Bugfixes, add a function to kill patroni daemon, make the feature description more concise.

features/basic_replication.feature

@@ -1,11 +1,5 @@
 Feature: basic replication
-  In order to check that basic replication works
-  As observers
-  We start the primary and the replica
-  add a table on the primary
-  and check that it gets replicated to the replica over time.
-  We stop the primary and check that the replica promotes itself to primary
-  We start the old primary and check that it rejoins as a replica.
+  We should check that the basic bootstrapping, replication and failover works.
 
   Scenario: check replication of a single table
     Given I start postgres0
@@ -14,9 +8,9 @@ Feature: basic replication
     Then table foo is present on postgres1 after 10 seconds
 
   Scenario: check the basic failover
-    When I shut down postgres0
-    Then postgres1 role is the primary after 10 seconds
+    When I kill postgres0
+    Then postgres1 role is the primary after 30 seconds
     When I start postgres0
     Then postgres0 role is the secondary after 10 seconds
     When I add the table bar to postgres1
-    Then table bar is present on postgres1 after 10 seconds
+    Then table bar is present on postgres0 after 10 seconds

features/basic_replication.py

@@ -16,6 +16,18 @@ class BasicReplicationSteps(object):
         '''I start (\w+)'''
         return world.pctl.start_patroni(pg_name)
 
+    def stop_patroni(self, step, pg_name):
+        '''I shut down (\w+)'''
+        return world.pctl.stop_patroni(pg_name)
+
+    def kill_patroni(self, step, pg_name):
+        '''I kill (\w+)'''
+        return world.pctl.stop_patroni(pg_name, kill=True)
+
+    def do_sleep(self, step, sleep_seconds):
+        '''I sleep (\w+)'''
+        sleep(int(sleep_seconds))
+
     def add_table(self, step, table_name, pg_name):
         '''I add the table (\w+) to (\w+)'''
         # parse the configuration file and get the port
@@ -36,7 +48,8 @@ class BasicReplicationSteps(object):
 
     def check_role(self, step, pg_name, pg_role, max_promotion_timeout):
         '''(\w+) role is the (\w+) after (\d+) seconds'''
-        return world.pctl.check_role_has_changed_to(pg_name, pg_role, timeout=int(max_promotion_timeout))
+        if not world.pctl.check_role_has_changed_to(pg_name, pg_role, timeout=int(max_promotion_timeout)):
+            assert False, "pg_name role didn't change to {0} after {1} seconds".format(pg_role, max_promotion_timeout)
 
 
 BasicReplicationSteps(world)

features/terrain.py

@@ -47,9 +47,12 @@ class PatroniController(object):
     def patroni_is_running(self, pg_name):
         return pg_name in self.processes and self.processes[pg_name].pid and (self.processes[pg_name].poll() is None)
 
-    def stop_patroni(self, pg_name):
+    def stop_patroni(self, pg_name, kill=False):
         while self.patroni_is_running(pg_name):
-            self.processes[pg_name].terminate()
+            if not kill:
+                self.processes[pg_name].terminate()
+            else:
+                self.processes[pg_name].kill()
             time.sleep(1)
         self.log.get('pg_name') and self.log[pg_name].close()
         del self.processes[pg_name]
@@ -142,16 +145,17 @@ class PatroniController(object):
 
     def check_role_has_changed_to(self, pg_name, new_role, timeout=10):
         bound_time = time.time() + timeout
-        current_role = 't' if new_role == 'primary' else 'f'
+        recovery_status = False if new_role == 'primary' else True
         role_has_changed = False
         while not role_has_changed:
             cur = self.query(pg_name, "SELECT pg_is_in_recovery()", fail_ok=True)
             if cur:
                 row = cur.fetchone()
-                if row and len(row) > 0 and row[0] != current_role:
+                if row and len(row) > 0 and row[0] == recovery_status:
                     role_has_changed = True
             if time.time() > bound_time:
                 break
+            time.sleep(1)
         return role_has_changed
 
     def stop_all(self):