0813

2026-08-13 22:49:50 +08:00
parent 5e296bfe19
commit 2aaaa16ae2
24 changed files with 1415 additions and 327 deletions
--- a/auto-check/scripts/cpu.sh
+++ b/auto-check/scripts/cpu.sh
@@ -1,13 +1,18 @@
 echo "===TEST:cpu==="

+# 读取温度（sensors 不可用时输出空）
+read_temp() {
+	sensors 2>/dev/null | grep -oP '\+?\d+(\.\d+)?(?=°C)' | head -1
+}
+
 # 后台采样 CPU 使用率 + 温度
 SAMPLE_LOG=/tmp/auto-check-cpu-sample.log
-> "$SAMPLE_LOG"
+: > "$SAMPLE_LOG"
 (
 	while true; do
 		TS=$(date +%H:%M:%S)
-		CPU=$(top -bn1 2>/dev/null | grep "Cpu(s)" | sed 's/.*,\s*\([0-9.]*\)%*\s*id.*/\1/' | awk '{printf "%.1f", 100-$1}' || echo 0)
-		TEMP=$(sensors 2>/dev/null | grep -iE 'Package|Core\s*0|temp1|CPU' | head -1 | grep -oP '\d+\.?\d*' | head -1 || echo 0)
+		CPU=$(top -bn1 2>/dev/null | grep "Cpu(s)" | sed -n 's/.*,\s*\([0-9.]*\)%*\s*id.*/\1/p' | awk '{printf "%.1f", 100-$1}' 2>/dev/null)
+		TEMP=$(read_temp)
 		echo "${TS},${CPU},${TEMP}" >> "$SAMPLE_LOG"
 		sleep "${AUTOCHECK_SAMPLE_INTERVAL:-2}"
 	done
@@ -16,7 +21,7 @@ SAMPLE_PID=$!

 START_TIME=$(date +%s%N)
 set +e
-OUTPUT=$(stress-ng --cpu "${AUTOCHECK_THREADS:-4}" --cpu-method all --timeout "${AUTOCHECK_DURATION:-30}s" --metrics-brief --temp-path /tmp 2>&1)
+OUTPUT=$(stress-ng --cpu "${AUTOCHECK_THREADS:-4}" --cpu-method all --timeout "${AUTOCHECK_DURATION:-30}s" --temp-path /tmp --metrics-brief 2>&1)
 EXIT_CODE=$?
 set -e
 END_TIME=$(date +%s%N)
@@ -27,10 +32,27 @@ wait $SAMPLE_PID 2>/dev/null || true
 DURATION_MS=$(( (END_TIME - START_TIME) / 1000000 ))
 if [ "$EXIT_CODE" -eq 0 ]; then echo 'status:pass'; else echo 'status:fail'; fi
 echo "duration:${DURATION_MS}ms"
+
+# 跑分：各 stressor 的 bogo ops/s 之和
+SCORE=$(echo "$OUTPUT" | awk '
+/^stress-ng:/ && ($4=="cpu" || $4=="vm" || $4=="iomix" || $4=="memrate") {
+	v=$(NF-1)+0; if (v>0) s+=v
+}
+END { printf "%.1f", s+0 }')
+echo "score:${SCORE}"
+
+# 关键指标
+AVG_CPU=$(awk -F',' '$2!="" {s+=$2; n++} END{printf "%.1f", (n?s/n:0)}' "$SAMPLE_LOG")
+echo "metric:avg_cpu=${AVG_CPU}"
+MAX_TEMP=$(awk -F',' '$3!="" && $3+0>m {m=$3} END{printf "%.1f", m+0}' "$SAMPLE_LOG")
+HAS_TEMP=$(awk -F',' '$3!="" && $3+0>0{f=1} END{print f+0}' "$SAMPLE_LOG")
+if [ "$HAS_TEMP" = "1" ]; then echo "metric:max_temp=${MAX_TEMP}"; fi
+
 echo "output:${OUTPUT}"

-# 输出采样数据
+# 采样数据（首行表头）
 echo "===SAMPLES==="
+echo "time,cpu,temp"
 cat "$SAMPLE_LOG"
 echo "===END_SAMPLES==="

--- a/auto-check/scripts/disk.sh
+++ b/auto-check/scripts/disk.sh
@@ -1,14 +1,77 @@
 echo "===TEST:disk==="
-TEST_DIR=${AUTOCHECK_DISK_DIR:-/tmp/stress-disk-test}
-DISK_SIZE_MB=${AUTOCHECK_DISK_SIZE_MB:-1024}
+
+TEST_DIR=${AUTOCHECK_DISK_DIR:-/tmp/auto-check-disk}
+mkdir -p "$TEST_DIR"
+
 START_TIME=$(date +%s%N)
-set +e
-OUTPUT=$(mkdir -p "${TEST_DIR}" && stress-ng --iomix 2 --iomix-bytes "${DISK_SIZE_MB}M" --timeout "${AUTOCHECK_DURATION:-30}s" --metrics-brief 2>&1; rm -rf "${TEST_DIR}")
-EXIT_CODE=$?
-set -e
-END_TIME=$(date +%s%N)
-DURATION_MS=$(( (END_TIME - START_TIME) / 1000000 ))
-if [ "$EXIT_CODE" -eq 0 ]; then echo 'status:pass'; else echo 'status:fail'; fi
-echo "duration:${DURATION_MS}ms"
-echo "output:${OUTPUT}"
+
+if command -v fio >/dev/null 2>&1; then
+	FIO_IMG="$TEST_DIR/fio.img"
+	BW_LOG="$TEST_DIR/ac-bw"
+	IOPS_LOG="$TEST_DIR/ac-iops"
+	rm -f "$BW_LOG"*.log "$IOPS_LOG"*.log "$FIO_IMG"
+
+	set +e
+	OUTPUT=$(fio --name=autocheck --rw=randrw --rwmixread=70 --bs=4k \
+		--size="${AUTOCHECK_DISK_SIZE_MB:-512}M" --iodepth=32 --numjobs=1 \
+		--runtime="${AUTOCHECK_DURATION:-30}" --time_based --direct=1 \
+		--group_reporting --output-format=normal \
+		--write_bw_log="$BW_LOG" --write_iops_log="$IOPS_LOG" \
+		--log_avg_msec=1000 --filename="$FIO_IMG" 2>&1)
+	EXIT_CODE=$?
+	set -e
+
+	READ_LINE=$(echo "$OUTPUT" | grep -E '^[[:space:]]*read:' | head -1)
+	WRITE_LINE=$(echo "$OUTPUT" | grep -E '^[[:space:]]*write:' | head -1)
+	READ_IOPS=$(echo "$READ_LINE" | grep -oP 'IOPS=\K[0-9.]+' | head -1)
+	READ_BW=$(echo "$READ_LINE" | grep -oP 'BW=\K[0-9.]+' | head -1)
+	WRITE_IOPS=$(echo "$WRITE_LINE" | grep -oP 'IOPS=\K[0-9.]+' | head -1)
+	WRITE_BW=$(echo "$WRITE_LINE" | grep -oP 'BW=\K[0-9.]+' | head -1)
+	SCORE=$(awk -v r="${READ_BW:-0}" -v w="${WRITE_BW:-0}" 'BEGIN{printf "%.1f", r+w}')
+	IOPS_TOTAL=$(awk -v r="${READ_IOPS:-0}" -v w="${WRITE_IOPS:-0}" 'BEGIN{printf "%.1f", r+w}')
+
+	END_TIME=$(date +%s%N)
+	DURATION_MS=$(( (END_TIME - START_TIME) / 1000000 ))
+	if [ "$EXIT_CODE" -eq 0 ]; then echo 'status:pass'; else echo 'status:fail'; fi
+	echo "duration:${DURATION_MS}ms"
+	echo "score:${SCORE}"
+	echo "metric:iops_total=${IOPS_TOTAL}"
+	[ -n "$READ_IOPS" ] && echo "metric:read_iops=${READ_IOPS}"
+	[ -n "$WRITE_IOPS" ] && echo "metric:write_iops=${WRITE_IOPS}"
+	[ -n "$READ_BW" ] && echo "metric:read_bw=${READ_BW}"
+	[ -n "$WRITE_BW" ] && echo "metric:write_bw=${WRITE_BW}"
+	[ -n "$READ_BW" ] && [ -n "$WRITE_BW" ] && echo "metric:bw_total=${SCORE}"
+	echo "output:${OUTPUT}"
+
+	# IOPS / 吞吐 曲线（按秒聚合）
+	BW_LOG_FILE=$(ls "$BW_LOG"*.log 2>/dev/null | head -1)
+	IOPS_LOG_FILE=$(ls "$IOPS_LOG"*.log 2>/dev/null | head -1)
+	echo "===SAMPLES==="
+	echo "time,iops_total,bw_total"
+	if [ -n "$BW_LOG_FILE" ] && [ -n "$IOPS_LOG_FILE" ]; then
+		awk -F',' '
+		FNR==NR { t=int($1/1000); if($3==0) ri[t]=$2; else wi[t]=$2; seen[t]=1; next }
+		{ t=int($1/1000); if($3==0) rb[t]=$2; else wb[t]=$2; seen[t]=1 }
+		END { for (k in seen) printf "%d,%.0f,%.1f\n", k, ri[k]+wi[k], (rb[k]+wb[k])/1024 }
+		' "$IOPS_LOG_FILE" "$BW_LOG_FILE" | sort -n
+	fi
+	echo "===END_SAMPLES==="
+
+	rm -f "$FIO_IMG" "$BW_LOG"*.log "$IOPS_LOG"*.log
+else
+	# 无 fio，降级为 stress-ng 磁盘 IO 压测（无 IOPS/吞吐曲线）
+	set +e
+	OUTPUT=$(stress-ng --iomix "${AUTOCHECK_THREADS:-2}" --iomix-bytes "${AUTOCHECK_DISK_MB:-512}M" --timeout "${AUTOCHECK_DURATION:-30}s" --temp-path /tmp --metrics-brief 2>&1)
+	EXIT_CODE=$?
+	set -e
+	SCORE=$(echo "$OUTPUT" | awk '/^stress-ng:/ && $4=="iomix" { v=$(NF-1)+0; if(v>0) s+=v } END{printf "%.1f", s+0}')
+
+	END_TIME=$(date +%s%N)
+	DURATION_MS=$(( (END_TIME - START_TIME) / 1000000 ))
+	if [ "$EXIT_CODE" -eq 0 ]; then echo 'status:pass'; else echo 'status:fail'; fi
+	echo "duration:${DURATION_MS}ms"
+	echo "score:${SCORE}"
+	echo "output:${OUTPUT}"
+fi
+
 echo ''
--- a/auto-check/scripts/full.sh
+++ b/auto-check/scripts/full.sh
@@ -1,12 +1,71 @@
 echo "===TEST:full==="
+
+# 读取温度（sensors 不可用时输出空）
+read_temp() {
+	sensors 2>/dev/null | grep -oP '\+?\d+(\.\d+)?(?=°C)' | head -1
+}
+
+# 后台采样 CPU + 内存 + 温度
+SAMPLE_LOG=/tmp/auto-check-full-sample.log
+: > "$SAMPLE_LOG"
+(
+	while true; do
+		TS=$(date +%H:%M:%S)
+		CPU=$(top -bn1 2>/dev/null | grep "Cpu(s)" | sed -n 's/.*,\s*\([0-9.]*\)%*\s*id.*/\1/p' | awk '{printf "%.1f", 100-$1}' 2>/dev/null)
+		MEM_TOTAL=$(awk '/^MemTotal:/{print $2}' /proc/meminfo)
+		MEM_AVAIL=$(awk '/^MemAvailable:/{print $2}' /proc/meminfo)
+		if [ -z "$MEM_AVAIL" ] || [ "$MEM_AVAIL" = "0" ]; then
+			MEM_AVAIL=$(awk '/^MemFree:/{print $2}' /proc/meminfo)
+		fi
+		MEM_PCT=$(awk -v t="$MEM_TOTAL" -v a="$MEM_AVAIL" 'BEGIN{if(t>0) printf "%.1f", (t-a)/t*100}')
+		TEMP=$(read_temp)
+		echo "${TS},${CPU},${MEM_PCT},${TEMP}" >> "$SAMPLE_LOG"
+		sleep "${AUTOCHECK_SAMPLE_INTERVAL:-2}"
+	done
+) &
+SAMPLE_PID=$!
+
 START_TIME=$(date +%s%N)
 set +e
-OUTPUT=$(stress-ng --cpu "${AUTOCHECK_THREADS:-4}" --vm 2 --vm-bytes 128M --iomix 1 --iomix-bytes 256M --timeout "${AUTOCHECK_DURATION:-30}s" --metrics-brief 2>&1)
+OUTPUT=$(stress-ng \
+	--cpu "${AUTOCHECK_THREADS:-4}" --cpu-method all \
+	--vm "${AUTOCHECK_THREADS:-4}" --vm-bytes "${AUTOCHECK_MEM_MB:-256}M" --vm-method all \
+	--iomix 2 --iomix-bytes "${AUTOCHECK_DISK_SIZE_MB:-512}M" \
+	--timeout "${AUTOCHECK_DURATION:-30}s" --temp-path /tmp --metrics-brief 2>&1)
 EXIT_CODE=$?
 set -e
 END_TIME=$(date +%s%N)
+
+kill $SAMPLE_PID 2>/dev/null || true
+wait $SAMPLE_PID 2>/dev/null || true
+
 DURATION_MS=$(( (END_TIME - START_TIME) / 1000000 ))
 if [ "$EXIT_CODE" -eq 0 ]; then echo 'status:pass'; else echo 'status:fail'; fi
 echo "duration:${DURATION_MS}ms"
+
+# 跑分：各 stressor 的 bogo ops/s 之和
+SCORE=$(echo "$OUTPUT" | awk '
+/^stress-ng:/ && ($4=="cpu" || $4=="vm" || $4=="iomix" || $4=="memrate") {
+	v=$(NF-1)+0; if (v>0) s+=v
+}
+END { printf "%.1f", s+0 }')
+echo "score:${SCORE}"
+
+# 关键指标
+AVG_CPU=$(awk -F',' '$2!="" {s+=$2; n++} END{printf "%.1f", (n?s/n:0)}' "$SAMPLE_LOG")
+echo "metric:avg_cpu=${AVG_CPU}"
+MAX_MEM=$(awk -F',' '$3!="" && $3+0>m {m=$3} END{printf "%.1f", m+0}' "$SAMPLE_LOG")
+echo "metric:mem_used_pct=${MAX_MEM}"
+MAX_TEMP=$(awk -F',' '$4!="" && $4+0>m {m=$4} END{printf "%.1f", m+0}' "$SAMPLE_LOG")
+HAS_TEMP=$(awk -F',' '$4!="" && $4+0>0{f=1} END{print f+0}' "$SAMPLE_LOG")
+if [ "$HAS_TEMP" = "1" ]; then echo "metric:max_temp=${MAX_TEMP}"; fi
+
 echo "output:${OUTPUT}"
+
+# 采样数据（首行表头）
+echo "===SAMPLES==="
+echo "time,cpu,mem_used_pct,temp"
+cat "$SAMPLE_LOG"
+echo "===END_SAMPLES==="
+
 echo ''
--- a/auto-check/scripts/memory.sh
+++ b/auto-check/scripts/memory.sh
@@ -1,14 +1,64 @@
 echo "===TEST:memory==="
-VM_WORKERS=${AUTOCHECK_THREADS:-4}
-if [ "$VM_WORKERS" -gt 4 ]; then VM_WORKERS=4; fi
+
+# 读取温度（sensors 不可用时输出空）
+read_temp() {
+	sensors 2>/dev/null | grep -oP '\+?\d+(\.\d+)?(?=°C)' | head -1
+}
+
+# 后台采样内存使用率 + 温度
+SAMPLE_LOG=/tmp/auto-check-memory-sample.log
+: > "$SAMPLE_LOG"
+(
+	while true; do
+		TS=$(date +%H:%M:%S)
+		MEM_TOTAL=$(awk '/^MemTotal:/{print $2}' /proc/meminfo)
+		MEM_AVAIL=$(awk '/^MemAvailable:/{print $2}' /proc/meminfo)
+		if [ -z "$MEM_AVAIL" ] || [ "$MEM_AVAIL" = "0" ]; then
+			MEM_AVAIL=$(awk '/^MemFree:/{print $2}' /proc/meminfo)
+		fi
+		MEM_PCT=$(awk -v t="$MEM_TOTAL" -v a="$MEM_AVAIL" 'BEGIN{if(t>0) printf "%.1f", (t-a)/t*100}')
+		TEMP=$(read_temp)
+		echo "${TS},${MEM_PCT},${TEMP}" >> "$SAMPLE_LOG"
+		sleep "${AUTOCHECK_SAMPLE_INTERVAL:-2}"
+	done
+) &
+SAMPLE_PID=$!
+
 START_TIME=$(date +%s%N)
 set +e
-OUTPUT=$(stress-ng --vm "${VM_WORKERS}" --vm-bytes 256M --vm-method all --timeout "${AUTOCHECK_DURATION:-30}s" --metrics-brief 2>&1)
+OUTPUT=$(stress-ng --vm "${AUTOCHECK_THREADS:-4}" --vm-bytes "${AUTOCHECK_MEM_MB:-256}M" --vm-method all --timeout "${AUTOCHECK_DURATION:-30}s" --temp-path /tmp --metrics-brief 2>&1)
 EXIT_CODE=$?
 set -e
 END_TIME=$(date +%s%N)
+
+kill $SAMPLE_PID 2>/dev/null || true
+wait $SAMPLE_PID 2>/dev/null || true
+
 DURATION_MS=$(( (END_TIME - START_TIME) / 1000000 ))
 if [ "$EXIT_CODE" -eq 0 ]; then echo 'status:pass'; else echo 'status:fail'; fi
 echo "duration:${DURATION_MS}ms"
+
+# 跑分：各 stressor 的 bogo ops/s 之和
+SCORE=$(echo "$OUTPUT" | awk '
+/^stress-ng:/ && ($4=="cpu" || $4=="vm" || $4=="iomix" || $4=="memrate") {
+	v=$(NF-1)+0; if (v>0) s+=v
+}
+END { printf "%.1f", s+0 }')
+echo "score:${SCORE}"
+
+# 关键指标
+MAX_MEM=$(awk -F',' '$2!="" && $2+0>m {m=$2} END{printf "%.1f", m+0}' "$SAMPLE_LOG")
+echo "metric:mem_used_pct=${MAX_MEM}"
+MAX_TEMP=$(awk -F',' '$3!="" && $3+0>m {m=$3} END{printf "%.1f", m+0}' "$SAMPLE_LOG")
+HAS_TEMP=$(awk -F',' '$3!="" && $3+0>0{f=1} END{print f+0}' "$SAMPLE_LOG")
+if [ "$HAS_TEMP" = "1" ]; then echo "metric:max_temp=${MAX_TEMP}"; fi
+
 echo "output:${OUTPUT}"
+
+# 采样数据（首行表头）
+echo "===SAMPLES==="
+echo "time,mem_used_pct,temp"
+cat "$SAMPLE_LOG"
+echo "===END_SAMPLES==="
+
 echo ''
--- a/auto-check/scripts/temp_end.sh
+++ b/auto-check/scripts/temp_end.sh
@@ -1,12 +1,15 @@
 echo '===MONITOR:temp==='
 if [ -f "$TEMP_LOG" ] && [ -s "$TEMP_LOG" ]; then
+	MAX_T=$(sort -n "$TEMP_LOG" | tail -1)
+	AVG_T=$(awk '{s+=$1; n++} END{printf "%.1f", (n?s/n:0)}' "$TEMP_LOG")
+	CNT=$(wc -l < "$TEMP_LOG")
 	echo 'status:pass'
-	TEMP_MAX=$(grep -oP '\d+\.?\d*°C' "$TEMP_LOG" | sort -t. -k1 -n | tail -1 || echo 'unknown')
-	echo "samples:$(wc -l < "$TEMP_LOG")"
-	echo "max:${TEMP_MAX}"
-	echo "output:$(tail -10 "$TEMP_LOG")"
+	echo "metric:max_temp=${MAX_T}"
+	echo "metric:avg_temp=${AVG_T}"
+	echo "output:采样 ${CNT} 次, 最高 ${MAX_T}°C, 平均 ${AVG_T}°C"
 else
 	echo 'status:skip'
+	echo 'output:未采集到温度数据'
 fi
 echo ''
 kill $TEMP_PID 2>/dev/null || true
--- a/auto-check/scripts/temp_start.sh
+++ b/auto-check/scripts/temp_start.sh
@@ -1,4 +1,8 @@
 TEMP_LOG=/tmp/auto-check-temp.log
 > "$TEMP_LOG"
-(while true; do sensors 2>/dev/null | grep -i 'temp\|core\|cpu' | head -5 >> "$TEMP_LOG"; sleep "${AUTOCHECK_TEMP_INTERVAL:-10}"; done) &
+(while true; do
+	T=$(sensors 2>/dev/null | grep -oP '\+?\d+(\.\d+)?(?=°C)' | head -1)
+	[ -n "$T" ] && echo "$T" >> "$TEMP_LOG"
+	sleep "${AUTOCHECK_TEMP_INTERVAL:-10}"
+done) &
 TEMP_PID=$!