Merge branch '9001-summary-skip-malformed-lines' closes #9001
[arvados.git] / services / api / app / models / container.rb
index 787047df68e877bc8944b3c23186cc400c3ae227..a60ea427b78649ae15977b69bea99060e9c465c9 100644 (file)
@@ -5,6 +5,7 @@ class Container < ArvadosModel
   include KindAndEtag
   include CommonApiTemplate
   include WhitelistUpdate
+  extend CurrentApiClient
 
   serialize :environment, Hash
   serialize :mounts, Hash
@@ -16,9 +17,13 @@ class Container < ArvadosModel
   validates :command, :container_image, :output_path, :cwd, :priority, :presence => true
   validate :validate_state_change
   validate :validate_change
+  validate :validate_lock
+  after_validation :assign_auth
+  before_save :sort_serialized_attrs
   after_save :handle_completed
 
   has_many :container_requests, :foreign_key => :container_uuid, :class_name => 'ContainerRequest', :primary_key => :uuid
+  belongs_to :auth, :class_name => 'ApiClientAuthorization', :foreign_key => :auth_uuid, :primary_key => :uuid
 
   api_accessible :user, extend: :common do |t|
     t.add :command
@@ -27,6 +32,7 @@ class Container < ArvadosModel
     t.add :environment
     t.add :exit_code
     t.add :finished_at
+    t.add :locked_by_uuid
     t.add :log
     t.add :mounts
     t.add :output
@@ -36,12 +42,14 @@ class Container < ArvadosModel
     t.add :runtime_constraints
     t.add :started_at
     t.add :state
+    t.add :auth_uuid
   end
 
   # Supported states for a container
   States =
     [
      (Queued = 'Queued'),
+     (Locked = 'Locked'),
      (Running = 'Running'),
      (Complete = 'Complete'),
      (Cancelled = 'Cancelled')
@@ -49,7 +57,8 @@ class Container < ArvadosModel
 
   State_transitions = {
     nil => [Queued],
-    Queued => [Running, Cancelled],
+    Queued => [Locked, Cancelled],
+    Locked => [Queued, Running, Cancelled],
     Running => [Complete, Cancelled]
   }
 
@@ -58,20 +67,105 @@ class Container < ArvadosModel
   end
 
   def update_priority!
-    if [Queued, Running].include? self.state
+    if [Queued, Locked, Running].include? self.state
       # Update the priority of this container to the maximum priority of any of
       # its committed container requests and save the record.
-      max = 0
-      ContainerRequest.where(container_uuid: uuid).each do |cr|
-        if cr.state == ContainerRequest::Committed and cr.priority > max
-          max = cr.priority
-        end
+      self.priority = ContainerRequest.
+        where(container_uuid: uuid,
+              state: ContainerRequest::Committed).
+        maximum('priority')
+      self.save!
+    end
+  end
+
+  def self.find_reusable(attrs)
+    candidates = Container.
+      where('command = ?', attrs[:command].to_yaml).
+      where('cwd = ?', attrs[:cwd]).
+      where('environment = ?', self.deep_sort_hash(attrs[:environment]).to_yaml).
+      where('output_path = ?', attrs[:output_path]).
+      where('container_image = ?', attrs[:container_image]).
+      where('mounts = ?', self.deep_sort_hash(attrs[:mounts]).to_yaml).
+      where('runtime_constraints = ?', self.deep_sort_hash(attrs[:runtime_constraints]).to_yaml)
+
+    # Check for Completed candidates that had consistent outputs.
+    completed = candidates.where(state: Complete).where(exit_code: 0)
+    outputs = completed.select('output').group('output').limit(2)
+    if outputs.count.count != 1
+      Rails.logger.debug("Found #{outputs.count.length} different outputs")
+    elsif Collection.
+        readable_by(current_user).
+        where(portable_data_hash: outputs.first.output).
+        count < 1
+      Rails.logger.info("Found reusable container(s) " +
+                        "but output #{outputs.first} is not readable " +
+                        "by user #{current_user.uuid}")
+    else
+      # Return the oldest eligible container whose log is still
+      # present and readable by current_user.
+      readable_pdh = Collection.
+        readable_by(current_user).
+        select('portable_data_hash')
+      completed = completed.
+        where("log in (#{readable_pdh.to_sql})").
+        order('finished_at asc').
+        limit(1)
+      if completed.first
+        return completed.first
+      else
+        Rails.logger.info("Found reusable container(s) but none with a log " +
+                          "readable by user #{current_user.uuid}")
       end
-      self.priority = max
+    end
+
+    # Check for Running candidates and return the most likely to finish sooner.
+    running = candidates.where(state: Running).
+      order('progress desc, started_at asc').limit(1).first
+    return running if not running.nil?
+
+    # Check for Locked or Queued ones and return the most likely to start first.
+    locked_or_queued = candidates.where("state IN (?)", [Locked, Queued]).
+      order('state asc, priority desc, created_at asc').limit(1).first
+    return locked_or_queued if not locked_or_queued.nil?
+
+    # No suitable candidate found.
+    nil
+  end
+
+  def lock
+    with_lock do
+      if self.state == Locked
+        raise AlreadyLockedError
+      end
+      self.state = Locked
+      self.save!
+    end
+  end
+
+  def unlock
+    with_lock do
+      if self.state == Queued
+        raise InvalidStateTransitionError
+      end
+      self.state = Queued
       self.save!
     end
   end
 
+  def self.readable_by(*users_list)
+    if users_list.select { |u| u.is_admin }.any?
+      return self
+    end
+    user_uuids = users_list.map { |u| u.uuid }
+    uuid_list = user_uuids + users_list.flat_map { |u| u.groups_i_can(:read) }
+    uuid_list.uniq!
+    permitted = "(SELECT head_uuid FROM links WHERE link_class='permission' AND tail_uuid IN (:uuids))"
+    joins(:container_requests).
+      where("container_requests.uuid IN #{permitted} OR "+
+            "container_requests.owner_uuid IN (:uuids)",
+            uuids: uuid_list)
+  end
+
   protected
 
   def fill_field_defaults
@@ -102,52 +196,112 @@ class Container < ArvadosModel
   end
 
   def validate_change
-    permitted = []
+    permitted = [:state]
 
     if self.new_record?
-      permitted.push :owner_uuid, :command, :container_image, :cwd, :environment,
-                     :mounts, :output_path, :priority, :runtime_constraints, :state
+      permitted.push(:owner_uuid, :command, :container_image, :cwd,
+                     :environment, :mounts, :output_path, :priority,
+                     :runtime_constraints)
     end
 
     case self.state
-    when Queued
-      # permit priority change only.
+    when Queued, Locked
       permitted.push :priority
 
     when Running
+      permitted.push :priority, :progress
       if self.state_changed?
-        # At point of state change, can set state and started_at
-        permitted.push :state, :started_at
-      else
-        # While running, can update priority and progress.
-        permitted.push :priority, :progress
+        permitted.push :started_at
       end
 
     when Complete
-      if self.state_changed?
-        permitted.push :state, :finished_at, :output, :log, :exit_code
-      else
-        errors.add :state, "cannot update record"
+      if self.state_was == Running
+        permitted.push :finished_at, :output, :log, :exit_code
       end
 
     when Cancelled
-      if self.state_changed?
-        if self.state_was == Running
-          permitted.push :state, :finished_at, :output, :log
-        elsif self.state_was == Queued
-          permitted.push :state, :finished_at
-        end
-      else
-        errors.add :state, "cannot update record"
+      case self.state_was
+      when Running
+        permitted.push :finished_at, :output, :log
+      when Queued, Locked
+        permitted.push :finished_at
       end
 
     else
-      errors.add :state, "invalid state"
+      # The state_transitions check will add an error message for this
+      return false
     end
 
     check_update_whitelist permitted
   end
 
+  def validate_lock
+    # If the Container is already locked by someone other than the
+    # current api_client_auth, disallow all changes -- except
+    # priority, which needs to change to reflect max(priority) of
+    # relevant ContainerRequests.
+    if locked_by_uuid_was
+      if locked_by_uuid_was != Thread.current[:api_client_authorization].uuid
+        check_update_whitelist [:priority]
+      end
+    end
+
+    if [Locked, Running].include? self.state
+      # If the Container was already locked, locked_by_uuid must not
+      # changes. Otherwise, the current auth gets the lock.
+      need_lock = locked_by_uuid_was || Thread.current[:api_client_authorization].uuid
+    else
+      need_lock = nil
+    end
+
+    # The caller can provide a new value for locked_by_uuid, but only
+    # if it's exactly what we expect. This allows a caller to perform
+    # an update like {"state":"Unlocked","locked_by_uuid":null}.
+    if self.locked_by_uuid_changed?
+      if self.locked_by_uuid != need_lock
+        return errors.add :locked_by_uuid, "can only change to #{need_lock}"
+      end
+    end
+    self.locked_by_uuid = need_lock
+  end
+
+  def assign_auth
+    if self.auth_uuid_changed?
+      return errors.add :auth_uuid, 'is readonly'
+    end
+    if not [Locked, Running].include? self.state
+      # don't need one
+      self.auth.andand.update_attributes(expires_at: db_current_time)
+      self.auth = nil
+      return
+    elsif self.auth
+      # already have one
+      return
+    end
+    cr = ContainerRequest.
+      where('container_uuid=? and priority>0', self.uuid).
+      order('priority desc').
+      first
+    if !cr
+      return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
+    end
+    self.auth = ApiClientAuthorization.
+      create!(user_id: User.find_by_uuid(cr.modified_by_user_uuid).id,
+              api_client_id: 0)
+  end
+
+  def sort_serialized_attrs
+    if self.environment_changed?
+      self.environment = self.class.deep_sort_hash(self.environment)
+    end
+    if self.mounts_changed?
+      self.mounts = self.class.deep_sort_hash(self.mounts)
+    end
+    if self.runtime_constraints_changed?
+      self.runtime_constraints = self.class.deep_sort_hash(self.runtime_constraints)
+    end
+  end
+
   def handle_completed
     # This container is finished so finalize any associated container requests
     # that are associated with this container.
@@ -155,13 +309,13 @@ class Container < ArvadosModel
       act_as_system_user do
         # Notify container requests associated with this container
         ContainerRequest.where(container_uuid: uuid,
-                               :state => ContainerRequest::Committed).each do |cr|
+                               state: ContainerRequest::Committed).each do |cr|
           cr.container_completed!
         end
 
         # Try to cancel any outstanding container requests made by this container.
         ContainerRequest.where(requesting_container_uuid: uuid,
-                               :state => ContainerRequest::Committed).each do |cr|
+                               state: ContainerRequest::Committed).each do |cr|
           cr.priority = 0
           cr.save
         end