diff --git a/mistral/engine1/commands.py b/mistral/engine1/commands.py
index 7b6baf5d..fe516e45 100644
--- a/mistral/engine1/commands.py
+++ b/mistral/engine1/commands.py
@@ -44,9 +44,13 @@ def _log_execution_state_change(name, from_state, to_state):
 class EngineCommand(object):
     """Engine command interface."""
 
-    @abc.abstractmethod
-    def run(self, exec_db, wf_handler, cause_task_db=None):
-        """Runs the command.
+    def run_local(self, exec_db, wf_handler, cause_task_db=None):
+        """Runs local part of the command.
+
+        "Local" means that the code can be performed within a scope
+        of an opened DB transaction. For example, for all commands
+        that simply change a state of execution (e.g. depending on
+        some conditions) it's enough to implement only this method.
 
         :param exec_db: Workflow execution DB object.
         :param wf_handler: Workflow handler currently being used.
@@ -54,12 +58,32 @@ class EngineCommand(object):
         :return False if engine should stop further command processing,
             True otherwise.
         """
-        raise NotImplementedError
+        return True
+
+    def run_remote(self, exec_db, wf_handler, cause_task_db=None):
+        """Runs remote part of the command.
+
+        "Remote" means that the code cannot be performed within a scope
+        of an opened DB transaction. All commands that deal with remote
+        invocations should implement this method. However, they may also
+        need to implement "run_local" if they need to do something with
+        DB state of execution and/or tasks.
+
+        :param exec_db: Workflow execution DB object.
+        :param wf_handler: Workflow handler currently being used.
+        :param cause_task_db: Task that caused the command to run.
+        :return False if engine should stop further command processing,
+            True otherwise.
+        """
+        return True
 
 
 class Noop(EngineCommand):
     """No-op command."""
-    def run(self, exec_db, wf_handler, cause_task_db=None):
+    def run_local(self, exec_db, wf_handler, cause_task_db=None):
+        pass
+
+    def run_remote(self, exec_db, wf_handler, cause_task_db=None):
         pass
 
 
@@ -68,12 +92,14 @@ class RunTask(EngineCommand):
         self.task_spec = task_spec
         self.task_db = task_db
 
-    def run(self, exec_db, wf_handler, cause_task_db=None):
+        if task_db:
+            self.exec_db = task_db.execution
+
+    def run_local(self, exec_db, wf_handler, cause_task_db=None):
         LOG.debug('Running workflow task: %s' % self.task_spec)
 
         self._prepare_task(exec_db, wf_handler, cause_task_db)
         self._before_task_start(wf_handler.wf_spec)
-        self._run_task()
 
         return True
 
@@ -82,6 +108,7 @@ class RunTask(EngineCommand):
             return
 
         self.task_db = self._create_db_task(exec_db)
+        self.exec_db = self.task_db.execution
 
         # Evaluate Data Flow properties ('input', 'in_context').
         data_flow.prepare_db_task(
@@ -111,6 +138,11 @@ class RunTask(EngineCommand):
             'project_id': exec_db.project_id
         })
 
+    def run_remote(self, exec_db, wf_handler, cause_task_db=None):
+        self._run_task()
+
+        return True
+
     def _run_task(self):
         # Policies could possibly change task state.
         if self.task_db.state != states.RUNNING:
@@ -130,7 +162,7 @@ class RunTask(EngineCommand):
             self._run_workflow()
 
     def _run_action(self):
-        exec_db = self.task_db.execution
+        exec_db = self.exec_db
         wf_spec = spec_parser.get_workflow_spec(exec_db.wf_spec)
 
         action_spec_name = self.task_spec.get_action_name()
@@ -178,8 +210,10 @@ class RunTask(EngineCommand):
             action_input.update(a_m.get_action_context(self.task_db))
 
         for_each = self.task_spec.get_for_each()
+
         if for_each:
             action_input_collection = self._calc_for_each_input(action_input)
+
             for a_input in action_input_collection:
                 rpc.get_executor_client().run_action(
                     self.task_db.id,
@@ -204,11 +238,13 @@ class RunTask(EngineCommand):
                 targets
             )
 
-    def _calc_for_each_input(self, action_input):
+    @staticmethod
+    def _calc_for_each_input(action_input):
         # In case of for-each iterate over action_input and send
         # each part of data to executor.
         # Calculate action input collection for separating input.
         action_input_collection = []
+
         for key, value in action_input.items():
             for index, item in enumerate(value):
                 iter_context = {key: item}
@@ -221,7 +257,7 @@ class RunTask(EngineCommand):
         return action_input_collection
 
     def _run_workflow(self):
-        parent_exec_db = self.task_db.execution
+        parent_exec_db = self.exec_db
         parent_wf_spec = spec_parser.get_workflow_spec(parent_exec_db.wf_spec)
 
         wf_spec_name = self.task_spec.get_workflow_name()
@@ -251,7 +287,7 @@ class RunTask(EngineCommand):
 
 
 class FailWorkflow(EngineCommand):
-    def run(self, exec_db, wf_handler, cause_task_db=None):
+    def run_local(self, exec_db, wf_handler, cause_task_db=None):
         _log_execution_state_change(
             exec_db.wf_name,
             exec_db.state,
@@ -262,9 +298,12 @@ class FailWorkflow(EngineCommand):
 
         return False
 
+    def run_remote(self, exec_db, wf_handler, cause_task_db=None):
+        return False
+
 
 class SucceedWorkflow(EngineCommand):
-    def run(self, exec_db, wf_handler, cause_task_db=None):
+    def run_local(self, exec_db, wf_handler, cause_task_db=None):
         _log_execution_state_change(
             exec_db.wf_name,
             exec_db.state,
@@ -275,18 +314,32 @@ class SucceedWorkflow(EngineCommand):
 
         return False
 
+    def run_remote(self, exec_db, wf_handler, cause_task_db=None):
+        return False
+
 
 class PauseWorkflow(EngineCommand):
-    def run(self, exec_db, wf_handler, cause_task_db=None):
+    def run_local(self, exec_db, wf_handler, cause_task_db=None):
+        _log_execution_state_change(
+            exec_db.wf_name,
+            exec_db.state,
+            states.PAUSED
+        )
 
         wf_handler.pause_workflow()
 
         return False
 
+    def run_remote(self, exec_db, wf_handler, cause_task_db=None):
+        return False
+
 
 class RollbackWorkflow(EngineCommand):
-    def run(self, exec_db, wf_handler, cause_task_db=None):
-        pass
+    def run_local(self, exec_db, wf_handler, cause_task_db=None):
+        return True
+
+    def run_remote(self, exec_db, wf_handler, cause_task_db=None):
+        return True
 
 
 RESERVED_COMMANDS = {
diff --git a/mistral/engine1/default_engine.py b/mistral/engine1/default_engine.py
index 0d0d1f8f..0391c65e 100644
--- a/mistral/engine1/default_engine.py
+++ b/mistral/engine1/default_engine.py
@@ -69,7 +69,9 @@ class DefaultEngine(base.Engine):
             # Calculate commands to process next.
             cmds = wf_handler.start_workflow(**params)
 
-            self._run_commands(cmds, exec_db, wf_handler)
+            self._run_local_commands(cmds, exec_db, wf_handler)
+
+        self._run_remote_commands(cmds, exec_db, wf_handler)
 
         return exec_db
 
@@ -95,9 +97,16 @@ class DefaultEngine(base.Engine):
             # Calculate commands to process next.
             cmds = wf_handler.on_task_result(task_db, raw_result)
 
-            self._run_commands(cmds, exec_db, wf_handler, task_db)
+            self._run_local_commands(
+                cmds,
+                exec_db,
+                wf_handler,
+                task_db
+            )
 
-            self._check_subworkflow_completion(exec_db)
+        self._run_remote_commands(cmds, exec_db, wf_handler)
+
+        self._check_subworkflow_completion(exec_db)
 
         return task_db
 
@@ -122,7 +131,9 @@ class DefaultEngine(base.Engine):
             # Calculate commands to process next.
             cmds = wf_handler.resume_workflow()
 
-            self._run_commands(cmds, exec_db, wf_handler)
+            self._run_local_commands(cmds, exec_db, wf_handler)
+
+        self._run_remote_commands(cmds, exec_db, wf_handler)
 
         return exec_db
 
@@ -131,13 +142,26 @@ class DefaultEngine(base.Engine):
         raise NotImplementedError
 
     @staticmethod
-    def _run_commands(cmds, exec_db, wf_handler, cause_task_db=None):
+    def _run_local_commands(cmds, exec_db, wf_handler, cause_task_db=None):
         if not cmds:
             return
 
         for cmd in cmds:
-            if not cmd.run(exec_db, wf_handler, cause_task_db):
-                break
+            if not cmd.run_local(exec_db, wf_handler, cause_task_db):
+                return False
+
+        return True
+
+    @staticmethod
+    def _run_remote_commands(cmds, exec_db, wf_handler, cause_task_db=None):
+        if not cmds:
+            return
+
+        for cmd in cmds:
+            if not cmd.run_remote(exec_db, wf_handler, cause_task_db):
+                return False
+
+        return True
 
     @staticmethod
     def _create_db_execution(wf_db, wf_spec, wf_input, params):
@@ -180,7 +204,11 @@ class DefaultEngine(base.Engine):
 
             wf_handler = wfh_factory.create_workflow_handler(exec_db)
 
-            commands.RunTask(task_spec, task_db).run(exec_db, wf_handler)
+            cmd = commands.RunTask(task_spec, task_db)\
+
+            cmd.run_local(exec_db, wf_handler)
+
+        cmd.run_remote(exec_db, wf_handler)
 
     def _check_subworkflow_completion(self, exec_db):
         if not exec_db.parent_task_id:
diff --git a/mistral/services/action_manager.py b/mistral/services/action_manager.py
index d296f81f..24394ff3 100644
--- a/mistral/services/action_manager.py
+++ b/mistral/services/action_manager.py
@@ -28,6 +28,8 @@ from mistral import utils
 from mistral.utils import inspect_utils as i_utils
 
 
+# TODO(rakhmerov): Make methods more consistent and granular.
+
 LOG = logging.getLogger(__name__)
 
 ACTIONS_PATH = '../resources/actions'
@@ -48,11 +50,11 @@ def get_registered_actions(**kwargs):
     return db_api.get_actions(**kwargs)
 
 
-def _register_action_in_db(name, action_class, attributes,
-                           description=None, input_str=None):
+def register_action_class(name, action_class_str, attributes,
+                          description=None, input_str=None):
     values = {
         'name': name,
-        'action_class': action_class,
+        'action_class': action_class_str,
         'attributes': attributes,
         'description': description,
         'input': input_str,
@@ -74,6 +76,7 @@ def _clear_system_action_db():
 
 def sync_db():
     _clear_system_action_db()
+
     register_action_classes()
     register_standard_actions()
 
@@ -90,7 +93,7 @@ def _register_dynamic_action_classes():
         for action in actions:
             attrs = i_utils.get_public_fields(action['class'])
 
-            _register_action_in_db(
+            register_action_class(
                 action['name'],
                 action_class_str,
                 attrs,
@@ -114,9 +117,13 @@ def register_action_classes():
 
             attrs = i_utils.get_public_fields(mgr[name].plugin)
 
-            _register_action_in_db(name, action_class_str, attrs,
-                                   description=description,
-                                   input_str=input_str)
+            register_action_class(
+                name,
+                action_class_str,
+                attrs,
+                description=description,
+                input_str=input_str
+            )
 
         _register_dynamic_action_classes()
 
@@ -134,8 +141,10 @@ def get_action_class(action_full_name):
     action_db = get_action_db(action_full_name)
 
     if action_db:
-        return action_factory.construct_action_class(action_db.action_class,
-                                                     action_db.attributes)
+        return action_factory.construct_action_class(
+            action_db.action_class,
+            action_db.attributes
+        )
 
 
 def get_action_context(task_db):
diff --git a/mistral/tests/unit/engine1/test_dataflow.py b/mistral/tests/unit/engine1/test_dataflow.py
index e93e11f8..b2da8e15 100644
--- a/mistral/tests/unit/engine1/test_dataflow.py
+++ b/mistral/tests/unit/engine1/test_dataflow.py
@@ -21,6 +21,7 @@ from mistral.services import workbooks as wb_service
 from mistral.tests.unit.engine1 import base
 
 LOG = logging.getLogger(__name__)
+
 # Use the set_default method to set value otherwise in certain test cases
 # the change in value is not permanent.
 cfg.CONF.set_default('auth_enable', False, group='pecan')
@@ -28,7 +29,9 @@ cfg.CONF.set_default('auth_enable', False, group='pecan')
 WORKBOOK = """
 ---
 version: '2.0'
+
 name: wb
+
 workflows:
   wf1:
     type: direct
@@ -72,10 +75,33 @@ class DataFlowEngineTest(base.EngineTestCase):
         self.assertEqual(states.SUCCESS, exec_db.state)
 
         tasks = exec_db.tasks
+
+        task1 = self._assert_single_item(tasks, name='task1')
+        task2 = self._assert_single_item(tasks, name='task2')
         task3 = self._assert_single_item(tasks, name='task3')
 
         self.assertEqual(states.SUCCESS, task3.state)
 
+        self.assertDictEqual(
+            {
+                'task': {
+                    'task1': {'hi': 'Hi,'},
+                },
+                'hi': 'Hi,',
+            },
+            task1.output
+        )
+
+        self.assertDictEqual(
+            {
+                'task': {
+                    'task2': {'username': 'Morpheus'},
+                },
+                'username': 'Morpheus',
+            },
+            task2.output
+        )
+
         self.assertDictEqual(
             {
                 'task': {
diff --git a/mistral/tests/unit/engine1/test_direct_workflow.py b/mistral/tests/unit/engine1/test_direct_workflow.py
index f887d931..afa4323b 100644
--- a/mistral/tests/unit/engine1/test_direct_workflow.py
+++ b/mistral/tests/unit/engine1/test_direct_workflow.py
@@ -22,6 +22,7 @@ from mistral.tests.unit.engine1 import base
 # TODO(nmakhotkin) Need to write more tests.
 
 LOG = logging.getLogger(__name__)
+
 # Use the set_default method to set value otherwise in certain test cases
 # the change in value is not permanent.
 cfg.CONF.set_default('auth_enable', False, group='pecan')
@@ -89,6 +90,7 @@ class DirectWorkflowEngineTest(base.EngineTestCase):
         exec_db = db_api.get_execution(exec_db.id)
 
         tasks = exec_db.tasks
+
         task3 = self._assert_single_item(tasks, name='task3')
         task4 = self._assert_single_item(tasks, name='task4')
 
diff --git a/mistral/tests/unit/engine1/test_race_condition.py b/mistral/tests/unit/engine1/test_race_condition.py
new file mode 100644
index 00000000..074e66ed
--- /dev/null
+++ b/mistral/tests/unit/engine1/test_race_condition.py
@@ -0,0 +1,208 @@
+# Copyright 2014 - Mirantis, Inc.
+#
+#    Licensed under the Apache License, Version 2.0 (the "License");
+#    you may not use this file except in compliance with the License.
+#    You may obtain a copy of the License at
+#
+#        http://www.apache.org/licenses/LICENSE-2.0
+#
+#    Unless required by applicable law or agreed to in writing, software
+#    distributed under the License is distributed on an "AS IS" BASIS,
+#    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#    See the License for the specific language governing permissions and
+#    limitations under the License.
+
+from eventlet import corolocal
+from eventlet import semaphore
+from oslo.config import cfg
+
+from mistral.actions import base as action_base
+from mistral.db.v2 import api as db_api
+from mistral.openstack.common import log as logging
+from mistral.services import action_manager as a_m
+from mistral.services import workflows as wf_service
+from mistral.tests.unit.engine1 import base
+from mistral.workflow import states
+
+LOG = logging.getLogger(__name__)
+
+# Use the set_default method to set value otherwise in certain test cases
+# the change in value is not permanent.
+cfg.CONF.set_default('auth_enable', False, group='pecan')
+
+
+WF_LONG_ACTION = """
+---
+version: '2.0'
+
+wf:
+  type: direct
+
+  description: |
+    The idea is to use action that runs longer than engine.start_workflow()
+    method. And we need to check that engine handles this situation.
+
+  output:
+    result: $.result
+
+  tasks:
+    task1:
+      action: std.block
+      publish:
+        result: $
+"""
+
+WF_SHORT_ACTION = """
+---
+version: '2.0'
+
+wf:
+  type: direct
+
+  description: |
+    The idea is to use action that runs faster than engine.start_workflow().
+    And we need to check that engine handles this situation as well. This was
+    a situation previously that led to a race condition in engine, method
+    on_task_result() was called while DB transaction in start_workflow() was
+    still active (not committed yet).
+    To emulate a short action we use a workflow with two start tasks so they
+    run both in parallel on the first engine iteration when we call method
+    start_workflow(). First task has a short action that just returns a
+    predefined result and the second task blocks until the test explicitly
+    unblocks it. So the first action will always end before start_workflow()
+    methods ends.
+
+  output:
+    result: $.result
+
+  tasks:
+    task1:
+      action: std.echo output=1
+      publish:
+        result1: $
+
+    task2:
+      action: std.block
+"""
+
+ACTION_SEMAPHORE = None
+TEST_SEMAPHORE = None
+
+
+class BlockingAction(action_base.Action):
+    def __init__(self):
+        pass
+
+    @staticmethod
+    def unblock_test():
+        TEST_SEMAPHORE.release()
+
+    @staticmethod
+    def wait_for_test():
+        ACTION_SEMAPHORE.acquire()
+
+    def run(self):
+        self.unblock_test()
+        self.wait_for_test()
+
+        print('Action completed [eventlet_id=%s]' % corolocal.get_ident())
+
+        return 'test'
+
+    def test(self):
+        pass
+
+
+class LongActionTest(base.EngineTestCase):
+    def setUp(self):
+        super(LongActionTest, self).setUp()
+
+        global ACTION_SEMAPHORE
+        global TEST_SEMAPHORE
+
+        ACTION_SEMAPHORE = semaphore.Semaphore(1)
+        TEST_SEMAPHORE = semaphore.Semaphore(0)
+
+        a_m.register_action_class(
+            'std.block',
+            '%s.%s' % (BlockingAction.__module__, BlockingAction.__name__),
+            None
+        )
+
+    @staticmethod
+    def block_action():
+        ACTION_SEMAPHORE.acquire()
+
+    @staticmethod
+    def unblock_action():
+        ACTION_SEMAPHORE.release()
+
+    @staticmethod
+    def wait_for_action():
+        TEST_SEMAPHORE.acquire()
+
+    def test_long_action(self):
+        wf_service.create_workflows(WF_LONG_ACTION)
+
+        self.block_action()
+
+        exec_db = self.engine.start_workflow('wf', None)
+
+        exec_db = db_api.get_execution(exec_db.id)
+
+        self.assertEqual(states.RUNNING, exec_db.state)
+        self.assertEqual(states.RUNNING, exec_db.tasks[0].state)
+
+        self.wait_for_action()
+
+        # Here's the point when the action is blocked but already running.
+        # Do the same check again, it should always pass.
+        exec_db = db_api.get_execution(exec_db.id)
+
+        self.assertEqual(states.RUNNING, exec_db.state)
+        self.assertEqual(states.RUNNING, exec_db.tasks[0].state)
+
+        self.unblock_action()
+
+        self._await(lambda: self.is_execution_success(exec_db.id))
+
+        exec_db = db_api.get_execution(exec_db.id)
+
+        self.assertDictEqual({'result': 'test'}, exec_db.output)
+
+    def test_short_action(self):
+        wf_service.create_workflows(WF_SHORT_ACTION)
+
+        self.block_action()
+
+        exec_db = self.engine.start_workflow('wf', None)
+
+        exec_db = db_api.get_execution(exec_db.id)
+
+        self.assertEqual(states.RUNNING, exec_db.state)
+
+        tasks = exec_db.tasks
+
+        task1 = self._assert_single_item(exec_db.tasks, name='task1')
+        task2 = self._assert_single_item(
+            tasks,
+            name='task2',
+            state=states.RUNNING
+        )
+
+        self._await(lambda: self.is_task_success(task1.id))
+
+        self.unblock_action()
+
+        self._await(lambda: self.is_task_success(task2.id))
+        self._await(lambda: self.is_execution_success(exec_db.id))
+
+        task1 = db_api.get_task(task1.id)
+
+        self.assertDictEqual(
+            {
+                'result1': 1,
+                'task': {'task1': {'result1': 1}}
+            },
+            task1.output
+        )
diff --git a/mistral/utils/inspect_utils.py b/mistral/utils/inspect_utils.py
index 59281940..ba5c96dc 100644
--- a/mistral/utils/inspect_utils.py
+++ b/mistral/utils/inspect_utils.py
@@ -22,6 +22,7 @@ def get_public_fields(obj):
                          if not attr.startswith("_")]
 
     public_fields = {}
+
     for attribute_str in public_attributes:
         attr = getattr(obj, attribute_str)
         is_field = not (inspect.isbuiltin(attr)
diff --git a/mistral/workflow/base.py b/mistral/workflow/base.py
index 31d98a5e..73f11632 100644
--- a/mistral/workflow/base.py
+++ b/mistral/workflow/base.py
@@ -129,7 +129,8 @@ class WorkflowHandler(object):
 
         return cmds
 
-    def _determine_task_output(self, task_spec, task_db, raw_result):
+    @staticmethod
+    def _determine_task_output(task_spec, task_db, raw_result):
         for_each = task_spec.get_for_each()
         t_name = task_spec.get_name()
 
@@ -172,7 +173,8 @@ class WorkflowHandler(object):
         else:
             return data_flow.evaluate_task_output(task_spec, raw_result)
 
-    def _determine_task_state(self, task_db, task_spec, raw_result):
+    @staticmethod
+    def _determine_task_state(task_db, task_spec, raw_result):
         state = states.ERROR if raw_result.is_error() else states.SUCCESS
 
         for_each = task_spec.get_for_each()
diff --git a/mistral/workflow/direct_workflow.py b/mistral/workflow/direct_workflow.py
index 10939e28..b18c4d93 100644
--- a/mistral/workflow/direct_workflow.py
+++ b/mistral/workflow/direct_workflow.py
@@ -88,7 +88,7 @@ class DirectWorkflowHandler(base.WorkflowHandler):
         Expression 'on_complete' is not mutually exclusive to 'on_success'
          and 'on_error'.
         :param task_db: Task DB model.
-        :param remove_incomplete_joins: True if incomplete "join"
+        :param remove_unsatisfied_joins: True if incomplete "join"
             tasks must be excluded from the list of commands.
         :return: List of task specifications.
         """