1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: AGPL-3.0
5 require 'whitelist_update'
6 require 'arvados/collection'
8 class ContainerRequest < ArvadosModel
9 include ArvadosModelUpdates
12 include CommonApiTemplate
13 include WhitelistUpdate
15 belongs_to :container, foreign_key: :container_uuid, primary_key: :uuid
16 belongs_to :requesting_container, {
17 class_name: 'Container',
18 foreign_key: :requesting_container_uuid,
22 # Posgresql JSONB columns should NOT be declared as serialized, Rails 5
23 # already know how to properly treat them.
24 attribute :properties, :jsonbHash, default: {}
25 attribute :secret_mounts, :jsonbHash, default: {}
26 attribute :output_storage_classes, :jsonbArray, default: lambda { Rails.configuration.DefaultStorageClasses }
27 attribute :output_properties, :jsonbHash, default: {}
29 serialize :environment, Hash
30 serialize :mounts, Hash
31 serialize :runtime_constraints, Hash
32 serialize :command, Array
33 serialize :scheduling_parameters, Hash
35 after_find :fill_container_defaults_after_find
36 after_initialize { @state_was_when_initialized = self.state_was } # see finalize_if_needed
37 before_validation :fill_field_defaults, :if => :new_record?
38 before_validation :fill_container_defaults
39 validates :command, :container_image, :output_path, :cwd, :presence => true
40 validates :output_ttl, numericality: { only_integer: true, greater_than_or_equal_to: 0 }
41 validates :priority, numericality: { only_integer: true, greater_than_or_equal_to: 0, less_than_or_equal_to: 1000 }
42 validate :validate_datatypes
43 validate :validate_runtime_constraints
44 validate :validate_scheduling_parameters
45 validate :validate_state_change
46 validate :check_update_whitelist
47 validate :secret_mounts_key_conflict
48 validate :validate_runtime_token
49 after_validation :scrub_secrets
50 after_validation :set_preemptible
51 after_validation :set_container
52 before_create :set_requesting_container_uuid
53 before_destroy :set_priority_zero
54 after_save :update_priority
55 after_save :finalize_if_needed
57 api_accessible :user, extend: :common do |t|
59 t.add :container_count
60 t.add :container_count_max
61 t.add :container_image
77 t.add :requesting_container_uuid
78 t.add :runtime_constraints
79 t.add :scheduling_parameters
82 t.add :output_storage_classes
83 t.add :output_properties
84 t.add :cumulative_cost
87 # Supported states for a container request
90 (Uncommitted = 'Uncommitted'),
91 (Committed = 'Committed'),
96 nil => [Uncommitted, Committed],
97 Uncommitted => [Committed],
101 AttrsPermittedAlways = [:owner_uuid, :state, :name, :description, :properties]
102 AttrsPermittedBeforeCommit = [:command, :container_count_max,
103 :container_image, :cwd, :environment, :filters, :mounts,
104 :output_path, :priority, :runtime_token,
105 :runtime_constraints, :state, :container_uuid, :use_existing,
106 :scheduling_parameters, :secret_mounts, :output_name, :output_ttl,
107 :output_storage_classes, :output_properties]
109 def self.any_preemptible_instances?
110 Rails.configuration.InstanceTypes.any? do |k, v|
115 def self.limit_index_columns_read
119 def logged_attributes
120 super.except('secret_mounts', 'runtime_token')
123 def state_transitions
127 def skip_uuid_read_permission_check
128 # The uuid_read_permission_check prevents users from making
129 # references to objects they can't view. However, in this case we
130 # don't want to do that check since there's a circular dependency
131 # where user can't view the container until the user has
132 # constructed the container request that references the container.
136 def finalize_if_needed
137 return if state != Committed
139 # get container lock first, then lock current container request
140 # (same order as Container#handle_completed). Locking always
141 # reloads the Container and ContainerRequest records.
142 c = Container.find_by_uuid(container_uuid)
146 if !c.nil? && container_uuid != c.uuid
147 # After locking, we've noticed a race, the container_uuid is
148 # different than the container record we just loaded. This
149 # can happen if Container#handle_completed scheduled a new
150 # container for retry and set container_uuid while we were
151 # waiting on the container lock. Restart the loop and get the
157 if state == Committed && c.final?
158 # The current container is
159 act_as_system_user do
160 leave_modified_by_user_alone do
165 elsif state == Committed
166 # Behave as if the container is cancelled
167 update_attributes!(state: Final)
173 # Finalize the container request after the container has
174 # finished/cancelled.
176 container = Container.find_by_uuid(container_uuid)
178 # We don't want to add the container cost if the container was
179 # already finished when this CR was committed. But we are
180 # running in an after_save hook after a lock/reload, so
181 # state_was has already been updated to Committed regardless.
182 # Hence the need for @state_was_when_initialized.
183 if @state_was_when_initialized == Committed
184 # Add the final container cost to our cumulative cost (which
185 # may already be non-zero from previous attempts if
186 # container_count_max > 1).
187 self.cumulative_cost += container.cost + container.subrequests_cost
190 # Add our cumulative cost to the subrequests_cost of the
191 # requesting container, if any.
192 if self.requesting_container_uuid
194 uuid: self.requesting_container_uuid,
195 state: Container::Running,
197 c.subrequests_cost += self.cumulative_cost
202 update_collections(container: container)
203 # update_collections makes a log collection that includes all of the logs
204 # for all of the containers associated with this request. For requests
205 # that are retried, this is the primary way users can get logs for
207 # The code below makes a log collection that is a verbatim copy of the
208 # container's logs. This is required for container reuse: a container
209 # will not be reused if the owner cannot read a collection with its logs.
210 # See the "readable log" section of Container.find_reusable().
211 if container.state == Container::Complete
212 log_col = Collection.where(portable_data_hash: container.log).first
214 # Need to save collection
215 completed_coll = Collection.new(
216 owner_uuid: self.owner_uuid,
217 name: "Container log for container #{container_uuid}",
220 'container_request' => self.uuid,
221 'container_uuid' => container_uuid,
223 portable_data_hash: log_col.portable_data_hash,
224 manifest_text: log_col.manifest_text,
225 storage_classes_desired: self.output_storage_classes
227 completed_coll.save_with_unique_name!
231 update_attributes!(state: Final)
234 def update_collections(container:, collections: ['log', 'output'])
236 # Check if parent is frozen or trashed, in which case it isn't
237 # valid to create new collections in the project, so return
238 # without creating anything.
239 owner = Group.find_by_uuid(self.owner_uuid)
240 return if owner && !owner.admin_change_permitted
242 collections.each do |out_type|
243 pdh = container.send(out_type)
245 c = Collection.where(portable_data_hash: pdh).first
247 manifest = c.manifest_text
249 coll_name = "Container #{out_type} for request #{uuid}"
251 if out_type == 'output'
252 if self.output_name and self.output_name != ""
253 coll_name = self.output_name
255 if self.output_ttl > 0
256 trash_at = db_current_time + self.output_ttl
260 coll_uuid = self.send(out_type + '_uuid')
261 coll = coll_uuid.nil? ? nil : Collection.where(uuid: coll_uuid).first
263 coll = Collection.new(
264 owner_uuid: self.owner_uuid,
267 storage_classes_desired: self.output_storage_classes)
271 # Copy the log into a merged collection
272 src = Arv::Collection.new(manifest)
273 dst = Arv::Collection.new(coll.manifest_text)
274 dst.cp_r("./", ".", src)
275 dst.cp_r("./", "log for container #{container.uuid}", src)
276 manifest = dst.manifest_text
279 merged_properties = {}
280 merged_properties['container_request'] = uuid
282 if out_type == 'output' and !requesting_container_uuid.nil?
283 # output of a child process, give it "intermediate" type by
285 merged_properties['type'] = 'intermediate'
287 merged_properties['type'] = out_type
290 if out_type == "output"
291 merged_properties.update(container.output_properties)
292 merged_properties.update(self.output_properties)
295 coll.assign_attributes(
296 portable_data_hash: Digest::MD5.hexdigest(manifest) + '+' + manifest.bytesize.to_s,
297 manifest_text: manifest,
300 properties: merged_properties)
301 coll.save_with_unique_name!
302 self.send(out_type + '_uuid=', coll.uuid)
306 def self.full_text_searchable_columns
307 super - ["mounts", "secret_mounts", "secret_mounts_md5", "runtime_token", "output_storage_classes"]
310 def set_priority_zero
311 self.update_attributes!(priority: 0) if self.priority > 0 && self.state != Final
316 def fill_field_defaults
317 self.state ||= Uncommitted
318 self.environment ||= {}
319 self.runtime_constraints ||= {}
321 self.secret_mounts ||= {}
323 self.container_count_max ||= Rails.configuration.Containers.MaxRetryAttempts
324 self.scheduling_parameters ||= {}
325 self.output_ttl ||= 0
330 if (container_uuid_changed? and
331 not current_user.andand.is_admin and
332 not container_uuid.nil?)
333 errors.add :container_uuid, "can only be updated to nil."
336 if self.container_count_changed?
337 errors.add :container_count, "cannot be updated directly."
340 if state_changed? and state == Committed and container_uuid.nil?
341 if self.command.length > 0 and self.command[0] == "arvados-cwl-runner"
342 # Special case, arvados-cwl-runner processes are always considered "supervisors"
343 self.scheduling_parameters['supervisor'] = true
346 c = Container.resolve(self)
348 if c.state == Container::Cancelled
349 # Lost a race, we have a lock on the container but the
350 # container was cancelled in a different request, restart
351 # the loop and resolve request to a new container.
354 self.container_uuid = c.uuid
358 if self.container_uuid != self.container_uuid_was
359 self.container_count += 1
360 return if self.container_uuid_was.nil?
362 old_container_uuid = self.container_uuid_was
363 old_container_log = Container.where(uuid: old_container_uuid).pluck(:log).first
364 return if old_container_log.nil?
366 old_logs = Collection.where(portable_data_hash: old_container_log).first
367 return if old_logs.nil?
369 log_coll = self.log_uuid.nil? ? nil : Collection.where(uuid: self.log_uuid).first
370 if self.log_uuid.nil?
371 log_coll = Collection.new(
372 owner_uuid: self.owner_uuid,
373 name: coll_name = "Container log for request #{uuid}",
375 storage_classes_desired: self.output_storage_classes)
378 # copy logs from old container into CR's log collection
379 src = Arv::Collection.new(old_logs.manifest_text)
380 dst = Arv::Collection.new(log_coll.manifest_text)
381 dst.cp_r("./", "log for container #{old_container_uuid}", src)
382 manifest = dst.manifest_text
384 log_coll.assign_attributes(
385 portable_data_hash: Digest::MD5.hexdigest(manifest) + '+' + manifest.bytesize.to_s,
386 manifest_text: manifest)
387 log_coll.save_with_unique_name!
388 self.log_uuid = log_coll.uuid
393 if (new_record? || state_changed?) &&
394 state == Committed &&
395 Rails.configuration.Containers.AlwaysUsePreemptibleInstances &&
396 get_requesting_container_uuid() &&
397 self.class.any_preemptible_instances?
398 self.scheduling_parameters['preemptible'] = true
402 def validate_runtime_constraints
405 ['vcpus', 'ram'].each do |k|
406 v = runtime_constraints[k]
407 if !v.is_a?(Integer) || v <= 0
408 errors.add(:runtime_constraints,
409 "[#{k}]=#{v.inspect} must be a positive integer")
412 if runtime_constraints['cuda']
413 ['device_count'].each do |k|
414 v = runtime_constraints['cuda'][k]
415 if !v.is_a?(Integer) || v < 0
416 errors.add(:runtime_constraints,
417 "[cuda.#{k}]=#{v.inspect} must be a positive or zero integer")
420 ['driver_version', 'hardware_capability'].each do |k|
421 v = runtime_constraints['cuda'][k]
422 if !v.is_a?(String) || (runtime_constraints['cuda']['device_count'] > 0 && v.to_f == 0.0)
423 errors.add(:runtime_constraints,
424 "[cuda.#{k}]=#{v.inspect} must be a string in format 'X.Y'")
431 def validate_datatypes
434 errors.add(:command, "must be an array of strings but has entry #{c.class}")
437 environment.each do |k,v|
438 if !k.is_a?(String) || !v.is_a?(String)
439 errors.add(:environment, "must be an map of String to String but has entry #{k.class} to #{v.class}")
442 [:mounts, :secret_mounts].each do |m|
443 self[m].each do |k, v|
444 if !k.is_a?(String) || !v.is_a?(Hash)
445 errors.add(m, "must be an map of String to Hash but is has entry #{k.class} to #{v.class}")
448 errors.add(m, "each item must have a 'kind' field")
450 [[String, ["kind", "portable_data_hash", "uuid", "device_type",
451 "path", "commit", "repository_name", "git_url"]],
452 [Integer, ["capacity"]]].each do |t, fields|
454 if !v[f].nil? && !v[f].is_a?(t)
455 errors.add(m, "#{k}: #{f} must be a #{t} but is #{v[f].class}")
459 ["writable", "exclude_from_output"].each do |f|
460 if !v[f].nil? && !v[f].is_a?(TrueClass) && !v[f].is_a?(FalseClass)
461 errors.add(m, "#{k}: #{f} must be a #{t} but is #{v[f].class}")
468 def validate_scheduling_parameters
469 if self.state == Committed
470 if scheduling_parameters.include? 'partitions' and
471 (!scheduling_parameters['partitions'].is_a?(Array) ||
472 scheduling_parameters['partitions'].reject{|x| !x.is_a?(String)}.size !=
473 scheduling_parameters['partitions'].size)
474 errors.add :scheduling_parameters, "partitions must be an array of strings"
476 if scheduling_parameters['preemptible'] &&
477 (new_record? || state_changed?) &&
478 !self.class.any_preemptible_instances?
479 errors.add :scheduling_parameters, "preemptible instances are not configured in InstanceTypes"
481 if scheduling_parameters.include? 'max_run_time' and
482 (!scheduling_parameters['max_run_time'].is_a?(Integer) ||
483 scheduling_parameters['max_run_time'] < 0)
484 errors.add :scheduling_parameters, "max_run_time must be positive integer"
489 def check_update_whitelist
490 permitted = AttrsPermittedAlways.dup
492 if self.new_record? || self.state_was == Uncommitted
493 # Allow create-and-commit in a single operation.
494 permitted.push(*AttrsPermittedBeforeCommit)
495 elsif mounts_changed? && mounts_was.keys.sort == mounts.keys.sort
496 # Ignore the updated mounts if the only changes are default/zero
497 # values as added by controller, see 17774
499 mounts.each do |path, mount|
500 (mount.to_a - mounts_was[path].to_a).each do |k, v|
501 if ![0, "", false, nil].index(v)
502 only_defaults = false
507 clear_attribute_change("mounts")
513 permitted.push :priority, :container_count_max, :container_uuid, :cumulative_cost
515 if self.priority.nil?
516 self.errors.add :priority, "cannot be nil"
519 # Allow container count to increment (not by client, only by us
520 # -- see set_container)
521 permitted.push :container_count
523 if current_user.andand.is_admin
524 permitted.push :log_uuid
528 if self.state_was == Committed
529 # "Cancel" means setting priority=0, state=Committed
530 permitted.push :priority, :cumulative_cost
532 if current_user.andand.is_admin
533 permitted.push :output_uuid, :log_uuid
542 def secret_mounts_key_conflict
543 secret_mounts.each do |k, v|
544 if mounts.has_key?(k)
545 errors.add(:secret_mounts, 'conflict with non-secret mounts')
551 def validate_runtime_token
552 if !self.runtime_token.nil? && self.runtime_token_changed?
553 if !runtime_token[0..2] == "v2/"
554 errors.add :runtime_token, "not a v2 token"
557 if ApiClientAuthorization.validate(token: runtime_token).nil?
558 errors.add :runtime_token, "failed validation"
564 if self.state == Final
565 self.secret_mounts = {}
566 self.runtime_token = nil
571 return unless saved_change_to_state? || saved_change_to_priority? || saved_change_to_container_uuid?
572 update_priorities container_uuid_before_last_save if !container_uuid_before_last_save.nil? and container_uuid_before_last_save != self.container_uuid
573 update_priorities self.container_uuid if self.container_uuid
576 def set_requesting_container_uuid
577 if (self.requesting_container_uuid = get_requesting_container_uuid())
578 # Determine the priority of container request for the requesting
580 self.priority = ContainerRequest.where(container_uuid: self.requesting_container_uuid).maximum("priority") || 0
584 def get_requesting_container_uuid
585 return self.requesting_container_uuid || Container.for_current_token.andand.uuid