[sglang] fix: HTTP server startup issues for Prometheus and Grafana integration (verl-project#4408)

### What does this PR do?

> Resolve the issue that Prometheus and Grafana cannot monitor relevant
metrics during rollout using SGLang.

### Problems

When using SGLang as the rollout engine, the HTTP server fails to start
properly. As a result, Prometheus cannot scrape metrics and Grafana is
unable to display them.

### Details

Set the **warmup_thread_args** property: Configure
app.warmup_thread_args before starting the server to satisfy SGLang’s
lifespan requirements.

Skip server warmup: Enable **skip_server_warmup=True** to avoid
connection issues caused by the warmup thread. Since the model is
already loaded in _launch_subprocesses, no additional warmup is
required.

Add **Prometheus middleware** early: Manually call
add_prometheus_middleware before invoking run_unvicorn to ensure that
the /metrics endpoint is available immediately when the server starts.

### Checklist Before Starting

- [ ] Search for similar PRs. Paste at least one query link here:
verl-project#4088
- [ ] Format the PR title as `[{modules}] {type}: {description}` (This
will be checked by the CI)
- `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`,
`trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`,
`ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`,
`env`, `tool`, `ckpt`, `doc`, `data`
- If this PR involves multiple modules, separate them with `,` like
`[megatron, fsdp, doc]`
  - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test`
- If this PR breaks any API (CLI arguments, config, function signature,
etc.), add `[BREAKING]` to the beginning of the title.
  - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching`

### Test

> For changes that can not be tested by CI (e.g., algorithm
implementation, new model support), validate by experiment(s) and show
results like training curve plots, evaluation results, etc.
<img width="595" height="246" alt="截屏2025-12-04 下午3 00 59"
src="https://github.com/user-attachments/assets/3aed1690-172c-4857-b880-5af5371646bf"
/>
<img width="595" height="246" alt="截屏2025-12-04 下午3 01 05"
src="https://github.com/user-attachments/assets/517077d7-7c96-4ac9-a2c5-0f06ea6701b3"
/>


### API and Usage Example

> Demonstrate how the API changes if any, and provide usage example(s)
if possible.

```python
# Add code snippet or script demonstrating how to use this
```

### Design & Code Changes

> Demonstrate the high-level design if this PR is complex, and list the
specific changes.

### Checklist Before Submitting

> [!IMPORTANT]
> Please check all the following items before requesting a review,
otherwise the reviewer might deprioritize this PR for review.

- [ ] Read the [Contribute
Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md).
- [ ] Apply [pre-commit
checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting):
`pre-commit install && pre-commit run --all-files --show-diff-on-failure
--color=always`
- [ ] Add / Update [the
documentation](https://github.com/volcengine/verl/tree/main/docs).
- [ ] Add unit or end-to-end test(s) to [the CI
workflow](https://github.com/volcengine/verl/tree/main/.github/workflows)
to cover all the code. If not feasible, explain why: ...
- [ ] Once your PR is ready for CI, send a message in [the `ci-request`
channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the
`verl` Slack
workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ).
(If not accessible, please try [the Feishu group
(飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

---------

Co-authored-by: jsfanfanfan <2981866535@qq.com>
Co-authored-by: AniZpZ <aniz1905@gmail.com>

Author: 3 people

docs/advance/grafana_prometheus.md

@@ -2,18 +2,13 @@
 
 **Author:** `https://github.com/meituan-search`
 
-Last updated: 11/12/2025.
+Last updated: 12/05/2025.
 
 Monitor the rollout computation process using Prometheus and Grafana when using verl to enhance system observability and facilitate further performance optimization.
 
 We provide an additional training monitoring capability, leveraging Prometheus and Grafana to display rollout information during training and enhance system observability to facilitate further performance optimization.
 
 The system automatically configures Prometheus to scrape metrics from rollout servers, eliminating manual configuration steps.
-    
-> Note: SGLang integration is currently under development and debugging. Some features may not
-> work as expected and the implementation is subject to change. Please use with caution
-> in production environments.
-
 
 
 ## Overview

verl/workers/rollout/sglang_rollout/async_sglang_server.py

@@ -194,8 +194,21 @@ async def launch_server(self, master_address: str = None, master_port: int = Non
             )
         )
         app.is_single_tokenizer_mode = True
-        app.server_args = server_args
-        app.warmup_thread_args = (server_args, None, None)
+
+        # Set warmup_thread_args to avoid AttributeError in lifespan function
+        app.warmup_thread_args = (
+            server_args,
+            None,
+            None,
+        )
+
+        # Manually add Prometheus middleware before starting server
+        # This ensures /metrics endpoint is available immediately
+        if server_args.enable_metrics:
+            from sglang.srt.utils.common import add_prometheus_middleware
+
+            add_prometheus_middleware(app)
+
         self._server_port, self._server_task = await run_unvicorn(app, server_args, self._server_address)
         self.tokenizer_manager.server_status = ServerStatus.Up