Merge pull request #4 from a-thomas-22/sigterm-handler

Improve SIGTERM handling

exporter.py

@@ -7,6 +7,7 @@
 import subprocess
 import json
 import datetime
+import signal
 
 from prometheus_client import start_http_server, Gauge
 from psycopg2.extras import DictCursor
@@ -20,6 +21,13 @@
 else:
     logging.basicConfig(format='%(asctime)s - %(message)s', level=logging.INFO)
 
+terminate = False;
+
+def signal_handler(sig, frame):
+    global terminate
+    logging.info('SIGTERM received, preparing to shut down...')
+    terminate = True
+
 # Class definition
 class Exporter():
     def __init__(self):
@@ -207,6 +215,9 @@ def get_archive_status(self):
     logging.info("Startup...")
     logging.info('My PID is: %s', os.getpid())
 
+    # Register the signal handler for SIGTERM
+    signal.signal(signal.SIGTERM, signal_handler)
+
     logging.info("Reading environment configuration")
 
     # Read the configuration
@@ -232,6 +243,11 @@ def get_archive_status(self):
     # Check if this is a primary instance
     # with while True and try catch this is how reconnect already should work.
     while True:
+
+        if terminate:
+            logging.info("Received SIGTERM, shutting down...")
+            break
+
         try:
             with psycopg2.connect(
                 host = pg_host,
@@ -283,6 +299,9 @@ def get_archive_status(self):
                         raise Exception(
                             "Unable to execute SELECT NOT pg_is_in_recovery()" + str(e))
         except Exception as e:
+            if terminate:
+                logging.info("Received SIGTERM during exception, shutting down...")
+                break
             logging.error(
                 "Error occured, retrying in 60sec..." + str(e))
             time.sleep(60)

scripts/entrypoint.sh

@@ -1,4 +1,4 @@
 #!/bin/bash
 
 sleep 5
-envdir /run/etc/wal-e.d/env wal-g-prometheus-exporter
+exec envdir /run/etc/wal-e.d/env wal-g-prometheus-exporter