fix: convert conditional disagg sglang warning to httperror 400

[karen-sy]

Overview:

The SGLang backend correctly rejects the conditional-disaggregation annotation x-bypass-remote-prefill and raises a message written specifically for the user:

Detected request annotation 'x-bypass-remote-prefill', but SGLang backend does not support conditional disaggregation yet. Use vLLM or TensorRT-LLM for conditional disaggregation.

That message is present in full in both the worker log and the frontend log, but the client receives:

{"message":"Internal server error","type":"Internal Server Error","code":500} The backend expresses the rejection as a bare RuntimeError, which the frontend classifies as 5xx, and 5xx bodies are deliberately replaced with a generic message by SanitizedError. The result: the one sentence that tells the user which backend to switch to never reaches them.

From the caller's point of view a self-correctable configuration mistake presents as an unattributable internal server error.

Details:

Where should the reviewer start?

Related Issues

⚠️ This section is required. Choose one path below and delete the other.

🔗 This PR is linked to an issue:

• closes DYN-3727

---

Summary by CodeRabbit

• Bug Fixes
  • Improved error handling for unsupported conditional disaggregation bypass requests.
  • Requests now return a clear HTTP 400 response with guidance to use vLLM or TensorRT-LLM instead.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[datadog-official]

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 15.21% (-32.37%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 75b192d | Docs | Datadog PR Page | Give us feedback!}

[coderabbitai]

Walkthrough

Changes

Decode handler error handling

Layer / File(s)	Summary
Return HTTP validation errors components/src/dynamo/sglang/request_handlers/llm/decode_handler.py	The handler raises HttpError(400, ...) instead of RuntimeError and identifies vLLM and TensorRT-LLM as supported backends.

Estimated code review effort: 1 (Trivial) | ~3 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly describes converting the SGLang conditional disaggregation failure to an HTTP 400 error.
Description check	✅ Passed	The description explains the problem and intended fix, and identifies the related issue, but leaves Details and reviewer-start guidance incomplete.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

components/src/dynamo/sglang/request_handlers/llm/decode_handler.py (1)

50-54: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Document the new HttpError path.

generate now raises HttpError for this request condition, but its Raises section still lists only RuntimeError on Lines 354-356. Add the HTTP 400 behavior to the docstring.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@components/src/dynamo/sglang/request_handlers/llm/decode_handler.py` around
lines 50 - 54, Update the generate method’s docstring Raises section to document
the new HttpError behavior: it raises HTTP 400 when
BYPASS_REMOTE_PREFILL_ANNOTATION is requested but unsupported by the SGLang
backend. Preserve the existing RuntimeError documentation.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@components/src/dynamo/sglang/request_handlers/llm/decode_handler.py`:
- Around line 50-54: Update the streaming HTTP configuration so the
status-preserving pre-commit error peek is enabled by default, allowing the
HttpError raised in DecodeWorkerHandler.generate for
BYPASS_REMOTE_PREFILL_ANNOTATION to produce HTTP 400 instead of a committed 200.
Add an endpoint test covering x-bypass-remote-prefill that asserts the 400
response and includes the conditional-disaggregation guidance message.

---

Nitpick comments:
In `@components/src/dynamo/sglang/request_handlers/llm/decode_handler.py`:
- Around line 50-54: Update the generate method’s docstring Raises section to
document the new HttpError behavior: it raises HTTP 400 when
BYPASS_REMOTE_PREFILL_ANNOTATION is requested but unsupported by the SGLang
backend. Preserve the existing RuntimeError documentation.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: e12f2c19-4edc-4794-94e1-1d2e6413c840

📥 Commits

Reviewing files that changed from the base of the PR and between 74bf051 and 0db2a74.

📒 Files selected for processing (1)

• components/src/dynamo/sglang/request_handlers/llm/decode_handler.py