From 631d428708839aa067c27be352374ce8ed0bb974 Mon Sep 17 00:00:00 2001
From: Lars Michelsen <lm@checkmk.com>
Date: Thu, 30 Jan 2025 20:02:49 +0100
Subject: [PATCH] Improve job-scheduler start and stop logging

Change-Id: I23268da616a39cd0e7566ead54781280e2d16a14
---
 cmk/gui/job_scheduler/_background_jobs/_server.py | 10 ++++++++--
 cmk/gui/job_scheduler/_scheduler.py               |  4 ++++
 cmk/gui/job_scheduler/main.py                     | 10 ++++++++++
 3 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/cmk/gui/job_scheduler/_background_jobs/_server.py b/cmk/gui/job_scheduler/_background_jobs/_server.py
index 1e58817ecb9..034b9723a15 100644
--- a/cmk/gui/job_scheduler/_background_jobs/_server.py
+++ b/cmk/gui/job_scheduler/_background_jobs/_server.py
@@ -3,14 +3,20 @@
 # This file is part of Checkmk (https://checkmk.com). It is subject to the terms and
 # conditions defined in the file COPYING, which is part of this source code package.
 
+from logging import Logger
+
 import gunicorn.app.base  # type: ignore[import-untyped]
 from fastapi import FastAPI
 
 from ._config import ServerConfig
 
 
-def run_server(config: ServerConfig, app: FastAPI) -> None:
-    _ApplicationServer(app, config).run()
+def run_server(config: ServerConfig, app: FastAPI, logger: Logger) -> None:
+    logger.info("Starting background job server")
+    try:
+        _ApplicationServer(app, config).run()
+    finally:
+        logger.info("Stopped background job server")
 
 
 class _ApplicationServer(gunicorn.app.base.BaseApplication):  # type: ignore[misc] # pylint: disable=abstract-method
diff --git a/cmk/gui/job_scheduler/_scheduler.py b/cmk/gui/job_scheduler/_scheduler.py
index d21e6f6c08e..d025514c94f 100644
--- a/cmk/gui/job_scheduler/_scheduler.py
+++ b/cmk/gui/job_scheduler/_scheduler.py
@@ -26,6 +26,7 @@
 def run_scheduler_threaded(
     crash_report_callback: Callable[[Exception], str], stop_event: threading.Event
 ) -> threading.Thread:
+    logger.info("Starting scheduler thread")
     t = threading.Thread(
         target=_run_scheduler,
         args=(crash_report_callback, stop_event),
@@ -39,6 +40,7 @@ def _run_scheduler(
     crash_report_callback: Callable[[Exception], str], stop_event: threading.Event
 ) -> None:
     job_threads: dict[str, threading.Thread] = {}
+    logger.info("Started scheduler")
     while not stop_event.is_set():
         try:
             cycle_start = time.time()
@@ -59,7 +61,9 @@ def _run_scheduler(
             # in case there were some locks left over
             store.release_all_locks()
 
+    logger.info("Waiting for jobs to finish")
     _wait_for_job_threads(job_threads)
+    logger.info("Stopped scheduler")
 
 
 def _load_last_job_runs() -> dict[str, datetime.datetime]:
diff --git a/cmk/gui/job_scheduler/main.py b/cmk/gui/job_scheduler/main.py
index 8825c9aaae2..7f9e2bc6d15 100644
--- a/cmk/gui/job_scheduler/main.py
+++ b/cmk/gui/job_scheduler/main.py
@@ -71,11 +71,13 @@ def main(crash_report_callback: Callable[[Exception], str]) -> int:
 
         _setup_console_logging()
 
+        # This is only an intermediate handler until gunicorn run_server sets its own handler
         signal.signal(signal.SIGTERM, lambda signum, frame: sys.exit(0))
 
         daemonize()
 
         _setup_file_logging(log_path / "ui-job-scheduler.log")
+        logger.info("--- Starting ui-job-scheduler (Checkmk %s) ---", cmk_version.__version__)
 
         with pid_file_lock(_pid_file(omd_root)):
             init_span_processor(
@@ -113,10 +115,17 @@ def main(crash_report_callback: Callable[[Exception], str]) -> int:
                         registered_jobs=dict(job_registry.items()),
                         executor=ThreadedJobExecutor(logger),
                     ),
+                    logger,
                 )
+            except SystemExit as exc:
+                logger.info("Process terminated (Exit code: %d)", exc.code)
+                raise
             finally:
+                logger.info("Stopping application")
                 stop_event.set()
                 scheduler_thread.join()
+    except SystemExit:
+        raise
     except Exception as exc:
         crash_msg = crash_report_callback(exc)
         logger.error("Unhandled exception (Crash ID: %s)", crash_msg, exc_info=True)
@@ -136,5 +145,6 @@ def _setup_file_logging(log_file: Path) -> None:
         logging.Formatter("%(asctime)s [%(levelno)s] [%(process)d/%(threadName)s] %(message)s")
     )
     logger = logging.getLogger()
+    del logger.handlers[:]  # Remove all previously existing handlers
     logger.addHandler(handler)
     logger.setLevel(logging.INFO)