Remove POST restart API server

[epompeii]

This changeset remove the POST restart API server endpoint.

[github-actions]

🤖 Claude Code Review

PR: #714
Base: devel
Head: u/ep/post-restart
Commit: cf66e6ed7a7e64147484c8180832a64281578007

---

Here's my review:

---

PR Review: Remove POST /v0/server/restart endpoint and replace select! with Race

Commits: rm_post_restart, rm_tokio_select, claude_1

Summary

This PR does two things:

1. Removes the /v0/server/restart endpoint, CLI command, JSON type, and restart_tx channel from ApiContext
2. Replaces tokio::select! with futures_concurrency::future::Race in the API server's run function

Positive

• Clean removal across all layers (API, CLI, JSON types, OpenAPI spec, console docs, tests)
• Changelog entry correctly marks it as a BREAKING CHANGE
• Replacing select! with Race aligns with the CLAUDE.md guideline: "Avoid select! macros - use futures_concurrency::stream::Merge::merge"
• Removes the restart_tx channel from ApiContext, simplifying the shared context struct
• The loop {} with restart is gone, making the server lifecycle straightforward

Issues

1. Bug fix in backup.rs OPTIONS handler (good catch, but worth highlighting)

// Before: OPTIONS body type didn't match the POST endpoint
-    _body: TypedBody<JsonRestart>,
+    _body: TypedBody<JsonBackup>,

This was a pre-existing bug where the OPTIONS handler for /v0/server/backup used JsonRestart instead of JsonBackup. Good fix, but it's buried in the restart removal — consider calling it out in the PR description.

2. futures-concurrency and futures-util appear unused at the crate root (services/api/src/lib.rs:10-11)

use futures_concurrency as _;
use futures_util as _;

These deps are only used in main.rs (the binary), not in lib.rs. The use ... as _ lines suppress unused-crate warnings, but the dependencies are only needed by the binary target. This is fine for compilation but slightly misleading — a comment or moving these to main.rs would be cleaner. Since main.rs already has #![expect(unused_crate_dependencies)], these as _ imports in lib.rs are the workaround for the lib target. This is acceptable.

3. Race semantics — first future to complete wins, others are dropped

The Race tuple will resolve as soon as any future completes. If ctrl_c resolves first, the API server JoinHandle is dropped (which does NOT abort the spawned task — it just detaches it). This is the same behavior as the old select! so it's not a regression, but worth noting: the spawned API server task will continue running briefly after ctrl-c until the tokio runtime shuts down. This is fine.

4. The #[expect(clippy::too_many_arguments)] was removed from into_context (config_tx.rs:178)

Only clippy::too_many_lines remains. Verify this doesn't trigger a clippy warning now that the function has one fewer argument. With restart_tx removed, the arg count likely dropped below the threshold, so this is correct.

5. tokio moved to dev-dependency only in lib/api_server (lib.rs:6-7)

#[cfg(test)]
use tokio as _;

Good — tokio is no longer needed at runtime in api_server since the restart module (which used tokio::spawn and tokio::time::sleep) was removed. The #[cfg(test)] use tokio as _; suppresses the unused dev-dep warning.

6. Missing: cargo gen-types verification

Per CLAUDE.md: "When reviewing code, also check cargo gen-types (if the API changed at all)". The openapi.json diff shows the restart endpoint and JsonRestart schema were removed. Verify that the checked-in openapi.json matches what cargo gen-spec produces — it looks correct from the diff but should be confirmed.

7. Missing: bencher_client regeneration

Since the OpenAPI spec changed (removed /v0/server/restart), the auto-generated bencher_client should be regenerated. The CLI already removed its restart.rs module, but confirm bencher_client types are in sync.

Minor

• The BodyParametersInner.astro cleanup correctly removes the delay default for the now-removed restart/config endpoints.
• The server.mdx API docs correctly remove the restart endpoint entry.

Verdict

The changes are clean and well-scoped. The two action items are:

1. Confirm cargo gen-types produces the same openapi.json that's checked in
2. Confirm bencher_client is regenerated/consistent with the new spec

---

_{Model: claude-opus-4-6}

[github-actions]

Bencher Report

Branch	u/ep/post-restart
Testbed	ubuntu-22.04

Click to view all benchmark results

Benchmark	Latency	Benchmark Result microseconds (µs) (Result Δ%)	Upper Boundary microseconds (µs) (Limit %)
Adapter::Json	📈 view plot 🚷 view threshold	3.78 µs (+7.66%) Baseline: 3.51 µs	4.64 µs (81.36%)
Adapter::Magic (JSON)	📈 view plot 🚷 view threshold	3.78 µs (+8.43%) Baseline: 3.48 µs	4.56 µs (82.89%)
Adapter::Magic (Rust)	📈 view plot 🚷 view threshold	25.54 µs (-0.01%) Baseline: 25.54 µs	31.38 µs (81.37%)
Adapter::Rust	📈 view plot 🚷 view threshold	2.79 µs (-1.44%) Baseline: 2.83 µs	3.31 µs (84.27%)
Adapter::RustBench	📈 view plot 🚷 view threshold	2.79 µs (-1.44%) Baseline: 2.83 µs	3.30 µs (84.50%)
head_version_insert/batch/10	📈 view plot 🚷 view threshold	99.44 µs (-0.88%) Baseline: 100.32 µs	118.68 µs (83.79%)
head_version_insert/batch/100	📈 view plot 🚷 view threshold	238.71 µs (+0.26%) Baseline: 238.08 µs	267.22 µs (89.33%)
head_version_insert/batch/255	📈 view plot 🚷 view threshold	459.48 µs (-0.77%) Baseline: 463.03 µs	500.02 µs (91.89%)
head_version_insert/batch/50	📈 view plot 🚷 view threshold	160.84 µs (-0.29%) Baseline: 161.30 µs	184.24 µs (87.30%)
threshold_query/join/10	📈 view plot 🚷 view threshold	143.67 µs (-0.98%) Baseline: 145.09 µs	170.51 µs (84.26%)
threshold_query/join/20	📈 view plot 🚷 view threshold	158.88 µs (-0.43%) Baseline: 159.57 µs	187.90 µs (84.55%)
threshold_query/join/5	📈 view plot 🚷 view threshold	136.07 µs (-0.88%) Baseline: 137.28 µs	161.31 µs (84.35%)
threshold_query/join/50	📈 view plot 🚷 view threshold	199.60 µs (-0.66%) Baseline: 200.93 µs	233.36 µs (85.53%)

🐰 View full continuous benchmarking report in Bencher