6277: add default_empty_manifest before_validation filter and update the tests accord...
[arvados.git] / services / api / app / models / job.rb
index 32f28e3582d509930a4bdf3339659035f6d9b5d5..6c24293334f6d4cc5af371b1b2b9a0d370466530 100644 (file)
@@ -2,17 +2,25 @@ class Job < ArvadosModel
   include HasUuid
   include KindAndEtag
   include CommonApiTemplate
-  attr_protected :docker_image_locator
+  attr_protected :arvados_sdk_version, :docker_image_locator
   serialize :script_parameters, Hash
   serialize :runtime_constraints, Hash
   serialize :tasks_summary, Hash
   before_create :ensure_unique_submit_id
   after_commit :trigger_crunch_dispatch_if_cancelled, :on => :update
   before_validation :set_priority
+  before_validation :update_state_from_old_state_attrs
   validate :ensure_script_version_is_commit
   validate :find_docker_image_locator
+  validate :find_arvados_sdk_version
+  validate :validate_status
+  validate :validate_state_change
+  validate :ensure_no_collection_uuids_in_script_params
+  before_save :tag_version_in_internal_repository
+  before_save :update_timestamps_when_state_changes
 
   has_many :commit_ancestors, :foreign_key => :descendant, :primary_key => :script_version
+  has_many(:nodes, foreign_key: :job_uuid, primary_key: :uuid)
 
   class SubmitIdReused < StandardError
   end
@@ -31,28 +39,51 @@ class Job < ArvadosModel
     t.add :output
     t.add :success
     t.add :running
+    t.add :state
     t.add :is_locked_by_uuid
     t.add :log
     t.add :runtime_constraints
     t.add :tasks_summary
-    t.add :dependencies
     t.add :nondeterministic
     t.add :repository
     t.add :supplied_script_version
+    t.add :arvados_sdk_version
     t.add :docker_image_locator
+    t.add :queue_position
+    t.add :node_uuids
     t.add :description
   end
 
+  # Supported states for a job
+  States = [
+            (Queued = 'Queued'),
+            (Running = 'Running'),
+            (Cancelled = 'Cancelled'),
+            (Failed = 'Failed'),
+            (Complete = 'Complete'),
+           ]
+
   def assert_finished
-    update_attributes(finished_at: finished_at || Time.now,
+    update_attributes(finished_at: finished_at || db_current_time,
                       success: success.nil? ? false : success,
                       running: false)
   end
 
+  def node_uuids
+    nodes.map(&:uuid)
+  end
+
   def self.queue
-    self.where('started_at is ? and is_locked_by_uuid is ? and cancelled_at is ? and success is ?',
-               nil, nil, nil, nil).
-      order('priority desc, created_at')
+    self.where('state = ?', Queued).order('priority desc, created_at')
+  end
+
+  def queue_position
+    Job::queue.each_with_index do |job, index|
+      if job[:uuid] == self.uuid
+        return index
+      end
+    end
+    nil
   end
 
   def self.running
@@ -60,6 +91,18 @@ class Job < ArvadosModel
       order('priority desc, created_at')
   end
 
+  def lock locked_by_uuid
+    transaction do
+      self.reload
+      unless self.state == Queued and self.is_locked_by_uuid.nil?
+        raise AlreadyLockedError
+      end
+      self.state = Running
+      self.is_locked_by_uuid = locked_by_uuid
+      self.save!
+    end
+  end
+
   protected
 
   def foreign_key_attributes
@@ -82,21 +125,43 @@ class Job < ArvadosModel
   end
 
   def ensure_script_version_is_commit
-    if self.is_locked_by_uuid and self.started_at
+    if state == Running
       # Apparently client has already decided to go for it. This is
       # needed to run a local job using a local working directory
       # instead of a commit-ish.
       return true
     end
-    if new_record? or script_version_changed?
-      sha1 = Commit.find_commit_range(current_user, self.repository, nil, self.script_version, nil)[0] rescue nil
-      if sha1
-        self.supplied_script_version = self.script_version if self.supplied_script_version.nil? or self.supplied_script_version.empty?
-        self.script_version = sha1
-      else
-        self.errors.add :script_version, "#{self.script_version} does not resolve to a commit"
+    if new_record? or repository_changed? or script_version_changed?
+      sha1 = Commit.find_commit_range(repository,
+                                      nil, script_version, nil).first
+      if not sha1
+        errors.add :script_version, "#{script_version} does not resolve to a commit"
         return false
       end
+      if supplied_script_version.nil? or supplied_script_version.empty?
+        self.supplied_script_version = script_version
+      end
+      self.script_version = sha1
+    end
+    true
+  end
+
+  def tag_version_in_internal_repository
+    if state == Running
+      # No point now. See ensure_script_version_is_commit.
+      true
+    elsif errors.any?
+      # Won't be saved, and script_version might not even be valid.
+      true
+    elsif new_record? or repository_changed? or script_version_changed?
+      uuid_was = uuid
+      begin
+        assign_uuid
+        Commit.tag_in_internal_repository repository, script_version, uuid
+      rescue
+        uuid = uuid_was
+        raise
+      end
     end
   end
 
@@ -109,47 +174,50 @@ class Job < ArvadosModel
     true
   end
 
-  def find_docker_image_locator
-    # Find the Collection that holds the Docker image specified in the
-    # runtime constraints, and store its locator in docker_image_locator.
-    unless runtime_constraints.is_a? Hash
-      # We're still in validation stage, so we can't assume
-      # runtime_constraints isn't something horrible like an array or
-      # a string. Treat those cases as "no docker image supplied";
-      # other validations will fail anyway.
-      self.docker_image_locator = nil
-      return true
+  def resolve_runtime_constraint(key, attr_sym)
+    if ((runtime_constraints.is_a? Hash) and
+        (search = runtime_constraints[key]))
+      ok, result = yield search
+    else
+      ok, result = true, nil
     end
-    image_search = runtime_constraints['docker_image']
-    image_tag = runtime_constraints['docker_image_tag']
-    if image_search.nil?
-      self.docker_image_locator = nil
-      true
-    elsif coll = Collection.for_latest_docker_image(image_search, image_tag)
-      self.docker_image_locator = coll.portable_data_hash
-      true
+    if ok
+      send("#{attr_sym}=".to_sym, result)
     else
-      errors.add(:docker_image_locator, "not found for #{image_search}")
-      false
+      errors.add(attr_sym, result)
     end
+    ok
   end
 
-  def dependencies
-    deps = {}
-    queue = self.script_parameters.values
-    while not queue.empty?
-      queue = queue.flatten.compact.collect do |v|
-        if v.is_a? Hash
-          v.values
-        elsif v.is_a? String
-          v.match(/^(([0-9a-f]{32})\b(\+[^,]+)?,?)*$/) do |locator|
-            deps[locator.to_s] = true
-          end
-          nil
-        end
+  def find_arvados_sdk_version
+    resolve_runtime_constraint("arvados_sdk_version",
+                               :arvados_sdk_version) do |git_search|
+      commits = Commit.find_commit_range("arvados",
+                                         nil, git_search, nil)
+      if commits.empty?
+        [false, "#{git_search} does not resolve to a commit"]
+      elsif not runtime_constraints["docker_image"]
+        [false, "cannot be specified without a Docker image constraint"]
+      else
+        [true, commits.first]
+      end
+    end
+  end
+
+  def find_docker_image_locator
+    runtime_constraints['docker_image'] =
+        Rails.configuration.default_docker_image_for_jobs if ((runtime_constraints.is_a? Hash) and
+                                                              (runtime_constraints['docker_image']).nil? and
+                                                              Rails.configuration.default_docker_image_for_jobs)
+    resolve_runtime_constraint("docker_image",
+                               :docker_image_locator) do |image_search|
+      image_tag = runtime_constraints['docker_image_tag']
+      if coll = Collection.for_latest_docker_image(image_search, image_tag)
+        [true, coll.portable_data_hash]
+      else
+        [false, "not found for #{image_search}"]
       end
     end
-    deps.keys
   end
 
   def permission_to_update
@@ -169,7 +237,8 @@ class Job < ArvadosModel
           success_changed? or
           output_changed? or
           log_changed? or
-          tasks_summary_changed?
+          tasks_summary_changed? or
+          state_changed?
         logger.warn "User #{current_user.uuid if current_user} tried to change protected job attributes on locked #{self.class.to_s} #{uuid_was}"
         return false
       end
@@ -197,7 +266,7 @@ class Job < ArvadosModel
       # Ensure cancelled_at cannot be set to arbitrary non-now times,
       # or changed once it is set.
       if self.cancelled_at and not self.cancelled_at_was
-        self.cancelled_at = Time.now
+        self.cancelled_at = db_current_time
         self.cancelled_by_user_uuid = current_user.uuid
         self.cancelled_by_client_uuid = current_api_client.andand.uuid
         @need_crunch_dispatch_trigger = true
@@ -217,4 +286,133 @@ class Job < ArvadosModel
       end
     end
   end
+
+  def update_timestamps_when_state_changes
+    return if not (state_changed? or new_record?)
+
+    case state
+    when Running
+      self.started_at ||= db_current_time
+    when Failed, Complete
+      self.finished_at ||= db_current_time
+    when Cancelled
+      self.cancelled_at ||= db_current_time
+    end
+
+    # TODO: Remove the following case block when old "success" and
+    # "running" attrs go away. Until then, this ensures we still
+    # expose correct success/running flags to older clients, even if
+    # some new clients are writing only the new state attribute.
+    case state
+    when Queued
+      self.running = false
+      self.success = nil
+    when Running
+      self.running = true
+      self.success = nil
+    when Cancelled, Failed
+      self.running = false
+      self.success = false
+    when Complete
+      self.running = false
+      self.success = true
+    end
+    self.running ||= false # Default to false instead of nil.
+
+    @need_crunch_dispatch_trigger = true
+
+    true
+  end
+
+  def update_state_from_old_state_attrs
+    # If a client has touched the legacy state attrs, update the
+    # "state" attr to agree with the updated values of the legacy
+    # attrs.
+    #
+    # TODO: Remove this method when old "success" and "running" attrs
+    # go away.
+    if cancelled_at_changed? or
+        success_changed? or
+        running_changed? or
+        state.nil?
+      if cancelled_at
+        self.state = Cancelled
+      elsif success == false
+        self.state = Failed
+      elsif success == true
+        self.state = Complete
+      elsif running == true
+        self.state = Running
+      else
+        self.state = Queued
+      end
+    end
+    true
+  end
+
+  def validate_status
+    if self.state.in?(States)
+      true
+    else
+      errors.add :state, "#{state.inspect} must be one of: #{States.inspect}"
+      false
+    end
+  end
+
+  def validate_state_change
+    ok = true
+    if self.state_changed?
+      ok = case self.state_was
+           when nil
+             # state isn't set yet
+             true
+           when Queued
+             # Permit going from queued to any state
+             true
+           when Running
+             # From running, may only transition to a finished state
+             [Complete, Failed, Cancelled].include? self.state
+           when Complete, Failed, Cancelled
+             # Once in a finished state, don't permit any more state changes
+             false
+           else
+             # Any other state transition is also invalid
+             false
+           end
+      if not ok
+        errors.add :state, "invalid change from #{self.state_was} to #{self.state}"
+      end
+    end
+    ok
+  end
+
+  def ensure_no_collection_uuids_in_script_params
+    # recursive_hash_search searches recursively through hashes and
+    # arrays in 'thing' for string fields matching regular expression
+    # 'pattern'.  Returns true if pattern is found, false otherwise.
+    def recursive_hash_search thing, pattern
+      if thing.is_a? Hash
+        thing.each do |k, v|
+          return true if recursive_hash_search v, pattern
+        end
+      elsif thing.is_a? Array
+        thing.each do |k|
+          return true if recursive_hash_search k, pattern
+        end
+      elsif thing.is_a? String
+        return true if thing.match pattern
+      end
+      false
+    end
+
+    # Fail validation if any script_parameters field includes a string containing a
+    # collection uuid pattern.
+    if self.script_parameters_changed?
+      if recursive_hash_search(self.script_parameters, Collection.uuid_regex)
+        self.errors.add :script_parameters, "must use portable_data_hash instead of collection uuid"
+        return false
+      end
+    end
+    true
+  end
 end