Redis connection failure in notify_push can abort unrelated operations and leave user deletion partially completed

Summary

When notify_push is installed but Redis is unavailable or not configured, notify_push attempts to connect to 127.0.0.1:6379 during app/command initialization. The resulting RedisException: Connection refused can abort unrelated Nextcloud operations.

In my case, deleting a user from the admin UI returned an Internal Server Error. Afterward, the user no longer existed according to Nextcloud (user not found), but stale/ghost entries were still visible in the Accounts admin UI. This suggests the user deletion process was interrupted after only part of the cleanup had completed.

Removing notify_push stopped the Redis connection errors.

Environment

• Nextcloud: 33.0.8.2
• Deployment: Kubernetes / k3s
• Database: PostgreSQL
• memcache.local: \OC\Memcache\APCu
• No redis, memcache.distributed, or memcache.locking configuration in the Nextcloud system config
• Redis was not running on 127.0.0.1:6379

Steps to reproduce

1. Run Nextcloud without Redis configured/running.
2. Install/enable notify_push.
3. Attempt to delete a user from the Nextcloud admin Accounts page.
4. The request fails with an Internal Server Error caused by RedisException: Connection refused.
5. The deletion may be left partially completed: occ user:info <uid> reports that the user is not found, while the Accounts UI can still show a stale entry for that UID.

The Redis error can also be reproduced by running an unrelated occ command such as:

php /var/www/html/occ config:list system

It prints Connection refused before the normal command output.

Relevant stack trace

RedisException: Connection refused

#0 /var/www/html/lib/private/RedisFactory.php(104):
   Redis->pconnect('127.0.0.1', 6379, ...)
#1 /var/www/html/lib/private/RedisFactory.php(148):
   OC\RedisFactory->create()
#2 /var/www/html/custom_apps/notify_push/lib/Queue/PushRedisFactory.php(38):
   OC\RedisFactory->getInstance()
#3 /var/www/html/custom_apps/notify_push/lib/AppInfo/Application.php(46):
   OCA\NotifyPush\Queue\PushRedisFactory->getRedis()
#4 ...
#15 /var/www/html/lib/private/Console/Application.php(209)
#16 /var/www/html/lib/private/Console/Application.php(101)
#17 /var/www/html/console.php(86)
#18 /var/www/html/occ(33)

The connection parameters shown in the original log were:

127.0.0.1:6379

Expected behavior

If Redis is unavailable or notify_push is not fully configured, notify_push should fail gracefully and report its own configuration problem without aborting unrelated Nextcloud operations.

In particular, a Redis connection failure in notify_push should not be able to interrupt a destructive core operation such as user deletion after that operation has already partially modified state.

It would also be preferable if merely loading occ commands did not establish a Redis connection that can throw an exception before the requested command runs.

Actual behavior

notify_push initializes its Redis-backed queue while Nextcloud is loading the app/commands. If Redis refuses the connection, the exception propagates into unrelated requests/commands. In the user-deletion case this appears capable of leaving the deletion only partially completed.

Additional note

After removing notify_push, the RedisException: Connection refused errors stopped.

Start with custom_apps/notify_push/lib/Queue/PushRedisFactory.php and lib/AppInfo/Application.php, then trace the RedisFactory calls in lib/private/RedisFactory.php and Console/Application.php. Reproduce the failure with Redis unavailable by running occ config:list system and attempting user deletion. Done means notify_push reports or handles its configuration problem without aborting unrelated commands or leaving deletion partially completed.