Merge lp:~hloeung/ubuntu-repository-cache/health-check-logging-improvements into lp:ubuntu-repository-cache

[Haw Loeung]

Log overrall check result and URLs for errors

[🤖 Canonical IS Merge Bot (canonical-is-mergebot)]

This merge proposal is being monitored by mergebot. Change the status to Approved to merge.

[Colin Misare (cmisare)]

LGTM

review: Approve

[🤖 Canonical IS Merge Bot (canonical-is-mergebot)]

Change successfully merged at revision 324

=== modified file 'files/health_check.py'
--- files/health_check.py	2021-03-03 23:01:10 +0000
+++ files/health_check.py	2021-03-03 23:09:02 +0000
@@ -97,6 +97,8 @@
         send_to_influx(render_influx(METRIC_NAME, 'status=DISABLED', 4.04))
         return
 
+    start_total = time.time()
+
     successes = {}
     failures = {}
     for url in CHECK_URLS:
@@ -114,7 +116,7 @@
                 resp.raise_for_status()
         except requests.exceptions.HTTPError as err:
             duration = time.time() - start
-            logging.error('{} ({:.3f}s)'.format(err, duration))
+            logging.error('ERROR: {}: {} ({:.3f}s)'.format(url, err, duration))
             failures[url] = [status_code, duration, err]
         except (
             requests.exceptions.ConnectTimeout,
@@ -122,18 +124,19 @@
             requests.exceptions.Timeout,
         ) as err:
             duration = time.time() - start
-            logging.error('{} ({:.3f}s)'.format(err, duration))
+            logging.error('ERROR: {}: {} ({:.3f}s)'.format(url, err, duration))
             failures[url] = [0, duration, err]
         except Exception as err:
             duration = time.time() - start
             # We don't want to show exactly what the error is, but we
             # do want to log it.
-            logging.critical('{} ({:.3f}s)'.format(err, duration))
+            logging.critical('ERROR: {}: {} ({:.3f}s)'.format(url, err, duration))
             failures[url] = [0, duration, 'Unknown error']
 
     metrics = []
     if failures:
-        print('Status: 503 Service Unavailable')
+        status = '503 Service Unavailable'
+        print('Status: {}'.format(status))
         print('Content-type: text/html\n')
         for url, val in failures.items():
             status_code = val[0]
@@ -143,7 +146,8 @@
             labels = 'url={},status=FAIL,status_code={}'.format(url, status_code)
             metrics.append(render_influx(METRIC_NAME, labels, duration))
     else:
-        print('Status: 200 OK')
+        status = '200 OK'
+        print('Status: {}'.format(status))
         print('Content-type: text/html\n')
 
     # Either way, output all successful checks.
@@ -158,6 +162,9 @@
     # request/connection on sending metrics to influx.
     print('', flush=True)
 
+    duration = time.time() - start_total
+    logging.info('{} ({:.3f}s)'.format(status, duration))
+
     for metric in metrics:
         send_to_influx(metric)