Bug fixes for 2.0.0a2

simvue/client.py

@@ -990,17 +990,23 @@ def get_alerts(
         RuntimeError
             if there was a failure retrieving data from the server
         """
-
         if not run_id:
+            if critical_only:
+                raise RuntimeError(
+                    "critical_only is ambiguous when returning alerts with no run ID specified."
+                )
             return [alert.name if names_only else alert for _, alert in Alert.get()]  # type: ignore
 
-        return [
-            alert.get("name")
-            if names_only
-            else Alert(identifier=alert.get("id"), **alert)
+        _alerts = [
+            Alert(identifier=alert.get("id"), **alert)
             for alert in Run(identifier=run_id).get_alert_details()
-            if not critical_only or alert["status"].get("current") == "critical"
-        ]  # type: ignore
+        ]
+
+        return [
+            alert.name if names_only else alert
+            for alert in _alerts
+            if not critical_only or alert.get_status(run_id) == "critical"
+        ]
 
     @prettify_pydantic
     @pydantic.validate_call

simvue/executor.py

@@ -348,9 +348,11 @@ def _update_alerts(self) -> None:
                 if self._runner._dispatcher:
                     self._runner._dispatcher.purge()
 
-                self._runner.log_alert(self._alert_ids[proc_id], "critical")
+                self._runner.log_alert(
+                    identifier=self._alert_ids[proc_id], state="critical"
+                )
             else:
-                self._runner.log_alert(self._alert_ids[proc_id], "ok")
+                self._runner.log_alert(identifier=self._alert_ids[proc_id], state="ok")
 
             _current_time: float = 0
             while (

simvue/metrics.py

@@ -36,14 +36,18 @@ def get_process_memory(processes: list[psutil.Process]) -> int:
     return rss
 
 
-def get_process_cpu(processes: list[psutil.Process]) -> int:
+def get_process_cpu(
+    processes: list[psutil.Process], interval: float | None = None
+) -> int:
     """
     Get the CPU usage
+
+    If first time being called, use a small interval to collect initial CPU metrics.
     """
     cpu_percent: int = 0
     for process in processes:
         with contextlib.suppress(Exception):
-            cpu_percent += process.cpu_percent()
+            cpu_percent += process.cpu_percent(interval=interval)
 
     return cpu_percent
 

simvue/run.py

@@ -308,14 +308,20 @@ def processes(self) -> list[psutil.Process]:
     def _get_sysinfo(self) -> dict[str, typing.Any]:
         """Retrieve system administration
 
+        Parameters
+        ----------
+        interval : float | None
+            The interval to use for collection of CPU metrics, by default None (non blocking)
+
         Returns
         -------
         dict[str, typing.Any]
             retrieved system specifications
         """
-        cpu = get_process_cpu(self.processes)
-        memory = get_process_memory(self.processes)
-        gpu = get_gpu_metrics(self.processes)
+        processes = self.processes
+        cpu = get_process_cpu(processes, interval=0.1)
+        memory = get_process_memory(processes)
+        gpu = get_gpu_metrics(processes)
         data: dict[str, typing.Any] = {}
 
         if memory is not None and cpu is not None:
@@ -351,6 +357,9 @@ def _heartbeat(
             last_heartbeat = time.time()
             last_res_metric_call = time.time()
 
+            if self._resources_metrics_interval:
+                self._add_metrics_to_dispatch(self._get_sysinfo(), join_on_fail=False)
+
             while not heartbeat_trigger.is_set():
                 time.sleep(0.1)
 
@@ -699,6 +708,7 @@ def init(
         self._sv_obj.alerts = []
         self._sv_obj.created = time.time()
         self._sv_obj.notifications = notification
+        self._sv_obj._staging["folder_id"] = self._folder.id
 
         if self._status == "running":
             self._sv_obj.system = get_system()
@@ -931,7 +941,7 @@ def reconnect(self, run_id: str) -> bool:
         self._status = "running"
 
         self._id = run_id
-        self._sv_obj = RunObject(identifier=self._id)
+        self._sv_obj = RunObject(identifier=self._id, _read_only=False)
         self._start(reconnect=True)
 
         return True
@@ -947,6 +957,7 @@ def set_pid(self, pid: int) -> None:
             PID of the process to be monitored
         """
         self._pid = pid
+        self._parent_process = psutil.Process(self._pid)
 
     @skip_if_failed("_aborted", "_suppress_errors", False)
     @pydantic.validate_call
@@ -1602,15 +1613,13 @@ def set_folder_details(
             return False
 
         try:
-            self._folder.read_only(False)
             if metadata:
                 self._folder.metadata = metadata
             if tags:
                 self._folder.tags = tags
             if description:
                 self._folder.description = description
             self._folder.commit()
-            self._folder.read_only(True)
         except (RuntimeError, ValueError, pydantic.ValidationError) as e:
             self._error(f"Failed to update folder '{self._folder.name}' details: {e}")
             return False
@@ -1918,16 +1927,21 @@ def create_user_alert(
     @check_run_initialised
     @pydantic.validate_call
     def log_alert(
-        self, identifier: str, state: typing.Literal["ok", "critical"]
+        self,
+        identifier: str | None = None,
+        name: str | None = None,
+        state: typing.Literal["ok", "critical"] = "critical",
     ) -> bool:
-        """Set the state of an alert
+        """Set the state of an alert - either specify the alert by ID or name.
 
         Parameters
         ----------
-        identifier : str
-            identifier of alert to update
+        identifier : str | None
+            ID of alert to update, by default None
+        name : str | None
+            Name of the alert to update, by default None
         state : Literal['ok', 'critical']
-            state to set alert to
+            state to set alert to, by default 'critical'
 
         Returns
         -------
@@ -1938,13 +1952,33 @@ def log_alert(
             self._error('state must be either "ok" or "critical"')
             return False
 
+        if (identifier and name) or (not identifier and not name):
+            self._error("Please specify alert to update either by ID or by name.")
+            return False
+
+        if name:
+            try:
+                if alerts := Alert.get(offline=self._user_config.run.mode == "offline"):
+                    identifier = next(
+                        (id for id, alert in alerts if alert.name == name), None
+                    )
+                else:
+                    self._error("No existing alerts")
+                    return False
+            except RuntimeError as e:
+                self._error(f"{e.args[0]}")
+                return False
+
+        if not identifier:
+            self._error(f"Alert with name '{name}' could not be found.")
+
         _alert = UserAlert(identifier=identifier)
-        # if not isinstance(_alert, UserAlert):
-        #     self._error(
-        #         f"Cannot update state for alert '{identifier}' "
-        #         f"of type '{_alert.__class__.__name__.lower()}'"
-        #     )
-        #     return False
+        if not isinstance(_alert, UserAlert):
+            self._error(
+                f"Cannot update state for alert '{identifier}' "
+                f"of type '{_alert.__class__.__name__.lower()}'"
+            )
+            return False
         _alert.read_only(False)
         _alert.set_status(run_id=self._id, status=state)
         _alert.commit()