Question PHP-FPM (pm.max_requests) Query

[maximus99]

Server operating system version

Centos7.9

Plesk version and microupdate number

18.0.52

PHP-FPM (pm.max_requests) Query

We are running plesk 18.0.52 using PHP-FPM for a domain . but as per plesk php config for pm.max_requests it shows Standard in Panel what it means ? default value 0 ? But still its not showing in configuration as show below



pm = ondemand
pm.max_children = 20
pm.max_spare_servers = 1
pm.min_spare_servers = 1
pm.process_idle_timeout = 10s
pm.start_servers = 1

Please advise what is my current value or what is standard ? And for this config what could be a ideal value to set . I agree it has some calculations we have some hug traffic and memory consumption

 

[Kaspar@Plesk]

The default value for pm.max_requests is 0 (as per PHP docs).

The optimal value depends on your needs an available server resources (primarily memory). There is an calculation that can be used a general guide line which is explained here: https://support.plesk.com/hc/en-us/articles/12377086904471-How-to-calculate-pm-max-children-value-on-a-Plesk-server

 

[Hangover2]

The Plesk article referenced does not address pm.max_requests - only pm.max_children.

The default value of 0 can be dangerous if there is a memory leak in a third-party library. We experienced this issue once in the past. As a result, the servers became unstable because the available free memory was gradually consumed (a small memory leak, but each PHP request led to a growing PHP process).

However, a common recommendation is to set it between 500 and 1,000 requests. This range is often effective in balancing performance and preventing memory leaks.

 

[maximus99]

Thanks a lot for your update . Will check and update

 

[maximus99]

No that didnt fixed my issue . One of our particular domain is going down multiple times between 12:00 AM to 4:00 AM . We are not sure what is causing that

1. No errors in DB side
2. didnt see errors in apache logs or the OS logs

this is the php-fpm parameters we have set for that particular domain , Any advise would be helpful


php_value[error_reporting] = 22519
php_value[max_execution_time] = 400
php_value[max_input_time] = 150
php_value[max_input_vars] = 1500
php_value[memory_limit] = 1024M
php_value[open_basedir] = none
php_value[post_max_size] = 128M
php_value[upload_max_filesize] = 128M


pm = ondemand
pm.max_children = 20
pm.max_requests = 800
pm.max_spare_servers = 1
pm.min_spare_servers = 1
pm.process_idle_timeout = 60
pm.start_servers = 1
request_terminate_timeout = 500

 

[Bitpalast]

What is meant by "going down"?
Is there a lot of activity in the Apache logs between midnight an 4 am?
Why is PHP-FPM suspected as a root cause?

 

[maximus99]

What is meant by "going down"? -- In our monitoring tool we have set the polling like if we dont get a status 200 in web check . it will notify and mark us its not reachable.
Is there a lot of activity in the Apache logs between midnight an 4 am? -- Well i could see in access_ssl_logs some 404 and 500 of our monitoring tool trying to check particular domain . I see its generic as its throwing after 4 AM as well.
We could see this message in error log : [proxy_fcgi:error] AH01068: Got bogus version 34 , And this as well generic.

Finally the proxy error log :

2024/07/31 00:39:01 [error] 6921#0: *9907 upstream prematurely closed connection while reading response header from upstream,
2024/07/31 00:39:03 [error] 6919#0: *12580 connect() failed (111: Connection refused) while connecting to upstream,
We see these upstream connection close errors

And to avoid this we have to fine tune the php-fpm is what the other KB or document says so we have suspected it could be because of that as its only issue with this particular domain.

 

[Bitpalast]

Would it not be a better approach to examine what the website scripts are doing and specifically where they spend most of the execution time instead of increasing server resources?