THRIFT-3544 Let cross test fail when server process died unexpectedly

Client: Test
Patch: Nobuaki Sukegawa

This closes #793

test/crossrunner/report.py

@@ -139,6 +139,11 @@ class ExecReporter(TestReporter):
       self._lock.release()
 
   def killed(self):
+    self.out.write('Process is killed.\n')
+    self.end(None)
+
+  def died(self):
+    self.out.write('Process is died unexpectedly.\n')
     self.end(None)
 
   _init_failure_exprs = {

test/crossrunner/run.py

@@ -47,6 +47,7 @@ class ExecutionContext(object):
     self.env = env
     self.timer = None
     self.expired = False
+    self.killed = False
 
   def _expire(self):
     self._log.info('Timeout')
@@ -55,6 +56,7 @@ class ExecutionContext(object):
 
   def kill(self):
     self._log.debug('Killing process : %d' % self.proc.pid)
+    self.killed = True
     if platform.system() != 'Windows':
       try:
         os.killpg(self.proc.pid, signal.SIGKILL)
@@ -64,7 +66,6 @@ class ExecutionContext(object):
       self.proc.kill()
     except Exception as err:
       self._log.info('Failed to kill process : %s' % str(err))
-    self.report.killed()
 
   def _popen_args(self):
     args = {
@@ -97,7 +98,12 @@ class ExecutionContext(object):
   def _scoped(self):
     yield self
     self._log.debug('Killing scoped process')
-    self.kill()
+    if self.proc.poll() is None:
+      self.kill()
+      self.report.killed()
+    else:
+      self._log.debug('Process died unexpectedly')
+      self.report.died()
 
   def wait(self):
     self.proc.communicate()
@@ -147,6 +153,8 @@ def run_test(testdir, logdir, test_dict, async=True, max_retry=3):
             if not cl.report.maybe_false_positive() or cl_retry_count >= cl_max_retry:
               if cl_retry_count > 0 and cl_retry_count < cl_max_retry:
                 logger.warn('[%s]: Connected after %d retry (%.2f sec each)' % (test.server.name, cl_retry_count, cl_retry_wait))
+              # Wait for 50 ms to see if server does not die at the end.
+              time.sleep(0.05)
               break
             logger.debug('Server may not be ready, waiting %.2f second...' % cl_retry_wait)
             time.sleep(cl_retry_wait)
@@ -154,7 +162,13 @@ def run_test(testdir, logdir, test_dict, async=True, max_retry=3):
 
       if not sv.report.maybe_false_positive() or retry_count >= max_retry:
         logger.debug('Finish')
-        return RESULT_TIMEOUT if cl.expired else cl.proc.returncode
+        if cl.expired:
+          return RESULT_TIMEOUT
+        elif not sv.killed and cl.proc.returncode == 0:
+          # Server should be alive at the end.
+          return RESULT_ERROR
+        else:
+          return cl.proc.returncode
       logger.warn('[%s]: Detected socket bind failure, retrying...' % test.server.name)
       retry_count += 1
   except (KeyboardInterrupt, SystemExit):
@@ -163,11 +177,10 @@ def run_test(testdir, logdir, test_dict, async=True, max_retry=3):
       raise
     stop.set()
     return None
-  except Exception as ex:
-    logger.warn('%s', ex)
+  except:
     if not async:
       raise
-    logger.debug('Error executing [%s]', test.name, exc_info=sys.exc_info())
+    logger.warn('Error executing [%s]', test.name, exc_info=sys.exc_info())
     return RESULT_ERROR
 
 

test/features/known_failures_Linux.json

@@ -1,6 +1,8 @@
 [
   "c_glib-limit_container_length_binary_buffered-ip",
   "c_glib-limit_string_length_binary_buffered-ip",
+  "cpp-theader_framed_binary_header_buffered-ip",
+  "cpp-theader_unframed_binary_header_buffered-ip",
   "csharp-limit_container_length_binary_buffered-ip",
   "csharp-limit_container_length_compact_buffered-ip",
   "csharp-limit_string_length_binary_buffered-ip",

test/known_failures_Linux.json

@@ -1,12 +1,29 @@
 [
+  "cpp-cpp_binary_buffered-ip-ssl",
+  "cpp-cpp_binary_framed-ip-ssl",
   "cpp-cpp_binary_http-domain",
   "cpp-cpp_binary_http-ip",
+  "cpp-cpp_binary_http-ip-ssl",
+  "cpp-cpp_compact_buffered-ip-ssl",
+  "cpp-cpp_compact_framed-ip-ssl",
   "cpp-cpp_compact_http-domain",
   "cpp-cpp_compact_http-ip",
+  "cpp-cpp_compact_http-ip-ssl",
+  "cpp-cpp_header_buffered-domain",
+  "cpp-cpp_header_buffered-ip",
+  "cpp-cpp_header_buffered-ip-ssl",
+  "cpp-cpp_header_framed-domain",
+  "cpp-cpp_header_framed-ip",
+  "cpp-cpp_header_framed-ip-ssl",
   "cpp-cpp_header_http-domain",
   "cpp-cpp_header_http-ip",
+  "cpp-cpp_header_http-ip-ssl",
+  "cpp-cpp_json_buffered-ip-ssl",
+  "cpp-cpp_json_framed-ip",
+  "cpp-cpp_json_framed-ip-ssl",
   "cpp-cpp_json_http-domain",
   "cpp-cpp_json_http-ip",
+  "cpp-cpp_json_http-ip-ssl",
   "cpp-dart_binary_http-ip",
   "cpp-dart_json_http-ip",
   "cpp-java_binary_http-ip",
@@ -17,6 +34,8 @@
   "cpp-java_json_http-ip-ssl",
   "cpp-perl_binary_buffered-ip-ssl",
   "cpp-perl_binary_framed-ip-ssl",
+  "cpp-py_binary-accel_framed-ip-ssl",
+  "cpp-rb_binary-accel_framed-ip",
   "csharp-cpp_binary_buffered-ip-ssl",
   "csharp-cpp_binary_framed-ip-ssl",
   "csharp-cpp_compact_buffered-ip-ssl",
@@ -71,6 +90,7 @@
   "erl-rb_compact_framed-ip",
   "go-dart_binary_framed-ip",
   "go-dart_json_framed-ip",
+  "go-nodejs_json_framed-ip",
   "go-perl_binary_buffered-ip-ssl",
   "go-perl_binary_framed-ip-ssl",
   "hs-csharp_binary_framed-ip",
@@ -90,6 +110,7 @@
   "perl-perl_binary_buffered-ip-ssl",
   "perl-perl_binary_framed-ip-ssl",
   "perl-php_binary_framed-ip",
+  "py-cpp_compact_buffered-ip",
   "py-perl_accel-binary_buffered-ip-ssl",
   "py-perl_accel-binary_framed-ip-ssl",
   "py-perl_binary_buffered-ip-ssl",