#!/usr/bin/env bash
# 采集器健康速查。跨地部署时两台都跑一遍，对着看。
set -uo pipefail

NAME="${NAME:-shadow}"
REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../../.." && pwd)"
OUT="$REPO_ROOT/research/out"

say() { printf '\n\033[1m==> %s\033[0m\n' "$*"; }

say "容器"
docker ps -a --filter "name=^${NAME}$" \
	--format 'table {{.Names}}\t{{.Status}}\t{{.RunningFor}}' || true

say "时钟"
if command -v chronyc >/dev/null 2>&1; then
	chronyc tracking | grep -E 'System time|Last offset|Leap status'
fi

say "最近心跳"
docker logs "$NAME" 2>&1 | grep '\[心跳\]' | tail -3 || echo "还没到第一次心跳（每 5 分钟一次）"

say "告警与异常"
docker logs "$NAME" 2>&1 \
	| grep -E '⚠|错误|失效|损坏|停滞|Traceback|退化' | tail -10 \
	|| echo "无"

say "输出规模"
for f in shadow_latency.csv shadow_drift.csv shadow_signals.csv; do
	p="$OUT/$f"
	if [[ -s "$p" ]]; then
		printf '  %-22s %8d 行\n' "$f" "$(( $(wc -l <"$p") - 1 ))"
	elif [[ -f "$p" ]]; then
		# 已建但表头还没冲刷：signals 只在有信号时才 flush，门控后每天仅 4~6 个
		printf '  %-22s %8s\n' "$f" "0（待首条）"
	else
		printf '  %-22s %8s\n' "$f" "无"
	fi
done
for f in shadow_books.jsonl.gz shadow_tape.jsonl.gz; do
	p="$OUT/$f"
	[[ -f "$p" ]] && printf '  %-22s %8s\n' "$f" "$(du -h "$p" | cut -f1)" \
		|| printf '  %-22s %8s\n' "$f" "无"
done

say "各站点行数（确认 site 列生效）"
p="$OUT/shadow_latency.csv"
if [[ -f "$p" ]]; then
	awk -F, 'NR>1{c[$1]++} END{for(s in c) printf "  %-16s %8d 行\n", s, c[s]}' "$p"
else
	echo "  尚无数据"
fi

say "到达延迟中位（本站，跨地对比的主指标）"
# 按列名取下标，不写死数字：加了 site 列之后字段整体右移过一次，
# 写死 $8 会静默变成读 lag_signal_ms
if [[ -s "$p" ]]; then
	for sym in BTC ETH SOL; do
		med=$(awk -F, -v s="$sym" '
			NR==1 { for (i=1;i<=NF;i++) { if ($i=="sym") si=i; if ($i=="lag_data_ms") li=i } ; next }
			$si==s && $li!="" { print $li }' "$p" | sort -n | awk '
			{ v[NR]=$1 } END { if (NR) printf "%.0f %d", v[int((NR+1)/2)], NR }')
		[[ -n "$med" ]] && printf '  %-5s 中位 %6sms  (n=%s)\n' "$sym" ${med} \
			|| printf '  %-5s 尚无数据\n' "$sym"
	done
	echo "  参考：本机（新加坡）补丁后实测 350~650ms，理论下限约 500ms"
fi