Multiple weird effects after upgrading to NC32? Try replacing APCu!

ℹ️ Support
1

Setup and Config

I run 6 NC instances bare metal on Arch using NGINX/PHP-FPM, MariaDB and Valkey. I upgraded to 32.0.0.13 using the downloaded archive, did upgrade and maintenance. My config.php of each cloud yielded

‘memcache.local’ => ‘\OC\Memcache\APCu’,
‘memcache.distributed’ => ‘\OC\Memcache\Redis’,
‘memcache.locking’ => ‘\OC\Memcache\Redis’,

as proposed by the memory caching page of the admin manual for the last NC versions – and that worked perfectly fine, up until now.

Issue summary

After upgrading to NC32 I started having weird glitches, like text or colors from one cloud in another. And strange log errors.

Steps to replicate (difficult)

  1. Have MULTIPLE NC32 instances use ONE APCu cache

  2. Use them a while (glitches only occur randomly)

Possible fix

After spending a whole day debugging all components, services, configs and pulling hair, my son had the idea that cache collisions in APCu could be the culprit. So I just replaced APCu with Redis in the above config snippet. Et voilà, everything seems to be working correctly again.

Nextcloud 32 "Seite nicht gefunden" bei Apps
2

Hey, we just discussed this today and did fix(cache): Ensure unique global prefix per instanceid by SystemKeeper · Pull Request #55383 · nextcloud/server · GitHub

Maybe you can try if it solves the issue?

3

I reverted my “APCu → Redis” workaround after applying the patch to all my instances, restarting PHP-FPM and clearing all caches. I don’t know how to systematically test but for the time being I mark it as solved. If I still encounter errors that might be related, I’ll post here.

1 Like
4

Just wondering why you didn’t have the same issues with Redis.

Did you configure “dbindex” for Redis? Otherwise you have had the same glitches with Redis.
Another solution with Redis is to use a seperate Redis instance for every Nextcloud instance.

5

No, I did not have the issue described here when I switched to Redis/Valkey since I use a different Valkey slot for each NC instance. I sure had the issue with that now obsolete groupfolder version since it did not add prefixes to the local APCu cache. As a workaround, I switched local caching to Valkey (which felt just as quick as APCu). Once the fix was available, I added the patch to my instances and switched back to APCu and confirmed that it fixed the original issue.

6

This topic was automatically closed 8 days after the last reply. New replies are no longer allowed.