references/hyperdrive/patterns.md
Hyperdrive design patterns
See api.md for maintained driver and ORM examples. Use the following decisions to select a pattern, then fetch its linked documentation for implementation.
| Workload or decision | Guidance and documentation | |----------------------|----------------------------| | Popular content or analytics dashboards | Cache only when the product can tolerate the configured stale window. Use query caching for eligibility, parameters, and settings. | | Mixed cached reads and fresh reads | Route authentication, permissions, and reads after writes through a cache-disabled configuration. Writes do not invalidate cached results; see read-after-write behavior. | | Multi-tenant queries | Derive tenant scope from authenticated application context and apply it to every query. A cache is not an authorization boundary. Review query caching for the selected query's behavior. | | Globally distributed callers | Understand the distinction between fast connection setup and the remaining query round trip in how Hyperdrive works. | | Multiple sequential database queries | For one regional database, configure an explicit region hint using placement setup. Compare deployed request duration and Hyperdrive metrics with representative cache behavior. | | Transactions or connection-local state | Keep transactions short and do not assume state survives across transactions. Fetch connection pooling and supported features before relying on session settings. | | Client lifetime and pool sizing | Create clients per handler invocation; Hyperdrive owns the origin pool. Use connection lifecycle and pool tuning instead of a global driver pool or copied connection counts. |
Separate application correctness from acceleration: use parameterized queries, enforce tenant access in the application, and select freshness before tuning cache hit rate. See gotchas.md when observed behavior differs from the design.