Testing & Cleanup¶

Between tests you can clean up data using the cleanup script or per-database commands.

cleanup.py — Complete Reference¶

WARNING: All cleanup operations are irreversible. Data cannot be recovered once deleted. Run cleanup only when you intend to re-ingest from scratch or switch to a different database backend.

The cleanup.py script lives in scripts/ and is run from the flexible-graphrag/ directory:

cd flexible-graphrag
python ../scripts/cleanup.py [flags]

Full cleanup (all stores)¶

python ../scripts/cleanup.py

Prompts for confirmation then wipes everything: PostgreSQL incremental tables and CocoIndex monitoring tables (cocoindex_ingest_log, cocoindex_run_log), vector store, search store, property graph store, RDF stores, CocoIndex LMDB state, and log files.

To skip the prompt:

python ../scripts/cleanup.py --yes

Use with care. Full cleanup deletes all ingested documents, embeddings, graph nodes, relationships, and incremental-update state. You must re-ingest all documents afterward.

Selective cleanup (one or more stores)¶

Selective flags skip the confirmation prompt and touch only the specified store(s):

Flags can be combined freely:

# Re-ingest graph only — keep vector embeddings
python ../scripts/cleanup.py --graph

# Re-ingest everything except incremental state
python ../scripts/cleanup.py --graph --vector --search --rdf --cocoindex --logs

# Switch graph backend (wipe graph + CocoIndex state, keep embeddings)
python ../scripts/cleanup.py --graph --cocoindex

# Wipe vector + graph for a full re-ingest (keep postgres incremental state)
python ../scripts/cleanup.py --graph --vector

Automated / CI mode¶

python ../scripts/cleanup.py --matrix-clean

Non-interactive mode for the integration test matrix runner. Cleans vector, search, graph, RDF, and CocoIndex stores without prompting. If ENABLE_INCREMENTAL_UPDATES=true is set in the environment, also clears the PostgreSQL incremental tables.

Common workflows¶

Vector Database Cleanup¶

When switching embedding models, you must delete existing vector indexes due to dimension incompatibility. See Vector Dimensions for per-database cleanup instructions.

Graph Database Cleanup¶

For SurrealDB-specific queries and cleanup, see SurrealDB Guide. For Neo4j-specific commands, see Neo4j Setup.

ArcadeDB Cleanup¶

The cleanup.py script includes ArcadeDB-specific handling — it directly connects via arcadedb_python, queries schema types, and issues DELETE FROM <type> statements (avoiding index-already-exists errors from the LlamaIndex factory).

RDF Store Cleanup¶

Use scripts/rdf_cleanup.py to manage RDF store data:

# List ingested documents and triple counts
python scripts/rdf_cleanup.py list-docs

# Show total triple count in named graph
python scripts/rdf_cleanup.py count

# Delete all triples for a specific document
python scripts/rdf_cleanup.py clear-doc <ref_doc_id>

# Wipe entire named graph (with confirmation)
python scripts/rdf_cleanup.py clear-all --yes

# Target a specific store
python scripts/rdf_cleanup.py list-docs --fuseki
python scripts/rdf_cleanup.py list-docs --graphdb
python scripts/rdf_cleanup.py list-docs --oxigraph

BM25 Index Cleanup¶

The BM25 index is file-based. Delete the directory configured in SEARCH_DB_CONFIG:

# Default location
rm -rf ./bm25_index

Incremental State Cleanup¶

To reset incremental update state, you can clear the PostgreSQL tables directly:

-- Connect to flexible_graphrag_incremental database
TRUNCATE TABLE document_state;
TRUNCATE TABLE datasource_config;

-- CocoIndex monitoring tables (if CocoIndex integration is enabled)
TRUNCATE TABLE cocoindex_ingest_log;
TRUNCATE TABLE cocoindex_run_log;

Or use the cleanup script: python ../scripts/cleanup.py --postgres

Or use pgAdmin at http://localhost:5050 (master password: admin, server password: password).