php8ts crashes in php_clear_stat_cache()

[jjdunn]

Description

debug-analysis-1.txt

very sorry: I don't know how to reproduce this bug. but I can see the crash dump files on the web server.

using WinDbg I opened and analyzed one of the files. the analysis including stack-dump is in the attached txt file.

this is PHP 8.3.19 thread-safe running in Apache 2.4.63 on Windows Server 2019

PHP Version

PHP 8.3.19

Operating System

Windows Server 2019

[devnexen]

cc @bukka, might have some idea whether it is related to last change in this area.

[bukka]

I have been looking into this and it seems like the new cache clearance is causing this. But after going couple of time through the code paths, I can't understand why this is crashing. The stack text is not that helpful as it's not proper stack trace so I can't really see the full path from it. But it looks like it's happening during sapi activate but the strange thing that all the new stream clearances happen only if EG(active) is true. But from what I see if EG(active) is true, then BG should always be initialized. It seems to me more like this is uncovering some other issue or I'm missing something. Maybe @nielsdos might have an idea as he was fixing things in TSRM in past.

The solution would be revert that stream part but I would like to first try to figure out what's going on.

@jjdunn Can you see some other issues in the logs (e.g. some memory limit or other fatal error around the same time)? Could you you also share your httpd.conf and tell us how often do you see those crashes compare to the other traffic?

Btw. there are some known issue with mod_event (assuming that's what you use) so it might be also related to this but not sure.

[cmb69]

@jjdunn, please also install the suitable debug pack (likely https://windows.php.net/downloads/releases/php-debug-pack-8.3.19-Win32-vs16-x64.zip), so we get a better stack backtrace.

[jjdunn]

Hi @bukka and @cmb69 - Thank you very much for investigating this problem !

there are some known issue with mod_event (assuming that's what you use)

we are not using mod_event

Could you you also share your httpd.conf

attached please find sanitized versions of httpd.conf and httpd-vhosts.conf, with only the host which is crashing included (suffix .txt added so this ticket system will allow upload)

httpd.conf.txt
httpd-vhosts.conf.txt

Can you see some other issues in the logs (e.g. some memory limit or other fatal error around the same time)?

there is nothing unusual in the apache Access logs
the apache Error log contains three lines for each crash:

[Wed Mar 26 07:27:00.943981 2025] [mpm_winnt:notice] [pid 2764:tid 144] AH00428: Parent: child process 3860 exited with status 3221225477 -- Restarting.
[Wed Mar 26 07:27:01.186187 2025] [mpm_winnt:notice] [pid 2764:tid 144] AH00455: Apache/2.4.63 (Win64) OpenSSL/3.3.2 mod_log_rotate/1.02 configured -- resuming normal operations
[Wed Mar 26 07:27:01.186187 2025] [mpm_winnt:notice] [pid 2764:tid 144] AH00456: Apache Lounge VS17 Server built: Feb 3 2025 10:10:51

there is nothing relevant in the logs generated by the PHP application

we have a page-based PHP / HTML / CSS app built using the Yii 1.1 framework; but there's an embedded single-page javascript app (no framework). The js app does server callbacks to get/save data. We just installed a new version of the js app (yesterday on the Test server, today on the Prod server). When the js app opens it fetches the data, and immediately tries to save. One of these operations is failing. It's unclear if there's a bug in the js app - it works fine on the lead developer's local environment; and works fine in my development environment; and now (weirdly) works fine in Test. (I reverted Prod to the prior version). We don't yet know if there's an application-code error; but in any case I would not expect PHP to crash even if there were.

@cmb69 wrote:

please also install the suitable debug pack.....

I can do this in my local dev environment, and on the Test server. but since we cannot reliably reproduce the problem, we may have to wait a while to see what happens...

I got the mini-dumps for the crashes on prod; but since you've explained there is insufficient information I won't send the WinDbg analysis unless you indicate this would be helpful. BTW I am not a systems-developer, just an application-land developer; so my ability to use WinDbg and similar tools is a bit limited. I can follow any instructions if you send them.