Files
ethereum-rpc-docker/check-health.sh

337 lines
15 KiB
Bash
Executable File

#!/bin/bash
BASEPATH="$(dirname "$0")"
if [ $# -lt 2 ]; then
echo "Error: At least two parameters are required."
exit 1
fi
RPC_URL=$1
shift
# Check for --starknet / --aztec / --cosmos flag
is_starknet=false
is_aztec=false
is_cosmos=false
if [ "$1" == "--starknet" ]; then
is_starknet=true
shift
elif [ "$1" == "--aztec" ]; then
is_aztec=true
shift
elif [ "$1" == "--cosmos" ]; then
is_cosmos=true
shift
fi
REF=""
for url in "$@"; do
REF+="--url $url "
done
# Optional: You can remove the trailing space if needed
ref=${REF% }
# echo "ref: $ref"
timeout=3 # seconds
# CometBFT / cosmos (gaiad and the cosmos batch): no EVM RPC. Use the chain's own
# sync_info from the CometBFT /status method — catching_up=false means caught up to head.
# Short-circuits here; the EVM/starknet/aztec block-comparison path below is not used.
if $is_cosmos; then
status=$(curl -L --ipv4 -m $timeout -s -X POST -H "Content-Type: application/json" --data '{"jsonrpc":"2.0","id":1,"method":"status"}' "$RPC_URL")
if [ $? -ne 0 ] || [ -z "$status" ]; then echo "timeout"; exit 1; fi
catching_up=$(echo "$status" | jq -r '.result.sync_info.catching_up // .sync_info.catching_up' 2>/dev/null)
node_height=$(echo "$status" | jq -r '.result.sync_info.latest_block_height // .sync_info.latest_block_height' 2>/dev/null)
if [ -z "$node_height" ] || [ "$node_height" = "null" ]; then echo "error"; exit 1; fi
if [ "$catching_up" = "true" ]; then echo "syncing"; exit 1; fi
# catching_up=false => synced. If a reference endpoint is given, sanity-check head gap.
if [ -n "$ref" ]; then
ref_status=$($BASEPATH/multicurl.sh -L --ipv4 -m $timeout -s -X POST -H "Content-Type: application/json" --data '{"jsonrpc":"2.0","id":1,"method":"status"}' $ref)
ref_height=$(echo "$ref_status" | jq -r '.result.sync_info.latest_block_height // .sync_info.latest_block_height' 2>/dev/null)
if [ -n "$ref_height" ] && [ "$ref_height" != "null" ] && [ "$ref_height" -gt 0 ] 2>/dev/null; then
gap=$(( ref_height - node_height ))
if [ "$gap" -gt 100 ]; then echo "behind ($gap)"; exit 1; fi
fi
fi
echo "online"; exit 0
fi
response_file=$(mktemp)
# Use appropriate RPC method based on chain type
if $is_starknet; then
rpc_method='{"jsonrpc":"2.0","method":"starknet_getBlockWithTxHashes","params":["latest"],"id":1}'
elif $is_aztec; then
rpc_method='{"jsonrpc":"2.0","method":"node_getBlock","params":["latest"],"id":1}'
else
rpc_method='{"jsonrpc":"2.0","method":"eth_getBlockByNumber","params":["latest", false],"id":1}'
fi
http_status_code=$(curl -L --ipv4 -m $timeout -s -X POST -w "%{http_code}" -o "$response_file" -H "Content-Type: application/json" --data "$rpc_method" $RPC_URL)
if [ $? -eq 0 ]; then
if [[ $http_status_code -eq 200 ]]; then
response=$(cat "$response_file")
if $is_starknet; then
# Starknet returns decimal timestamp and block_number
latest_block_timestamp_decimal=$(echo "$response" | jq -r '.result.timestamp')
latest_block_number=$(echo "$response" | jq -r '.result.block_number')
latest_block_hash=$(echo "$response" | jq -r '.result.block_hash')
elif $is_aztec; then
# Aztec: node_getBlock("latest") returns blockHash, header.globalVariables.blockNumber, header.globalVariables.timestamp
latest_block_number=$(echo "$response" | jq -r '.result.header.globalVariables.blockNumber')
latest_block_timestamp_decimal=$(echo "$response" | jq -r '.result.header.globalVariables.timestamp')
latest_block_hash=$(echo "$response" | jq -r '.result.blockHash')
if [ "$latest_block_number" = "null" ] || [ "$latest_block_timestamp_decimal" = "null" ] || [ -z "$latest_block_timestamp_decimal" ]; then
echo "error"
exit 1
fi
else
# Ethereum returns hex timestamp and number
latest_block_timestamp=$(echo "$response" | jq -r '.result.timestamp')
latest_block_timestamp_decimal=$((16#${latest_block_timestamp#0x}))
latest_block_number=$(echo "$response" | jq -r '.result.number')
latest_block_hash=$(echo "$response" | jq -r '.result.hash')
fi
current_time=$(date +%s)
time_difference=$((current_time - latest_block_timestamp_decimal))
rm "$response_file"
if [ -n "$ref" ]; then
MAX_RETRIES=3
attempt=1
while [ $attempt -le $MAX_RETRIES ]; do
# Re-fetch local latest block for retries > 1
if [ $attempt -gt 1 ]; then
sleep 3
# Re-query local latest block and update variables
response_file=$(mktemp)
http_status_code=$(curl -L --ipv4 -m $timeout -s -X POST -w "%{http_code}" -o "$response_file" -H "Content-Type: application/json" --data "$rpc_method" $RPC_URL)
if [ $? -eq 0 ] && [[ $http_status_code -eq 200 ]]; then
response=$(cat "$response_file")
if $is_starknet; then
latest_block_timestamp_decimal=$(echo "$response" | jq -r '.result.timestamp')
latest_block_number=$(echo "$response" | jq -r '.result.block_number')
latest_block_hash=$(echo "$response" | jq -r '.result.block_hash')
elif $is_aztec; then
latest_block_number=$(echo "$response" | jq -r '.result.header.globalVariables.blockNumber')
latest_block_timestamp_decimal=$(echo "$response" | jq -r '.result.header.globalVariables.timestamp')
latest_block_hash=$(echo "$response" | jq -r '.result.blockHash')
if [ "$latest_block_number" = "null" ] || [ "$latest_block_timestamp_decimal" = "null" ] || [ -z "$latest_block_timestamp_decimal" ]; then
echo "error"
exit 1
fi
else
latest_block_timestamp=$(echo "$response" | jq -r '.result.timestamp')
latest_block_timestamp_decimal=$((16#${latest_block_timestamp#0x}))
latest_block_number=$(echo "$response" | jq -r '.result.number')
latest_block_hash=$(echo "$response" | jq -r '.result.hash')
fi
current_time=$(date +%s)
time_difference=$((current_time - latest_block_timestamp_decimal))
rm "$response_file"
else
rm "$response_file"
echo "error"
exit 1
fi
fi
response_file2=$(mktemp)
if [ $attempt -eq 1 ]; then
sleep 3 # to give the reference node more time to import the block if it is very current
fi
if $is_starknet; then
# Starknet uses block_id object with block_number
rpc_method2="{\"jsonrpc\":\"2.0\",\"method\":\"starknet_getBlockWithTxHashes\",\"params\":[{\"block_number\":$latest_block_number}],\"id\":1}"
elif $is_aztec; then
rpc_method2="{\"jsonrpc\":\"2.0\",\"method\":\"node_getBlock\",\"params\":[$latest_block_number],\"id\":1}"
else
rpc_method2="{\"jsonrpc\":\"2.0\",\"method\":\"eth_getBlockByNumber\",\"params\":[\"$latest_block_number\", false],\"id\":1}"
fi
http_status_code2=$($BASEPATH/multicurl.sh -L --ipv4 -m $timeout -s -X POST -w "%{http_code}" -o "$response_file2" -H "Content-Type: application/json" --data "$rpc_method2" $ref)
curl_code2=$?
if [ $curl_code2 -eq 0 ]; then
if [[ $http_status_code2 -eq 200 ]]; then
response2=$(cat "$response_file2")
if $is_starknet; then
latest_block_hash2=$(echo "$response2" | jq -r '.result.block_hash')
elif $is_aztec; then
latest_block_hash2=$(echo "$response2" | jq -r '.result.blockHash')
else
latest_block_hash2=$(echo "$response2" | jq -r '.result.hash')
fi
rm "$response_file2"
# Proceed if hashes match (or both empty for Aztec when API omits hash)
if [ "$latest_block_hash" == "$latest_block_hash2" ]; then
response_file3=$(mktemp)
status_file3=$(mktemp)
if $is_aztec; then
# Aztec: node_getBlock("latest") - same single-request pattern as eth/starknet
rpc_method_latest='{"jsonrpc":"2.0","method":"node_getBlock","params":["latest"],"id":1}'
{
$BASEPATH/multicurl.sh -L --ipv4 -m $timeout -s -X POST -w "%{http_code} %{time_total}" -o "$response_file3" -H "Content-Type: application/json" --data "$rpc_method_latest" $ref > "$status_file3"
} &
pid3=$!
response_file4=$(mktemp)
status_file4=$(mktemp)
{
curl -L --ipv4 -m $timeout -s -X POST -w "%{http_code} %{time_total}" -o "$response_file4" -H "Content-Type: application/json" --data "$rpc_method_latest" $RPC_URL > "$status_file4"
} &
pid4=$!
else
if $is_starknet; then
rpc_method_latest='{"jsonrpc":"2.0","method":"starknet_getBlockWithTxHashes","params":["latest"],"id":1}'
else
rpc_method_latest='{"jsonrpc":"2.0","method":"eth_getBlockByNumber","params":["latest", false],"id":1}'
fi
{
$BASEPATH/multicurl.sh -L --ipv4 -m $timeout -s -X POST -w "%{http_code} %{time_total}" -o "$response_file3" -H "Content-Type: application/json" --data "$rpc_method_latest" $ref > "$status_file3"
} &
pid3=$!
response_file4=$(mktemp)
status_file4=$(mktemp)
{
curl -L --ipv4 -m $timeout -s -X POST -w "%{http_code} %{time_total}" -o "$response_file4" -H "Content-Type: application/json" --data "$rpc_method_latest" $RPC_URL > "$status_file4"
} &
pid4=$!
fi
wait $pid3
curl_code3=$?
http_status_code3=$(cat "$status_file3" | cut -d ' ' -f 1)
request_time3=$(cat "$status_file3" | cut -d ' ' -f 2)
rm "$status_file3"
wait $pid4
curl_code4=$?
http_status_code4=$(cat "$status_file4" | cut -d ' ' -f 1)
request_time4=$(cat "$status_file4" | cut -d ' ' -f 2)
rm "$status_file4"
# echo "lets check"
if [ $curl_code3 -eq 0 ]; then
if [[ $http_status_code3 -eq 200 ]]; then
response3=$(cat "$response_file3")
if $is_starknet; then
ref_num=$(echo "$response3" | jq -r '.result.block_number // empty')
elif $is_aztec; then
ref_num=$(echo "$response3" | jq -r '.result.header.globalVariables.blockNumber // empty')
else
ref_num_hex=$(echo "$response3" | jq -r '.result.number // empty')
ref_num=$([ -n "$ref_num_hex" ] && printf '%d' "$ref_num_hex" 2>/dev/null)
fi
# echo "refer: $latest_block_timestamp_decimal3"
rm "$response_file3"
if [ $curl_code4 -eq 0 ]; then
if [[ $http_status_code4 -eq 200 ]]; then
response4=$(cat "$response_file4")
if $is_starknet; then
local_num=$(echo "$response4" | jq -r '.result.block_number // empty')
elif $is_aztec; then
local_num=$(echo "$response4" | jq -r '.result.header.globalVariables.blockNumber // empty')
else
local_num_hex=$(echo "$response4" | jq -r '.result.number // empty')
local_num=$([ -n "$local_num_hex" ] && printf '%d' "$local_num_hex" 2>/dev/null)
fi
#echo "local: $latest_block_timestamp_decimal4"
rm "$response_file4"
# Lag in BLOCKS between the reference head and the local head
# (positive => local behind). Compare against dRPC's own per-chain
# thresholds (LAGGING_LAG / SYNCING_LAG from chains.yaml via
# sync-status.sh) so our status matches the dRPC gateway's view.
# dRPC uses the two thresholds inconsistently (sometimes
# lagging<syncing, sometimes the reverse), so treat the smaller as
# the online boundary and the larger as the syncing/drop boundary.
if [ -z "$ref_num" ] || [ -z "$local_num" ]; then
echo "error"
exit 1
fi
lag=$(( ref_num - local_num ))
lo=${LAGGING_LAG:-2}; hi=${SYNCING_LAG:-6}
if [ "$lo" -gt "$hi" ]; then tmp=$lo; lo=$hi; hi=$tmp; fi
if [ "$lag" -le "$lo" ]; then
echo "online"
exit 0
elif [ "$lag" -le "$hi" ]; then
echo "lagging"
exit 0
else
echo "syncing"
exit 1
fi
else
echo "error"
exit 1
fi
fi
fi
fi
else
# Hash mismatch - retry if we have attempts left
if [ $attempt -lt $MAX_RETRIES ]; then
rm "$response_file2"
attempt=$((attempt + 1))
continue
else
rm "$response_file2"
echo "forked"
exit 1
fi
fi
else
echo "unverified ($http_status_code2)"
exit 1
fi
fi
done
echo "unverified ($curl_code)"
exit 0
elif [ $time_difference -lt 60 ]; then
echo "online"
exit 0
else
echo "behind"
exit 1
fi
elif [[ $http_status_code -eq 404 ]]; then
echo "offline"
exit 1
elif [[ $http_status_code -eq 401 ]]; then
echo "unauthorized"
exit 1
elif [[ $http_status_code -eq 500 ]]; then
echo "error"
exit 1
else
echo "Unexpected HTTP status code: $http_status_code"
exit 1
fi
else
echo "timeout"
exit 1
fi