fix: run the frappe scheduler outside lite and worker-pool benches

docs/production.md

@@ -38,6 +38,7 @@ Supported managers are `systemd` and `supervisor`.
 A new bench deploys one bench process plus admin and the two redis servers, because
 `[lite_mode] enabled` is the default. Turn lite mode off and the set becomes web,
 socketio, admin, workers, and redis - see [Lite Mode](configuration.md#lite-mode).
+Supervisor also runs a `schedule` process; systemd's worker pool runs the scheduler itself.
 
 Each workload unit sets `LimitNOFILE=65535`. A systemd user unit gets 1024
 descriptors by default, which is too few for a lite-mode bench process and for

pilot/managers/processes/definitions.py

@@ -53,6 +53,7 @@ def prod_process_definitions(self) -> list[ProcessDefinition]:
                 self.socketio_definition(),
                 self.admin_definition(),
                 *worker_defs,
+                self.schedule_definition(),
             ]
         defs.append(self.redis_definition("redis_cache", "redis_cache.conf"))
         defs.append(self.redis_definition("redis_queue", "redis_queue.conf"))
@@ -218,6 +219,16 @@ def worker_definitions(self, queue: str, count: int) -> list[ProcessDefinition]:
             for i in range(1, count + 1)
         ]
 
+    def schedule_definition(self) -> ProcessDefinition:
+        """Plain `frappe worker` has no scheduler; lite and worker-pool run their own."""
+        return ProcessDefinition(
+            name="schedule",
+            argv=[str(self.python), "-m", "frappe.utils.bench_helper", "frappe", "schedule"],
+            log_file=self.bench.logs_path / "schedule.log",
+            env=self.python_env(),
+            working_dir=self.bench.sites_path,
+        )
+
     def redis_definition(self, name: str, config_filename: str) -> ProcessDefinition:
         from pilot.managers.redis import redis_server_binary
 

tests/pilot/core/test_core.py

@@ -699,12 +699,12 @@ def test_bench_init_apps_comes_from_config(tmp_path: Path) -> None:
 def test_process_definitions_returns_correct_count(tmp_path: Path) -> None:
     bench = make_bench(tmp_path)
     # workers: default=2, short=1, long=1 => 4 worker processes
-    # plus web, socketio, redis_cache, redis_queue = 4
+    # plus web, socketio, schedule, redis_cache, redis_queue = 5
     # plus admin, watch (on by default in dev) = 2
-    # total = 10
+    # total = 11
     process_manager = ProcessManager(bench)
     definitions = process_manager._process_definitions()
-    assert len(definitions) == 10
+    assert len(definitions) == 11
     assert "watch" in [pd.name for pd in definitions]
     assert "admin-ui" not in [pd.name for pd in definitions]
 
@@ -713,15 +713,15 @@ def test_process_definitions_watch_admin_js_adds_vite_ui(tmp_path: Path) -> None
     bench = make_bench(tmp_path)
     definitions = ProcessManager(bench, watch_admin_js=True)._process_definitions()
     assert "admin-ui" in [pd.name for pd in definitions]
-    assert len(definitions) == 11
+    assert len(definitions) == 12
 
 
 def test_process_definitions_can_disable_app_watch(tmp_path: Path) -> None:
     bench = make_bench(tmp_path)
     bench.config.watch_apps_js = False
     definitions = ProcessManager(bench)._process_definitions()
     assert "watch" not in [pd.name for pd in definitions]
-    assert len(definitions) == 9
+    assert len(definitions) == 10
 
 
 def test_run_processes_survives_noncritical_exit(tmp_path: Path) -> None:
@@ -750,6 +750,22 @@ def test_watch_definition_is_noncritical_frappe_watch(tmp_path: Path) -> None:
     assert all(pd.critical for pd in definitions if pd.name != "watch")
 
 
+def test_process_definitions_run_frappe_scheduler(tmp_path: Path) -> None:
+    bench = make_bench(tmp_path)
+    definitions = ProcessManager(bench)._process_definitions()
+    schedule = next(pd for pd in definitions if pd.name == "schedule")
+    assert "frappe schedule" in shlex.join(schedule.argv)
+    assert schedule.working_dir == bench.sites_path
+
+
+def test_systemd_definitions_leave_scheduling_to_worker_pool(tmp_path: Path) -> None:
+    bench = make_bench(tmp_path)
+    bench.config.production.process_manager = "systemd"
+    names = [pd.name for pd in ProcessManager(bench)._prod_process_definitions()]
+    assert "worker_pool" in names
+    assert "schedule" not in names
+
+
 def test_process_definitions_worker_names_are_numbered(tmp_path: Path) -> None:
     bench = make_bench(tmp_path)
     process_manager = ProcessManager(bench)