Exception occurred in preexec_fn

[alexander-zhang]

When setting memory_limit_MB to be not -1, we encounter exception:

{"status":"SandboxError","message":"exception on running command python /tmp/tmpmn1cjrq6/tmpzw8a2ydp.py: Exception occurred in preexec_fn. | None","compile_result":null,"run_result":{"status":"Error","execution_time":null,"return_code":null,"stdout":null,"stderr":"exception on running command python /tmp/tmpmn1cjrq6/tmpzw8a2ydp.py: Exception occurred in preexec_fn. | None"},"executor_pod_name":null,"files":{}}%

To determine the root cause, I put a try/except block inside the preexec_fn to catch the error.
We see:

2025-12-11 21:09:52 [debug    ] running command python /tmp/tmpvwksj3in/tmpy4x_jjk6.py [sandbox.runners.base]
/root/miniconda3/lib/python3.11/site-packages/structlog/stdlib.py:1087: UserWarning: Remove `format_exc_info` from your processor chain if you want pretty exceptions.
  ed = p(logger, meth_name, cast(EventDict, ed))
2025-12-11 21:09:52 [error    ] memory_limit_preexec failed    [sandbox.runners.base] error=current limit exceeds maximum limit error_type=ValueError
Traceback (most recent call last):
  File "/root/sandbox/sandbox/runners/base.py", line 136, in memory_limit_preexec
    resource.setrlimit(resource.RLIMIT_STACK, (soft_memory_limit, hard_memory_limit_STACK))
ValueError: current limit exceeds maximum limit
  File "/root/sandbox/sandbox/runners/base.py", line 58, in run_command_bare
    p = await asyncio.create_subprocess_shell(command,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/miniconda3/lib/python3.11/asyncio/subprocess.py", line 208, in create_subprocess_shell
    transport, protocol = await loop.subprocess_shell(
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "uvloop/loop.pyx", line 2808, in subprocess_shell
  File "uvloop/loop.pyx", line 2778, in __subprocess_run
  File "uvloop/handles/process.pyx", line 611, in uvloop.loop.UVProcessTransport.new
  File "uvloop/handles/process.pyx", line 172, in uvloop.loop.UVProcess._init
2025-12-11 21:09:52 [warning  ] exception on running command python /tmp/tmpvwksj3in/tmpy4x_jjk6.py: Exception occurred in preexec_fn. | None [sandbox.runners.base]

The issue lies here:

SandboxFusion/sandbox/runners/base.py

Lines 126 to 137 in 5e37f71

	if args.memory_limit_MB > 0:
	def memory_limit_preexec():
	_, hard_memory_limit_AS = resource.getrlimit(resource.RLIMIT_AS)
	_, hard_memory_limit_DATA = resource.getrlimit(resource.RLIMIT_DATA)
	soft_memory_limit = args.memory_limit_MB * 1024 * 1024
	resource.setrlimit(resource.RLIMIT_AS, (soft_memory_limit, hard_memory_limit_AS))
	resource.setrlimit(resource.RLIMIT_DATA, (soft_memory_limit, hard_memory_limit_DATA))
	if platform.uname().system != "Darwin":
	_, hard_memory_limit_STACK = resource.getrlimit(resource.RLIMIT_STACK)
	resource.setrlimit(resource.RLIMIT_STACK, (soft_memory_limit, hard_memory_limit_STACK))
	preexec_steps.insert(0, memory_limit_preexec)
	preexec_fn = lambda: [step() for step in preexec_steps] if preexec_steps else None

It seems SandboxFusion tries to allocated the desired memory in the Address Space, Data, and Stack.

However, the Stack typically has a process-set maximum of much less than 1GB for example.
Therefore, exception is thrown due to an inability to update the Stack memory limit.

To fix this, I replaced:

def memory_limit_preexec():
                _, hard_memory_limit_AS = resource.getrlimit(resource.RLIMIT_AS)
                _, hard_memory_limit_DATA = resource.getrlimit(resource.RLIMIT_DATA)
                soft_memory_limit = args.memory_limit_MB * 1024 * 1024
                resource.setrlimit(resource.RLIMIT_AS, (soft_memory_limit, hard_memory_limit_AS))
                resource.setrlimit(resource.RLIMIT_DATA, (soft_memory_limit, hard_memory_limit_DATA))
                if platform.uname().system != "Darwin":
                    _, hard_memory_limit_STACK = resource.getrlimit(resource.RLIMIT_STACK)
                    resource.setrlimit(resource.RLIMIT_STACK, (soft_memory_limit, hard_memory_limit_STACK))

with

def memory_limit_preexec():
                try:
                    _, hard_memory_limit_AS = resource.getrlimit(resource.RLIMIT_AS)
                    _, hard_memory_limit_DATA = resource.getrlimit(resource.RLIMIT_DATA)
                    soft_memory_limit = args.memory_limit_MB * 1024 * 1024
                    resource.setrlimit(resource.RLIMIT_AS, (soft_memory_limit, hard_memory_limit_AS))
                    resource.setrlimit(resource.RLIMIT_DATA, (soft_memory_limit, hard_memory_limit_DATA))
                    if platform.uname().system != "Darwin":
                        _, hard_memory_limit_STACK = resource.getrlimit(resource.RLIMIT_STACK)
                        soft_stack_limit = min(soft_memory_limit, hard_memory_limit_STACK)
                        resource.setrlimit(resource.RLIMIT_STACK, (soft_stack_limit, hard_memory_limit_STACK))
                except Exception as e:
                    logger.error(
                        "memory_limit_preexec failed",
                        error_type=type(e).__name__,
                        error=str(e),
                        exc_info=True,
                    )
                    raise