# Crash in libEOSSDK-Linux-Shipping.so 

**URL:** <https://forums.unrealengine.com/t/crash-in-libeossdk-linux-shipping-so/2858474>\
**Category:** Developer Portal\
**Tags:** epic-online-services, question\
**Created:** [May 29, 2026, 7:39am UTC](https://forums.unrealengine.com/t/crash-in-libeossdk-linux-shipping-so/2858474 "2026-05-29T07:39:10Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![anonymous-edc](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@anonymous-edc](https://forums.unrealengine.com/u/anonymous-edc)\
**Post date:** [May 29, 2026, 7:39am UTC](https://forums.unrealengine.com/t/crash-in-libeossdk-linux-shipping-so/2858474/1 "2026-05-29T07:39:10Z")

</div>

Hi, we encountered a crash in the Linux version of the EOS SDK.

The stack trace (from v1.19.0.7 SDK, but I reproduced the same issue on latest version):

```

Fatal error: [File:Unknown] [Line: 50]

#

# Fatal error in: ../../../rtc\_base/platform\_thread.cc, line 219

# last system error: 0

# Check failed: 0 == pthread\_create(&handle, &attr, &RunPlatformThread, start\_thread\_function\_ptr) (0 vs. 22)

#

```

```

Thread 109 “EOSP2PNetwork” received signal SIGABRT, Aborted.

\_\_pthread\_kill\_implementation (no\_tid=0, signo=6, threadid=4850133759552) at ./nptl/pthread\_kill.c:44

44 ./nptl/pthread\_kill.c: No such file or directory.

(gdb) bt

#0 \_\_pthread\_kill\_implementation (no\_tid=0, signo=6, threadid=4850133759552) at ./nptl/pthread\_kill.c:44

#1 \_\_pthread\_kill\_internal (signo=6, threadid=4850133759552) at ./nptl/pthread\_kill.c:78

#2 \_\_GI\_\_\_pthread\_kill (threadid=4850133759552, signo=signo@entry=6) at ./nptl/pthread\_kill.c:89

#3 0x7fffe3e42476 in \_\_GI\_raise (sig=sig@entry=6) at ../sysdeps/posix/raise.c:26

#4 0x7fffe3e287f3 in \_\_GI\_abort () at ./stdlib/abort.c:79

#5 0x7ffff483ee39 in () at /libEOSSDK-Linux-Shipping.so

…

#15 0x7fffe3e94ac3 in start\_thread (arg=\<optimized out\>) at ./nptl/pthread\_create.c:442

#16 0x7fffe3f268d0 in clone3 () at ../sysdeps/unix/sysv/linux/x86\_64/clone3.S:81

```

The hardcoded stack size of 1MB in rtc\_base/platform\_thread.cc is too little probably because of TLS we use, I managed to locally work around this by the LD\_PRELOAD trick with this file:

```cpp

int pthread\_attr\_setstacksize(pthread\_attr\_t \*attr, size\_t stacksize) {

if (stacksize \< 4 \* 1024 \* 1024) {

stacksize = 4 \* 1024 \* 1024;

}

int (\*orig\_setstacksize)(pthread\_attr\_t \*, size\_t) = dlsym(RTLD\_NEXT, “pthread\_attr\_setstacksize”);

return orig\_setstacksize(attr, stacksize);

}

```

But we cannot use the LD\_PRELOAD trick in customer builds, can you increase the stack size request to 4MB on your side, or expose a C API we can use to set the limit ourselves? Thank you.

Adam

---

<div class="post-metadata">

**Author:** ![SebBergy2.0](https://d1ap1mz92jnks1.cloudfront.net/user_avatar/forums.unrealengine.com/sebbergy2.0/32/1214708_2.png) [@SebBergy2.0](https://forums.unrealengine.com/u/SebBergy2.0)\
**Post date:** [October 15, 2026, 7:55pm UTC](https://forums.unrealengine.com/t/crash-in-libeossdk-linux-shipping-so/2858474/2 "2026-10-15T19:55:55Z")

</div>

Your analysis is right, and this was a bug on our side. The SDK’s own threads already add glibc’s TLS-aware minimum on top of the stack size they request, but the threads WebRTC creates for P2P used a fixed 1MB request without that adjustment, so pthread\_create fails with EINVAL once your static TLS goes past about 1MB.

Rather than a fixed 4MB or a new stack size API, we’ve fixed it by applying the same TLS adjustment to the WebRTC threads inside the SDK. We reproduced your crash and confirmed the fix with up to 7MB of static TLS. It’s going through review now and will ship in a future SDK release; it will be listed in the release notes.

In the meantime you don’t need LD\_PRELOAD. Compile the same pthread\_attr\_setstacksize override into your game executable and export it to its dynamic symbol table (link with -Wl,–export-dynamic, or -Wl,–dynamic-list with just that symbol). It takes precedence over glibc for calls from libEOSSDK, the same way your preload does. Keep in mind it applies to every thread in the process. I reproduced your crash on 1.19.1.2 and confirmed this workaround avoids it there, but please validate it against your own build before shipping, and let me know if it doesn’t hold.
