Skip to content

[sglang,rollout] fix: sglang port race condition #3977

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Nov 4, 2025
Merged

[sglang,rollout] fix: sglang port race condition #3977

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
487b352
pre assign nccl port
theely Oct 31, 2025
356052c
Update verl/workers/rollout/sglang_rollout/sglang_rollout.py
theely Oct 31, 2025
8bd7055
fix format
theely Nov 3, 2025
a6c16a4
Merge branch 'Fix-sglang-port-race-condition' of github.com:theely/ve…
theely Nov 3, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions verl/workers/rollout/sglang_rollout/sglang_rollout.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -434,8 +434,8 @@ def _init_inference_engine(self, trust_remote_code, actor_module, port):
if self.config.mode == "async" and not self.config.skip_tokenizer_init:
raise ValueError("async mode requires skip_tokenizer_init to be True")
backend = attention_backend if attention_backend is not None else "fa3"
sglang_port = int(os.getenv("SGLANG_PORT", "30000")) + (dist.get_rank() * 2)
if effective_first:
rank = dist.get_rank()
os.environ["SGLANG_BLOCK_NONZERO_RANK_CHILDREN"] = "0"
args = {
"model_path": actor_module,
Expand All @@ -453,7 +453,8 @@ def _init_inference_engine(self, trust_remote_code, actor_module, port):
"max_running_requests": max_running_requests,
# NOTE(linjunrong): add rank to prevent SGLang generate same port inside PortArgs.init_new
# when random.seed is being set during training
"port": 30000 + rank,
"port": sglang_port,
"nccl_port": sglang_port + 1,
# NOTE(Chenyang): if you want to debug the SGLang engine output
# please set the following parameters
# Otherwise, it will make the engine run too slow
Expand Down
Loading