add metric for kernel restarts

jupyter_server/prometheus/metrics.py

@@ -5,7 +5,16 @@
 conventions for metrics & labels.
 """
 
+from prometheus_client import Counter
+
 try:
+    import notebook  # type: ignore
+
+    if notebook.__name__ != "notebook":
+        # avoid double-importing myself if nbclassic is shimming jupyter_server into notebook,
+        # in which case notebook.__name__ will be 'jupyter_server'
+        _msg = "Not importing jupyter_server metrics under two names"
+        raise ImportError(_msg)
     # Jupyter Notebook also defines these metrics.  Re-defining them results in a ValueError.
     # Try to de-duplicate by using the ones in Notebook if available.
     # See https://github.com/jupyter/jupyter_server/issues/209
@@ -34,3 +43,9 @@
         "counter for how many kernels are running labeled by type",
         ["type"],
     )
+
+KERNEL_RESTARTS = Counter(
+    "jupyter_kernel_restarts",
+    "counter for how many kernel restarts, labeled by kernel_name and source (user or restarter)",
+    ["kernel_name", "source"],
+)

jupyter_server/services/kernels/handlers.py

@@ -16,6 +16,7 @@
 from tornado import web
 
 from jupyter_server.auth import authorized
+from jupyter_server.prometheus.metrics import KERNEL_RESTARTS
 from jupyter_server.utils import url_escape, url_path_join
 
 from ...base.handlers import APIHandler
@@ -104,6 +105,7 @@ async def post(self, kernel_id, action):
                 self.set_status(500)
             else:
                 model = await ensure_async(km.kernel_model(kernel_id))
+                KERNEL_RESTARTS.labels(kernel_name=model["name"], source="user").inc()
                 self.write(json.dumps(model, default=json_default))
         self.finish()
 

jupyter_server/services/kernels/kernelmanager.py

@@ -38,7 +38,7 @@
 )
 
 from jupyter_server._tz import isoformat, utcnow
-from jupyter_server.prometheus.metrics import KERNEL_CURRENTLY_RUNNING_TOTAL
+from jupyter_server.prometheus.metrics import KERNEL_CURRENTLY_RUNNING_TOTAL, KERNEL_RESTARTS
 from jupyter_server.utils import ApiPath, import_item, to_os_path
 
 
@@ -179,6 +179,10 @@
     # Methods for managing kernels and sessions
     # -------------------------------------------------------------------------
 
+    def _handle_kernel_restart(self, kernel_id, kernel_name):
+        """notice that a kernel restarted"""
+        KERNEL_RESTARTS.labels(kernel_name=kernel_name, source="restarter").inc()
+
     def _handle_kernel_died(self, kernel_id):
         """notice that a kernel died"""
         self.log.warning("Kernel %s died, removing from map.", kernel_id)
@@ -279,6 +283,11 @@
             lambda: self._handle_kernel_died(kernel_id),
             "dead",
         )
+        # register callback to count restarts
+        self.add_restart_callback(
+            kernel_id,
+            lambda: self._handle_kernel_restart(kernel_id, km.kernel_name),
+        )
 
     def ports_changed(self, kernel_id):
         """Used by ZMQChannelsHandler to determine how to coordinate nudge and replays.