[Bug] Segfault for sd-server on musl

[atomic-carpenter]

Git commit

c97702e

Operating System & Version

Alpine Linux 3.23

GGML backends

Vulkan

Command-line arguments used

sd-server -m juggernautXL_version2.safetensors

Steps to reproduce

• Run server
• Generate any image

What you expected to happen

Image generation

What actually happened

Segfault

Logs / error messages / stack trace

The stack tracke has 1887 lines of:

No symbol table info available.
#1878 0x0000555555bc58cb in ggml_visit_parents_graph ()

The default stack size is 128KB. It's unlikely that this can fit this many recursive calls. Did the stack run out?

#1888 0x00005555557f4f77 in std::_Function_handler<ggml_cgraph* (), UNetModelRunner::compute(int, sd::Tensor<float> const&, sd::Tensor<float> const&, sd::Tensor<float> const&, sd::Tensor<float> const&, sd::Tensor<float> const&, int, std::vector<sd::Tensor<float>, std::allocator<sd::Tensor<float> > > const&, float)::{lambda()#1}>::_M_invoke(std::_Any_data const&) ()
No symbol table info available.
#1889 0x000055555575d680 in std::optional<sd::Tensor<float> > GGMLRunner::compute<float>(std::function<ggml_cgraph* ()>, int, bool, bool) ()
No symbol table info available.
#1890 0x000055555575e52d in UNetModel::compute(int, DiffusionParams const&) ()
No symbol table info available.
--Type <RET> for more, q to quit, c to continue without paging--
#1891 0x0000555555747940 in StableDiffusionGGML::sample(std::shared_ptr<DiffusionModel> const&, bool, sd::Tensor<float> const&, sd::Tensor<float>, SDCondition const&, SDCondition const&, SDCondition const&, SDCondition const&, sd::Tensor<float> const&, float, sd_guidance_params_t const&, float, int, sample_method_t, bool, std::vector<float, std::allocator<float> > const&, int, std::vector<sd::Tensor<float>, std::allocator<sd::Tensor<float> > > const&, bool, sd::Tensor<float> const&, sd::Tensor<float> const&, float, sd_cache_params_t const*)::{lambda(sd::Tensor<float> const&, float, int)#1}::operator()(sd::Tensor<float> const&, float, int) const::{lambda(SDCondition const&, sd::Tensor<float> const*, std::vector<int, std::allocator<int> > const*)#1}::operator()(SDCondition const&, sd::Tensor<float> const*, std::vector<int, std::allocator<int> > const*) const ()
No symbol table info available.
#1892 0x00005555557a8755 in StableDiffusionGGML::sample(std::shared_ptr<DiffusionModel> const&, bool, sd::Tensor<float> const&, sd::Tensor<float>, SDCondition const&, SDCondition const&, SDCondition const&, SDCondition const&, sd::Tensor<float> const&, float, sd_guidance_params_t const&, float, int, sample_method_t, bool, std::vector<float, std::allocator<float> > const&, int, std::vector<sd::Tensor<float>, std::allocator<sd::Tensor<float> > > const&, bool, sd::Tensor<float> const&, sd::Tensor<float> const&, float, sd_cache_params_t const*)::{lambda(sd::Tensor<float> const&, float, int)#1}::operator()(sd::Tensor<float> const&, float, int) const ()
No symbol table info available.
#1893 0x00005555557aa4f8 in std::_Function_handler<sd::Tensor<float> (sd::Tensor<float> const&, float, int), StableDiffusionGGML::sample(std::shared_ptr<DiffusionModel> const&, bool, sd::Tensor<float> const&, sd::Tensor<float>, SDCondition const&, SDCondition const&, SDCondition const&, SDCondition const&, sd::Tensor<float> const&, float, sd_guidance_params_t const&, float, int, sample_method_t, bool, std::vector<float, std::allocator<float> > const&, int, std::vector<sd::Tensor<float>, std::allocator<sd::Tensor<float> > > const&, bool, sd::Tensor<float> const&, sd::Tensor<float> const&, float, sd_cache_params_t const*)::{lambda(sd::Tensor<float> const&, float, int)#1}>::_M_invoke(std::_Any_data const&, sd::Tensor<float> const&, float&&, int&&) ()
No symbol table info available.
#1894 0x0000555555795a37 in sample_euler_ancestral(std::function<sd::Tensor<float> (sd::Tensor<float> const&, float, int)>, sd::Tensor<float>, std::vector<float, std::allocator<float> > const&, std::shared_ptr<RNG>, float) ()
No symbol table info available.
#1895 0x00005555557a633b in StableDiffusionGGML::sample(std::shared_ptr<DiffusionModel> const&, bool, sd::Tensor<float> const&, sd::Tensor<float>, SDCondition const&, SDCondition const&, SDCondition const&, SDCondition const&, sd::Tensor<float> const&, float, sd_guidance_params_t const&, float, int, sample_method_t, bool, std::vector<float, std::allocator<float> > const&, int, std::vector<sd::Tensor<float>, std::allocator<sd::Tensor<float> > > const&, bool, sd::Tensor<float> const&, sd::Tensor<float> const&, float, sd_cache_params_t const*) ()
No symbol table info available.
#1896 0x0000555555733708 in generate_image ()
No symbol table info available.
#1897 0x00005555556c758d in execute_img_gen_job(ServerRuntime&, AsyncGenerationJob&, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&) ()
No symbol table info available.
#1898 0x00005555556c9b77 in async_job_worker(ServerRuntime&) ()
No symbol table info available.
#1899 0x00007ffff7afa680 in ?? () from /lib/libstdc++.so.6
No symbol table info available.
#1900 0x00007ffff7fbef6d in start (p=<optimized out>) at src/thread/pthread_create.c:207
        args = <optimized out>
        state = <optimized out>
#1901 0x00007ffff7fc08c0 in __clone () at src/thread/x86_64/clone.s:22

Stack size for the worker should be much larger. This can be set with pthread_attr_setstacksize.

Additional context / environment details

No response

[Green-Sky]

Doesn't a thread inherit the value from the parent process/thread?
Try setting your stack size larger using eg ulimit -s 8192, which is the default for me. (number in kb)