Introduction

LakeQL never talks to a query engine directly. Everything layered on top of the adapter works against the neutral contracts in @lakeql/datasource. That includes the GraphQL server, the CLI, and the query builder. An adapter implements those contracts for one engine.

What an Adapter Owns #

An adapter owns everything engine-specific:

• Connection — how to reach the engine and authenticate.
• SQL dialect — how neutral query intent is rendered into the engine's SQL.
• Read strategy — how a read is executed and validated.
• Write strategies — which write paths the adapter supports.
• Response decoding — how raw protocol responses become keyed records.
• Errors — mapping engine errors to the normalized DataSourceError categories.

LakeQL ships two adapters: Trino and Fabric. The layer is neutral, so one project can register both and route each named data source to a different engine.

Registration #

Register adapters in defineConfig, then declare data sources that select one by name:

1
2
3
4
5
6
7
8
9
10
11
12
import { defineConfig } from "@lakeql/config"
import { trinoAdapterDefinition } from "@lakeql/adapter-trino"
import { fabricAdapterDefinition } from "@lakeql/adapter-fabric"

export default defineConfig({
  adapters: [trinoAdapterDefinition, fabricAdapterDefinition],
  dataSources: {
    warehouse: { adapter: "trino", connection: {/* ... */} },
    reporting: { adapter: "fabric", connection: {/* ... */} },
  },
})

Adapters never depend on @lakeql/config. You register them into it. See the runtime config contract for how registered adapters and data sources compose into a runtime.