Cloudron instance scaling issues after a few hours / couple of days, apps responsive but showing a permanent "Starting..." status

[nebulon]

@uwcrbc please send a mail with your Cloudron ID / dashbaord domain info to support@cloudron.io and enable remote ssh support for us, so we can take a closer look to get this resolved.

[uwcrbc]

@nebulon Hi @nebulon - Thanks for this. Email sent.

[nebulon]

We got the info and are looking into this. The findings so far are, that the calcom app is constantly restarting and causing the server to become very slow. Not sure why this seems to have cascading effects, since the resources are plenty, but depending on the server provider, the virtual CPUs might get throttled if one core is used 100% over a longer period of time. Will keep this updated on the findings.

[uwcrbc]

@nebulon Interesting.
The server provider is ssdnodes - So far I have had nothing to complain about (just about the opposite). Happy to play the liaison part should we need to relay info / ask questions of them.
Many thanks,

[uwcrbc]

Also: interestingly the moment the server started playing up is indeed the moment that I had roughly linked to our latest cal.com deployment.
I say latest because this Cloudron server had a previous cal.com app instance that we used as a test a while back - not sure if relevant but maybe it can help inform the situation.

[nebulon]

I've updated the calcom instance on that server to latest, which has some fixes in the package. So far things seem fine.

[uwcrbc]

Happy to report that the updated cal.com package seems to have fixed the underlying server issue.
Many thanks,

[uwcrbc]

At last - I think I was too eager to see this resolved.
The problem is still there with the exact same symptoms.
How can we proceed further? @nebulon @james Would another remote troubleshooting/debugging be helpful?

[nebulon]

Do you see a constant process showing up in htop or so on your system, which may constantly hog some core like calcom restart did?

[uwcrbc]

@nebulon Thanks - To the untrained eye, there is nothing in htop that pops up as flagrantly hogging resources.
The swap is high but I am not sure if this is a relevant indicator

[nebulon]

And the systemctl restart box still fixes the issue temporarily?

[uwcrbc]

@nebulon Yes - this fixes the issue temporarily at least

[girish]

This is probably happening because the apphealthmonitor is getting stuck (for some reason) and thus unable to update the health of apps. @uwcrbc in the box.log , do you see a line printed every 10 seconds like box:apphealthmonitor app health: xx ?

[uwcrbc]

@girish Thanks for this - After looking into the box.log:

• post systemctl restart box I do see indeed box:apphealthmonitor app health: xx running / 0 stopped / 0 unresponsive entry type every 10 seconds or so.
• pre systemctl restart box (when we experience the issue) I do not see much of the box:apphealthmonitor app health:xx entries. Rather, I do have a few rare box:apphealthmonitor setHealth: <<CONTAINER_UID>> (<<URL>>) waiting for 1192.461 to update health entries

Hopefully it helps?

[girish]

@uwcrbc I think there is a bug that the apphealthmonitor is getting stuck (for some reason). We have seen this happen but very rarely and not reproducibly . If you can give us access, can you send me a mail at support@cloudron.io ? I can debug this further .

[Neiluj]

This post is deleted!

[uwcrbc]

@girish Thanks for this. Access given / mail sent. Appreciate the help.

[robi]

I think I have another instance of this bug with the health monitor.

I have installed an app that does not have health checks and it shows as Not Responding in the dashboard even though it works fine.

However, it seems that because of that app and the health monitor getting stuck, any new apps that get installed or upgraded also fail their health checks and remain in Starting... mode in the dashboard.

I have rebooted the server and all the apps come up, other than the one mentioned above, then after this updated apps show the Starting... message.

P.S.
It would be really nice to add the button for the 'cloudron' service, like we have for all other services.

[girish]

My home server got into a situation like this just now. It seems it's because the eventlog got flooded with many entries.

+----------+
| count(*) |
+----------+
|   563547 |
+----------+

mysql> SELECT action, COUNT(*) AS count
    -> FROM eventlog
    -> GROUP BY action
    -> ORDER BY count DESC;
+---------------------------+--------+
| action                    | count  |
+---------------------------+--------+
| app.up                    | 446592 |
| app.down                  | 106588 |
| backup.cleanup.finish     |   3664 |
| app.update.finish         |   1354 |
| app.update                |   1354 |
| backup.finish             |    920 |
| backup.start              |    920 |
| cloudron.update.finish    |    847 |
| cloudron.update           |    833 |
| cloudron.start            |    110 |
| dyndns.update             |     78 |

Those app up/down eventlogs are out of hand!

I nuked them manually:

mysql> DELETE from eventlog WHERE action='app.up';
Query OK, 446782 rows affected (8 min 1,59 sec)

mysql> DELETE from eventlog WHERE action='app.down';
Query OK, 106588 rows affected (9 min 16,27 sec)

That took a whopping 17min just to delete entries!

[robi]

Oh wow, thanks for finding that Girish!

Are you pushing fixes?
Should we do the same manually?
Does box need a restart after or will apphealthmonitor just work better w/o half a million app.up entries?