Files
ethereum-rpc-docker/peer-capture.sh
rob 2919960697 Peer persistence hooks: capture before stop, reinject after start
Clients (esp. discovery-disabled OP-stack EL) lose peers across recreate.
Generic peer-capture/peer-reinject hooks + force-recreate wiring restore
the peerset from local peer-state via in-network admin_addPeer.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-19 05:25:27 +00:00

121 lines
3.8 KiB
Bash
Executable File

#!/bin/bash
# peer-capture.sh — generic .before-stop hook: snapshot admin peers for reinject.
# In-network vantage (rpc_chains curl): public gateway rejects admin; some hosts
# firewall host→bridge-IP. Client-aware peer list → /root/rpc/peer-state/<path>.json
#
# Usage: peer-capture.sh <compose-path>
# compose-path: e.g. op/reth/lisk-mainnet-op-reth-pruned-trace
set -euo pipefail
COMPOSE_PATH="${1:-}"
if [ -z "$COMPOSE_PATH" ]; then
echo "Usage: $0 <compose-path>" >&2
exit 1
fi
BASEPATH="$(cd "$(dirname "$0")" && pwd)"
STATE_DIR="$BASEPATH/peer-state"
mkdir -p "$STATE_DIR"
SAFE="$(echo "$COMPOSE_PATH" | sed 's|[^a-zA-Z0-9._-]|_|g')"
OUT="$STATE_DIR/${SAFE}.json"
CURL_IMAGE="${PEER_CURL_IMAGE:-curlimages/curl:8.5.0}"
LOG="${PEER_HOOK_LOG:-/var/log/peer-hooks.log}"
log() { echo "$(date '+%F %T') peer-capture[$COMPOSE_PATH]: $*" | tee -a "$LOG" >&2; }
# Documented skips
case "$COMPOSE_PATH" in
*nitro*|*zksync*|*zk-sync*)
log "skip (nitro/zksync)"
exit 0
;;
esac
BASE="$(basename "$COMPOSE_PATH" | sed 's/--.*//')"
# Find execution container (same heuristics as characterize-node exec_container)
EL=""
while read -r n; do
[ -z "$n" ] && continue
case "$n" in
rpc-*) ;;
*) continue ;;
esac
case "$n" in
*-node-*|*-relay-*) continue ;;
esac
core="$(echo "$n" | sed -E 's/^rpc-(.+)-[0-9]+$/\1/')"
if [ "$BASE" = "$core" ] || [[ "$BASE" == "$core"-* ]]; then
EL="$n"
break
fi
done < <(docker ps --format '{{.Names}}')
if [ -z "$EL" ]; then
log "no execution container for $BASE"
exit 0
fi
IP="$(docker inspect -f '{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' "$EL" 2>/dev/null || true)"
if [ -z "$IP" ]; then
log "no IP for $EL"
exit 0
fi
docker rm -f peer-hook-probe >/dev/null 2>&1 || true
docker image inspect "$CURL_IMAGE" >/dev/null 2>&1 || docker pull -q "$CURL_IMAGE" >/dev/null
docker run -d --name peer-hook-probe --network rpc_chains "$CURL_IMAGE" sleep 90 >/dev/null
rpc() {
local method="$1"
docker exec peer-hook-probe curl -s --max-time 8 -X POST \
-H 'Content-Type: application/json' \
--data "{\"jsonrpc\":\"2.0\",\"method\":\"$method\",\"params\":[],\"id\":1}" \
"http://${IP}:8545" 2>/dev/null || true
}
INFO="$(rpc admin_nodeInfo)"
CLIENT_NAME="$(echo "$INFO" | jq -r '.result.name // empty' 2>/dev/null || true)"
PEERS_JSON="$(rpc admin_peers)"
PEER_COUNT="$(echo "$PEERS_JSON" | jq -r '.result | length' 2>/dev/null || echo 0)"
# Detect client type (connect-peers.sh logic)
NAME_LC="$(echo "$CLIENT_NAME" | tr '[:upper:]' '[:lower:]')"
CLIENT_TYPE="unknown"
case "$NAME_LC" in
*geth*) CLIENT_TYPE="geth" ;;
*reth*) CLIENT_TYPE="reth" ;;
*erigon*) CLIENT_TYPE="erigon" ;;
*nethermind*) CLIENT_TYPE="nethermind" ;;
*besu*) CLIENT_TYPE="besu" ;;
esac
TS="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
echo "$PEERS_JSON" | jq -c --arg path "$COMPOSE_PATH" --arg ts "$TS" \
--arg el "$EL" --arg cn "$CLIENT_NAME" --arg ct "$CLIENT_TYPE" \
--argjson pc "${PEER_COUNT:-0}" '
{
node_path: $path,
captured_at: $ts,
container: $el,
client_name: $cn,
client_type: $ct,
peer_count: $pc,
peers: [(.result // [])[] | select(.enode != null) | {
id: .enode,
client: (.name // ""),
layer: "execution"
}]
}' > "$OUT.tmp" 2>/dev/null || {
log "admin_peers failed; writing empty seed"
cat > "$OUT.tmp" <<EOF
{"node_path":"$COMPOSE_PATH","captured_at":"$TS","container":"$EL","client_name":"$CLIENT_NAME","client_type":"$CLIENT_TYPE","peer_count":0,"peers":[]}
EOF
}
mv -f "$OUT.tmp" "$OUT"
# also keep a plain enode list for easy playback
jq -r '.peers[].id' "$OUT" > "${OUT%.json}.enodes" 2>/dev/null || true
docker rm -f peer-hook-probe >/dev/null 2>&1 || true
log "captured ${PEER_COUNT:-0} peer(s) → $OUT"
exit 0