nimbus-eth1/fluffy/scripts/launch_local_testnet.sh

390 lines
11 KiB
Bash
Executable File

#!/usr/bin/env bash
# Copyright (c) 2021-2024 Status Research & Development GmbH. Licensed under
# either of:
# - Apache License, version 2.0
# - MIT license
# at your option. This file may not be copied, modified, or distributed except
# according to those terms.
# This script is for a big part a copy of the nimbus-eth2 launch_local_testnet
# script. This script however does not expect fluffy nodes to exit 0 in the good
# case, but instead the json-rpc interface is used to check whether certain
# values are what we expect them to be.
set -e
cd "$(dirname "${BASH_SOURCE[0]}")"/../..
####################
# argument parsing #
####################
GETOPT_BINARY="getopt"
if uname | grep -qi darwin; then
# macOS
GETOPT_BINARY=$(find /opt/homebrew/opt/gnu-getopt/bin/getopt /usr/local/opt/gnu-getopt/bin/getopt 2> /dev/null | head -n1 || true)
[[ -f "$GETOPT_BINARY" ]] || { echo "GNU getopt not installed. Please run 'brew install gnu-getopt'. Aborting."; exit 1; }
fi
! ${GETOPT_BINARY} --test > /dev/null
if [ ${PIPESTATUS[0]} != 4 ]; then
echo '`getopt --test` failed in this environment.'
exit 1
fi
OPTS="h:n:d"
LONGOPTS="help,nodes:,data-dir:,run-tests,log-level:,base-port:,base-rpc-port:,trusted-block-root:,portal-bridge,base-metrics-port:,reuse-existing-data-dir,timeout:,kill-old-processes"
# default values
NUM_NODES="64"
DATA_DIR="local_testnet_data"
RUN_TESTS="0"
LOG_LEVEL="INFO"
BASE_PORT="9000"
BASE_METRICS_PORT="8008"
BASE_RPC_PORT="10000"
REUSE_EXISTING_DATA_DIR="0"
TIMEOUT_DURATION="0"
KILL_OLD_PROCESSES="0"
SCRIPTS_DIR="fluffy/scripts/"
PORTAL_BRIDGE="0"
TRUSTED_BLOCK_ROOT=""
# REST_URL="http://127.0.0.1:5052"
REST_URL="http://testing.mainnet.beacon-api.nimbus.team"
print_help() {
cat <<EOF
Usage: $(basename "$0") [OPTIONS] -- [FLUFFY OPTIONS]
E.g.: $(basename "$0") --nodes ${NUM_NODES} --data-dir "${DATA_DIR}" # defaults
-h, --help this help message
-n, --nodes number of nodes to launch (default: ${NUM_NODES})
-d, --data-dir directory where all the node data and logs will end up
(default: "${DATA_DIR}")
--base-port bootstrap node's discv5 port (default: ${BASE_PORT})
--base-rpc-port bootstrap node's RPC port (default: ${BASE_RPC_PORT})
--base-metrics-port bootstrap node's metrics server port (default: ${BASE_METRICS_PORT})
--portal-bridge run a portal bridge attached to the bootstrap node
--trusted-block-root recent trusted finalized block root to initialize the consensus light client from
--run-tests when enabled run tests else use "htop" to see the fluffy processes without doing any tests
--log-level set the log level (default: ${LOG_LEVEL})
--reuse-existing-data-dir instead of deleting and recreating the data dir, keep it and reuse everything we can from it
--timeout timeout in seconds (default: ${TIMEOUT_DURATION} - no timeout)
--kill-old-processes if any process is found listening on a port we use, kill it (default: disabled)
EOF
}
! PARSED=$(${GETOPT_BINARY} --options=${OPTS} --longoptions=${LONGOPTS} --name "$0" -- "$@")
if [ ${PIPESTATUS[0]} != 0 ]; then
# getopt has complained about wrong arguments to stdout
exit 1
fi
# read getopt's output this way to handle the quoting right
eval set -- "$PARSED"
while true; do
case "$1" in
-h|--help)
print_help
exit
;;
-n|--nodes)
NUM_NODES="$2"
shift 2
;;
-d|--data-dir)
DATA_DIR="$2"
shift 2
;;
--run-tests)
RUN_TESTS="1"
shift
;;
--log-level)
LOG_LEVEL="$2"
shift 2
;;
--base-port)
BASE_PORT="$2"
shift 2
;;
--base-rpc-port)
BASE_RPC_PORT="$2"
shift 2
;;
--trusted-block-root)
TRUSTED_BLOCK_ROOT="$2"
shift 2
;;
--portal-bridge)
PORTAL_BRIDGE="1"
shift
;;
--base-metrics-port)
BASE_METRICS_PORT="$2"
shift 2
;;
--reuse-existing-data-dir)
REUSE_EXISTING_DATA_DIR="1"
shift
;;
--timeout)
TIMEOUT_DURATION="$2"
shift 2
;;
--kill-old-processes)
KILL_OLD_PROCESSES="1"
shift
;;
--)
shift
break
;;
*)
echo "argument parsing error"
print_help
exit 1
esac
done
# when sourcing env.sh, it will try to execute $@, so empty it
EXTRA_ARGS="$@"
if [[ $# != 0 ]]; then
shift $#
fi
if [[ "$REUSE_EXISTING_DATA_DIR" == "0" ]]; then
rm -rf "${DATA_DIR}"
fi
"${SCRIPTS_DIR}"/makedir.sh "${DATA_DIR}"
HAVE_LSOF=0
# Windows detection
if uname | grep -qiE "mingw|msys"; then
MAKE="mingw32-make"
else
MAKE="make"
which lsof &>/dev/null && HAVE_LSOF=1 || { echo "'lsof' not installed and we need it to check for ports already in use. Aborting."; exit 1; }
fi
# number of CPU cores
if uname | grep -qi darwin; then
NPROC="$(sysctl -n hw.logicalcpu)"
else
NPROC="$(nproc)"
fi
# kill lingering processes from a previous run
if [[ "${HAVE_LSOF}" == "1" ]]; then
for NUM_NODE in $(seq 0 $(( NUM_NODES - 1 ))); do
for PORT in $(( BASE_PORT + NUM_NODE )) $(( BASE_METRICS_PORT + NUM_NODE )) $(( BASE_RPC_PORT + NUM_NODE )); do
for PID in $(lsof -n -i tcp:${PORT} -sTCP:LISTEN -t); do
echo -n "Found old process listening on port ${PORT}, with PID ${PID}. "
if [[ "${KILL_OLD_PROCESSES}" == "1" ]]; then
echo "Killing it."
kill -9 ${PID} || true
else
echo "Aborting."
exit 1
fi
done
done
done
fi
# Build the binaries
BINARIES="fluffy"
if [[ "${PORTAL_BRIDGE}" == "1" ]]; then
BINARIES="${BINARIES} portal_bridge"
fi
$MAKE -j ${NPROC} LOG_LEVEL=TRACE ${BINARIES}
if [[ "$RUN_TESTS" == "1" ]]; then
TEST_BINARIES="test_portal_testnet"
$MAKE -j ${NPROC} LOG_LEVEL=INFO ${TEST_BINARIES}
fi
# Kill child processes on Ctrl-C/SIGTERM/exit, passing the PID of this shell
# instance as the parent and the target process name as a pattern to the
# "pkill" command.
cleanup() {
for BINARY in ${BINARIES}; do
pkill -f -P $$ ${BINARY} &>/dev/null || true
done
sleep 2
for BINARY in ${BINARIES}; do
pkill -f -9 -P $$ ${BINARY} &>/dev/null || true
done
# Delete the binaries we just built, because these are with none default logs.
# TODO: When fluffy gets run time log options a la nimbus-eth2 we can keep
# the binaries around.
for BINARY in ${BINARIES}; do
rm build/${BINARY}
done
}
trap 'cleanup' SIGINT SIGTERM EXIT
# timeout - implemented with a background job
timeout_reached() {
echo -e "\nTimeout reached. Aborting.\n"
cleanup
}
trap 'timeout_reached' SIGALRM
# TODO: This doesn't seem to work in Windows CI as it can't find the process
# with WATCHER_PID when doing the taskkill later on.
if [[ "${TIMEOUT_DURATION}" != "0" ]]; then
export PARENT_PID=$$
( sleep ${TIMEOUT_DURATION} && kill -ALRM ${PARENT_PID} ) 2>/dev/null & WATCHER_PID=$!
fi
PIDS=""
NUM_JOBS=$(( NUM_NODES + PORTAL_BRIDGE ))
dump_logs() {
LOG_LINES=20
for LOG in "${DATA_DIR}"/log*.txt; do
echo "Last ${LOG_LINES} lines of ${LOG}:"
tail -n ${LOG_LINES} "${LOG}"
echo "======"
done
}
BOOTSTRAP_NODE=0
BOOTSTRAP_TIMEOUT=5 # in seconds
BOOTSTRAP_ENR_FILE="${DATA_DIR}/node${BOOTSTRAP_NODE}/fluffy_node.enr"
TRUSTED_BLOCK_ROOT_ARG=""
if [[ -n ${TRUSTED_BLOCK_ROOT} ]]; then
TRUSTED_BLOCK_ROOT_ARG="--trusted-block-root=${TRUSTED_BLOCK_ROOT}"
fi
for NUM_NODE in $(seq 0 $(( NUM_NODES - 1 ))); do
NODE_DATA_DIR="${DATA_DIR}/node${NUM_NODE}"
rm -rf "${NODE_DATA_DIR}"
"${SCRIPTS_DIR}"/makedir.sh "${NODE_DATA_DIR}" 2>&1
done
echo "Starting ${NUM_NODES} nodes."
for NUM_NODE in $(seq 0 $(( NUM_NODES - 1 ))); do
# Reset arguments
BOOTSTRAP_ARG=""
NODE_DATA_DIR="${DATA_DIR}/node${NUM_NODE}"
if [[ ${NUM_NODE} != ${BOOTSTRAP_NODE} ]]; then
BOOTSTRAP_ARG="--bootstrap-file=${BOOTSTRAP_ENR_FILE}"
# All nodes but bootstrap node run with log. radius of 254 which should
# result in ~1/4th of the data set stored.
RADIUS_ARG="--radius=static:254"
# Wait for the bootstrap node to write out its enr file
START_TIMESTAMP=$(date +%s)
while [[ ! -f "${BOOTSTRAP_ENR_FILE}" ]]; do
sleep 0.1
NOW_TIMESTAMP=$(date +%s)
if [[ "$(( NOW_TIMESTAMP - START_TIMESTAMP - GENESIS_OFFSET ))" -ge "$BOOTSTRAP_TIMEOUT" ]]; then
echo "Bootstrap node failed to start in ${BOOTSTRAP_TIMEOUT} seconds. Aborting."
dump_logs
exit 1
fi
done
fi
# Running with bits-per-hop of 1 to make the lookups more likely requiring
# to request to nodes over the network instead of having most of them in the
# own routing table.
./build/fluffy \
--listen-address:127.0.0.1 \
--nat:extip:127.0.0.1 \
--log-level="${LOG_LEVEL}" \
--udp-port=$(( BASE_PORT + NUM_NODE )) \
--data-dir="${NODE_DATA_DIR}" \
--portal-network="none" \
${BOOTSTRAP_ARG} \
--rpc \
--rpc-address="127.0.0.1" \
--rpc-port="$(( BASE_RPC_PORT + NUM_NODE ))" \
--metrics \
--metrics-address="127.0.0.1" \
--metrics-port="$(( BASE_METRICS_PORT + NUM_NODE ))" \
--table-ip-limit=1024 \
--bucket-ip-limit=24 \
--bits-per-hop=1 \
--networks:beacon,history,state \
${TRUSTED_BLOCK_ROOT_ARG} \
${RADIUS_ARG} \
${EXTRA_ARGS} \
> "${DATA_DIR}/log${NUM_NODE}.txt" 2>&1 &
if [[ "${PIDS}" == "" ]]; then
PIDS="$!"
else
PIDS="${PIDS},$!"
fi
done
if [[ "$PORTAL_BRIDGE" == "1" ]]; then
# Give the nodes time to connect before the bridge (node 0) starts gossip
sleep 10
echo "Starting portal bridge for beacon network."
./build/portal_bridge beacon \
--rest-url="${REST_URL}" \
--rpc-address="127.0.0.1" \
--rpc-port="${BASE_RPC_PORT}" \
--backfill-amount=128 \
${TRUSTED_BLOCK_ROOT_ARG} \
> "${DATA_DIR}/log_portal_bridge.txt" 2>&1 &
PIDS="${PIDS},$!"
fi
# give the regular nodes time to crash
sleep 5
BG_JOBS="$(jobs | wc -l | tr -d ' ')"
if [[ "${TIMEOUT_DURATION}" != "0" ]]; then
BG_JOBS=$(( BG_JOBS - 1 )) # minus the timeout bg job
fi
if [[ "$BG_JOBS" != "$NUM_JOBS" ]]; then
echo "$(( NUM_JOBS - BG_JOBS )) fluffy instance(s) exited early. Aborting."
dump_logs
exit 1
fi
# launch htop and run until `TIMEOUT_DURATION` or check the nodes and quit.
if [[ "$RUN_TESTS" == "0" ]]; then
htop -p "$PIDS"
cleanup
else
# Need to let to settle the network a bit, as currently at start discv5 and
# the Portal networks all send messages at once to the same nodes, causing
# messages to drop when handshakes are going on.
sleep 5
./build/test_portal_testnet --node-count:${NUM_NODES}
FAILED=$?
if [[ "$FAILED" != "0" ]]; then
dump_logs
if [[ "${TIMEOUT_DURATION}" != "0" ]]; then
if uname | grep -qiE "mingw|msys"; then
echo ${WATCHER_PID}
taskkill //F //PID ${WATCHER_PID}
else
pkill -HUP -P ${WATCHER_PID}
fi
fi
exit 1
fi
fi
if [[ "${TIMEOUT_DURATION}" != "0" ]]; then
if uname | grep -qiE "mingw|msys"; then
taskkill //F //PID ${WATCHER_PID}
else
pkill -HUP -P ${WATCHER_PID}
fi
fi