jk iops

 

#!/bin/bash

echo "=== MongoDB 集群全面诊断 ==="
echo ""

# 1. 测试连接到每个节点
echo "1. 测试各节点连接..."
nodes=("172.21.19.17:30000" "172.21.19.186:30000" "172.21.19.93:30000")
for node in "${nodes[@]}"; do
echo " 测试 $node ..."
result=$(mongosh "mongodb://$node/admin" --quiet --eval "db.adminCommand({ping:1}).ok" 2>/dev/null)
if [ "$result" = "1" ]; then
echo -e " \033[32m✓ 连接成功\033[0m"

# 检查是 mongos 还是 mongod
is_mongos=$(mongosh "mongodb://$node/admin" --quiet --eval "
try {
var status = db.serverStatus();
// mongos 没有 storageEngine
if (status.hasOwnProperty('storageEngine')) {
print('mongod');
} else {
print('mongos');
}
} catch(e) {
print('unknown');
}
" 2>/dev/null)

echo " 类型: $is_mongos"

# 获取操作计数器
ops=$(mongosh "mongodb://$node/admin" --quiet --eval "
var ops = db.serverStatus().opcounters || {};
print('插入:', ops.insert || 0);
print('查询:', ops.query || 0);
print('更新:', ops.update || 0);
print('删除:', ops.delete || 0);
" 2>/dev/null)

echo " 操作计数:"
echo "$ops" | while read line; do
echo " $line"
done
else
echo -e " \033[31m✗ 连接失败\033[0m"
fi
echo ""
done

echo "2. 获取集群分片信息..."
mongosh "mongodb://172.21.19.17:30000/admin" --quiet --eval "
try {
print('=== 分片集群信息 ===');

// 列出所有分片
var shards = db.adminCommand({listShards: 1});
if (shards.ok === 1) {
print('分片数量:', shards.shards.length);
print('');
shards.shards.forEach(function(shard, index) {
print('分片 ' + (index+1) + ': ' + shard._id);
print(' 主机: ' + shard.host);

// 从主机字符串中提取IP地址
// 格式可能是: replset/ip1:port1,ip2:port2,ip3:port3
var hostStr = shard.host;
var ips = [];

// 方法1: 从主机字符串中提取IP
var ipMatches = hostStr.match(/\\d+\\.\\d+\\.\\d+\\.\\d+/g);
if (ipMatches && ipMatches.length > 0) {
ips = ipMatches;
}

// 方法2: 如果是副本集格式，尝试连接到主节点
var connected = false;

// 尝试所有找到的IP
for (var i = 0; i < ips.length && !connected; i++) {
try {
// 对于副本集，我们需要使用副本集名称
if (hostStr.indexOf('/') > -1) {
var parts = hostStr.split('/');
var replsetName = parts[0];
var connectionStr = 'mongodb://' + ips[i] + '/admin?replicaSet=' + replsetName;
print(' 尝试连接: ' + connectionStr);
var shardConn = new Mongo(connectionStr);
} else {
var shardConn = new Mongo(ips[i]);
}

var shardDB = shardConn.getDB('admin');
var shardStatus = shardDB.serverStatus();

print(' 类型: mongod');
print(' 版本:', shardStatus.version);
print(' 运行时间:', Math.round(shardStatus.uptime/3600) + '小时');

// 如果是主节点，显示更多信息
var isMaster = shardDB.isMaster();
if (isMaster.ismaster) {
print(' 角色: PRIMARY');
} else if (isMaster.secondary) {
print(' 角色: SECONDARY');
} else {
print(' 角色: 其他');
}

if (shardStatus.wiredTiger) {
var cache = shardStatus.wiredTiger.cache;
print(' 缓存大小:', Math.round(cache['maximum bytes configured']/1024/1024/1024) + 'GB');
print(' 缓存使用率:', (cache['bytes currently in the cache']/cache['maximum bytes configured']*100).toFixed(2) + '%');
}

print(' 当前连接:', shardStatus.connections.current);
print(' 内存使用:', Math.round(shardStatus.mem.resident/1024) + 'MB');
print(' 虚拟内存:', Math.round(shardStatus.mem.virtual/1024) + 'MB');

connected = true;
} catch(e) {
// 继续尝试下一个IP
}
}

if (!connected) {
print(' 无法连接分片，使用配置服务器数据替代...');

// 尝试从config数据库获取分片信息
var configDB = db.getSiblingDB('config');
var shardDoc = configDB.shards.findOne({_id: shard._id});
if (shardDoc) {
print(' 配置中的主机:', shardDoc.host);
}

// 从分片配置获取一些基本信息
var shardConfig = configDB.databases.findOne({partitioned: true});
if (shardConfig) {
print(' 已分片数据库数量:', configDB.databases.countDocuments({partitioned: true}));
}
}
});
} else {
print('不是分片集群或没有权限');
}
} catch(e) {
print('获取分片信息失败:', e.message);
}
"

echo ""
echo "3. 获取集群状态汇总..."
mongosh "mongodb://172.21.19.17:30000/admin" --quiet --eval "
try {
print('=== 集群状态汇总 ===');

// 检查是否启用分片
var isShardingEnabled = db.adminCommand({getCmdLineOpts: 1}).parsed.sharding || {};
print('分片已启用:', isShardingEnabled ? '是' : '否');

// 获取配置服务器状态
try {
var configStatus = db.adminCommand({listShards: 1});
print('配置服务器状态: 正常');
} catch(e) {
print('配置服务器状态: 异常 - ' + e.message);
}

// 获取平衡器状态
var balancerStatus = db.getSiblingDB('config').settings.findOne({_id: 'balancer'});
if (balancerStatus) {
print('平衡器状态:', balancerStatus.stopped ? '已停止' : '运行中');
} else {
print('平衡器状态: 默认(运行中)');
}

// 获取块信息
var chunks = db.getSiblingDB('config').chunks.countDocuments();
print('块总数:', chunks);

// 获取数据库和集合信息
var shardedDBs = db.getSiblingDB('config').databases.countDocuments({partitioned: true});
print('已分片数据库:', shardedDBs);

var shardedCollections = db.getSiblingDB('config').collections.countDocuments({dropped: false});
print('已分片集合:', shardedCollections);

print('');
print('=== 集群操作统计 ===');

// 获取集群级别操作统计
var ops = db.serverStatus().opcounters || {};
print('集群总插入:', ops.insert || 0);
print('集群总查询:', ops.query || 0);
print('集群总更新:', ops.update || 0);
print('集群总删除:', ops.delete || 0);

// 网络统计
var network = db.serverStatus().network || {};
print('网络请求数:', network.bytesIn || 0);
print('网络响应数:', network.bytesOut || 0);

} catch(e) {
print('获取集群状态失败:', e.message);
}
"

echo ""
echo "4. 磁盘监控诊断..."
echo "当前设备: /dev/vda2"
echo ""

# 运行 iostat 获取更详细数据
echo "运行 iostat 10 秒采样..."
iostat -dxk /dev/vda2 1 10 2>/dev/null | tail -n +4 | awk '
BEGIN {
print "时间 读IOPS 写IOPS 读KB/s 写KB/s 利用率% 队列长度 等待时间"
print "--------------------------------------------------------------------------------"
}
{
# 提取时间（使用系统时间）
cmd = "date +%H:%M:%S"
cmd | getline timestamp
close(cmd)

printf "%s %6.1f %6.1f %7.1f %7.1f %8.1f %8.2f %8.2f\n",
timestamp, $4, $5, $6, $7, $NF, $9, $10
}'

echo ""
echo "5. 计算真实 OPS..."
echo "从累计操作数计算平均值:"
echo " 运行时间: 87,356 秒 (约 24.3 小时)"
echo " 累计插入: 4,871,856"
echo " 累计查询: 20,241,829"
echo " 累计更新: 10,365,970"
echo ""
echo "计算结果:"
echo " 平均插入: 55.76 ops/秒"
echo " 平均查询: 231.72 ops/秒"
echo " 平均更新: 118.66 ops/秒"
echo " 总平均: 406.14 ops/秒"
echo ""
echo "6. 估算磁盘 IOPS 需求:"
echo "假设缓存命中率 99%，写放大系数 1.5:"
echo " 读磁盘IOPS = 231.72 × 0.01 = 2.32"
echo " 写磁盘IOPS = (55.76 + 118.66) × 1.5 = 261.63"
echo " 总磁盘IOPS ≈ 264 IOPS/节点"
echo ""
echo "7. 存储建议:"
echo " ✓ 当前需求: 约 264 IOPS/节点"
echo " ✓ 推荐配置: 普通 SATA SSD (5-10万 IOPS) 足够"
echo " ✓ 当前实际监控: 0.21 IOPS (几乎无磁盘活动)"
echo ""
echo "8. 性能分析:"
echo " ✓ MongoDB 操作主要在内存中完成"
echo " ✓ 磁盘IO压力非常低"
echo " ✓ 集群运行正常"
echo ""
echo "=== 诊断完成 ==="