Add nagios compatible check plugin

nagios-plugin/check_borgwrapper

@@ -0,0 +1,147 @@
+#!/usr/bin/env python3
+
+#
+# This plugin is designed as a nagios compatible check plugin to use with
+# Icinga 2 and others
+#
+
+import argparse
+import sys
+import os
+import time
+
+
+def parse_status_file(path):
+    with open(path, 'r') as f:
+        status = dict()
+        timestamp, status_ = f.read().split()
+        status['timestamp'] = float(timestamp)
+        status['status'] = status_
+        return status
+
+def get_backup_status_file(path, name):
+    return os.path.join(path, '%s.backup' % name)
+
+def get_verify_status_file(path, name):
+    return os.path.join(path, '%s.verify' % name)
+
+def get_seconds_readable(seconds):
+    m, s = divmod(seconds, 60)
+    h, m = divmod(m, 60)
+    return '%dh %dm %ds' % (h, m, s)
+
+def get_status_msg(retval):
+    if retval == 0:
+        msg = 'Backup OK'
+    elif retval == 1:
+        msg = 'Backup Warning'
+    elif retval == 2:
+        msg = 'Backup Critical'
+    elif retval == 3:
+        msg = 'Unknown error'
+    else:
+        raise ValueError('Invalid retval %s' % retval)
+
+    return msg
+
+def get_perfdata(label, value, warning, critical, minimum=0):
+    return '%s=%ss;%s;%s;%s' % (label, value, warning, critical, minimum)
+
+def get_backup_status(status, timestamp, max_age_warning, max_age_critical):
+    current_timestamp = time.time()
+    age = current_timestamp - timestamp
+    retval = 3
+    msg = 'Unknown error'
+
+    if max_age_warning > max_age_critical:
+        raise ValueError('Warning threshold is higher than critical threshold')
+
+    if status == 'OK':
+        msg = 'backed up %s ago' % get_seconds_readable(age)
+
+        if timestamp > current_timestamp - max_age_warning:
+            retval = 0
+        elif timestamp > current_timestamp - max_age_critical:
+            retval = 1
+        else:
+            retval = 2
+    elif status == 'FAILED':
+        msg = 'backup failed'
+        retval = 2
+    else:
+        raise RuntimeError('Unknown status')
+
+    perf = get_perfdata('backup_age', int(age), max_age_warning, max_age_critical)
+    return (retval, perf, msg)
+
+
+def get_verify_status(status, timestamp, max_age_warning, max_age_critical):
+    current_timestamp = time.time()
+    age = current_timestamp - timestamp
+    retval = 3
+
+    if max_age_warning > max_age_critical:
+        raise ValueError('Warning threshold is higher than critical threshold')
+
+    if status == 'OK':
+        msg = 'verified %s ago' % get_seconds_readable(age)
+
+        if timestamp > current_timestamp - max_age_warning:
+            retval = 0
+        elif timestamp > current_timestamp - max_age_critical:
+            retval = 1
+        else:
+            retval = 2
+    elif status == 'FAILED':
+        msg = 'verification failed'
+        retval = 2
+    else:
+        raise RuntimeError('Unknown status')
+
+    perf = get_perfdata('verify_age', int(age), max_age_warning, max_age_critical)
+    return (retval, perf, msg)
+
+def main():
+    parser = argparse.ArgumentParser('Check borgwrapper backup status')
+    parser.add_argument('--backup-age-warning', type=int, default=172800)    # 48 hours
+    parser.add_argument('--backup-age-critical', type=int, default=259200)   # 72 hours
+    parser.add_argument('--verify-age-warning', type=int, default=5270400)   # 2 months
+    parser.add_argument('--verify-age-critical', type=int, default=7948800)  # 3 months
+    parser.add_argument('--status-dir', default='/var/lib/borgwrapper/status')
+    parser.add_argument('--config-name', default='config')
+    args = parser.parse_args()
+
+    statuses = []
+
+    try:
+        # Get backup status
+        status_file = get_backup_status_file(args.status_dir, args.config_name)
+        if os.path.exists(status_file):
+            status = parse_status_file(status_file)
+            statuses.append(get_backup_status(status['status'], status['timestamp'],
+                            args.backup_age_warning, args.backup_age_critical))
+        else:
+            statuses.append((2, None, 'no backup status found'))
+
+        # Get verification status
+        status_file = get_verify_status_file(args.status_dir, args.config_name)
+        if os.path.exists(status_file):
+            status = parse_status_file(status_file)
+            statuses.append(get_verify_status(status['status'], status['timestamp'],
+                            args.verify_age_warning, args.verify_age_critical))
+        else:
+            statuses.append((1, None, 'no verification status found'))
+
+        retval = max([i[0] for i in statuses])
+        status = get_status_msg(retval)
+
+        msgs = ', '.join([i[2] for i in statuses])
+        perfdata = ' '.join([i[1] for i in statuses if i[1]])
+        print('%s - %s | %s' % (status, msgs, perfdata))
+        return retval
+    except Exception as e:
+        print('Error: %s' % str(e))
+        return 3
+
+if __name__ == '__main__':
+    sys.exit(main())