[Inductor] [Distributed] DDP torch.compile model hangs on exit (python 3.8/3.9)

[jataylo]

🐛 Describe the bug

On python3.8/3.9 environments hangs are being observed with DDP model+torch.compile on model exit. This is observed both on AMD/NV platforms on nightly and does not occur in equivalent python3.10+ environment.

The workaround for this so far has been to set TORCHINDUCTOR_COMPILE_THREADS=1 in the same approach used in some distributed inductor unit tests e.g.

pytorch/test/distributed/test_inductor_collectives.py

Line 98 in 761a7b8

@patch.object(torch._inductor.config, "compile_threads", 1)

which mentions # TODO: somehow inductor bg compile threads are causing hangs at exit with distributed work

Reproducer:
https://gist.github.com/jataylo/50a88745acfc817e2e2f89a41232f308

Versions

PyTorch version: 2.4.0a0+git9dded14
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 6.0.32831-204d35d16

OS: Ubuntu 20.04.6 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
Clang version: 17.0.0 (https://github.com/RadeonOpenCompute/llvm-project roc-6.0.2 24012 af27734ed982b52a9f1be0f035ac91726fc697e4)
CMake version: version 3.26.4
Libc version: glibc-2.31

Python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.19.0-42-generic-x86_64-with-glibc2.17
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: AMD Instinct MI250X/MI250 (gfx90a:sramecc+:xnack-)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: 6.0.32831
MIOpen runtime version: 3.0.0
Is XNNPACK available: True

CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 48 bits physical, 48 bits virtual
CPU(s): 128
On-line CPU(s) list: 0-127
Thread(s) per core: 1
Core(s) per socket: 64
Socket(s): 2
NUMA node(s): 2
Vendor ID: AuthenticAMD
CPU family: 25
Model: 1
Model name: AMD EPYC 7713 64-Core Processor
Stepping: 1
Frequency boost: enabled
CPU MHz: 1500.000
CPU max MHz: 3720.7029
CPU min MHz: 1500.0000
BogoMIPS: 3992.54
Virtualization: AMD-V
L1d cache: 4 MiB
L1i cache: 4 MiB
L2 cache: 64 MiB
L3 cache: 512 MiB
NUMA node0 CPU(s): 0-63
NUMA node1 CPU(s): 64-127
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines, IBPB conditional, IBRS_FW, STIBP disabled, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin brs arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm

Versions of relevant libraries:
[pip3] mypy==1.9.0
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.21.2
[pip3] optree==0.11.0
[pip3] torch==2.4.0a0+git9dded14
[pip3] torchvision==0.19.0a0+96640af
[pip3] triton==2.1.0
[conda] No relevant packages

cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @mrshenli @pritamdamania87 @zhaojuanmao @satgera @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang @kwen2501 @awgu @penguinwu @fegin @XilunWu @wanchaol @fduwjj @wz337 @tianyu-l @wconstab @yf225 @chauhang @d4l3k @bdhirsh @anijain2305 @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @chenyang78 @muchulee8 @ColinPeppler @amjames @desertfire @rohan-varma

[jataylo]

cc: @wconstab @ezyang @jansel

[LucasLLC]

Assigning high priority so we don't lose track of this issue. If this breaks some one's workflow, please feel free to remove