Nudge kernel with info request until we receive IOPub messages

Author: SylvainCorlay

notebook/services/kernels/handlers.py

@@ -127,6 +127,64 @@ def create_stream(self):
             meth = getattr(km, 'connect_' + channel)
             self.channels[channel] = stream = meth(self.kernel_id, identity=identity)
             stream.channel = channel
+        
+        shell_channel = self.channels['shell']
+        iopub_channel = self.channels['iopub']
+        
+        future = Future()
+        info_future = Future()
+        iopub_future = Future()
+
+        def finish():
+            """Common cleanup"""
+            loop.remove_timeout(timeout)
+            loop.remove_timeout(nudge_handle)
+            iopub_channel.stop_on_recv()
+            shell_channel.stop_on_recv()
+
+        def on_shell_reply(msg):
+            if not info_future.done():
+                self.log.debug("Nudge: shell info reply received: %s", self.kernel_id)
+                shell_channel.stop_on_recv()
+                self.log.debug("Nudge: resolving shell future")
+                info_future.set_result(msg)
+                if iopub_future.done():
+                    finish()
+                    self.log.debug("Nudge: resolving main future in shell handler")
+                    future.set_result(info_future.result())
+
+        def on_iopub(msg):
+            if not iopub_future.done():
+                self.log.debug("Nudge: first IOPub received: %s", self.kernel_id)
+                iopub_channel.stop_on_recv()
+                self.log.debug("Nudge: resolving iopub future")
+                iopub_future.set_result(None)
+                if info_future.done():
+                    finish()
+                    self.log.debug("Nudge: resolving main future in iopub handler")
+                    future.set_result(info_future.result())
+
+        def on_timeout():
+            self.log.warning("Nudge: Timeout waiting for kernel_info_reply: %s", self.kernel_id)
+            finish()
+            if not future.done():
+                future.set_exception(TimeoutError("Timeout waiting for nudge"))
+
+        iopub_channel.on_recv(on_iopub)
+        shell_channel.on_recv(on_shell_reply)
+        loop = IOLoop.current()
+
+        # Nudge the kernel with kernel info requests until we get an IOPub message
+        def nudge():
+            self.log.debug("Nudge")
+            if not future.done():
+                self.log.debug("nudging")
+                self.session.send(shell_channel, "kernel_info_request")
+                nudge_handle = loop.call_later(0.5, nudge)
+        nudge_handle = loop.call_later(0, nudge)
+
+        timeout = loop.add_timeout(loop.time() + self.kernel_info_timeout, on_timeout)
+        return future
 
     def request_kernel_info(self):
         """send a request for kernel_info"""
@@ -253,6 +311,7 @@ def _register_session(self):
             yield stale_handler.close()
         self._open_sessions[self.session_key] = self
 
+    @gen.coroutine
     def open(self, kernel_id):
         super().open()
         km = self.kernel_manager
@@ -269,9 +328,11 @@ def open(self, kernel_id):
                 for channel, msg_list in replay_buffer:
                     stream = self.channels[channel]
                     self._on_zmq_reply(stream, msg_list)
+            connected = Future()
+            connected.set_result(None)
         else:
             try:
-                self.create_stream()
+                connected = self.create_stream()
             except web.HTTPError as e:
                 self.log.error("Error opening stream: %s", e)
                 # WebSockets don't response to traditional error codes so we
@@ -285,8 +346,13 @@ def open(self, kernel_id):
         km.add_restart_callback(self.kernel_id, self.on_kernel_restarted)
         km.add_restart_callback(self.kernel_id, self.on_restart_failed, 'dead')
 
-        for channel, stream in self.channels.items():
-            stream.on_recv_stream(self._on_zmq_reply)
+        def subscribe(value):
+            for channel, stream in self.channels.items():
+                stream.on_recv_stream(self._on_zmq_reply)
+
+        connected.add_done_callback(subscribe)
+
+        return connected
 
     def on_message(self, msg):
         if not self.channels:

notebook/services/kernels/kernelmanager.py

@@ -310,27 +310,52 @@ async def restart_kernel(self, kernel_id, now=False):
         await maybe_future(self.pinned_superclass.restart_kernel(self, kernel_id, now=now))
         kernel = self.get_kernel(kernel_id)
         # return a Future that will resolve when the kernel has successfully restarted
-        channel = kernel.connect_shell()
+        shell_channel = kernel.connect_shell()  # TODO: do not use transient channels
+        iopub_channel = kernel.connect_iopub()
+        
         future = Future()
+        info_future = Future()
+        iopub_future = Future()
 
         def finish():
-            """Common cleanup when restart finishes/fails for any reason."""
-            if not channel.closed():
-                channel.close()
+            """Common cleanup"""
             loop.remove_timeout(timeout)
+            loop.remove_timeout(nudge_handle)
+            # iopub_channel.stop_on_recv()  # TODO: do not use transient channels
+            # shell_channel.stop_on_recv()
+            if not iopub_channel.closed():	
+                iopub_channel.close()
+            if not shell_channel.closed():	
+                shell_channel.close()
             kernel.remove_restart_callback(on_restart_failed, 'dead')
 
-        def on_reply(msg):
-            self.log.debug("Kernel info reply received: %s", kernel_id)
-            finish()
-            if not future.done():
-                future.set_result(msg)
+        def on_shell_reply(msg):
+            if not info_future.done():
+                self.log.debug("Nudge: shell info reply received: %s", kernel_id)
+                shell_channel.stop_on_recv()
+                self.log.debug("Nudge: resolving shell future")
+                info_future.set_result(msg)
+                if iopub_future.done():
+                    finish()
+                    self.log.debug("Nudge: resolving main future in shell handler")
+                    future.set_result(info_future.result())
+
+        def on_iopub(msg):
+            if not iopub_future.done():
+                self.log.debug("Nudge: first IOPub received: %s", kernel_id)
+                iopub_channel.stop_on_recv()
+                self.log.debug("Nudge: resolving iopub future")
+                iopub_future.set_result(None)
+                if info_future.done():
+                    finish()
+                    self.log.debug("Nudge: resolving main future in iopub handler")
+                    future.set_result(info_future.result())
 
         def on_timeout():
-            self.log.warning("Timeout waiting for kernel_info_reply: %s", kernel_id)
+            self.log.warning("Nudge: Timeout waiting for kernel_info_reply: %s", kernel_id)
             finish()
             if not future.done():
-                future.set_exception(TimeoutError("Timeout waiting for restart"))
+                future.set_exception(TimeoutError("Timeout waiting for nudge"))
 
         def on_restart_failed():
             self.log.warning("Restarting kernel failed: %s", kernel_id)
@@ -339,10 +364,21 @@ def on_restart_failed():
                 future.set_exception(RuntimeError("Restart failed"))
 
         kernel.add_restart_callback(on_restart_failed, 'dead')
-        kernel.session.send(channel, "kernel_info_request")
-        channel.on_recv(on_reply)
+
+        iopub_channel.on_recv(on_iopub)
+        shell_channel.on_recv(on_shell_reply)
         loop = IOLoop.current()
+
+        # Nudge the kernel with kernel info requests until we get an IOPub message
+        def nudge():
+            self.log.debug("Nudge")
+            if not future.done():
+                self.log.debug("nudging")
+                kernel.session.send(shell_channel, "kernel_info_request")
+                nudge_handle = loop.call_later(0.5, nudge)
+
         timeout = loop.add_timeout(loop.time() + self.kernel_info_timeout, on_timeout)
+        nudge_handle = loop.call_later(0, nudge)
         return future
 
     def notify_connect(self, kernel_id):