Add Bencher Bare Metal Stripe product

[epompeii]

This changeset adds the Bencher Bare Metal Stripe product. Further, the Bencher Cloud annual license has been removed. All licenses are now for Bencher Self-Hosted only.

[github-actions]

🤖 Claude Code Review

PR: #715
Base: devel
Head: u/ep/prod-bare-metal
Commit: 6b721d94d3148db11097ec172cc0d3189cf3ada2

---

Here's my review:

---

PR Review: Billing Bare Metal + Licensed Plan Cleanup

Commits: 7 commits covering bare metal billing, licensed plan simplification, OCI timeout, and test improvements.

Overall Assessment

This is a well-structured set of changes that: (1) adds a bare_metal Stripe product as a mandatory line item on every subscription, (2) removes the CloudLicensed usage kind since licensed plans on Bencher Cloud are now exclusively self-hosted, and (3) simplifies the Licensor::into_json API by removing the self_hosted field.

Issues

1. bare_metal_price called before into_plus_price consumes self (biller.rs:217-218)

let bare_metal_price = plus_plan.bare_metal_price(&self.products)?;
let (plus_price, entitlements) = plus_plan.into_plus_price(&self.products)?;

bare_metal_price takes &self while into_plus_price takes self — this works because PlusPlan is Clone (derived), but the code relies on the borrow ending before the move. This is fine today, but worth noting that bare_metal_price re-extracts the usage variant and price_name that into_plus_price also extracts. Consider a single method that returns both prices to avoid the redundant pattern matching.

2. Bare metal line item added unconditionally for metered (non-licensed) plans

In new_checkout_session and create_subscription, the bare metal line item is always added regardless of whether the plan is metered or licensed. For metered-only cloud users who don't use bare metal, they'll see an extra line item in their Stripe checkout. Verify this is the intended billing model — that every subscriber pays for bare metal access.

3. bare_metal_price falls back to "default" key inconsistently

into_plus_price does a direct get(&price_name) with no fallback to "default". But bare_metal_price also does a direct get(price_name.as_str()) with no fallback. Meanwhile, Product::preferred_price_ids uses .or_else(|| self.metered.get(DEFAULT_PRICE_NAME)). The fallback behavior is inconsistent — if a preferred/non-default key is used, bare_metal_price will fail even if a "default" price exists.

4. Breaking API change: JsonLicense.self_hosted and UsageKind::CloudLicensed removed

The self_hosted field is removed from JsonLicense and CloudLicensed is removed from UsageKind in the OpenAPI spec and TypeScript types. This is a breaking change for any API consumers. Make sure this is coordinated with a version bump or migration plan.

5. Validation duplication (checkout.rs:66 vs plan.rs:167)

The same guard — entitlements.is_some() && self_hosted.is_none() — appears in both checkouts_post_inner and post_inner (plan.rs), but with a subtle difference: plan.rs additionally checks context.is_bencher_cloud. The checkout endpoint doesn't check is_bencher_cloud, meaning it would reject licensed-without-self-hosted even on self-hosted instances. If the checkout endpoint is only reachable on Bencher Cloud this is fine, but worth confirming.

Minor Notes

• Test data correctly updated: old "metrics" key removed, "default" now points to new price IDs, and bare_metal test product added.
• The filter_subscription_items_excludes_bare_metal test is a good addition.
• OCI timeout (timeout-minutes: 5) in deploy.yml is sensible — prevents a hung OCI conformance test from blocking the pipeline.
• Adding sentry and otel to bencher_api_tests default features is appropriate for test coverage.
• The #[expect(clippy::too_many_lines)] removal on get_inner (usage.rs) is valid since the function was shortened by removing the CloudLicensed branch.

Summary

The changes are clean and well-tested. The main concerns are: (1) confirming that bare metal billing on every subscription is intentional, (2) the missing is_bencher_cloud check in the checkout endpoint validation, and (3) awareness that this is a breaking API change.

---

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

[github-actions]

Bencher Report

Branch	u/ep/prod-bare-metal
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	3.74 µs (+6.49%) Baseline: 3.52 µs	4.66 µs (80.43%)
Adapter::Magic (JSON)	📈 view plot 🚷 view threshold	3.68 µs (+5.43%) Baseline: 3.49 µs	4.57 µs (80.58%)
Adapter::Magic (Rust)	📈 view plot 🚷 view threshold	25.82 µs (+1.12%) Baseline: 25.54 µs	31.38 µs (82.29%)
Adapter::Rust	📈 view plot 🚷 view threshold	2.79 µs (-1.46%) Baseline: 2.83 µs	3.30 µs (84.33%)
Adapter::RustBench	📈 view plot 🚷 view threshold	2.77 µs (-1.93%) Baseline: 2.82 µs	3.29 µs (84.14%)
head_version_insert/batch/10	📈 view plot 🚷 view threshold	105.56 µs (+5.16%) Baseline: 100.38 µs	118.45 µs (89.12%)
head_version_insert/batch/100	📈 view plot 🚷 view threshold	236.99 µs (-0.42%) Baseline: 237.99 µs	266.67 µs (88.87%)
head_version_insert/batch/255	📈 view plot 🚷 view threshold	463.83 µs (+0.22%) Baseline: 462.81 µs	499.49 µs (92.86%)
head_version_insert/batch/50	📈 view plot 🚷 view threshold	161.87 µs (+0.40%) Baseline: 161.22 µs	183.81 µs (88.06%)
threshold_query/join/10	📈 view plot 🚷 view threshold	143.71 µs (-0.89%) Baseline: 145.00 µs	170.04 µs (84.51%)
threshold_query/join/20	📈 view plot 🚷 view threshold	156.55 µs (-1.91%) Baseline: 159.59 µs	187.40 µs (83.54%)
threshold_query/join/5	📈 view plot 🚷 view threshold	137.67 µs (+0.35%) Baseline: 137.19 µs	160.87 µs (85.58%)
threshold_query/join/50	📈 view plot 🚷 view threshold	200.76 µs (-0.06%) Baseline: 200.89 µs	232.73 µs (86.26%)

🐰 View full continuous benchmarking report in Bencher