Merge branch '37-unexplained-mirror-timeout-causing-migrations-to-stall' into 'develop'

Resolve "Unexplained mirror timeout causing migrations to stall"

Closes #37

See merge request open-source/flexnbd-c!59

debian/changelog

@@ -2,6 +2,8 @@ flexnbd (0.4.0) UNRELEASED; urgency=medium
 
   * Explicitly close the server control socket, and wait for it to close, to
     prevent deadlocks during the server clean-up process (#40 !58)
+  * Ensure mirroring can be restarted after a break command is sent to the
+    source (#37, !59)
 
  -- Patrick J Cherry <patrick@bytemark.co.uk>  Fri, 07 Dec 2018 16:38:56 +0000
 

src/server/mirror.c

@@ -671,6 +671,7 @@ static void mirror_abandon_cb(struct ev_loop *loop, ev_io * w, int revents)
     debug("Abandon message received");
     mirror_set_state(ctrl->mirror, MS_ABANDONED);
     self_pipe_signal_clear(ctrl->mirror->abandon_signal);
+    ev_io_stop(loop, &ctrl->abandon_watcher);
     ev_break(loop, EVBREAK_ONE);
     return;
 }

tests/acceptance/test_write_during_migration.rb

@@ -86,6 +86,14 @@ class TestWriteDuringMigration < Test::Unit::TestCase
     end
   end
 
+  def stop_mirror
+    UNIXSocket.open(@source_sock) do |sock|
+      sock.write("break\x0A\x0A")
+      sock.flush
+      sock.readline
+    end
+  end
+
   def wait_for_quit
     Timeout.timeout(10) do
       Process.waitpid2(@dst_proc)
@@ -177,7 +185,6 @@ class TestWriteDuringMigration < Test::Unit::TestCase
     end
   end
 
-
   def test_status_call_after_cleanup
     Dir.mktmpdir do |tmpdir|
       Dir.chdir(tmpdir) do
@@ -195,4 +202,28 @@ class TestWriteDuringMigration < Test::Unit::TestCase
       end
     end
   end
+
+  def test_mirroring_can_be_restarted
+    @size = 100 * 1024 * 1024 # 100MB
+    Dir.mktmpdir do |tmpdir|
+      Dir.chdir(tmpdir) do
+        make_files
+
+        launch_servers
+
+        # This is a bit racy.  It needs to be slow enough that the migration
+        # isn't finished before the stop runs, and slow enough so that we can
+        # stop/start a few times.
+        3.times do
+          start_mirror
+          sleep 0.1
+          stop_mirror
+          sleep 0.1
+        end
+        start_mirror
+
+        wait_for_quit
+      end
+    end
+  end
 end