Merge branch '8840-lock-job-record' closes #8840
[arvados.git] / services / api / app / models / job.rb
index 6e01de9213b1d8329451602ff3ea8fdc996e5679..0ed53535778335d11b2b12d3007058e9ad76adfc 100644 (file)
@@ -2,6 +2,7 @@ class Job < ArvadosModel
   include HasUuid
   include KindAndEtag
   include CommonApiTemplate
+  serialize :components, Hash
   attr_protected :arvados_sdk_version, :docker_image_locator
   serialize :script_parameters, Hash
   serialize :runtime_constraints, Hash
@@ -11,10 +12,12 @@ class Job < ArvadosModel
   before_validation :set_priority
   before_validation :update_state_from_old_state_attrs
   validate :ensure_script_version_is_commit
-  validate :find_arvados_sdk_version
   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
@@ -42,7 +45,6 @@ class Job < ArvadosModel
     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
@@ -51,6 +53,7 @@ class Job < ArvadosModel
     t.add :queue_position
     t.add :node_uuids
     t.add :description
+    t.add :components
   end
 
   # Supported states for a job
@@ -63,7 +66,7 @@ class Job < ArvadosModel
            ]
 
   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
@@ -77,12 +80,13 @@ class Job < ArvadosModel
   end
 
   def queue_position
-    Job::queue.each_with_index do |job, index|
-      if job[:uuid] == self.uuid
-        return index
-      end
-    end
-    nil
+    # We used to report this accurately, but the implementation made queue
+    # API requests O(n**2) for the size of the queue.  See #8800.
+    # We've soft-disabled it because it's not clear we even want this
+    # functionality: now that we have Node Manager with support for multiple
+    # node sizes, "queue position" tells you very little about when a job will
+    # run.
+    state == Queued ? 0 : nil
   end
 
   def self.running
@@ -91,8 +95,7 @@ class Job < ArvadosModel
   end
 
   def lock locked_by_uuid
-    transaction do
-      self.reload
+    with_lock do
       unless self.state == Queued and self.is_locked_by_uuid.nil?
         raise AlreadyLockedError
       end
@@ -124,21 +127,43 @@ class Job < ArvadosModel
   end
 
   def ensure_script_version_is_commit
-    if self.state == Running
+    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
 
@@ -169,17 +194,23 @@ class Job < ArvadosModel
   def find_arvados_sdk_version
     resolve_runtime_constraint("arvados_sdk_version",
                                :arvados_sdk_version) do |git_search|
-      commits = Commit.find_commit_range(current_user, "arvados",
+      commits = Commit.find_commit_range("arvados",
                                          nil, git_search, nil)
-      if commits.andand.any?
-        [true, commits.first]
-      else
+      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']
@@ -191,24 +222,6 @@ class Job < ArvadosModel
     end
   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
-      end
-    end
-    deps.keys
-  end
-
   def permission_to_update
     if is_locked_by_uuid_was and !(current_user and
                                    (current_user.uuid == is_locked_by_uuid_was or
@@ -227,7 +240,8 @@ class Job < ArvadosModel
           output_changed? or
           log_changed? or
           tasks_summary_changed? or
-          state_changed?
+          state_changed? or
+          components_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
@@ -255,7 +269,7 @@ 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
@@ -281,11 +295,11 @@ class Job < ArvadosModel
 
     case state
     when Running
-      self.started_at ||= Time.now
+      self.started_at ||= db_current_time
     when Failed, Complete
-      self.finished_at ||= Time.now
+      self.finished_at ||= db_current_time
     when Cancelled
-      self.cancelled_at ||= Time.now
+      self.cancelled_at ||= db_current_time
     end
 
     # TODO: Remove the following case block when old "success" and
@@ -308,6 +322,8 @@ class Job < ArvadosModel
     end
     self.running ||= false # Default to false instead of nil.
 
+    @need_crunch_dispatch_trigger = true
+
     true
   end
 
@@ -372,4 +388,34 @@ class Job < ArvadosModel
     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