Merge branch 'fix-stuck-forked-project-import' into 'master'

Fix bug where projects would appear to be stuck in the forked import state

A race condition existed between when Rails committed the `import_status` to
`started` and when the Sidekiq worker forked a project. If this fork were quick,
it's possible that the worker would attempt to move into the `finished` state
before the `started` state took effect.

As mentioned in https://github.com/mperham/sidekiq/wiki/Problems-and-Troubleshooting#cannot-find-modelname-with-id12345,
we can either delay the worker to ensure the DB has a chance to update, or use the nice `after_commit_queue` gem to schedule the task after the state machine commit. See:

* https://github.com/pluginaweek/state_machine/issues/191 
* https://github.com/shellycloud/after_commit_queue


Closes #2736

See merge request !1434

1 files changed, 7 insertions, 2 deletions

diff --git a/app/models/project.rb b/app/models/project.rb
index 1a5c1c978c9..57c5f7369d7 100644
--- a/app/models/project.rb
+++ b/app/models/project.rb
@@ -39,6 +39,7 @@ class Project < ActiveRecord::Base
   include Gitlab::VisibilityLevel
   include Referable
   include Sortable
+  include AfterCommitQueue
 
   extend Gitlab::ConfigHelper
   extend Enumerize
@@ -191,7 +192,7 @@ class Project < ActiveRecord::Base
     state :finished
     state :failed
 
-    after_transition any => :started, do: :add_import_job
+    after_transition any => :started, do: :schedule_add_import_job
     after_transition any => :finished, do: :clear_import_data
   end
 
@@ -275,13 +276,17 @@ class Project < ActiveRecord::Base
     id && persisted?
   end
 
+  def schedule_add_import_job
+    run_after_commit(:add_import_job)
+  end
+
   def add_import_job
     if forked?
       unless RepositoryForkWorker.perform_async(id, forked_from_project.path_with_namespace, self.namespace.path)
         import_fail
       end
     else
-      RepositoryImportWorker.perform_in(2.seconds, id)
+      RepositoryImportWorker.perform_async(id)
     end
   end