# Caching & performance

> How Flax keeps queries fast — connection pooling, a per-tenant result cache, and the tiered source cache, plus when a query hits cache versus your warehouse.

*[View this page in the Flax docs](https://flax-analytics.com/docs/connections/caching-and-performance)*

Dashboards and explorations issue the same or overlapping queries repeatedly. Flax avoids paying warehouse latency and compute for every view by pooling connections and caching results (ADR-0029, ADR-0008), while never serving data past its freshness policy.

## Connection pooling

Flax keeps a bounded, reusable connection pool per warehouse connection rather than dialing a fresh connection for every query. A global cap with LRU eviction stops any one tenant from exhausting resources, and a pool is dropped automatically when its connection details change, so rotated credentials never leave a stale pool behind.

## Result cache

Each query result is cached per tenant with a time-to-live (TTL). The cache key is derived from the organization, connection, compiled SQL, and its arguments, so an identical repeat query is served from cache and a genuinely different query is not.

| Behaviour | Detail |
| --- | --- |
| Scope | Per tenant (organization) — never shared across organizations. |
| Key | Hash of `org`, `connection`, `SQL`, and `args`. |
| Freshness | Governed by a TTL; a short default keeps results current. |
| Bypass | An explicit refresh re-runs the query against the warehouse. |

## Tiered source cache

The result cache is the first tier of a broader tiered cache for customer source data (ADR-0008). A larger materialized tier for reusing pulled data across overlapping queries is planned; today the per-tenant result cache captures most of the repeat-query win.

## Cache vs. warehouse

- **Cache** — a repeat query whose key matches a fresh (within-TTL) cached result is served without touching the warehouse.
- **Warehouse** — a first-time query, a query past its TTL, or an explicit refresh is compiled and pushed down to your warehouse (ADR-0004), and the fresh result repopulates the cache.

> [!NOTE]
> Because a cached result can lag the source until its TTL expires, use refresh when you need to confirm the latest data.

## Related

- [Connecting data](/docs/connections/overview)
- [Connection security](/docs/connections/security)
- [Exploring overview](/docs/exploring/overview)