Add OCI registry URL to the API server config

[epompeii]

This changset adds a url to the the plus.registry section of the API server configuration.
If not specified, it will point to http://localhost:61016 by default for Bencher Self-Hosted instances.

[github-actions]

🤖 Claude Code Review

PR: #676
Base: main
Head: u/ep/registry-url
Commit: 8f8a1a3a4b9095dc20d076cbc5673eaac51e8d8a

---

I now have a thorough understanding of the changes. Here's my review:

---

PR Review: registry_url + optional_data_store

Commits: 6a99a337 (registry_url) and 8f8a1a3a (optional_data_store)

Summary

Two related changes to the OCI registry configuration:

1. Add a configurable url field to JsonRegistry for the externally-reachable registry URL
2. Make data_store optional and add an explicit Local variant to RegistryDataStore

Positives

• Good defaults: Self-hosted gets localhost:61016, Bencher Cloud gets registry.bencher.dev -- clean fallback logic in default_registry_url()
• Thorough documentation: All 9 language doc files updated consistently, example config updated
• Clean error handling: New PlusError::RegistryUrl variant with thiserror, consistent with project patterns
• OpenAPI spec regenerated: openapi.json updated correctly with nullable fields and the new Local variant
• PLAN.md updated: Completed tasks properly checked off and renumbered

Issues

1. LOCALHOST_BENCHER_REGISTRY_URL_STR reuses the API server address

lib/bencher_json/src/lib.rs:205:

pub const LOCALHOST_BENCHER_REGISTRY_URL_STR: &str = "http://localhost:61016";

This is the same as LOCALHOST_BENCHER_API_URL_STR (line 164). If the registry is served by the same API server (at /v2/), this is correct. But calling it a "registry URL" while pointing to the general API server could be confusing. A doc comment clarifying that the registry is hosted on the same origin (at /v2/ paths) would help future readers.

2. PROD_BENCHER_REGISTRY_URL is defined but never referenced

lib/bencher_json/src/lib.rs:228-232 defines PROD_BENCHER_REGISTRY_URL but nothing in the codebase uses it -- only BENCHER_REGISTRY_URL and LOCALHOST_BENCHER_REGISTRY_URL are actually consumed. Consider removing it to avoid dead code, or add #[expect(dead_code)] with a reason if it's intended for future use.

3. registry_url() accessor is defined but never called

lib/bencher_schema/src/context/mod.rs:134-137:

pub fn registry_url(&self) -> &Url {
    &self.registry_url
}

No code calls context.registry_url() yet. The PLAN.md indicates this is for a future step (building JsonJobConfig). This is fine as scaffolding, but worth noting that the accessor is currently dead code.

4. data_store.service documented as "Required: Yes" but data_store itself is optional

In the docs (e.g., en/plus-registry.mdx), the table says data_store is "Required: No" (local if omitted), but data_store.service is "Required: Yes". This is internally consistent (if you provide data_store, you must specify the service tag) but could confuse readers. Consider a note like: "Required when data_store is provided."

5. Minor: debug_assertions pattern for BENCHER_REGISTRY_URL_STR

#[cfg(debug_assertions)]
pub const BENCHER_REGISTRY_URL_STR: &str = LOCALHOST_BENCHER_REGISTRY_URL_STR;
#[cfg(not(debug_assertions))]
pub const BENCHER_REGISTRY_URL_STR: &str = PROD_BENCHER_REGISTRY_URL_STR;

This follows the existing pattern for BENCHER_API_URL_STR, so it's consistent. Just noting that this means the constant switches between localhost/production based on debug vs release mode, which is the established convention.

6. Breaking change to JsonRegistry serialization

data_store went from required (RegistryDataStore) to optional (Option<RegistryDataStore>). Existing config files with "data_store": { "service": "aws_s3", ... } will still deserialize fine. But if any code was generating configs that relied on data_store being non-optional, the serialization of None will differ (field absent vs. present). Since this is a config struct (not stored in DB), and Option fields are skipped by default in serde if using skip_serializing_if, this could be fine -- but note that there's no #[serde(skip_serializing_if = "Option::is_none")] on the data_store or url fields. This means serializing a default config will emit "data_store": null and "url": null in JSON. Consider adding skip_serializing_if for cleaner output.

No Security Concerns

• The registry URL is a configuration value with sensible defaults
• No new attack surface introduced
• Credentials (secret_access_key) handling unchanged

No Performance Concerns

• Static LazyLock URLs are initialized once
• No hot-path changes

Verdict

Clean, well-structured change that follows project conventions. The main actionable items are the unused PROD_BENCHER_REGISTRY_URL static and optionally adding skip_serializing_if to the new Option fields in JsonRegistry for cleaner serialized output. The dead registry_url() accessor is acceptable as scaffolding for the next step in the PLAN.

---

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

[github-actions]

Bencher Report

Branch	u/ep/registry-url
Testbed	ubuntu-22.04

⚠️ WARNING: No Threshold found!

Without a Threshold, no Alerts will ever be generated.

• Latency (nanoseconds (ns))

Click here to create a new Threshold
For more information, see the Threshold documentation.
To only post results if a Threshold exists, set the --ci-only-thresholds flag.

Click to view all benchmark results

Benchmark	Latency	microseconds (µs)
Adapter::Json	📈 view plot ⚠️ NO THRESHOLD	3.70 µs
Adapter::Magic (JSON)	📈 view plot ⚠️ NO THRESHOLD	3.64 µs
Adapter::Magic (Rust)	📈 view plot ⚠️ NO THRESHOLD	25.38 µs
Adapter::Rust	📈 view plot ⚠️ NO THRESHOLD	2.81 µs
Adapter::RustBench	📈 view plot ⚠️ NO THRESHOLD	2.83 µs
head_version_insert/batch/10	📈 view plot ⚠️ NO THRESHOLD	94.99 µs
head_version_insert/batch/100	📈 view plot ⚠️ NO THRESHOLD	228.77 µs
head_version_insert/batch/255	📈 view plot ⚠️ NO THRESHOLD	455.46 µs
head_version_insert/batch/50	📈 view plot ⚠️ NO THRESHOLD	154.69 µs
threshold_query/join/10	📈 view plot ⚠️ NO THRESHOLD	138.84 µs
threshold_query/join/20	📈 view plot ⚠️ NO THRESHOLD	156.29 µs
threshold_query/join/5	📈 view plot ⚠️ NO THRESHOLD	131.32 µs
threshold_query/join/50	📈 view plot ⚠️ NO THRESHOLD	194.15 µs

🐰 View full continuous benchmarking report in Bencher