add cli to list/clear failures and fix if-alerted alerting

src/sysalert/db.py

@@ -2,7 +2,7 @@ import datetime
 import os
 import sqlite3
 
-def register_success(name, db):
+def clear_failure(name, db):
     cur = db.cursor()
     cur.execute('''
             DELETE FROM alert
@@ -86,6 +86,20 @@ def get_failures(name, db):
     failures.sort(key=lambda x: x['timestamp'])
     return failures
 
+def get_services(db):
+    cur = db.cursor()
+    cur.execute('''SELECT
+        service.name, COUNT(*)
+        FROM service
+            JOIN failure ON
+                failure.service=service.id
+        GROUP BY
+            service.name;
+        ''')
+    db.commit()
+    ret = {name: failures for (name, failures) in cur.fetchall()}
+    return ret
+
 
 def init(path):
     con = sqlite3.connect(path, detect_types=sqlite3.PARSE_DECLTYPES | sqlite3.PARSE_COLNAMES)

src/sysalert/util.py

@@ -1,4 +1,4 @@
-
+import argparse
 import configparser
 import datetime
 import importlib
@@ -37,11 +37,12 @@ def register_exit(config, db):
     if os.environ['MONITOR_SERVICE_RESULT'] == 'success':
         # exit with success status
         failures = sysalert.db.get_failures(service_name, db)
-        sysalert.db.register_success(service_name, db)
+        sysalert.db.clear_failure(service_name, db)
         try:
             do_alert = config.getboolean(section_name, 'recovery_alert')
         except ValueError:
-            if config.get(section_name, 'recovery_alert') == 'if-alerted' and failures:
+            if config.get(section_name, 'recovery_alert') == 'if-alerted' and \
+                    any([f['alert_method'] for f in failures]):
                 do_alert = True
             else:
                 do_alert = False
@@ -75,6 +76,27 @@ def register_exit(config, db):
     alert.failure(service_name, failures, alert_config)
     return 0
 
+def cli_args():
+    parser = argparse.ArgumentParser(
+            description='Manage active sysalerts',
+            epilog='''Note that when called from systemd this argument parser
+is not used. Use the provided sysalert-* services and see
+documentation for details.''')
+    parser.add_argument('--service', '-s',
+                        help='Only operate on specific service')
+    subparsers = parser.add_subparsers()
+
+    parser_list = subparsers.add_parser('list')
+    parser_list.set_defaults(cmd='list')
+
+    parser_clear = subparsers.add_parser('clear')
+    parser_clear.set_defaults(cmd='clear')
+    parser_clear.add_argument('--force', '-f',
+                              action='store_true',
+                              help='''Force clear for all services if service is
+                              not specified''')
+
+    return parser.parse_args()
 
 def cli():
     config = configparser.ConfigParser()
@@ -85,9 +107,37 @@ def cli():
     if _test_env():
         # invoked by systemd
         ret = register_exit(config, db)
+    else:
+        args = cli_args()
+        if args.service and not args.service.endswith('.service'):
+            args.service = f'{args.service}.service'
+
+        if args.cmd == 'list':
+            cli_list(args, db)
+        elif args.cmd == 'clear':
+            cli_clear(args, db)
 
     sysalert.db.close(db)
     return ret
 
+def cli_list(args, db):
+    for service, failures in sysalert.db.get_services(db).items():
+        if args.service and args.service == service:
+            print(f'{service}: {failures} failures')
+    return 0
+
+def cli_clear(args, db):
+    if not args.service and not args.force:
+        print(f'No service specified, use "--force" to clear all services', sys.stderr)
+        return 1
+    if args.service:
+        sysalert.db.clear_failure(args.service, db)
+        print(f'Cleared failures for {args.service}')
+    else:
+        for service, failures in sysalert.db.get_services(db).items():
+            sysalert.db.clear_failure(service, db)
+        print(f'Cleared all failures')
+    return 0
+
 if __name__ == '__main__':
   sys.exit(cli())