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: {}
27 serialize :environment, Hash
28 serialize :mounts, Hash
29 serialize :runtime_constraints, Hash
30 serialize :command, Array
31 serialize :scheduling_parameters, Hash
33 before_validation :fill_field_defaults, :if => :new_record?
34 before_validation :validate_runtime_constraints
35 before_validation :set_default_preemptible_scheduling_parameter
36 before_validation :set_container
37 validates :command, :container_image, :output_path, :cwd, :presence => true
38 validates :output_ttl, numericality: { only_integer: true, greater_than_or_equal_to: 0 }
39 validates :priority, numericality: { only_integer: true, greater_than_or_equal_to: 0, less_than_or_equal_to: 1000 }
40 validate :validate_datatypes
41 validate :validate_scheduling_parameters
42 validate :validate_state_change
43 validate :check_update_whitelist
44 validate :secret_mounts_key_conflict
45 validate :validate_runtime_token
46 before_save :scrub_secrets
47 before_create :set_requesting_container_uuid
48 before_destroy :set_priority_zero
49 after_save :update_priority
50 after_save :finalize_if_needed
52 api_accessible :user, extend: :common do |t|
54 t.add :container_count
55 t.add :container_count_max
56 t.add :container_image
72 t.add :requesting_container_uuid
73 t.add :runtime_constraints
74 t.add :scheduling_parameters
79 # Supported states for a container request
82 (Uncommitted = 'Uncommitted'),
83 (Committed = 'Committed'),
88 nil => [Uncommitted, Committed],
89 Uncommitted => [Committed],
93 AttrsPermittedAlways = [:owner_uuid, :state, :name, :description, :properties]
94 AttrsPermittedBeforeCommit = [:command, :container_count_max,
95 :container_image, :cwd, :environment, :filters, :mounts,
96 :output_path, :priority, :runtime_token,
97 :runtime_constraints, :state, :container_uuid, :use_existing,
98 :scheduling_parameters, :secret_mounts, :output_name, :output_ttl]
100 def self.limit_index_columns_read
104 def logged_attributes
105 super.except('secret_mounts', 'runtime_token')
108 def state_transitions
112 def skip_uuid_read_permission_check
113 # The uuid_read_permission_check prevents users from making
114 # references to objects they can't view. However, in this case we
115 # don't want to do that check since there's a circular dependency
116 # where user can't view the container until the user has
117 # constructed the container request that references the container.
121 def finalize_if_needed
122 return if state != Committed
124 # get container lock first, then lock current container request
125 # (same order as Container#handle_completed). Locking always
126 # reloads the Container and ContainerRequest records.
127 c = Container.find_by_uuid(container_uuid)
131 if container_uuid != c.uuid
132 # After locking, we've noticed a race, the container_uuid is
133 # different than the container record we just loaded. This
134 # can happen if Container#handle_completed scheduled a new
135 # container for retry and set container_uuid while we were
136 # waiting on the container lock. Restart the loop and get the
141 if state == Committed && c.final?
142 # The current container is
143 act_as_system_user do
144 leave_modified_by_user_alone do
153 # Finalize the container request after the container has
154 # finished/cancelled.
156 update_collections(container: Container.find_by_uuid(container_uuid))
157 update_attributes!(state: Final)
160 def update_collections(container:, collections: ['log', 'output'])
161 collections.each do |out_type|
162 pdh = container.send(out_type)
164 coll_name = "Container #{out_type} for request #{uuid}"
166 if out_type == 'output'
168 coll_name = self.output_name
170 if self.output_ttl > 0
171 trash_at = db_current_time + self.output_ttl
174 manifest = Collection.where(portable_data_hash: pdh).first.manifest_text
176 coll_uuid = self.send(out_type + '_uuid')
177 coll = coll_uuid.nil? ? nil : Collection.where(uuid: coll_uuid).first
179 coll = Collection.new(
180 owner_uuid: self.owner_uuid,
185 'container_request' => uuid,
190 src = Arv::Collection.new(manifest)
191 dst = Arv::Collection.new(coll.manifest_text)
192 dst.cp_r("./", ".", src)
193 dst.cp_r("./", "log for container #{container.uuid}", src)
194 manifest = dst.manifest_text
197 coll.assign_attributes(
198 portable_data_hash: Digest::MD5.hexdigest(manifest) + '+' + manifest.bytesize.to_s,
199 manifest_text: manifest,
202 coll.save_with_unique_name!
203 self.send(out_type + '_uuid=', coll.uuid)
207 def self.full_text_searchable_columns
208 super - ["mounts", "secret_mounts", "secret_mounts_md5", "runtime_token"]
213 def fill_field_defaults
214 self.state ||= Uncommitted
215 self.environment ||= {}
216 self.runtime_constraints ||= {}
218 self.secret_mounts ||= {}
220 self.container_count_max ||= Rails.configuration.Containers.MaxRetryAttempts
221 self.scheduling_parameters ||= {}
222 self.output_ttl ||= 0
227 if (container_uuid_changed? and
228 not current_user.andand.is_admin and
229 not container_uuid.nil?)
230 errors.add :container_uuid, "can only be updated to nil."
233 if state_changed? and state == Committed and container_uuid.nil?
235 c = Container.resolve(self)
237 if c.state == Container::Cancelled
238 # Lost a race, we have a lock on the container but the
239 # container was cancelled in a different request, restart
240 # the loop and resolve request to a new container.
243 self.container_uuid = c.uuid
247 if self.container_uuid != self.container_uuid_was
248 if self.container_count_changed?
249 errors.add :container_count, "cannot be updated directly."
252 self.container_count += 1
253 if self.container_uuid_was
254 old_container = Container.find_by_uuid(self.container_uuid_was)
255 old_logs = Collection.where(portable_data_hash: old_container.log).first
257 log_coll = self.log_uuid.nil? ? nil : Collection.where(uuid: self.log_uuid).first
258 if self.log_uuid.nil?
259 log_coll = Collection.new(
260 owner_uuid: self.owner_uuid,
261 name: coll_name = "Container log for request #{uuid}",
265 # copy logs from old container into CR's log collection
266 src = Arv::Collection.new(old_logs.manifest_text)
267 dst = Arv::Collection.new(log_coll.manifest_text)
268 dst.cp_r("./", "log for container #{old_container.uuid}", src)
269 manifest = dst.manifest_text
271 log_coll.assign_attributes(
272 portable_data_hash: Digest::MD5.hexdigest(manifest) + '+' + manifest.bytesize.to_s,
273 manifest_text: manifest)
274 log_coll.save_with_unique_name!
275 self.log_uuid = log_coll.uuid
282 def set_default_preemptible_scheduling_parameter
283 c = get_requesting_container()
284 if self.state == Committed
285 # If preemptible instances (eg: AWS Spot Instances) are allowed,
286 # ask them on child containers by default.
287 if Rails.configuration.Containers.UsePreemptibleInstances and !c.nil? and
288 self.scheduling_parameters['preemptible'].nil?
289 self.scheduling_parameters['preemptible'] = true
294 def validate_runtime_constraints
299 ['keep_cache_ram', false]].each do |k, required|
300 if !required && !runtime_constraints.include?(k)
303 v = runtime_constraints[k]
304 unless (v.is_a?(Integer) && v > 0)
305 errors.add(:runtime_constraints,
306 "[#{k}]=#{v.inspect} must be a positive integer")
312 def validate_datatypes
315 errors.add(:command, "must be an array of strings but has entry #{c.class}")
318 environment.each do |k,v|
319 if !k.is_a?(String) || !v.is_a?(String)
320 errors.add(:environment, "must be an map of String to String but has entry #{k.class} to #{v.class}")
323 [:mounts, :secret_mounts].each do |m|
324 self[m].each do |k, v|
325 if !k.is_a?(String) || !v.is_a?(Hash)
326 errors.add(m, "must be an map of String to Hash but is has entry #{k.class} to #{v.class}")
329 errors.add(m, "each item must have a 'kind' field")
331 [[String, ["kind", "portable_data_hash", "uuid", "device_type",
332 "path", "commit", "repository_name", "git_url"]],
333 [Integer, ["capacity"]]].each do |t, fields|
335 if !v[f].nil? && !v[f].is_a?(t)
336 errors.add(m, "#{k}: #{f} must be a #{t} but is #{v[f].class}")
340 ["writable", "exclude_from_output"].each do |f|
341 if !v[f].nil? && !v[f].is_a?(TrueClass) && !v[f].is_a?(FalseClass)
342 errors.add(m, "#{k}: #{f} must be a #{t} but is #{v[f].class}")
349 def validate_scheduling_parameters
350 if self.state == Committed
351 if scheduling_parameters.include? 'partitions' and
352 (!scheduling_parameters['partitions'].is_a?(Array) ||
353 scheduling_parameters['partitions'].reject{|x| !x.is_a?(String)}.size !=
354 scheduling_parameters['partitions'].size)
355 errors.add :scheduling_parameters, "partitions must be an array of strings"
357 if !Rails.configuration.Containers.UsePreemptibleInstances and scheduling_parameters['preemptible']
358 errors.add :scheduling_parameters, "preemptible instances are not allowed"
360 if scheduling_parameters.include? 'max_run_time' and
361 (!scheduling_parameters['max_run_time'].is_a?(Integer) ||
362 scheduling_parameters['max_run_time'] < 0)
363 errors.add :scheduling_parameters, "max_run_time must be positive integer"
368 def check_update_whitelist
369 permitted = AttrsPermittedAlways.dup
371 if self.new_record? || self.state_was == Uncommitted
372 # Allow create-and-commit in a single operation.
373 permitted.push(*AttrsPermittedBeforeCommit)
378 permitted.push :priority, :container_count_max, :container_uuid
380 if self.container_uuid.nil?
381 self.errors.add :container_uuid, "has not been resolved to a container."
384 if self.priority.nil?
385 self.errors.add :priority, "cannot be nil"
388 # Allow container count to increment by 1
389 if (self.container_uuid &&
390 self.container_uuid != self.container_uuid_was &&
391 self.container_count == 1 + (self.container_count_was || 0))
392 permitted.push :container_count
395 if current_user.andand.is_admin
396 permitted.push :log_uuid
400 if self.state_was == Committed
401 # "Cancel" means setting priority=0, state=Committed
402 permitted.push :priority
404 if current_user.andand.is_admin
405 permitted.push :output_uuid, :log_uuid
414 def secret_mounts_key_conflict
415 secret_mounts.each do |k, v|
416 if mounts.has_key?(k)
417 errors.add(:secret_mounts, 'conflict with non-secret mounts')
423 def validate_runtime_token
424 if !self.runtime_token.nil? && self.runtime_token_changed?
425 if !runtime_token[0..2] == "v2/"
426 errors.add :runtime_token, "not a v2 token"
429 if ApiClientAuthorization.validate(token: runtime_token).nil?
430 errors.add :runtime_token, "failed validation"
436 if self.state == Final
437 self.secret_mounts = {}
438 self.runtime_token = nil
443 return unless state_changed? || priority_changed? || container_uuid_changed?
444 act_as_system_user do
446 where('uuid in (?)', [self.container_uuid_was, self.container_uuid].compact).
447 map(&:update_priority!)
451 def set_priority_zero
452 self.update_attributes!(priority: 0) if self.state != Final
455 def set_requesting_container_uuid
456 c = get_requesting_container()
458 self.requesting_container_uuid = c.uuid
459 # Determine the priority of container request for the requesting
461 self.priority = ContainerRequest.where(container_uuid: self.requesting_container_uuid).maximum("priority") || 0
465 def get_requesting_container
466 return self.requesting_container_uuid if !self.requesting_container_uuid.nil?
467 Container.for_current_token