#!/bin/bash LOCAL_REPO_DIR="/home/limingjie/LMJ_Work/RobotCompetition/OmniSocketGo" KCP_PEER_BIN="./bin/kcppeer" OUTPUT_ROOT="/home/limingjie/LMJ_Work/RobotCompetition/KCPData/BCAlogs/" PEERB_POLL_INTERVAL_SEC=5 PEERB_MAX_POLLS=180 PEER_A_EXIT_WAIT_SEC=5 require_local_binary() { if [ ! -x "$1" ]; then echo "ERROR: 缺少可执行文件 $1" exit 1 fi } cleanup_remote_peerb() { ssh omni-peer bash -s <<'EOF' pids=$(ps -eo pid=,args= | awk '/[b]in\/kcppeer -id peer-b/ {print $1}') if [ -n "$pids" ]; then kill $pids 2>/dev/null || true fi pids=$(ps -eo pid=,args= | awk '/\/tmp\/peerb_batch\.sh/ {print $1}') if [ -n "$pids" ]; then kill $pids 2>/dev/null || true fi rm -f /tmp/peerb_batch_done /tmp/peerb_batch.sh /tmp/peerb_commands EOF } echo "=== 开始自动化测试 ===" cd "$LOCAL_REPO_DIR" require_local_binary "$KCP_PEER_BIN" echo ">>> 0. 清理上次残留进程..." pkill -f 'bin/kcppeer -id peer-a' 2>/dev/null || true cleanup_remote_peerb || exit 1 rm -rf logs rm -rf inbox/a mkdir -p logs inbox/a # 1. 清理残留 & 启动 Server D echo ">>> 1. 启动 Server D..." ssh bj-txy bash -s <<'EOF' pkill -f kcpserver 2>/dev/null || true pkill -f 'bin/kcpserver' 2>/dev/null || true sleep 1 cd /home/ubuntu/OmniSocketGo rm -rf logs mkdir -p logs if [ ! -x ./bin/kcpserver ]; then echo "ERROR: 缺少 ./bin/kcpserver" exit 1 fi setsid ./bin/kcpserver -listen 0.0.0.0:10909 \ -telemetry-peer peer-a-telemetry \ -kcp-ts-debug-log logs/d-kcp-ts.jsonl \ -kcp-session-stats-log logs/d-kcp-stats.jsonl > server_console.log 2>&1 /dev/null; then echo " Server D 就绪 (${i}s)" break fi if [ "$i" -eq 60 ]; then echo " ERROR: Server D 60s 内未就绪,退出" exit 1 fi sleep 1 done # 2. 启动本地 Peer-A echo ">>> 2. 启动本地 Peer-A..." PEER_A_CMD_FIFO="/tmp/peera_commands_$$" rm -f "$PEER_A_CMD_FIFO" mkfifo "$PEER_A_CMD_FIFO" nohup "$KCP_PEER_BIN" \ -id peer-a \ -server 81.70.156.140:10909 \ -inbox-dir inbox/a \ -latency-log logs/a-latency.jsonl \ -kcp-ts-debug-log logs/a-kcp-ts.jsonl \ -kcp-session-stats-log logs/a-kcp-stats.jsonl \ < "$PEER_A_CMD_FIFO" > logs/peera_console.log 2>&1 & PEER_A_PID=$! exec 4>"$PEER_A_CMD_FIFO" # 等待 peer-a 注册成功 echo " 等待 Peer-A 注册..." for i in $(seq 1 30); do if grep -Eq "opened KCP session as peer-a|connected to .* as peer-a( \\(KCP\\))?" logs/peera_console.log 2>/dev/null; then echo " Peer-A 就绪 (${i}s)" break fi if [ "$i" -eq 30 ]; then echo " WARNING: Peer-A 30s 内未就绪" fi sleep 1 done # 3. 在远端后台启动 peer-b 整个发送流程,不依赖长 SSH 连接 echo ">>> 3. 启动远端 Peer-B 并执行 50 轮打流测试..." ssh omni-peer "cd /home/boll/LMJWork/OmniSocketGo && rm -rf logs inbox/b && mkdir -p logs inbox/b" PEERB_DONE_FLAG="/tmp/peerb_batch_done" PEERB_BATCH_SCRIPT="/tmp/peerb_batch.sh" # 把整个发送脚本写到远端,setsid 后台执行 ssh omni-peer bash -s <<'DEPLOY_SCRIPT' DONE_FLAG="/tmp/peerb_batch_done" BATCH_SCRIPT="/tmp/peerb_batch.sh" rm -f "$DONE_FLAG" cat > "$BATCH_SCRIPT" <<'INNER_EOF' #!/bin/bash cd /home/boll/LMJWork/OmniSocketGo CMD_FIFO=/tmp/peerb_commands DONE_FLAG="/tmp/peerb_batch_done" rm -f "$CMD_FIFO" "$DONE_FLAG" mkfifo "$CMD_FIFO" if [ ! -x ./bin/kcppeer ]; then echo "ERROR: 缺少 ./bin/kcppeer" > logs/peerb_console.log echo "error" > "$DONE_FLAG" exit 1 fi # 启动 peer-b ./bin/kcppeer \ -id peer-b \ -server 81.70.156.140:10909 \ -inbox-dir inbox/b \ -latency-log logs/b-latency.jsonl \ -kcp-ts-debug-log logs/b-kcp-ts.jsonl \ -kcp-session-stats-log logs/b-kcp-stats.jsonl \ < "$CMD_FIFO" > logs/peerb_console.log 2>&1 & PEER_B_PID=$! exec 3>"$CMD_FIFO" # 等 peer-b 就绪 for i in $(seq 1 60); do if grep -Eq "opened KCP session as peer-b|connected to .* as peer-b( \\(KCP\\))?" logs/peerb_console.log 2>/dev/null; then break fi sleep 1 done # 50 轮发送 for i in $(seq 1 50); do echo "file peer-a /home/boll/test30.bin" >&3 sleep 1 echo "file peer-a /home/boll/test5.bin" >&3 sleep 1 done sleep 5 echo "quit" >&3 exec 3>&- rm -f "$CMD_FIFO" peer_b_exited=0 for i in $(seq 1 15); do if ! kill -0 $PEER_B_PID 2>/dev/null; then peer_b_exited=1 break fi sleep 1 done if [ "$peer_b_exited" -eq 0 ]; then kill $PEER_B_PID 2>/dev/null || true sleep 1 fi if kill -0 $PEER_B_PID 2>/dev/null; then kill -9 $PEER_B_PID 2>/dev/null || true fi wait $PEER_B_PID 2>/dev/null || true # 写完成标记 echo "done" > "$DONE_FLAG" INNER_EOF chmod +x "$BATCH_SCRIPT" setsid bash "$BATCH_SCRIPT" /dev/null 2>&1 & echo "peer-b batch launched in background" DEPLOY_SCRIPT # 本地轮询等待远端完成(短 SSH 连接,不怕断开) echo " 等待 peer-b 发送完成(预计 ~110 秒)..." for i in $(seq 1 "$PEERB_MAX_POLLS"); do PEERB_STATUS=$(ssh omni-peer "cat /tmp/peerb_batch_done 2>/dev/null || true") if [ "$PEERB_STATUS" = "done" ]; then ELAPSED_SEC=$(( (i - 1) * PEERB_POLL_INTERVAL_SEC )) echo " peer-b 发送完成(约 ${ELAPSED_SEC}s)" break fi if [ "$PEERB_STATUS" = "error" ]; then echo " ERROR: peer-b 后台任务启动失败,请检查远端 logs/peerb_console.log" exit 1 fi if [ "$i" -eq "$PEERB_MAX_POLLS" ]; then echo " ERROR: peer-b $((PEERB_MAX_POLLS * PEERB_POLL_INTERVAL_SEC))s 内未完成" exit 1 fi sleep "$PEERB_POLL_INTERVAL_SEC" done # 4. 清理 echo ">>> 4. 清理所有进程..." sleep 2 echo "quit" >&4 || true exec 4>&- for i in $(seq 1 "$PEER_A_EXIT_WAIT_SEC"); do if ! kill -0 "$PEER_A_PID" 2>/dev/null; then break fi sleep 1 done if kill -0 "$PEER_A_PID" 2>/dev/null; then kill "$PEER_A_PID" 2>/dev/null || true fi wait "$PEER_A_PID" 2>/dev/null || true rm -f "$PEER_A_CMD_FIFO" ssh bj-txy "pkill -f kcpserver || true; pkill -f 'bin/kcpserver' || true" # 获取当前时间戳,格式为 YYYYMMDD_HHMMSS TIMESTAMP=$(date +"%Y%m%d_%H%M%S") OUTPUT_DIR="$OUTPUT_ROOT/$TIMESTAMP" # 5. 拉取数据 & 生成报告 echo ">>> 5. 拉取数据并生成汇总报告..." mkdir -p "$OUTPUT_DIR" scp -o ServerAliveInterval=15 -P 10022 boll@175.178.116.187:/home/boll/LMJWork/OmniSocketGo/logs/b-latency.jsonl "$LOCAL_REPO_DIR/logs/b-latency.jsonl" || exit 1 (cd "$LOCAL_REPO_DIR/go" && go run ./cmd/latencysummary \ -input "$LOCAL_REPO_DIR/logs/a-latency.jsonl" \ -input "$LOCAL_REPO_DIR/logs/b-latency.jsonl" \ -output "$LOCAL_REPO_DIR/logs/latency-summary.jsonl") || exit 1 cd "$LOCAL_REPO_DIR/.." || exit 1 mv "$LOCAL_REPO_DIR/logs/a-latency.jsonl" "$OUTPUT_DIR/a-latency.jsonl" || exit 1 mv "$LOCAL_REPO_DIR/logs/b-latency.jsonl" "$OUTPUT_DIR/b-latency.jsonl" || exit 1 mv "$LOCAL_REPO_DIR/logs/latency-summary.jsonl" "$OUTPUT_DIR/latency-summary.jsonl" || exit 1 if [ -f "$LOCAL_REPO_DIR/logs/latency-summary.html" ]; then mv "$LOCAL_REPO_DIR/logs/latency-summary.html" "$OUTPUT_DIR/latency-summary.html" || exit 1 fi echo "=== 测试完成!==="