Erigon Test Rpc

QA RPC Integration Tests

Tests that run rpcdaemon against a pre-synced Erigon datadir and compare JSON-RPC responses against golden expected outputs from the erigontech/rpc-tests repository.

Prerequisites

• A synced Erigon mainnet datadir (the tests need specific blocks to exist — tests at block ~24.3M require sync to at least that height)
• Python 3 (for the rpc-tests runner)
• rpcdaemon and integration binaries built (make rpcdaemon integration)

How It Works

The CI workflow:

1. Backs up chaindata from the reference datadir
2. Runs integration run_migrations to apply pending migrations
3. Starts rpcdaemon --datadir <reference-datadir>
4. Runs run_rpc_tests.sh which clones erigontech/rpc-tests, sets up a Python venv, and runs run_tests.py against port 8545
5. Restores original chaindata from backup

The backup/restore pattern exists because run_migrations writes to the DB.

Local Setup

Step 1: Set your reference datadir

# Point to your synced mainnet Erigon node
export ERIGON_REFERENCE_DATA_DIR=/path/to/your/erigon/mainnet-datadir

# Optional: workspace for test output
export WORKSPACE=/tmp/rpc-test-run
mkdir -p $WORKSPACE

Step 2: Back up chaindata (protects your DB from migration writes)

cp -r $ERIGON_REFERENCE_DATA_DIR/chaindata /tmp/chaindata-backup

Step 3: Build binaries

make rpcdaemon integration

Step 4: Run migrations

./build/bin/integration run_migrations --datadir $ERIGON_REFERENCE_DATA_DIR --chain mainnet

Step 5: Start rpcdaemon

./build/bin/rpcdaemon --datadir $ERIGON_REFERENCE_DATA_DIR \
  --http.api admin,debug,eth,parity,erigon,trace,web3,txpool,ots,net \
  --ws > /tmp/rpcdaemon.log 2>&1 &

RPC_DAEMON_PID=$!

# Wait for port 8545
for i in {1..30}; do
  nc -z localhost 8545 && break || sleep 5
done

Step 6: Run the tests

.github/workflows/scripts/run_rpc_tests_ethereum.sh $WORKSPACE /tmp/rpc-results
echo "Exit: $?"

Step 7: Stop and restore

kill $RPC_DAEMON_PID

# Restore original chaindata
rm -rf $ERIGON_REFERENCE_DATA_DIR/chaindata
mv /tmp/chaindata-backup $ERIGON_REFERENCE_DATA_DIR/chaindata

Run a Single Test Manually

Clone rpc-tests and run one test directly:

git clone --depth 1 --branch v1.121.0 https://github.com/erigontech/rpc-tests /tmp/rpc-tests
cd /tmp/rpc-tests
python3 -m venv .venv && source .venv/bin/activate
pip3 install -r requirements.txt

cd integration
# Run just one test
python3 run_tests.py --blockchain mainnet --port 8545 \
  --include-api-list eth_getBlockByNumber/test_1.json \
  --display-only-fail --json-diff

When Tests Fail — Drill Down

The runner outputs diffs for failing tests. Each test is a .json file in the integration/<chain>/ directory of rpc-tests. To investigate:

# Show which tests failed
grep "FAILED\|PASS\|SKIP" /tmp/rpc-results/output.log | grep FAIL

# Re-run just the failing API group
cd /tmp/rpc-tests/integration
python3 run_tests.py --blockchain mainnet --port 8545 \
  --include-api-list eth_getBlockByNumber \
  --display-only-fail --json-diff

# Run with full response dump to see what rpcdaemon returned
python3 run_tests.py --blockchain mainnet --port 8545 \
  --include-api-list eth_getBlockByNumber/test_1.json \
  --dump-response

Disabled Tests

Some tests are intentionally disabled in the CI run (see run_rpc_tests_ethereum.sh). These include:

• Engine API tests (require active Erigon engine, not just rpcdaemon)
• Tests requiring specific peer state (net_peerCount, admin_peers, etc.)
• Known-broken tests with open issues

CI Equivalents

Remote dispatch:

gh workflow run "QA - RPC Integration Tests" --ref <branch>
gh workflow run "QA - RPC Integration Tests Remote" --ref <branch>
gh workflow run "QA - RPC Integration Tests (Gnosis)" --ref <branch>

TODO: Reference Datadir Setup

The CI runner maintains a pre-synced mainnet datadir at /opt/erigon-versions/reference-version/datadir. To create a local equivalent:

1. Sync Erigon mainnet to at least block ~24.5M (covers all test cases)
2. Stop the node cleanly
3. Set ERIGON_REFERENCE_DATA_DIR to that datadir
4. Follow the workflow above

A full mainnet sync takes ~2-4 days on NVMe with --prune.mode=minimal. Alternatively, ask the team for a snapshot of the reference datadir.