6 GraphDB Fundamentals Loading Data

Loading RDF data into GraphDB can be done through the Workbench interface for quick, interactive imports—or through command-line tools designed for large-scale, high-performance ingestion. The core takeaway is that GraphDB offers multiple loading paths depending on whether the goal is convenience (uploading files, importing from URLs, pasting snippets) or throughput and reliability (offline loaders that serialize data directly into internal indexes).

In GraphDB Workbench, local RDF files are imported by uploading them, selecting which files to include, and running an import that asks for a base IRI and target graphs. Import settings are saved for repeat runs, and an “advanced settings” area lets users adjust additional parameters before starting the import. Workbench also supports remote ingestion: users can choose an RDF import flow that pulls data from a provided URL. For smaller or ad-hoc datasets, RDF text snippets can be pasted directly into the Workbench import dialog.

Workbench further supports server-side file loading for arbitrary-size datasets. Files are read from a directory on the GraphDB server, with a default location of $user.home/graphdb import/. The import directory can be changed via the system property graphdb.workbench.importdirectorysystemprop, and the Workbench can list available files in that server directory so users can import them selectively.

For very large datasets, GraphDB relies on offline, batch-oriented tools. The load rdf tool is built for offline loading of RDF datasets; it cannot target a running server. Its performance rationale is straightforward: it serializes RDF directly into GraphDB’s internal indexes and produces a ready-to-use repository. It can create a new repository using a standard Turtle configuration template (configs/templates) or initialize from an existing repository, in which case repository data is overwritten. Execution happens from the command line using load rdf with parameters including repository name, serial/parallel mode, and the RDF data file.

When datasets are so large that initial indexing becomes a multi-hour process, preload becomes the critical step. Preload converts RDF files into GraphDB indexes at a low level and is positioned for initial loads of datasets larger than several billion RDF statements. It supports transactional behavior such as stop requests, resume, and consistent output even after failures. If a run ends abnormally—due to disk space, out-of-memory, or other interruptions—preload can resume from intermediate restore points rather than restarting from scratch, using collected data sufficient to reinitialize internal components and continue.

Finally, onto refine provides a transformation layer for mapping structured data into RDF. It can convert tabular and semi-structured formats—TSV, CSV, TSV, XLS/XLSX, JSON, XML, RDF/XML, and Google Sheets—into RDF using a locally stored RDF schema. The workflow runs inside the GraphDB Workbench visual interface: start Workbench mode, create a project, upload data (from computer, URLs, clipboard, or existing projects), then define RDF mappings. Users map table headers to RDF triple construction, configure value mappings (value source, value type, optional transformation), and apply transformations using language-specific expressions including GREL (Google Refine expression language). The mapping editor supports prefixes from common RDF vocabularies (e.g., FOAF, Geo, RDF, RDFS, SKOS, XSD), enabling users to build IRIs and literals without importing every schema into the repository.