Module config 

Runtime configuration loaded from datasets.toml.

Each instance binds to a list of datasets. A dataset’s [dataset.source] block selects the format (parquet or delta) and the location (a local path or an s3://bucket/key URL). When the location is on S3, an optional [dataset.s3] block carries non-secret connection details (region, endpoint, addressing style, …).

Credentials are resolved at runtime via DatasetConfig::resolved_creds in this precedence order:

1. Per-dataset env vars ${PREFIX}_AWS_ACCESS_KEY_ID, ${PREFIX}_AWS_SECRET_ACCESS_KEY, ${PREFIX}_AWS_SESSION_TOKEN where ${PREFIX} is the dataset name uppercased with non-alphanumeric characters replaced by _ (e.g. accidents → ACCIDENTS, sales.eu-1 → SALES_EU_1).
2. Inline access_key_id / secret_access_key / session_token in the [dataset.s3] block.
3. Plain AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY / AWS_SESSION_TOKEN.
4. None — fall back to the engine’s own provider chain (~/.aws/credentials, IMDS, …).

Structs

AppConfig

AuthConfig

OIDC bearer-token enforcement for the HTTP API ([auth] block).

DataFusionConfig

DataFusion backend performance tuning ([datafusion] block).

DatasetConfig

DocsConfig

Embedded MkDocs documentation site ([docs] block).

ExplorerConfig

Embedded dataset explorer UI ([explorer] block).

IndexConfig

MetricsConfig

Prometheus metrics endpoint ([metrics] block).

QuackConfig

Experimental DuckDB Quack remote protocol server.

ResolvedCreds

Resolved S3 credentials. None fields mean “let the engine’s default provider chain figure it out”.

S3Config

Non-secret S3 connection settings. Credentials are pulled from env / the AWS credential chain — see DatasetConfig::resolved_creds.

ServerConfig

SourceConfig

SqlConfig

Raw-SQL query endpoint ([sql] block).

SwaggerConfig

Swagger UI + embedded OpenAPI spec ([swagger] block).

SwaggerOAuth2Config

OIDC single-sign-on for the Swagger UI ([swagger.oauth2]).

Enums

AddressingStyle

Backend

BucketInHost

Whether the bucket name is folded into the endpoint hostname for virtual-hosted-style requests against a custom endpoint. This aligns the DataFusion object-store path with DuckDB, which builds the virtual host itself — so the same plain endpoint works on both backends.

IndexMode

Partitioning

How hive-style partition columns (key=value/ path segments) are handled for an S3 parquet source. Local parquet always auto-detects; this option brings S3 in line and lets you force or disable the behaviour.

SourceKind