[Error:38]Function not implemented. multiprocessing in graphene about graphene HOT 7 OPEN

From what I understand, Python's multiprocessing package uses POSIX semaphores (or maybe even older Sys-V semaphores). Check https://linux.die.net/man/7/sem_overview.

Gramine currently doesn't support semaphores at all. So this example unfortunately cannot run in Gramine.

from graphene.

You can actually add loader.log_level = "trace" in your manifest file, and re-run your simple example. Gramine will output a lot of additional info.

If you see some system call like sem_wait() or sem_open() or segmet(), and they return -38 ("not implemented"), then it's definitely the problem: Gramine doesn't support these system calls (Gramine doesn't support semaphores).

from graphene.

You can actually add loader.log_level = "trace" in your manifest file, and re-run your simple example. Gramine will output a lot of additional info.

If you see some system call like sem_wait() or sem_open() or segmet(), and they return -38 ("not implemented"), then it's definitely the problem: Gramine doesn't support these system calls (Gramine doesn't support semaphores).

i got it, so is there any workaround to deal with this problem?

from graphene.

i got it, so is there any workaround to deal with this problem?

The only workaround I can think of is: don't use multiprocessing package in Python.

from graphene.

i got it, so is there any workaround to deal with this problem?

The only workaround I can think of is: don't use multiprocessing package in Python.

I try to set num_workers=0 to ban the multiprocessing function ,  

data.DataLoader(val_dst, batch_size=opts.val_batch_size, shuffle=False, num_workers=0)

The original error 38 disappeared, but another bug happened, my program killed by singal 8. I checked that it means 8) SIGFPE.
I have tried without graphene, it ran successfully. Then i try to increase enclave_size, stack.size and pal_internal_mem_size,but all invalid. I have no idea what happened.

for i, (images) in enumerate(tqdm(val_loader)):

from graphene.

arithmetic fault is interesting!

Can you debug your program using gdb? You simply build all Gramine in debug mode and then run GDB=1 gramine-sgx <your app>. For more info, check https://gramine.readthedocs.io/en/latest/devel/debugging.html

The debugger should point you to the place in code / assembly instruction where SIGFPE (airthmetic fault) happens. This may immediately give you an idea of what goes wrong. Or at least you can copy-paste the assembly snippet where SIGFPE happens -- maybe we'll be able to help just by looking at the failing assembly.

from graphene.

arithmetic fault is interesting!

Can you debug your program using gdb? You simply build all Gramine in debug mode and then run GDB=1 gramine-sgx <your app>. For more info, check https://gramine.readthedocs.io/en/latest/devel/debugging.html

The debugger should point you to the place in code / assembly instruction where SIGFPE (airthmetic fault) happens. This may immediately give you an idea of what goes wrong. Or at least you can copy-paste the assembly snippet where SIGFPE happens -- maybe we'll be able to help just by looking at the failing assembly.

Sorry for not being able to check the message due to some things in the past two days, i will try GBD immediately.

from graphene.