distbuild: Add distbuild start and cancel functionality

Add command for distbuild-start to build_plugin in morphlib,
and create a boolean parameter to inform the initiator whether
to disconnect the controller and leave the build running remotely.

Add distbuild-cancel command to parse currently-running distbuild
build-request IDs and cancel the one matching the given argument

Change-Id: I458a5767bb768ceb2b4d8876adf1c86075d452bd

8 files changed, 198 insertions, 11 deletions

diff --git a/distbuild/__init__.py b/distbuild/__init__.py
index 271b5def..b9b5fbc5 100644
--- a/distbuild/__init__.py
+++ b/distbuild/__init__.py
@@ -54,7 +54,8 @@ from build_controller import (BuildController, BuildFailed, BuildProgress,
                               BuildOutput, BuildStepFinished, BuildStepFailed,
                               BuildFinished, BuildCancel,
                               build_step_name, map_build_graph)
-from initiator import (Initiator, InitiatorListJobs)
+from initiator import (Initiator, InitiatorStart, InitiatorCancel,
+                       InitiatorListJobs)
 from protocol import message
 
 from crashpoint import (crash_point, add_crash_condition, add_crash_conditions,
diff --git a/distbuild/build_controller.py b/distbuild/build_controller.py
index 3971fe68..27a6c862 100644
--- a/distbuild/build_controller.py
+++ b/distbuild/build_controller.py
@@ -175,6 +175,7 @@ class BuildController(distbuild.StateMachine):
         self._helper_id = None
         self.debug_transitions = False
         self.debug_graph_state = False
+        self.allow_detach = build_request_message['allow_detach']
 
     def __repr__(self):
         return '<BuildController at 0x%x, request-id %s>' % (id(self),
@@ -497,10 +498,17 @@ class BuildController(distbuild.StateMachine):
         logging.debug("BuildController %r: initiator id %s disconnected",
             self, event.id)
 
-        cancel_pending = distbuild.WorkerCancelPending(event.id)
-        self.mainloop.queue_event(distbuild.WorkerBuildQueuer, cancel_pending)
+        if self.allow_detach:
+            logging.debug('Detaching from client; build continuing remotely.')
+        else:
+            self.mainloop.queue_event(self, build_cancel(event.id))
+
+    def build_cancel(self, event_id):
+        cancel_pending = distbuild.WorkerCancelPending(event_id)
+        self.mainloop.queue_event(distbuild.WorkerBuildQueuer,
+                                  cancel_pending)
 
-        cancel = BuildCancel(event.id)
+        cancel = BuildCancel(event_id)
         self.mainloop.queue_event(BuildController, cancel)
 
         self.mainloop.queue_event(self, _Abort())
diff --git a/distbuild/initiator.py b/distbuild/initiator.py
index b53f4e20..391c8707 100644
--- a/distbuild/initiator.py
+++ b/distbuild/initiator.py
@@ -69,6 +69,7 @@ class Initiator(distbuild.StateMachine):
             self._partial = True
         self._step_outputs = {}
         self.debug_transitions = False
+        self.allow_detach = False
 
         if app.settings['initiator-step-output-dir'] == '':
             self._step_output_dir = create_build_directory()
@@ -107,7 +108,8 @@ class Initiator(distbuild.StateMachine):
             original_ref=self._original_ref,
             component_names=self._component_names,
             partial=self._partial,
-            protocol_version=distbuild.protocol.VERSION
+            protocol_version=distbuild.protocol.VERSION,
+            allow_detach=self.allow_detach,
         )
         self._jm.send(msg)
         logging.debug('Initiator: sent to controller: %s', repr(msg))
@@ -127,7 +129,7 @@ class Initiator(distbuild.StateMachine):
             'step-finished': self._handle_step_finished_message,
             'step-failed': self._handle_step_failed_message,
         }
-        
+
         handler = handlers[event.msg['type']]
         handler(event.msg)
 
@@ -170,6 +172,12 @@ class Initiator(distbuild.StateMachine):
         self._write_status_to_build_log(self._get_output(msg), status)
 
     def _handle_step_started_message(self, msg):
+        if self.allow_detach:
+            self._app.status(msg='Detaching distbuild from controller (build'
+                                 ' will continue on the distbuild network)')
+            self.mainloop.queue_event(self._cm, distbuild.StopConnecting())
+            self._jm.close()
+            self.allow_detach = False
         status = 'Started building %s on %s' % (
             msg['step_name'], msg['worker_name'])
         self._app.status(msg=status)
@@ -216,7 +224,7 @@ class Initiator(distbuild.StateMachine):
         self.mainloop.queue_event(self._cm, distbuild.StopConnecting())
         self._jm.close()
         logging.info('Build finished OK')
-        
+
         urls = event.msg['urls']
         if urls:
             for url in urls:
@@ -249,6 +257,72 @@ class Initiator(distbuild.StateMachine):
         self._step_outputs = {}
 
 
+class InitiatorStart(Initiator):
+
+    def __init__(self, cm, conn, app, repo_name, ref, morphology,
+                 original_ref, component_names):
+        super(InitiatorStart, self).__init__(cm, conn, app, repo_name, ref,
+                                             morphology, original_ref,
+                                             component_names)
+        self._step_outputs = {}
+        self.debug_transitions = False
+        self.allow_detach = True
+
+
+class InitiatorCancel(distbuild.StateMachine):
+
+    def __init__(self, cm, conn, app, job_id):
+        distbuild.StateMachine.__init__(self, 'waiting')
+        self._cm = cm
+        self._conn = conn
+        self._app = app
+        self._job_id = job_id
+
+    def setup(self):
+        distbuild.crash_point()
+
+        self._jm = distbuild.JsonMachine(self._conn)
+        self.mainloop.add_state_machine(self._jm)
+        logging.debug('initiator: _jm=%s' % repr(self._jm))
+
+        spec = [
+            # state, source, event_class, new_state, callback
+            ('waiting', self._jm, distbuild.JsonEof, None, self._terminate),
+            ('waiting', self._jm, distbuild.JsonNewMessage, None,
+                self._handle_json_message),
+        ]
+        self.add_transitions(spec)
+
+        self._app.status(msg='Sending cancel request for distbuild job.')
+        msg = distbuild.message('build-cancel',
+            id=self._job_id,
+            protocol_version=distbuild.protocol.VERSION,
+        )
+        self._jm.send(msg)
+        logging.debug('Initiator: sent to controller: %s', repr(msg))
+
+    def _handle_json_message(self, event_source, event):
+        distbuild.crash_point()
+
+        logging.debug('Initiator: from controller: %s', str(event.msg))
+
+        handlers = {
+            'request-output': self._handle_request_output,
+        }
+
+        handler = handlers[event.msg['type']]
+        handler(event.msg)
+
+    def _handle_request_output(self, msg):
+        self._app.status(msg=str(msg['message']))
+        self.mainloop.queue_event(self._cm, distbuild.StopConnecting())
+        self._jm.close()
+
+    def _terminate(self, event_source, event):
+        self.mainloop.queue_event(self._cm, distbuild.StopConnecting())
+        self._jm.close()
+
+
 class InitiatorListJobs(distbuild.StateMachine):
 
     def __init__(self, cm, conn, app):
@@ -277,7 +351,7 @@ class InitiatorListJobs(distbuild.StateMachine):
         self._app.status(msg='Requesting currently running distbuilds.')
         msg = distbuild.message('list-requests',
             id=msg_uuid,
-            protocol_version=distbuild.protocol.VERSION
+            protocol_version=distbuild.protocol.VERSION,
         )
         self._jm.send(msg)
         logging.debug('Initiator: sent to controller: %s', repr(msg))
diff --git a/distbuild/initiator_connection.py b/distbuild/initiator_connection.py
index fdb1dab5..601d8c5c 100644
--- a/distbuild/initiator_connection.py
+++ b/distbuild/initiator_connection.py
@@ -117,6 +117,8 @@ class InitiatorConnection(distbuild.StateMachine):
                 self._handle_build_request(event)
             elif event.msg['type'] == 'list-requests':
                 self._handle_list_requests(event)
+            elif event.msg['type'] == 'build-cancel':
+                self._handle_build_cancel(event)
             else:
                 logging.error('Invalid message type: %s', event.msg)
         except (KeyError, ValueError) as ex:
@@ -151,6 +153,26 @@ class InitiatorConnection(distbuild.StateMachine):
                                 message=('\n\n'.join(output_msg)))
         self.jm.send(msg)
 
+    def _handle_build_cancel(self, event):
+        requests = self.mainloop.state_machines_of_type(
+                   distbuild.BuildController)
+        for build in requests:
+            if build.get_request()['id'] == event.msg['id']:
+                self.mainloop.queue_event(InitiatorConnection,
+                                          InitiatorDisconnect(event.msg['id']))
+                self.mainloop.queue_event(distbuild.BuildController,
+                                          build.build_cancel(event.msg['id']))
+                msg = distbuild.message('request-output', message=(
+                                        'Cancelling build request with ID %s' %
+                                        event.msg['id']))
+                self.jm.send(msg)
+                break
+        else:
+            msg = distbuild.message('request-output', message=('Given '
+                                    'build-request ID does not match any '
+                                    'running build IDs.'))
+            self.jm.send(msg)
+
     def _disconnect(self, event_source, event):
         for id in self.our_ids:
             logging.debug('InitiatorConnection: %s: InitiatorDisconnect(%s)',
diff --git a/distbuild/protocol.py b/distbuild/protocol.py
index 2fbfa085..1924925d 100644
--- a/distbuild/protocol.py
+++ b/distbuild/protocol.py
@@ -33,6 +33,7 @@ _required_fields = {
         'morphology',
         'partial',
         'protocol_version',
+        'allow_detach',
     ],
     'build-progress': [
         'id',
@@ -92,6 +93,10 @@ _required_fields = {
     'request-output': [
         'message',
     ],
+    'build-cancel': [
+        'id',
+        'protocol_version',
+    ],
 }
 
 
diff --git a/morphlib/buildcommand.py b/morphlib/buildcommand.py
index fd5acdf5..f34b898c 100644
--- a/morphlib/buildcommand.py
+++ b/morphlib/buildcommand.py
@@ -519,10 +519,11 @@ class InitiatorBuildCommand(BuildCommand):
     RECONNECT_INTERVAL = 30 # seconds
     MAX_RETRIES = 1
 
-    def __init__(self, app, addr, port):
+    def __init__(self, app, addr, port, allow_detach):
         self.app = app
         self.addr = addr
         self.port = port
+        self.allow_detach = allow_detach
         self.app.settings['push-build-branches'] = True
         super(InitiatorBuildCommand, self).__init__(app)
 
@@ -540,10 +541,15 @@ class InitiatorBuildCommand(BuildCommand):
         loop = distbuild.MainLoop()
         args = [repo_name, ref, filename, original_ref or ref,
                 component_names]
+        if self.allow_detach:
+            initiator_type = distbuild.InitiatorStart
+        else:
+            initiator_type = distbuild.Initiator
+
         cm = distbuild.InitiatorConnectionMachine(self.app,
                                                   self.addr,
                                                   self.port,
-                                                  distbuild.Initiator,
+                                                  initiator_type,
                                                   [self.app] + args,
                                                   self.RECONNECT_INTERVAL,
                                                   self.MAX_RETRIES)
diff --git a/morphlib/plugins/build_plugin.py b/morphlib/plugins/build_plugin.py
index e5b35853..8c94b928 100644
--- a/morphlib/plugins/build_plugin.py
+++ b/morphlib/plugins/build_plugin.py
@@ -46,10 +46,14 @@ class BuildPlugin(cliapp.Plugin):
                                              '[COMPONENT...]')
         self.app.add_subcommand('distbuild', self.distbuild,
                                 arg_synopsis='SYSTEM [COMPONENT...]')
+        self.app.add_subcommand('distbuild-start', self.distbuild_start,
+                                arg_synopsis='SYSTEM [COMPONENT...]')
         self.use_distbuild = False
+        self.allow_detach = False
 
     def disable(self):
         self.use_distbuild = False
+        self.allow_detach = False
 
     def distbuild_morphology(self, args):
         '''Distbuild a system, outside of a system branch.
@@ -97,6 +101,12 @@ class BuildPlugin(cliapp.Plugin):
         your system, the system artifact will be copied from your trove
         and cached locally.
 
+        Log information can be found in the current working directory, in
+        directories called build-xx.
+
+        If you do not have a persistent connection to the server on which
+        the distbuild runs, consider using `morph distbuild-start` instead.
+
         Example:
 
             morph distbuild devel-system-x86_64-generic.morph
@@ -106,6 +116,25 @@ class BuildPlugin(cliapp.Plugin):
         self.use_distbuild = True
         self.build(args)
 
+    def distbuild_start(self, args):
+        '''Distbuild a system image without a lasting client-server connection.
+
+        This command launches a distributed build, and disconnects from the
+        distbuild cluster once the build starts, leaving the build running
+        remotely.
+
+        The command will return a build-ID which can be used to cancel the
+        distbuild via `morph distbuild-cancel`. Builds started in this manner
+        can be found via `morph distbuild-list-jobs`
+
+        See `morph help distbuild` for more information and example usage.
+
+        '''
+
+        self.use_distbuild = True
+        self.allow_detach = True
+        self.build(args)
+
     def build_morphology(self, args):
         '''Build a system, outside of a system branch.
 
@@ -210,7 +239,7 @@ class BuildPlugin(cliapp.Plugin):
             port = self.app.settings['controller-initiator-port']
 
             build_command = morphlib.buildcommand.InitiatorBuildCommand(
-                self.app, addr, port)
+                self.app, addr, port, self.allow_detach)
         else:
             build_command = morphlib.buildcommand.BuildCommand(self.app)
 
diff --git a/morphlib/plugins/distbuild_plugin.py b/morphlib/plugins/distbuild_plugin.py
index 8aaead10..d93c2214 100644
--- a/morphlib/plugins/distbuild_plugin.py
+++ b/morphlib/plugins/distbuild_plugin.py
@@ -40,6 +40,48 @@ class DistbuildOptionsPlugin(cliapp.Plugin):
         pass
 
 
+class DistbuildCancel(cliapp.Plugin):
+
+    RECONNECT_INTERVAL = 30 # seconds
+    MAX_RETRIES = 1
+
+    def enable(self):
+        self.app.add_subcommand('distbuild-cancel', self.distbuild_cancel,
+                                arg_synopsis='ID')
+
+    def disable(self):
+        pass
+
+    def distbuild_cancel(self, args):
+        '''Cancels a currently-running distbuild
+
+        Command line arguments:
+
+        `ID` of the running process that you wish to cancel
+        (this can be found via distbuild-list-jobs)
+
+        Example:
+
+            * morph distbuild-cancel InitiatorConnection-1
+
+        '''
+
+        if len(args) == 0:
+            raise cliapp.AppException(
+                'This command requires a build-request ID to run.')
+
+        addr = self.app.settings['controller-initiator-address']
+        port = self.app.settings['controller-initiator-port']
+        icm = distbuild.InitiatorConnectionMachine(self.app, addr, port,
+                                                   distbuild.InitiatorCancel,
+                                                   [self.app] + args,
+                                                   self.RECONNECT_INTERVAL,
+                                                   self.MAX_RETRIES)
+        loop = distbuild.MainLoop()
+        loop.add_state_machine(icm)
+        loop.run()
+
+
 class DistbuildListJobsPlugin(cliapp.Plugin):
 
     RECONNECT_INTERVAL = 30 # seconds