SGLang is a framework for running inference for large language models and multimodal models, meaning it processes inputs to produce model outputs such as text or other media. It is used to serve and accelerate open AI models and related workloads.
Borrowing it
Nothing to install: this file belongs to sgl-project/sglang. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/sgl-project/sglang/main/.agents/skills/debug-distributed-hang/SKILL.mdgit clone --depth 1 https://github.com/sgl-project/sglangWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/sgl-project/sglang/debug-distributed-hang)<a href="https://agentmods.dev/skills/sgl-project/sglang/debug-distributed-hang"><img src="https://agentmods.dev/badge/skills/sgl-project/sglang/debug-distributed-hang/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/sgl-project/sglang/debug-distributed-hang"><img src="https://agentmods.dev/badge/skills/sgl-project/sglang/debug-distributed-hang.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- Snyk pass
- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00085 | $0.02427 |
| Opus 5.5 | $0.00034 | $0.00971 |
| Sonnet 5.5 | $0.00017 | $0.00485 |
| Haiku 4.5 | $0.00009 | $0.00243 |
Grade A, and why
debug-distributed-hang scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 249 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Debugging Distributed Hangs in SGLang
Overview
Hangs in distributed inference happen when ranks diverge in state, causing collective operations (AllGather, AllReduce, Broadcast, Barrier) to deadlock. Common causes:
- Size mismatch: ranks pass different tensor sizes to a collective
- Branch divergence: one rank enters a collective, another skips it
- Cascading state drift: a small non-determinism (e.g., floating-point) propagates into different batch structures
- Resource exhaustion: one rank OOMs or crashes, others wait forever
Prerequisites
- py-spy:
pip install py-spyor system package. Requires root orCAP_SYS_PTRACEto attach to running processes. - cuda-gdb: Ships with the CUDA toolkit. Ensure it's on your
PATH.
Step 1: Confirm and Locate the Hang
1a. Watchdog / py-spy
SGLang's watchdog automatically dumps py-spy traces on timeout. Look for:
Scheduler watchdog timeout (self.watchdog_timeout=300, self.soft=False)
The py-spy dump shows the stack trace of each thread. The hanging thread is typically blocked in a CUDA synchronize or NCCL collective:
Thread (active): "MainThread"
cuStreamSynchronize (libcuda.so)
...
forward_extend (model_runner.py)
SGLang has two watchdog modes (see python/sglang/srt/utils/watchdog.py):
- Hard watchdog (
soft=False, default): dumps py-spy traces then sendsSIGQUITto kill the parent process. - Soft watchdog (
soft=True): only logs the timeout without killing the process, giving you more time to manually attach debuggers or collect coredumps.
If the watchdog doesn't trigger, manually dump:
py-spy dump --pid <scheduler_pid>
1b. NCCL Debug Logging
export NCCL_DEBUG=INFO
export NCCL_DEBUG_SUBSYS=COLL
Look for the last collective logged before the hang. Mismatched sizes show up as one rank waiting and another never entering.
1c. CUDA Coredump
When a process hangs, you can trigger a GPU coredump on demand to see which kernel is stuck. Set these env vars before launching:
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2mo ago First seen · 249 lines · 85 tokens per session scan A b2287588e589
debug-distributed-hang is a skill published in the GitHub repository sgl-project/sglang (38,324 stars, last pushed today), licensed Apache-2.0. It adds 85 tokens to every session and 2,427 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
perf-analyze
Launch and operate this repo's perf-analyze workflow, which DIAGNOSES a TensorRT-LLM serving deployment without applying changes — benchmark at one concurrency or a Pareto curve of them (tok/s/user vs tok/s/gpu), analytical SOL projection on by default (via the internal-perf-sol-analysis skill), nsys + ncu per-kernel…
systematic-debugging
4-phase root cause debugging: understand bugs before fixing.
python-debugpy
Debug Python: pdb REPL + debugpy remote (DAP).
resolve-handoff
Emit the complete Resolve handoff with exact modes, outcomes, test evidence, and publication state.
iterate-implementation
Only use when the user explicitly invokes this skill by name.
debug-with-langwatch
Root-cause production errors and misbehaving agent runs with LangWatch. Finds errored traces, inspects spans, checks monitor and evaluator scores, then narrows to a root cause. Use when something is failing or misbehaving in production (errors, bad answers, latency spikes).