters])
255 candidates = candidates.where('script_parameters_digest = ?', digest)
256 log_reuse_info(candidates) { "after filtering on script_parameters_digest #{digest}" }
258 candidates = candidates.where('nondeterministic is distinct from ?', true)
259 log_reuse_info(candidates) { "after filtering on !nondeterministic" }
261 # prefer Running jobs over Queued
262 candidates = candidates.order('state desc, created_at')
264 candidates = apply_filters candidates, filters
265 log_reuse_info(candidates) { "after filtering on repo, script, and custom filters #{filters.inspect}" }
270 candidates.each do |j|
271 if j.state != Job::Complete
273 # We'll use this if we don't find a job that has completed
274 log_reuse_info { "job #{j.uuid} is reusable, but unfinished; continuing search for completed jobs" }
277 log_reuse_info { "job #{j.uuid} is unfinished and we already have #{incomplete_job.uuid}; ignoring" }
279 elsif chosen == false
280 # Ignore: we have already decided not to reuse any completed
282 log_reuse_info { "job #{j.uuid} with output #{j.output} ignored, see above" }
284 log_reuse_info { "job #{j.uuid} has nil output" }
286 log_reuse_info { "job #{j.uuid} has nil log" }
287 elsif Rails.configuration.reuse_job_if_outputs_differ
288 if !Collection.readable_by(current_user).find_by_portable_data_hash(j.output)
289 # Ignore: keep looking for an incomplete job or one whose
290 # output is readable.
291 log_reuse_info { "job #{j.uuid} output #{j.output} unavailable to user; continuing search" }
292 elsif !Collection.readable_by(current_user).find_by_portable_data_hash(j.log)
293 # Ignore: keep looking for an incomplete job or one whose
295 log_reuse_info { "job #{j.uuid} log #{j.log} unavailable to user; continuing search" }
297 log_reuse_info { "job #{j.uuid} with output #{j.output} is reusable; decision is final." }
301 if chosen_output != j.output
302 # If two matching jobs produced different outputs, run a new
303 # job (or use one that's already running/queued) instead of
304 # choosing one arbitrarily.
305 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)" }
308 log_reuse_info { "job #{j.uuid} output #{j.output} agrees with chosen #{chosen.uuid}; continuing search in case other candidates have different outputs" }
310 # ...and that's the only thing we need to do once we've chosen
312 elsif !Collection.readable_by(current_user).find_by_portable_data_hash(j.output)
313 # This user cannot read the output of this job. Any other
314 # completed job will have either the same output (making it
315 # unusable) or a different output (making it unusable because
316 # reuse_job_if_outputs_different is turned off). Therefore,
317 # any further investigation of reusable jobs is futile.
318 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)" }
320 elsif !Collection.readable_by(current_user).find_by_portable_data_hash(j.log)
321 # This user cannot read the log of this job, don't try to reuse the
322 # job but consider if the output is consistent.
323 log_reuse_info { "job #{j.uuid} log #{j.log} is unavailable to user; continuing search" }
324 chosen_output = j.output
326 log_reuse_info { "job #{j.uuid} with output #{j.output} can be reused; continuing search in case other candidates have different outputs" }
328 chosen_output = j.output
331 j = chosen || incomplete_job
333 log_reuse_info { "done, #{j.uuid} was selected" }
335 log_reuse_info { "done, nothing suitable" }
340 def self.default_git_filters(attr_name, repo_name, refspec)
341 # Add a filter to @filters for `attr_name` = the latest commit available
342 # in `repo_name` at `refspec`. No filter is added if refspec can't be
344 commits = Commit.find_commit_range(repo_name, nil, refspec, nil)
345 if commit_hash = commits.first
346 [[attr_name, "=", commit_hash]]
352 def cancel(cascade: false, need_transaction: true)
354 ActiveRecord::Base.transaction do
355 cancel(cascade: cascade, need_transaction: false)
360 if self.state.in?([Queued, Running])
361 self.state = Cancelled
363 elsif self.state != Cancelled
364 raise InvalidStateTransitionError
369 # cancel all children; they could be jobs or pipeline instances
370 children = self.components.andand.collect{|_, u| u}.compact
372 return if children.empty?
374 # cancel any child jobs
375 Job.where(uuid: children, state: [Queued, Running]).each do |job|
376 job.cancel(cascade: cascade, need_transaction: false)
379 # cancel any child pipelines
380 PipelineInstance.where(uuid: children, state: [PipelineInstance::RunningOnServer, PipelineInstance::RunningOnClient]).each do |pi|
381 pi.cancel(cascade: cascade, need_transaction: false)
387 def self.sorted_hash_digest h
388 Digest::MD5.hexdigest(Oj.dump(deep_sort_hash(h)))
391 def foreign_key_attributes
392 super + %w(output log)
395 def skip_uuid_read_permission_check
396 super + %w(cancelled_by_client_uuid)
399 def skip_uuid_existence_check
400 super + %w(output log)
404 if self.priority.nil?
410 def ensure_script_version_is_commit
412 # Apparently client has already decided to go for it. This is
413 # needed to run a local job using a local working directory
414 # instead of a commit-ish.
417 if new_record? or repository_changed? or script_version_changed?
418 sha1 = Commit.find_commit_range(repository,
419 nil, script_version, nil).first
421 errors.add :script_version, "#{script_version} does not resolve to a commit"
424 if supplied_script_version.nil? or supplied_script_version.empty?
425 self.supplied_script_version = script_version
427 self.script_version = sha1
432 def tag_version_in_internal_repository
434 # No point now. See ensure_script_version_is_commit.
437 # Won't be saved, and script_version might not even be valid.
439 elsif new_record? or repository_changed? or script_version_changed?
443 Commit.tag_in_internal_repository repository, script_version, uuid
451 def ensure_unique_submit_id
453 if Job.where('submit_id=?',self.submit_id).first
454 raise SubmitIdReused.new
460 def resolve_runtime_constraint(key, attr_sym)
461 if ((runtime_constraints.is_a? Hash) and
462 (search = runtime_constraints[key]))
463 ok, result = yield search
465 ok, result = true, nil
468 send("#{attr_sym}=".to_sym, result)
470 errors.add(attr_sym, result)
475 def find_arvados_sdk_version
476 resolve_runtime_constraint("arvados_sdk_version",
477 :arvados_sdk_version) do |git_search|
478 commits = Commit.find_commit_range("arvados",
479 nil, git_search, nil)
481 [false, "#{git_search} does not resolve to a commit"]
482 elsif not runtime_constraints["docker_image"]
483 [false, "cannot be specified without a Docker image constraint"]
485 [true, commits.first]
490 def find_docker_image_locator
491 if runtime_constraints.is_a? Hash
492 runtime_constraints['docker_image'] ||=
493 Rails.configuration.default_docker_image_for_jobs
496 resolve_runtime_constraint("docker_image",
497 :docker_image_locator) do |image_search|
498 image_tag = runtime_constraints['docker_image_tag']
499 if coll = Collection.for_latest_docker_image(image_search, image_tag)
500 [true, coll.portable_data_hash]
502 [false, "not found for #{image_search}"]
507 def permission_to_update
508 if is_locked_by_uuid_was and !(current_user and
509 (current_user.uuid == is_locked_by_uuid_was or
510 current_user.uuid == system_user.uuid))
511 if script_changed? or
512 script_parameters_changed? or
513 script_version_changed? or
514 (!cancelled_at_was.nil? and
515 (cancelled_by_client_uuid_changed? or
516 cancelled_by_user_uuid_changed? or
517 cancelled_at_changed?)) or
518 started_at_changed? or
519 finished_at_changed? or
524 tasks_summary_changed? or
525 (state_changed? && state != Cancelled) or
527 logger.warn "User #{current_user.uuid if current_user} tried to change protected job attributes on locked #{self.class.to_s} #{uuid_was}"
531 if !is_locked_by_uuid_changed?
535 logger.warn "Anonymous user tried to change lock on #{self.class.to_s} #{uuid_was}"
537 elsif is_locked_by_uuid_was and is_locked_by_uuid_was != current_user.uuid
538 logger.warn "User #{current_user.uuid} tried to steal lock on #{self.class.to_s} #{uuid_was} from #{is_locked_by_uuid_was}"
540 elsif !is_locked_by_uuid.nil? and is_locked_by_uuid != current_user.uuid
541 logger.warn "User #{current_user.uuid} tried to lock #{self.class.to_s} #{uuid_was} with uuid #{is_locked_by_uuid}"
549 def update_modified_by_fields
550 if self.cancelled_at_changed?
551 # Ensure cancelled_at cannot be set to arbitrary non-now times,
552 # or changed once it is set.
553 if self.cancelled_at and not self.cancelled_at_was
554 self.cancelled_at = db_current_time
555 self.cancelled_by_user_uuid = current_user.uuid
556 self.cancelled_by_client_uuid = current_api_client.andand.uuid
557 @need_crunch_dispatch_trigger = true
559 self.cancelled_at = self.cancelled_at_was
560 self.cancelled_by_user_uuid = self.cancelled_by_user_uuid_was
561 self.cancelled_by_client_uuid = self.cancelled_by_client_uuid_was
567 def trigger_crunch_dispatch_if_cancelled
568 if @need_crunch_dispatch_trigger
569 File.open(Rails.configuration.crunch_refresh_trigger, 'wb') do
570 # That's all, just create/touch a file for crunch-job to see.
575 def update_timestamps_when_state_changes
576 return if not (state_changed? or new_record?)
580 self.started_at ||= db_current_time
581 when Failed, Complete
582 self.finished_at ||= db_current_time
584 self.cancelled_at ||= db_current_time
587 # TODO: Remove the following case block when old "success" and
588 # "running" attrs go away. Until then, this ensures we still
589 # expose correct success/running flags to older clients, even if
590 # some new clients are writing only the new state attribute.
598 when Cancelled, Failed
605 self.running ||= false # Default to false instead of nil.
607 @need_crunch_dispatch_trigger = true
612 def update_state_from_old_state_attrs
613 # If a client has touched the legacy state attrs, update the
614 # "state" attr to agree with the updated values of the legacy
617 # TODO: Remove this method when old "success" and "running" attrs
619 if cancelled_at_changed? or
624 self.state = Cancelled
625 elsif success == false
627 elsif success == true
628 self.state = Complete
629 elsif running == true
639 if self.state.in?(States)
642 errors.add :state, "#{state.inspect} must be one of: #{States.inspect}"
647 def validate_state_change
649 if self.state_changed?
650 ok = case self.state_was
652 # state isn't set yet
655 # Permit going from queued to any state
658 # From running, may only transition to a finished state
659 [Complete, Failed, Cancelled].include? self.state
660 when Complete, Failed, Cancelled
661 # Once in a finished state, don't permit any more state changes
664 # Any other state transition is also invalid
668 errors.add :state, "invalid change from #{self.state_was} to #{self.state}"
674 def ensure_no_collection_uuids_in_script_params
675 # Fail validation if any script_parameters field includes a string containing a
676 # collection uuid pattern.
677 if self.script_parameters_changed?
678 if recursive_hash_search(self.script_parameters, Collection.uuid_regex)
679 self.errors.add :script_parameters, "must use portable_data_hash instead of collection uuid"
686 # recursive_hash_search searches recursively through hashes and
687 # arrays in 'thing' for string fields matching regular expression
688 # 'pattern'. Returns true if pattern is found, false otherwise.
689 def recursive_hash_search thing, pattern
692 return true if recursive_hash_search v, pattern
694 elsif thing.is_a? Array
696 return true if recursive_hash_search k, pattern
698 elsif thing.is_a? String
699 return true if thing.match pattern