docs: Rust servicer selection ships in the v1.12.0 wheels and engine images; docs still say current main only

[github-actions]

What this PR does

Can I select the Rust gRPC servicer for vLLM, SGLang and TokenSpeed workers with the released v1.12.0 wheels and engine images, and which selection surface does v1.12.0 have for each engine (vLLM: the --servicer-impl flag and SMG_VLLM_SERVICER_IMPL; SGLang and TokenSpeed: the environment variable only) — or is the selection still unreleased, main-only, as the page says?

Why we need it

Source change: smg-project/smg@dcc5cb0

Documentation targets the pinned published stable release.

Revisited pending concern, every claim re-verified tonight in the pinned v1.12.0 checkout. getting-started/grpc-workers.md states three times that the Rust servicer selection 'exists on current main but not in the v1.11.0 wheels or engine images' (:111 vLLM, :152 SGLang, :176 TokenSpeed). All of it shipped in v1.12.0: .nightly-source/grpc_servicer/smg_grpc_servicer/vllm/rust.py (SERVICER_IMPL_ENV='SMG_VLLM_SERVICER_IMPL' :70, resolve_servicer_impl :343-360 reading args.servicer_impl ahead of the environment and writing the decision back) with vllm/plugin.py (:27-40, the --servicer-impl {python,rust} flag added to parsers that define --grpc as a vllm.general_plugins plugin) and vllm/launcher_switch.py; sglang/rust.py:45 (SERVICER_IMPL_ENV='SMG_SGLANG_SERVICER_IMPL', env-only resolve :52-58) consulted by sglang/server.py:114; tokenspeed/rust.py:48-61 (SERVICER_IMPL_ENV='SMG_TOKENSPEED_SERVICER_IMPL', resolve_servicer_impl reads only the environment) consulted by tokenspeed/main.py:44 — no --servicer-impl argparse flag anywhere in the tokenspeed launcher. bindings/python/pyproject.toml version = 1.12.0; bindings/python/src/smg/servicer.py exports VllmGrpcServer/SglangGrpcServer/TokenSpeedGrpcServer (:26-32); bindings/python/src/smg/serve.py:827 defines smg serve --servicer-impl and :937-941 rejects it outside --backend vllm --connection-mode grpc. Primary commit 1 (dcc5cb0b, smg#2888, diff read) made the sentences stale: it sets every release-*-docker.yml smg_commit default from v1.11.0 to v1.12.0 and bumps the wheel, so the push-triggered release wheels/images were built from this tree; the servicer commits (IDs 110/99/100/104 = smg#2727/#2755/#2750/#2747) are all first-parent-reachable from the tag. TokenSpeed nuance verified for the same edit: tokenspeed/rust.py:155 resolves KV events only from --kv-events-config (kv_events.py:39-43 returns None without it — no auto-enable), and the only tuning envs in the release are HANDSHAKE_PORT/DRAIN_SECS/STARTUP_TIMEOUT_SECS (rust.py:49-51); SMG_TOKENSPEED_SERVICER_STARTUP_CEILING_SECS is absent from the release tree — so :176 must name the environment variable as v1.12.0's selection surface, and the :194 'Unreleased:' note (the TokenSpeed --servicer-impl flag, the automatic KV-events default, the startup ceiling) plus the :205/:207 sentences describing them as current must be reconciled to the page's newer-than-v1.12.0 idiom, with v1.12.0 as exactly the build that carries the switch but predates them. This is one independent concern: one release event made one page's three identical unreleased-availability claims false; per-engine selection semantics were already documented by merged PRs #105/#115/#126/#143 and are not re-litigated. Must not document the SGLang launcher --servicer-impl flag (closed release-deferred #144, post-v1.12.0 commit absent from the supplied history) or any post-release TokenSpeed addition as released. Dedup: no open PR holds grpc-workers.md (open #174 holds contributing/development.md; #168/#164 hold tls/logging/authentication/hardening/configuration files); merged #166 covered only getting-started/index.md's engine-image matrix, and index.md's servicer mentions (:172, :300, :323, :355, re-read tonight) use the 'newer than v1.11.0' idiom that stays literally true for a v1.12.0 feature.

Scope: cli-configuration / rust-servicer-selection-in-release-wheels. Other concerns are deferred.

How to test

• Passed the documentation path and size guard (under 1000 added plus deleted lines; no file-count limit).
• Passed an independent accuracy and single-concern review.
• Passed git diff --check and pnpm check and the production pnpm build.

Checklist

• Tests added/updated (if applicable)
• Docs updated (if applicable)
• Commit includes a DCO sign-off

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: f2f1dff1-a800-4afd-9248-5438c3c3aef2

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

[github-actions]

Documentation maintenance: ready. Unsuccessful content rounds: 0/3. Operational attempts: 0/3.

Head: f124601f258edd3b04ae6d262b14bb18aef978b9; reviewed main: 105d7201ae6dd49f23bb2f0c7e490505f6ccc54a.
Reviewed SMG source: dcc5cb0b8f2a64694938c10f02c6caaa0c40d5ef.

All edits serve the single concern (one release event made the page's three identical unreleased-availability claims stale) on the one allowlisted page, ~40 changed lines. Every claim verified in the pinned v1.12.0 checkout: vLLM selection via the smg-servicer vllm.general_plugins flag plus SMG_VLLM_SERVICER_IMPL with flag-ahead-of-env and write-back (vllm/plugin.py:27-49, vllm/rust.py:343-361); SGLang and TokenSpeed env-only (sglang/rust.py:45-59 via server.py:114; tokenspeed/rust.py:48-62 via __main__.py:44, no launcher flag); smg serve --servicer-impl scoped to vllm+grpc with fail-fast hook check (serve.py:826-837, 888-897, 937-948). The removed TokenSpeed material (launcher flag, flag precedence, auto KV-events default, SMG_TOKENSPEED_SERVICER_KV_EVENTS, SMG_TOKENSPEED_SERVICER_STARTUP_CEILING_SECS, the Unreleased note) is absent from the release tree; KV events resolve only from --kv-events-config (tokenspeed/kv_events.py:36-45). The tuning paragraph matches source defaults (free port/5/1800; rust_lifecycle.py:22-27) and states poll-based dead-engine detection without a wall-clock promise (rust_lifecycle.py:256-260, engine.rs:49-57). The 'requires SMG v1.12.0 or newer wheels and engine images' boundary is supported by the pinned tag checkout plus the source commit flipping release wheel/image defaults from v1.11.0 to v1.12.0, and sits in its own short paragraph in each section. Related pages (getting-started/index.md:172/300/323/355, concepts/architecture/multimodal.md:152) use the 'newer than v1.11.0' idiom, which stays literally true for a v1.12.0 feature and links to the corrected canonical sections, so no out-of-allowlist edit is needed. No threads exist in the PR feedback, so none are reported addressed. Minor note: the context.json file snapshot differs slightly from the actual PR head in the final tuning-paragraph wording; the patch and working tree agree and were reviewed as authoritative.

Workflow evidence

Human review threads and CODEOWNER approval remain under repository policy.