Issue Segfault error 4 in php-fpm likely on CPU 4

[AlexReiner]

Found a issue at github:

PHP 8.2.19: segfault at 7f377e872428 (sp 00007ffcd580fec0 error 4) · Issue #14261 · php/php-src

Description Hi, sorry for unclear report, I don't know why this happens: kernel: [269065.088395] php-fpm8.2[2107797]: segfault at 7f377e872428 ip 0000562b0b8d9ebd sp 00007ffcd580fec0 error 4 in php...

github.com

 

[Avox]

It looks like Plesk wouldn't care :/

 

[Kaspar@Plesk]

It looks like Plesk wouldn't care :/

Why not?

 

[Avox]

Why not?

The forum has issues since Jul 15, 2023?

Search results for query: Segfault

talk.plesk.com

There is currently no statement from Plesk :/ or maybe you can link here one?
Hopefully we can get started with Debian 11 or 12 soon, there is already good and active communication in Git.

PHP 8.2.19: segfault at 7f377e872428 (sp 00007ffcd580fec0 error 4) · Issue #14261 · php/php-src

Description Hi, sorry for unclear report, I don't know why this happens: kernel: [269065.088395] php-fpm8.2[2107797]: segfault at 7f377e872428 ip 0000562b0b8d9ebd sp 00007ffcd580fec0 error 4 in php...

github.com

 

[Kaspar@Plesk]

This is an community forum, available for users to discus Plesk and related technologies. Like you're doing in this thread. (But I you want to discus beer recipes that's okay too). The forums however are not an official support channel for Plesk.

As far as I can tell the PHP segfault issue is not caused by any configuration by Plesk, rather it seems be an issue directly in PHP. I could be wrong off course.

You are welcome to open an ticket with Plesk support so our support engineers can investigate the issue directly on your server.

 

[Avox]

@ChristophRo i have on question, maybe it's possible.
You write on debian 11 and 12 you don't have issues.

Maybe you have the chance to setup a new debian 11 or 12 system and can check your logs?
It would be interesting this error comes only on a new setup machine?

 

[ChristophRo]

I strongly doubt that a new setup does change anything, as it uses the same bits for PHP as lets say a server that was installed a couple months ago.

If anything, then our setup may slightly differ from yours, in regards to libraries and additional software packages that may or may not be installed.
Though that should not make a difference either, except if maybe some library is a different version, because we (or you) do use a custom apt source (like deb.sury.org in our case) that installs a different/newer version of a library than stock Debian provides. (libpcre3 comes to mind, that we get from deb.sury.org)

Another thing I can think off (also relative low chance) is that the PHP configuration has something to do with it. We do adjust some parameters for all PHP versions, but in our case I can't see anything that could remotely have any influence on a segfault.

Another thing I can think off (probability quite a bit higher) is that your PHP application does use a function/extension that none of our websites does.
I could think of stuff like LDAP, pgsql, or redis, that in our case is most likely not used anywhere.

 

[AlexReiner]

Hmm we‘ve tested on a new server with Debian 12.5 with the one click installer and don‘t change anything and had this error…

 

[AlexReiner]

And now - thats really strange - the server hangs an i trying to restart. Than following error occours:



Does anyone have a conclusion?

 

[ChristophRo]

Your image/screenshot is not accessible (possibly for anyone except yourself)

As for the segfaults, I do somehow suspect that this problem may only affect Hetzner customer at most, as otherwise I'm quite sure we would have much more similar reports right now.
I've never used any services from Hetzner, but I could imagine that they use custom Debian images with custom/own repositories and possibly even some own packages and libraries that come preinstalled.

 

[AlexReiner]

But if I use the original Debian image with netinstall, install it into a vm. after this i installed plesk with the one click installer and migrateing some websites. I think there are the original repos (not from hetzner)

 

[Avox]

The support does not see it as a Plesk error, everything works without Plesk.
It's time to change the panel, too bad.

 

[ChristophRo]

Just a heads up for everyone affected by it...
While I thought that none of our systems were affected, just yesterday we discovered that indeed we have one server showing the same errors for some days now.


It's an older system with Debian 10 (initially Debian9 i believe) that is using the PHP packages of deb.sury.org, BUT has never ever seen a Plesk panel or any other control panel software.
So no, this problem has absolutely nothing to do with Plesk. (as everyone with a little bit of understanding how computers works already guessed of course)

As for now, we have no solution yet.
I've disabled (the manually enabled/changed Jit compiler options) of PHP8 8.2 and will observe the situation, as this seems to be a likely candidate for these troubles.
And just for the record, these Jit options also get enabled if you ever use the "Plesk booster" option, so keep that in mind.

 

[Avox]

@ChristophRo thats funny, we have Debian 10 with Plesk and php 7.4 + 8.2 + 8.3 without any errors on a very high traffic server with big queries.
On a Debian 12 + Plesk has errors, on Debian 12 + ISPConfig - not, but you write you have errors on Debian 10 and me not.
It sounds more and more like a hardware thing, unfortunately I don't know :/

 

[ChristophRo]

Nope, this is not a hardware problem. (our affected system is a VMware based virtual machine anyway, that runs a single Typo3 website)

I suspect some obscure opcache/Jit problem, that may only trigger under certain conditions (possibly a combination of conditions), like:

- usage of a specific php extension (and I mean real usage, not simply the fact it is enabled)
- overflow of opcache.max_accelerated_files
- overflow of opcache.memory_consumption
- overflow of opcache.interned_strings_buffer

 

[msnet]

Same here.

Hetzner AX52
Debian 11.9 in Proxmox 8 VM
Version 18.0.60 Update #1
PHP 8.3.7, opcache enabled and in use

lots of and similiar:

12718609.284385] php[3456798]: segfault at 7f8f9fd00018 ip 00005640ac84f4c2 sp 00007fff132333d0 error 6 in php[5640ac5c4000+317000]

 

[ChristophRo]

We did not have any further segfaults so far, after we disabled "opcache.jit" and "opcache.jit_buffer_size" (i.e. reset it to server/php defaults) and also increased the "opcache.max_accelerated_files" from 8000 to 16000

But this may not be representative, as on that server the segfaults only started to happen like a week ago and out of the blue.
So, it may be a coincidence that after months of silence, then three days of segfaults, it's now "ok" again for days, weeks or months to come...

If anyone has a reliable way to trigger these segfaults, I'm happy to try out.

 

[ChristophRo]

Can you guys test if the segfaults still happen with the new php packages that got released today?

Possibly they are related to Segmentation fault on unknown address 0x0001ffff8041 with XML extension under certain memory limit · Issue #14124 · php/php-src

 

[AlexReiner]

We have still problems with sury and php8.3 (PHP 8.3.8 (built: Jun 6 2024 16:32:04)