使用mysql-proxy 快速实现mysql 集群 读写分离（转）

使用mysql-proxy 快速实现mysql 集群 读写分离

目前较为常见的mysql读写分离分为两种：
1、 基于程序代码内部实现：在代码中对select操作分发到从库；其它操作由主库执行；这类方法也是目前生产环境应用最广泛，知名的如DISCUZ X2。优点是性能较好，因为在程序代码中实现，不需要增加额外的设备作为硬件开支。缺点是需要开发人员来实现，运维人员无从下手。

2、 基于中间代理层实现：我们都知道代理一般是位于客户端和服务器之间，代理服务器接到客户端请求后通过判断然后转发到后端数据库。在这有两个代表性程序

mysql-proxy：mysql-proxy为mysql开源项目，通过其自带的lua脚本进行sql判断，虽然是mysql官方产品，但是mysql官方并不建议将mysql-proxy用到生产环境。
amoeba：由陈思儒开发，作者曾就职于阿里巴巴，现就职于盛大。该程序由java语言进行开发，目前只听说阿里巴巴将其用于生产环境。另外，此项目严重缺少维护和推广（作者有个官方博客，很多用户反馈的问题发现作者不理睬）
经过上述简单的比较，通过程序代码实现mysql读写分离自然是一个不错的选择。但是并不是所有的应用都适合在程序代码中实现读写分离，像大型SNS、B2C这类应用可以在代码中实现，因为这样对程序代码本身改动较小；像一些大型复杂的java应用，这种类型的应用在代码中实现对代码改动就较大了。所以，像这种应用一般就会考虑使用代理层来实现。

下面我们看一下如何搭建mysql-proxy来实现mysql读写分离

环境拓扑如下：

关于mysql、mysql主从的搭建，在此不再演示，如下的操作均在mysql-proxy（192.168.1.200）服务器进行
一、安装mysql-proxy
1、安装lua (mysql-proxy需要使用lua脚本进行数据转发)
#tar zxvf lua-5.1.4.tar.gz
#cd lua-5.1.4
#vi Makefile，修改INSTALL_TOP= /usr/local/lua
#make posix
#make install

2、安装libevent
#tar zxvf libevent-2.0.8-rc.tar.gz
#cd libevent-2.0.8-rc
#./configure --prefix=/usr/local/libevent
#make && make install

3、安装check
#tar zxvf check-0.9.8.tar.gz
#cd check-0.9.8
#./configure && make && make install

4、安装mysql客户端
#tar zxvf mysql-5.0.92.tar.gz
#cd mysql-5.0.92
#./configure --without-server && make && make install

5、设置环境变量 （安装mysql-proxy所需变量）
#vi /etc/profile
export LUA_CFLAGS="-I/usr/local/lua/include" LUA_LIBS="-L/usr/local/lua/lib -llua -ldl" LDFLAGS="-L/usr/local/libevent/lib -lm"
export CPPFLAGS="-I/usr/local/libevent/include"
export CFLAGS="-I/usr/local/libevent/include"
# source /etc/profile

6、安装mysql-proxy
#tar zxvf mysql-proxy-0.6.0.tar.gz
#cd mysql-proxy-0.6.0
# ./configure --prefix=/usr/local/mysql-proxy --with-mysql --with-lua
#make && make install

7、启动mysql-proxy
本次对两台数据库实现了读写分离；mysql-master为可读可写，mysql-slave为只读
#/usr/local/mysql-proxy/sbin/mysql-proxy --proxy-backend-addresses=192.168.1.201:3306 --proxy-read-only-backend-addresses=192.168.1.202:3306 --proxy-lua-script=/usr/local/mysql-proxy/share/mysql-proxy/rw-splitting.lua &

注：如果正常情况下启动后终端不会有任何提示信息，mysql-proxy启动后会启动两个端口4040和4041，4040用于SQL转发，4041用于管理mysql-proxy。如有多个mysql-slave可以依次在后面添加


二、测试
1、连接测试
因为默认情况下mysql数据库不允许用户在远程连接
mysql>grant all privileges on *.* to identified by '123456';
mysql>flush privileges;

客户端连接
#mysql -uroot -p123456 -h192.168.1.200 -P4040


2、读写分离测试
为了测试出mysql读写分离的真实性，在测试之前，需要开启两台mysql的log功能，然后在mysql-slave服务器停止复制
① 、在两台mysql配置文件my.cnf中加入log=query.log，然后重启

② 、在mysql-slave上执行SQL语句stop slave

③ 、在两台mysql上执行#tail -f /usr/local/mysql/var/query.log

④ 、在客户端上连接mysql（三个连接以上），然后执行create、select等SQL语句，观察两台mysql的日志有何变化

注：生产环境中除了进行程序调试外，其它不要开启mysql查询日志，因为查询日志记录了客户端的所有语句，频繁的IO操作将会导致mysql整体性能下降

总结：在上述环境中，mysql-proxy和mysql-master、mysql-slave三台服务器均存在单点故障。如果在可用性要求较高的场合，单点隐患是绝对不允许的。为了避免mysql-proxy单点隐患有两种方法，一种方法是mysql-proxy配合keepalived做双机，另一种方法是将mysql-proxy和应用服务安装到同一台服务器上；为了避免mysql-master单点故障可以使用DRBD+heartbear做双机；避免mysql-slave单点故障增加多台mysql-slave即可，因为mysql-proxy会自动屏蔽后端发生故障的mysql-slave。

 

附: mysql-proxy LUA 读写分离脚本代码:

--[[
--
-- author : KDr2
-- version 0.01
-- SYNOPSIS:
--- 1.维护了一个连接池
--- 2.读写分离，简单的将select开头的语句放到slave上执行
--- 3.事务支持，所有事务放到master上执行，事务中不更改连接
--- 4.简单日志
--
--]]

--- config vars
local min_idle_connections = 4
local max_idle_connections = 8
local log_level=1
local encoding="utf8"
--- end of config

-- 事务标识，在事务内不归还连接
local transaction_flags={}
setmetatable(transaction_flags,{__index=function() return 0 end})

-- log system
log={
level={debug=1,info=2,warn=3,error=4},
funcs={"debug","info","warn","error"},
}
function log.log(level,m)
if level >= log_level then
local msg="[" .. os.date("%Y-%m-%d %X") .."] ".. log.funcs[level] .. ": " .. tostring(m)
print(msg) -- TODO write msg into a log file.
end
end
for i,v in ipairs(log.funcs) do
log[v]=function(m) log.log(log.level[v],m) end
end

-- connect to server
function connect_server()
log.info(" starting connect_server ... ")
local least_idle_conns_ndx = 0
local least_idle_conns = 0

for i = 1, #proxy.backends do
local s = proxy.backends[i]
local pool = s.pool
local cur_idle = pool.users[""].cur_idle_connections

log.debug("[".. s.address .."].connected_clients = " .. s.connected_clients)
log.debug("[".. s.address .."].idling_connections = " .. cur_idle)
log.debug("[".. s.address .."].type = " .. s.type)
log.debug("[".. s.address .."].state = " .. s.state)

if s.state ~= proxy.BACKEND_STATE_DOWN then
-- try to connect to each backend once at least
if cur_idle == 0 then
proxy.connection.backend_ndx = i
log.info("server [".. proxy.backends[i].address .."] open new connection")
return
end
-- try to open at least min_idle_connections
if least_idle_conns_ndx == 0 or
( cur_idle < min_idle_connections and
cur_idle < least_idle_conns ) then
least_idle_conns_ndx = i
least_idle_conns = cur_idle
end
end
end

if least_idle_conns_ndx > 0 then
proxy.connection.backend_ndx = least_idle_conns_ndx
end

if proxy.connection.backend_ndx > 0 then
local s = proxy.backends[proxy.connection.backend_ndx]
local pool = s.pool
local cur_idle = pool.users[""].cur_idle_connections

if cur_idle >= min_idle_connections then
-- we have 4 idling connections in the pool, that's good enough
log.debug("using pooled connection from: " .. proxy.connection.backend_ndx)
return proxy.PROXY_IGNORE_RESULT
end
end
-- open a new connection
log.info("opening new connection on: " .. proxy.backends[proxy.connection.backend_ndx].address)
end

---

-- auth.packet is the packet
function read_auth_result( auth )
if auth.packet:byte() == proxy.MYSQLD_PACKET_OK then
-- 连接正常
proxy.connection.backend_ndx = 0
elseif auth.packet:byte() == proxy.MYSQLD_PACKET_EOF then
-- we received either a
-- * MYSQLD_PACKET_ERR and the auth failed or
-- * MYSQLD_PACKET_EOF which means a OLD PASSWORD (4.0) was sent
log.error("(read_auth_result) ... not ok yet");
elseif auth.packet:byte() == proxy.MYSQLD_PACKET_ERR then
log.error("auth failed!")
end
end

---
-- read/write splitting
function read_query( packet )
log.debug("[read_query]")
log.debug("authed backend = " .. proxy.connection.backend_ndx)
log.debug("used db = " .. proxy.connection.client.default_db)

if packet:byte() == proxy.COM_QUIT then
proxy.response = {
type = proxy.MYSQLD_PACKET_OK,
}
return proxy.PROXY_SEND_RESULT
end

if proxy.connection.backend_ndx == 0 then
local is_read=(string.upper(packet:sub(2))):match("^SELECT")
local target_type=proxy.BACKEND_TYPE_RW
if is_read then target_type=proxy.BACKEND_TYPE_RO end
for i = 1, #proxy.backends do
local s = proxy.backends[i]
local pool = s.pool
local cur_idle = pool.users[proxy.connection.client.username].cur_idle_connections

if cur_idle > 0 and
s.state ~= proxy.BACKEND_STATE_DOWN and
s.type == target_type then
proxy.connection.backend_ndx = i
break
end
end
end
-- sync the client-side default_db with the server-side default_db
if proxy.connection.server and proxy.connection.client.default_db ~= proxy.connection.server.default_db then
local server_db=proxy.connection.server.default_db
local client_db=proxy.connection.client.default_db
local default_db= (#client_db > 0) and client_db or server_db
if #default_db > 0 then
proxy.queries:append(2, string.char(proxy.COM_INIT_DB) .. default_db)
proxy.queries:append(2, string.char(proxy.COM_QUERY) .. "set names '" .. encoding .."'")
log.info("change database to " .. default_db);
end
end
if proxy.connection.backend_ndx > 0 then
log.debug("Query[" .. packet:sub(2) .. "] Target is [" .. proxy.backends[proxy.connection.backend_ndx].address .."]")
end
proxy.queries:append(1, packet)
return proxy.PROXY_SEND_QUERY
end

---
-- as long as we are in a transaction keep the connection
-- otherwise release it so another client can use it
function read_query_result( inj )
local res = assert(inj.resultset)
local flags = res.flags

if inj.id ~= 1 then
-- ignore the result of the USE <default_db>
return proxy.PROXY_IGNORE_RESULT
end
is_in_transaction = flags.in_trans

if flags.in_trans then
transaction_flags[proxy.connection.server.thread_id] = transaction_flags[proxy.connection.server.thread_id] + 1
elseif inj.query:sub(2):lower():match("^%s*commit%s*$") or inj.query:sub(2):lower():match("^%s*rollback%s*$") then
transaction_flags[proxy.connection.server.thread_id] = transaction_flags[proxy.connection.server.thread_id] - 1
if transaction_flags[proxy.connection.server.thread_id] < 0 then transaction_flags[proxy.connection.server.thread_id] = 0 end
end

log.debug("transaction res : " .. tostring(transaction_flags[proxy.connection.server.thread_id]));
if transaction_flags[proxy.connection.server.thread_id]==0 or transaction_flags[proxy.connection.server.thread_id] == nil then
-- isnot in a transaction, need to release the backend
proxy.connection.backend_ndx = 0
end
end

---
-- close the connections if we have enough connections in the pool
--
-- @return nil - close connection
-- IGNORE_RESULT - store connection in the pool
function disconnect_client()
log.debug("[disconnect_client]")
if proxy.connection.backend_ndx == 0 then
for i = 1, #proxy.backends do
local s = proxy.backends[i]
local pool = s.pool
local cur_idle = pool.users[proxy.connection.client.username].cur_idle_connections

if s.state ~= proxy.BACKEND_STATE_DOWN and
cur_idle > max_idle_connections then
-- try to disconnect a backend
proxy.connection.backend_ndx = i
log.info("[".. proxy.backends[i].address .."] closing connection, idling: " .. cur_idle)
return
end
end
return proxy.PROXY_IGNORE_RESULT
end
end