Merge pull request #238 from timj/u/timj/exit-when-all-dead

Stop looping if all workers have died

changelog/45.bugfix

@@ -0,0 +1 @@
+Fix hang when all worker nodes crash and restart limit is reached

testing/acceptance_test.py

@@ -701,6 +701,17 @@ class TestNodeFailure:
             "*2 failed*2 passed*",
         ])
 
+    def test_max_slave_restart_die(self, testdir):
+        f = testdir.makepyfile("""
+            import os
+            os._exit(1)
+        """)
+        res = testdir.runpytest(f, '-n4', '--max-slave-restart=0')
+        res.stdout.fnmatch_lines([
+            "*Unexpectedly no active workers*",
+            "*INTERNALERROR*"
+        ])
+
     def test_disable_restart(self, testdir):
         f = testdir.makepyfile("""
             import os

xdist/dsession.py

@@ -120,6 +120,10 @@ class DSession:
     def loop_once(self):
         """Process one callback from one of the slaves."""
         while 1:
+            if not self._active_nodes:
+                # If everything has died stop looping
+                self.triggershutdown()
+                raise RuntimeError("Unexpectedly no active workers available")
             try:
                 eventcall = self.queue.get(timeout=2.0)
                 break