Salta al contenuto principale

How can I prevent that # uses more workers than it should?

2022-03-21T11:14:12Z worker [NOTICE]: Load: 5.078125/12 - processes: 174861/37/105455 - jpm: 806/114/251 (0:36, 1:0/13, 20:0/1655, 30:0/200, 40:1/278448) - maximum: 38/25 [] - {"file":"Worker.php","line":819,"function":"tooMuchWorkers","uid":"fd72e1","process_id":741975}


38 instead of maximum 25 will always cause oom kill for PHP.

!Friendica Support@Friendica Admins
#friendica @Friendica Admins !Friendica Support

reshared this

in reply to utzer [Friendica]

utzer [Friendica]
utzer [Friendica]
• •
possibly this is not the reason for OOM, but maybe... who knows.

reshared this

in reply to utzer [Friendica]

Michael Vogel
Michael Vogel
• •
I guess that's an effect of the "fast lane" setting.

reshared this

in reply to Michael Vogel

utzer [Friendica]
utzer [Friendica]
• •
@Michael Vogel ok, will disable that, anyway just want the queue to get reduced. good idea.
@Michael Vogel

reshared this

in reply to utzer [Friendica]

Michael Vogel
Michael Vogel
• •
BTW: What do you mean with " oom kill"? Can you show the message?

reshared this

in reply to Michael Vogel

utzer [Friendica]
utzer [Friendica]
• •
@Michael Vogel@Friendica Support@Friendica Admins Too many worker processes running at the same time exausts my servers ram and then Linux kernel kills some applications, mostly PHP or MariaDB.
@Michael Vogel @Friendica Admins @Friendica Support

reshared this

in reply to utzer [Friendica]

Tobias
Tobias
• •
I think this needs some clarification (in the code?) @Michael Vogel at least once upon atime the !-notation was introduced in Friendica to have an exclusive distribution of a posting via a forum. At this time the problem was with @-mentioning two or more forums, that the distribution of the entire thread was not successful along the lines of all possible distribution ways.

This posting is addressed with the !-notation to the "Support Forum" which should be the exclusive path of the distribution. But it also contains the @-mention of the admin forum. The result seems to be that it is now distributed via both forums.

Which is either a bug, or the not updated documentation and change of expected distribution path'.
@Michael Vogel

reshared this

in reply to Tobias

Michael Vogel
Michael Vogel
• •
The distribution via two forums is not a bug but is the expected behaviour, due to the recent changes.

reshared this

in reply to utzer [Friendica]

Luca Nucifora
Luca Nucifora
•
@utzer did you try to focus on the server load too? "Maximum Load Average"
@utzer [Friendica]

reshared this

in reply to Luca Nucifora

utzer [Friendica]
utzer [Friendica]
• •
@Luca Nucifora yes, but the load average rarely exceeds 6 and there is 12 cores available for this host, also htop looks OK.
@Luca Nucifora

reshared this

Unknown parent

Michael Vogel
Michael Vogel
• •
Can you point me to the part of the documentation?

reshared this

in reply to utzer [Friendica]

Luca Nucifora
Luca Nucifora
•
@utzer ok :) disabling "fast lane" resolved your issue? (I'm curious because maybe I will need it too)
@utzer [Friendica]

reshared this

in reply to Luca Nucifora

utzer [Friendica]
utzer [Friendica]
• •
@Luca Nucifora there is more than one issue I have, 130000 jobs in the queue reduced to 96000 since yesterday and PHP fpm did not get killed by OOM anymore. I think fastlane is a good idea, but I then need to configure the maximum memory used better.
@Luca Nucifora

reshared this

in reply to utzer [Friendica]

utzer [Friendica]
utzer [Friendica]
• •
@Luca Nucifora something is odd anyway, with 1000 jpm (jobs per minute) the jobs in queue still increase slowly sometimes. But also 1000jpm is rare to be processed, often it drops to 10 jpm only, but I guess that is different priority jobs, which also take different amount of time to process.
@Luca Nucifora

reshared this

⇧