Limits & quotas
Default caps
Every cap below is enforced at creation: the request that would exceed one is refused with 409 Conflict and a message naming the cap and the way out. The count is taken against the scope in the middle column, so one project's databases never count against another's.
Suspension
A suspended organization keeps running what it already has, but provisioning is blocked: create, deploy and resize return 503 with an explanation, and the console replaces the affected actions with a "provisioning paused" notice. Data is never deleted on suspension.
Rate limits
The control plane allows 100 requests per minute per client IP, and answers 429 with retry-after (seconds) when that is exceeded. The limiter runs before authentication, so it is a floor against abuse rather than a per-tenant budget: callers behind one NAT share it. Use the SDK (which honours retry-after), batch reads, and raise the ceiling deliberately rather than polling in a tight loop. Route-specific limits are stricter where they need to be — sign-in allows 10 attempts per 5 minutes per IP.
Gateway rate-limit rules are separate and apply to application traffic, not to the control plane.
Raising limits
Caps are platform settings: a platform administrator raises one from Admin → Settings → Resource limits (the underlying keys are limits.projectsPerOrg, limits.databasesPerProject, limits.bucketsPerProject, limits.sandboxesPerProject and limits.edgeFunctionsPerProject), and the new value applies on the next request. Clearing a key restores the published default.
Because a service must belong to a project, hitting a per-project cap is usually a signal to split the workload across projects rather than to raise the cap.