petrova.smo1.io: Custom-Domain Links for petrova.codes Artifacts

ELI5

petrova.codes publishes artifacts and wants short links on its own branded subdomain, petrova.smo1.io. The same edge worker serves it. The difference is that the API remembers which hostname a link belongs to, so petrova.smo1.io/report and smo1.io/report can point to different places. The configuration for the worker is in place; on 2026-10-25 the hostname itself did not yet exist in DNS.

Technical Deep Dive

Source: zoomies-edge wrangler.toml:40-48, src/handlers/redirect.ts:79-101, src/index.ts:201-256 (origin/main 5bdde59); purr-api migrations/000007_custom_domains.up.sql, internal/repository/link_repository.go:148-163, internal/service/link_service.go:26-32, 67-73, 104-111, 198-222, internal/models/link.go:112-116, 169-170, internal/service/redirect_cache.go:58-63, cmd/server/main.go:156-159 (bb6cfc2); smo1-io docs/runbooks/petrova-smo1-io-domain.md §1-5, docs/runbooks/petrova-codes-build-auth.md §1 (1f19667); DNS lookup of petrova.smo1.io, read 2026-10-25.

Prerequisites and request path

flowchart TD
    DNS["1. Proxied DNS record for petrova in the smo1.io zone"] --> RT["2. Worker route petrova.smo1.io/* in env.lion"]
    RT --> ROW["3. custom_domains row: hostname petrova.smo1.io, user_id = enterprise account"]
    ROW --> REQ["GET petrova.smo1.io/report"]
    REQ --> W["Worker: host = petrova.smo1.io"]
    W --> RES["purr-api /api/links/resolve?host=petrova.smo1.io and slug=report"]
    RES --> M{"link with domain_id = that row ?"}
    M -->|Yes| R302[302 to destination]
    M -->|No| LP[landing proxy]

Data model

Migration 000007 adds custom_domains(id, user_id text, hostname unique, verified_at, created_at) and links.domain_id (NULL means the apex). Custom-slug uniqueness becomes per (user_id, COALESCE(domain_id, nil-uuid), custom_slug), so an apex link and a petrova link may share a slug.

Host-aware resolution

ResolveByHostAndSlug (link_repository.go:153-163) matches custom_slug or uid, then requires COALESCE(domain_id, nil) to equal the id of the custom_domains row for the host, or nil when the host is unknown. An unknown host therefore resolves only apex links.

Creating petrova links

The petrova.codes build authenticates with a developer API key (X-API-Key) belonging to one business-tier enterprise account (build-auth runbook §1). A create request carries domain: "petrova.smo1.io" (link.go:112-116); resolveDomain refuses an unknown host (ErrDomainNotFound), a host owned by another account (ErrDomainForbidden), and a service built without the domain repository (ErrDomainsUnavailable) rather than silently creating an apex link (link_service.go:198-222). Short URLs are rendered as https://petrova.smo1.io/<slug> (link.go:169-170).

Still host-blind

• The edge’s proxy-mode probe calls /api/links/slug/:slug, which is uid = $1 OR custom_slug = $1 LIMIT 1 with no host (index.ts:243, link_repository.go:148).
• Click tracking sends only the slug and resolves it the same host-blind way, so a colliding petrova custom slug can be credited to an apex link (smo1-008).
• The Redis redirect cache avoids the problem by never caching a custom-domain link under its bare slug (redirect_cache.go:58-63).

Key Terms

• custom_domains — hostnames smo1 serves links on, each owned by one user
• domain_id — the link’s custom domain; NULL for smo1.io
• resolveDomain — create-time check that the requester owns the hostname
• ResolveByHostAndSlug — the host-aware lookup behind /api/links/resolve

Q&A

Q: Which three things must exist before petrova.smo1.io/ can redirect? A: A proxied DNS record for petrova in the smo1.io zone, the petrova.smo1.io/* route on the deployed lion Worker, and a custom_domains row for the hostname owned by the account that created the links. On 2026-10-25 the DNS record was missing.

Q: How does purr-api pick between an apex link and a petrova.smo1.io link with the same custom slug? A: By host: /api/links/resolve joins the request’s hostname to custom_domains and only matches links whose domain_id equals that row’s id, or NULL for the apex.

Q: Which smo1 code paths are still host-blind for custom domains? A: The Worker’s proxy-mode probe (/api/links/slug/:slug) and click-tracking attribution (/api/analytics/track by slug). Both use the uid OR custom_slug LIMIT 1 lookup.

Examples

Runbook §5’s check: create a throwaway link with domain_id set to the petrova row and custom slug rbook-test, then curl -sI https://petrova.smo1.io/rbook-test should return a 302. Run today, that curl fails at DNS resolution, which is the step to fix first.