Cache CRD lookup and only retrieve required CRD #349

Merged

tobru merged 1 commit from feature/cache-crd-lookup into main

2026-01-16 14:31:26 +00:00

tobru commented

2026-01-16 12:47:54 +00:00

Owner

Now we:

Check Django cache first (24h TTL)
Use K8s API discovery (/apis/{group}/{version}) to get the correct plural name for the resource kind (handles irregular plurals like VSHNForgejo -> vshnforgejoes)
Use read_custom_resource_definition(name) to fetch a single CRD directly by its name ({plural}.{group})
Fall back to listing only if discovery or direct lookup fails

This reduces CRD-related latency from ~300ms to ~0ms (cached) or ~70ms (single fetch on cache miss).

Previously, every page load for a ServiceInstance would call list_custom_resource_definition() to fetch ALL CRDs from Kubernetes, then loop through them to find the matching one. This caused ~300ms+ latency per request, which was especially painful over the NetBird SOCKS proxy where each API call has ~66ms network overhead. Now we: 1. Check Django cache first (24h TTL) 2. Use K8s API discovery (/apis/{group}/{version}) to get the correct plural name for the resource kind (handles irregular plurals like VSHNForgejo -> vshnforgejoes) 3. Use read_custom_resource_definition(name) to fetch a single CRD directly by its name ({plural}.{group}) 4. Fall back to listing only if discovery or direct lookup fails This reduces CRD-related latency from ~300ms to ~0ms (cached) or ~70ms (single fetch on cache miss).