feat(grpc-router): use per-model retry config from WorkerRegistry

[CatherineSue]

Description

Part of the per-worker resilience refactor series: #799 → #803 → #821 → #836 → #875 → #881 → this PR.

Problem

gRPC router uses a single global retry_config for all requests, ignoring per-model retry config set by workers.

Solution

Look up per-model retry config from WorkerRegistry at request time for all 3 endpoints (chat, generate, messages), falling back to the router-level default.

Independent of #881 (HTTP router migration).

Changes

• 3 call sites in grpc/router.rs updated to read from WorkerRegistry

Test Plan

• cargo test -p smg --lib — all 450 tests pass
• Pre-commit hooks pass (rustfmt, clippy, codespell, DCO)

Checklist

• cargo +nightly fmt passes
• cargo clippy --all-targets --all-features -- -D warnings passes
• (Optional) Documentation updated
• (Optional) Please join us on Slack #sig-smg to discuss, review, and merge PRs

Summary by CodeRabbit

Refactor

• Retry configuration is now selected on a per-model basis for chat, generate, and messages routes instead of using a single router-level setting.

Summary by CodeRabbit

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: 01001d64-8349-48d6-b7f1-1b4c49196318

📥 Commits

Reviewing files that changed from the base of the PR and between 7310a1d and 60eff9a.

📒 Files selected for processing (1)

• model_gateway/src/routers/grpc/router.rs

---

📝 Walkthrough

Walkthrough

The gRPC router now implements per-model retry configuration selection across chat, generate, and messages routes. For each request, it queries the worker registry for model-specific retry configuration, uses it if available, otherwise falls back to the router's default configuration, and passes the resolved config to the retry executor.

Changes

Cohort / File(s)	Summary
gRPC Router Retry Config Selection model_gateway/src/routers/grpc/router.rs	Added per-model retry config resolution by querying worker_registry.get_retry_config(model_id) in chat, generate, and messages routes; routes now conditionally use model-specific retry config or fall back to self.retry_config.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• feat(core): add per-model retry config to WorkerRegistry #821: Adds per-model retry storage and get_retry_config/get_retry_enabled APIs in WorkerRegistry that this change directly depends on.
• feat(core): wire per-worker resilience and HTTP client into BasicWorker #803: Wires per-worker resilience and retry configuration into workers, exposing the retry-related APIs the router now uses.
• feat(core): add per-worker resilience and HTTP pool config types #799: Introduces per-worker resilience types and resolution that enable the per-model retry configuration selection implemented here.

Suggested labels

model-gateway

Suggested reviewers

• key4ng
• slin1237

Poem

🐰 A rabbit hops through retry schemes so fine,
Per-model configs? Now that's divine!
When one path fails, another shall do,
Flexible retries, tried and true! ✨

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately reflects the main change: implementing per-model retry configuration lookup from WorkerRegistry in the gRPC router, which is the primary focus of the changeset.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/grpc-router-retry-from-registry

---

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

[gemini-code-assist]

Code Review

This pull request introduces per-model retry configurations within the gRPC router, enabling the system to prioritize model-specific settings over global defaults. The logic for resolving the effective retry configuration is duplicated across several methods. It is recommended to extract this logic into a private helper method, possibly utilizing std::borrow::Cow to efficiently handle both owned and borrowed configuration data.