feat(vllm): adopt split gRPC services and Control discovery

[connorcarpenter15]

Overview:

Part 1 of a 4-PR stack. This foundation moves the Dynamo vLLM sidecar and mocker to vLLM's split native gRPC services and discovers runtime metadata through Control.

Details:

• Vendor the split vLLM Inference and Control protocols pinned to 2d2c3af18c52e8e4efa4b0b4903843b15c0dba0e.
• Move the vLLM mocker to split inference, control, and health services; all mocker changes in the stack are confined to this PR.
• Discover model identity, capacity, parallelism, KV-event sources, and readiness through Control.
• Remove required model-path configuration and avoid advertising parser defaults the protocol cannot honor.
• Use the shared sidecar transport while preserving existing launch, deployment, and documentation surfaces.

Deterministic DP/KV routing, cancellation handoff behavior, and multimodal conversion remain in later PRs.

Stack

1. #12734 — split gRPC services and Control discovery
2. #12736 — preserve decode handoff on cancellation
3. #12735 — deterministic DP and KV routing
4. #12214 — multimodal sidecar requests

Base: main

Validation

• cargo fmt --all -- --check
• cargo test -p dynamo-vllm-sidecar -p dynamo-vllm-mocker — 30 tests passed
• cargo clippy -p dynamo-vllm-sidecar -p dynamo-vllm-mocker --all-targets -- -D warnings

Where should the reviewer start?

• lib/sidecar/vllm/src/client.rs for health and startup retry behavior.
• lib/sidecar/vllm/src/model.rs for Control metadata validation and identity checks.
• lib/sidecar/vllm/src/engine.rs for bootstrap discovery and shared transport startup.
• lib/mocker/servers/vllm/src/server.rs for the split-service mock contract.

Related Issues

This PR is NOT linked to an issue:

• Confirmed — no related issue

[github-actions]

🌿 Fern Docs Preview: https://nvidia-preview-019fe2fb-a384-70ce-8261-6747b8f3e501.docs.buildwithfern.com/dynamo/dev

[datadog-official]

🎯 Code Coverage (details)
• Patch Coverage: 50.00%
• Overall Coverage: 47.07% (-5.28%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 1b1b290 | Docs | Datadog PR Page | Give us feedback!}

[devin-ai-integration]

Devin Review found 4 potential issues.

[coderabbitai]

Walkthrough

Changes

vLLM API migration and discovery

Layer / File(s)	Summary
Protocol contracts and code generation .github/workflows/copyright-check.ps1, .pre-commit-config.yaml, lib/sidecar/vllm/proto/*, lib/sidecar/vllm/build.rs, lib/*/Cargo.toml	Replaced the legacy protobuf with separate Inference and Control services. Added multimodal request fields, metadata messages, health support, and updated build and tooling configuration.
Mock Inference, Control, and health services lib/mocker/servers/vllm/src/*, lib/mocker/servers/vllm/tests/sidecar.rs, lib/mocker/servers/vllm/README.md	Updated the mock server and tests to register both services, report health, expose metadata, abort requests, and validate data-parallel ranks.
Health checks and metadata discovery lib/sidecar/vllm/src/client.rs, lib/sidecar/vllm/src/model.rs	Added deadline-bounded service health checks and Control API discovery for model and server metadata.
Discovered model engine startup lib/sidecar/vllm/src/engine.rs, lib/sidecar/vllm/src/convert.rs, lib/sidecar/vllm/src/tests.rs	The engine validates discovered metadata, checks identity consistency, propagates server limits, and sends the discovered served model name.
Launch, deployment, and executable updates lib/sidecar/vllm/README.md, lib/sidecar/vllm/deploy/*, lib/sidecar/vllm/launch/*, lib/sidecar/vllm/tests/executable.rs	Removed --model-path from launch commands and updated documentation and help checks for Control-based discovery and health checks.

Estimated code review effort: 4 (Complex) | ~60 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 27.12% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
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.
Title check	✅ Passed	The title clearly and concisely summarizes the main change: adopting split vLLM gRPC services and Control-based discovery.
Description check	✅ Passed	The description includes all required sections, explains the changes, identifies review starting points, and confirms that no issue is linked.

---

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

[coderabbitai]

Actionable comments posted: 3

🧹 Nitpick comments (2)

lib/sidecar/vllm/src/model.rs (1)

63-71: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Name the field that changed in the identity error.

ModelIdentity also compares aliases, reasoning_parser, and tool_call_parser. If only one of those changes, the message prints identical source and served names, which makes the failure hard to diagnose. Include the two identity values in the message, for example with {:?} on self.identity and observed.identity.

🤖 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 `@lib/sidecar/vllm/src/model.rs` around lines 63 - 71, Update
ModelIdentity::ensure_same_identity to include the complete expected and
observed identity values in the protocol error message, using debug formatting
for self.identity and observed.identity. Preserve the existing source and
served-name context while making changes to aliases, reasoning_parser, or
tool_call_parser visible.

lib/sidecar/vllm/src/tests.rs (1)

175-186: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Add an identity-mismatch startup test.

After bootstrap, set server.service.model_info_override to a different model and assert that engine.start(0) returns model identity changed between bootstrap and startup.

🤖 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 `@lib/sidecar/vllm/src/tests.rs` around lines 175 - 186, Add a test covering
identity mismatch after bootstrap: update server.service.model_info_override
with a different model than the bootstrapped one, then assert engine.start(0)
fails with the error message “model identity changed between bootstrap and
startup.” Reuse the existing bootstrap, server, and model-construction helpers
in the surrounding tests.

🤖 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 `@lib/mocker/servers/vllm/src/server.rs`:
- Around line 89-98: Update the ModelInfo construction to set
supports_text_input to false, matching PreparedRequest::new’s rejection of
Prompt::Text; leave token-ID input capability unchanged.

In `@lib/sidecar/vllm/src/engine.rs`:
- Around line 254-279: Document the synchronous execution constraint on
VllmSidecarEngine::from_args and its bootstrap_discover Runtime::block_on path.
State that it must be called before dynamo_backend_common::run in the production
entry point, while asynchronous callers must invoke it through spawn_blocking or
an equivalent dedicated thread.

In `@lib/sidecar/vllm/src/model.rs`:
- Around line 90-101: Update the required and nonempty helpers to return the
trimmed value after validation, while preserving their existing error and None
behavior for blank inputs. Ensure GenerateRequest.model receives the normalized
model name without surrounding whitespace.

---

Nitpick comments:
In `@lib/sidecar/vllm/src/model.rs`:
- Around line 63-71: Update ModelIdentity::ensure_same_identity to include the
complete expected and observed identity values in the protocol error message,
using debug formatting for self.identity and observed.identity. Preserve the
existing source and served-name context while making changes to aliases,
reasoning_parser, or tool_call_parser visible.

In `@lib/sidecar/vllm/src/tests.rs`:
- Around line 175-186: Add a test covering identity mismatch after bootstrap:
update server.service.model_info_override with a different model than the
bootstrapped one, then assert engine.start(0) fails with the error message
“model identity changed between bootstrap and startup.” Reuse the existing
bootstrap, server, and model-construction helpers in the surrounding tests.

🪄 Autofix

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: bd81be70-6b4b-4b97-8e51-4888cd03fadc

📥 Commits

Reviewing files that changed from the base of the PR and between b6d6480 and 663fa53.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock

📒 Files selected for processing (26)

• .github/workflows/copyright-check.ps1
• .pre-commit-config.yaml
• lib/mocker/servers/vllm/Cargo.toml
• lib/mocker/servers/vllm/README.md
• lib/mocker/servers/vllm/src/main.rs
• lib/mocker/servers/vllm/src/server.rs
• lib/mocker/servers/vllm/src/server_request.rs
• lib/mocker/servers/vllm/src/server_tests.rs
• lib/mocker/servers/vllm/tests/sidecar.rs
• lib/sidecar/vllm/Cargo.toml
• lib/sidecar/vllm/README.md
• lib/sidecar/vllm/build.rs
• lib/sidecar/vllm/deploy/agg.yaml
• lib/sidecar/vllm/deploy/disagg.yaml
• lib/sidecar/vllm/launch/agg.sh
• lib/sidecar/vllm/launch/disagg.sh
• lib/sidecar/vllm/proto/README.md
• lib/sidecar/vllm/proto/control.proto
• lib/sidecar/vllm/proto/inference.proto
• lib/sidecar/vllm/src/args.rs
• lib/sidecar/vllm/src/client.rs
• lib/sidecar/vllm/src/convert.rs
• lib/sidecar/vllm/src/engine.rs
• lib/sidecar/vllm/src/model.rs
• lib/sidecar/vllm/src/tests.rs
• lib/sidecar/vllm/tests/executable.rs

💤 Files with no reviewable changes (3)

• lib/sidecar/vllm/tests/executable.rs
• lib/sidecar/vllm/src/args.rs
• lib/sidecar/vllm/launch/disagg.sh

[PeaBrane]

✅ General approval, conditional on resolving the inline comments before merge. Please also replace the stale vLLM gRPC v0.25.1 validation wording in lib/sidecar/vllm/src/convert.rs with feature-based wording.

[tanmayv25]

Automated multi-agent code-review pass over this PR's diff (framing + expert lenses, each finding adversarially verified). Findings below are all minor / non-blocking.