Re: Anybody seeing NextCloud crash?

> On Nov 14, 2024, at 13:29, Ronald Klop <ronald@FreeBSD.org> wrote:
> 
> Op 04-11-2024 om 05:39 schreef Kevin P. Neal:
>> I'm seeing a problem with NextCloud's php-fpm instance crashes when trying
>> to handle a request over the web interface. My web server is Apache 2.4
>> if it matters. The MacOS desktop sync client works just fine, and so does
>> the app from iOS 16 (I have an iPhone 8 that I'm upgrading soon).
>> Apache responds with a "Service Unavailable" error page.
>> In my /var/log/php-fpm.log I find this:
>> [03-Nov-2024 11:06:45] WARNING: [pool www] child 68236 exited on signal 4 (SIGILL) after 5.975134 seconds from start
> 
> 
> SIGILL means that an instruction was executed that was not understood by the CPU.
> This can be something like the software using SSE2 instructions and your CPU does not support that.
> 
> I'm not sure if there are other reasons for SIGILL.
> 
> It helps if you could share what FreeBSD version your are running and on what architecture.
> 
> The output of 'uname -a' gives this.
> 
> /var/run/dmesg.boot gives information about your CPU. This helps in determining the supported instructions.
> Mine gives this:
> CPU: Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz (2400.14-MHz K8-class CPU)
>  Origin="GenuineIntel"  Id=0x406e3  Family=0x6  Model=0x4e  Stepping=3
>  Features=0x1783fbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,MMX,FXSR,SSE,SSE2,HTT>
>  Features2=0xdeda2203<SSE3,PCLMULQDQ,SSSE3,CX16,PCID,SSE4.1,SSE4.2,MOVBE,POPCNT,AESNI,XSAVE,OSXSAVE,AVX,RDRAND,HV>
>  AMD Features=0x28100800<SYSCALL,NX,RDTSCP,LM>
>  AMD Features2=0x121<LAHF,ABM,Prefetch>
>  Structured Extended Features=0x842421<FSGSBASE,AVX2,INVPCID,NFPUSG,RDSEED,CLFLUSHOPT>
>  Structured Extended Features3=0x10000400<MD_CLEAR,L1DFL>
>  TSC: P-state invariant
> 
> If php-fpm gives a core dump it is also possible to analyze on what instruction it is failing. I don't know the command on how to do that by the top of my head. Maybe somebody else can help.
> 
> If you can find out what package update resulted in the crash then you could ask the package maintainer to compile it for older CPUs by default.
> 
> Regards,
> Ronald.
> 
> 
> 
> 
>> Every time I tried to reload a page on the web site I get a failure like
>> that one. The php-fpm processes are dying.
>> I use the checkrestart package after every pkg upgrade and I restart
>> anything that gets mentioned. This handy utility isn't foolproof, but
>> I've had good luck with it.
>> Reinstalling all my packages with pkg upgrade -f changed nothing.
>> I rolled back my ZFS dataset to October 30's, rebooted (needed, don't know
>> why), and I have it back up and running now.
>> I did "service php_fpm restart" after the bad update, but I forgot to check
>> that everything was still working. That's why I had to rollback so far --
>> I have SMS messages on my phone that imply it was working on that date.
>> The problem was introduced with one of these updates:
>> The following 11 package(s) will be affected (of 0 checked):
>> New packages to be INSTALLED:
>> openh264: 2.3.0,2
>> openjph: 0.17.0
>> Installed packages to be UPGRADED:
>> ffmpeg: 6.1.2_4,1 -> 6.1.2_5,1
>> glib: 2.80.5,2 -> 2.80.5_1,2
>> libheif: 1.18.2_1 -> 1.19.1
>> libnghttp2: 1.63.0 -> 1.64.0
>> libssh2: 1.11.0_1,3 -> 1.11.1,3
>> libvpx: 1.14.1 -> 1.15.0
>> pciids: 20240920 -> 20241024
>> py311-redis: 5.1.1 -> 5.2.0
>> svt-av1: 2.2.0 -> 2.3.0
>> Number of packages to be installed: 2
>> Number of packages to be upgraded: 9
>> I do have my own poudriere setup, but I don't think I have any custom
>> settings at this point. I can doublecheck if anyone is curious, though.
>> Anyone else seen this? Or should I file a bug report?
> 
> 

Just for everyone else in the list, I have gone through some discussion
with the original author of the mail and it seems to have some sort of
regression with graphics/pecl-imagick. Whenever this php module is
loaded php-fpm crashes on at least php82 and maybe later. This module
actually may not support php82 and later. Or maybe there was some sort
of new regression with some other commits down the line of ImageMagic.

Kind regards,
Moin