Skip to content

Convert GDB to GeoParquet

Convert an ESRI file geodatabase (.gdb directory, .zip archive containing exactly one, or an HTTP(S) URL of such a .zip) into GeoParquet — one file per layer, each with its own schema, CRS (as PROJJSON), and geometry types (GeoParquet 1.1.0, WKB encoding, ZSTD). Layers are read and written in chunks, so memory stays flat on multi-GB geodatabases. No reprojection. Output is S3 by default (one file per layer under an s3://<bucket>/<target>/ prefix); local_output writes to a directory on the EPT service filesystem instead.

Request

To S3, using your personal bucket (oidc-<username>, derived from your bearer token) and its default endpoint:

curl --fail-with-body \
  -X POST "${EPT_API_URL}/v1/conversions/gdb/parquet" \
  -H "Authorization: Bearer ${ACCESS_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{
    "source_uri": "/data/MSFD_package.gdb",
    "target": "geo/out"
  }'

To a specific bucket/project prefix, with request-scoped S3 credentials from POST /v1/auth:

{
  "source_uri": "https://example.org/MSFD_package.zip",
  "bucket": "project-ept-projectsforum",
  "target": "arco-data/shipwrecks",
  "endpoint_url": "https://minio.dive.edito.eu",
  "access_key_id": "<s3_credentials.access_key_id>",
  "secret_access_key": "<s3_credentials.secret_access_key>",
  "session_token": "<s3_credentials.session_token>"
}

To a directory on the EPT service filesystem (mutually exclusive with target/bucket and S3 credentials):

{
  "source_uri": "/data/MSFD_package.gdb",
  "local_output": "/data/out"
}

Select specific layers (defaults to all feature classes and tables):

{
  "source_uri": "/data/pkg.zip",
  "local_output": "/data/out",
  "layers": ["points", "areas"],
  "overwrite": true
}

Response

{
  "source_uri": "/data/MSFD_package.gdb",
  "target_uri": "s3://oidc-<your-username>/geo/out",
  "layers": ["areas", "points"],
  "feature_counts": {"points": 2, "areas": 1},
  "crs": {"points": "EPSG:31370", "areas": "EPSG:31370"},
  "outputs": ["s3://oidc-<your-username>/geo/out/areas.parquet", "s3://oidc-<your-username>/geo/out/points.parquet"],
  "status": "completed"
}

target_uri and outputs appear only in the response; the request takes target/bucket/local_output.

Notes

  • Request validation is extra="forbid" — unknown fields (e.g. a request-level target_uri) fail with 422.
  • target is a prefix relative to bucket and must not include s3:// or the bucket name; bucket must be a bare bucket name. bucket defaults to your personal oidc-<username> bucket.
  • When endpoint_url is provided, target_uri/outputs are full https:// URLs; s3:// URIs otherwise.
  • Sources are local paths or HTTP(S) URLs, never s3://.
  • The target must not exist (local) or must not collide with existing objects (S3) unless overwrite is true.
  • Layer names are sanitized into file names ([^A-Za-z0-9._-] → _); collisions fail the request.
  • Non-spatial tables convert too; their files carry no geometry column and crs is null.
  • S3 credentials are request-scoped; they never enter the process environment.
  • The optional GeoParquet bbox key is not written (streaming writer cannot amend the footer); spatial-filter pushdown degrades but files remain spec-valid.