All fpm-processes are blocked results in 504

[SebastianRzk]

What does happen:

• Nextcloud is not accessible anymore, server returns status code 504

When does it happen:

• Often: When browsing the nextcloud web-ui
• Rarely: When using the nextcloud app

What does happen in the backend:

The processes are no longer freed and are permanently blocked (>1h).

app_1         | [16-Apr-2023 21:41:05] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 84 total children
app_1         | [16-Apr-2023 21:41:06] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 90 total children
app_1         | [16-Apr-2023 21:41:07] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 96 total children
app_1         | [16-Apr-2023 21:41:08] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 102 total children
app_1         | [16-Apr-2023 21:41:09] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 108 total children
app_1         | [16-Apr-2023 21:41:10] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 114 total children

I noticed, that the request_terminate_timeout is set to 0s = disabled.

 NOTICE: 	request_terminate_timeout = 0s

When I set the request_terminate_timeout to a value, I did not experience the problem again.

So there are two questions for me:

• Why are there processes that are not terminating?

• Why is the request_terminate_timeout set to 0?

[SebastianRzk]

Maybe this happens on apache aswell #1803

[SebastianRzk]

Maybe this bug is related aswell #1886

[J0WI]

Duplicate of #1766 and docker-library/php#1328

[rugk]

Given we have a lot of duplicates and a serious (DOS) bug here, what is the solution (for now) then? We're waiting for upstream docker (#1766 - which they seem reluctant to add/merge)? And why is this exactly closed as “not planned”?

Could not you reconfigure that setting in the Nextcloud docker, if it is causing problems?
Or is there any workaround we can apply for now?

[J0WI]

The solution is to mount a custom FPM config into /usr/local/etc/php-fpm.d/. The config depends on the capacity of your host.