graphbench

GraphBench

graphbench runs the scale benchmark corpus under benchmark/testdata/scale. It is meant for runtime gap accounting: query duration, returned row counts, PostgreSQL plan details, Neo4j plan operators, fallback reasons, and comparison summaries.

The current execution modes are:

• postgres_sql: runs DAWGS' PostgreSQL SQL translation against a PostgreSQL database.
• local_traversal: records explicit not_implemented placeholders until the local traversal executor lands.
• neo4j: runs the same corpus against Neo4j through the DAWGS Neo4j backend.

Apache AGE is not an execution mode in this harness yet. AGE behavior can be captured in corpus reference_design notes so DAWGS can use it as design input without treating it as a direct benchmark comparison.

Inputs

The command loads cases from benchmark/testdata/scale by default and imports the fixture datasets from integration/testdata.

Connection strings can be supplied as flags or environment variables:

• PostgreSQL: -pg-connection, PG_CONNECTION_STRING, -connection, or CONNECTION_STRING.
• Neo4j: -neo4j-connection, NEO4J_CONNECTION_STRING, -connection, or CONNECTION_STRING.

Examples

Run only PostgreSQL SQL translation:

go run ./cmd/graphbench \
  -modes postgres_sql \
  -pg-connection "$PG_CONNECTION_STRING" \
  -jsonl-output .coverage/graphbench-postgres.jsonl \
  -summary .coverage/graphbench-postgres.md \
  -summary-json .coverage/graphbench-postgres.json

Capture PostgreSQL, local traversal placeholders, and Neo4j in one report:

go run ./cmd/graphbench \
  -modes postgres_sql,local_traversal,neo4j \
  -pg-connection "$PG_CONNECTION_STRING" \
  -neo4j-connection "$NEO4J_CONNECTION_STRING" \
  -jsonl-output .coverage/graphbench.jsonl \
  -summary .coverage/graphbench.md \
  -summary-json .coverage/graphbench.json

Compare a run against a previous JSONL capture:

go run ./cmd/graphbench \
  -modes postgres_sql,neo4j \
  -pg-connection "$PG_CONNECTION_STRING" \
  -neo4j-connection "$NEO4J_CONNECTION_STRING" \
  -baseline .coverage/graphbench-baseline.jsonl \
  -jsonl-output .coverage/graphbench.jsonl \
  -summary .coverage/graphbench.md

Outputs

JSONL output contains one CaseResult record per case and execution mode. Markdown and JSON summaries aggregate mode status counts, per-case timings, row counts, fallback reasons, and baseline regressions or improvements when a baseline capture is supplied.

PostgreSQL records include translated SQL and EXPLAIN (ANALYZE, BUFFERS, TIMING OFF, FORMAT JSON) metrics. Neo4j records include plan operator names when an EXPLAIN plan can be captured.