最近看到關于redis故障轉移的文章，然後發現自己項目的redis并沒有應用故障轉移的機制，項目目前的機制是如果redis所在的伺服器當機或者redis服務異常/關閉等時候遊戲伺服器就擷取不到redis相關資訊，之前也有過幾次這種情況，但是當時的處理方式都是重新開機redis服務。
我按照這邊文章https://blog.csdn.net/guweiyu_thinker/article/details/79188281?tdsourcetag=s_pctim_aiomsg 來自己部署了下故障轉移，這個文章很詳細，具體的原理就不贅述了,他是在同一台伺服器上開啟不同的端口來嘗試的，我是在3台不同的伺服器上部署的，每個伺服器部署1個redis服務,1個sentinel服務,其中要提到一點 sentinel的myid不需要自己去設定 服務啟動的時候會自動生成一個myid 寫入到sentinel*.conf配置中，還有一些配置如
sentinel config-epoch mymaster 10
sentinel leader-epoch mymaster 10
sentinel known-slave mymaster 10.1.67.28 6379
sentinel known-slave mymaster 10.1.67.44 6399
sentinel known-sentinel mymaster 10.1.67.27 26379 0e86212b10f3e231376f84aa0315c7e1f9bcdb4f
sentinel known-sentinel mymaster 10.1.67.44 26379 c7d8950964042eb5ff3871f36929934ad46a8779
都是故障轉移後自動寫入到sentinel*.conf中裡面的，自己不需要配置
最後故障轉移成功，然後redis的master的host發生了變化,變成了某個slave的host，現在的問題是遊戲伺服器怎麼能擷取到最新的master的host,如果通過redis用戶端redis-cli可以直接SENTINEL get-master-addr-by-name mymaster來擷取,但是遊戲服中需要連接配接到某個sentinel然後擷取 每個項目架構不同連接配接的方法應該不一樣，我們用的是predis,predis中有ServerSentinel.php,然後自己寫了一個檔案sentinel.lua用來來連接配接到sentinel的cli然後進行查詢最新的master的host
sentinel.lua
-------------------------------------------
_G.Sentinel = class("ServerSentinel")
function Sentinel:__construct(host, onConnect)
local parent = self
local onconnect = function(net, ip, port, myip, myport)
parent:onConnect(net)
if onConnect then
onConnect(net, ip, port, myip, myport)
end
end
local onclose = function()
self.__net = nil
self:onClose(net, timeout, notconn, err)
printi("Sentinel:onClose", timeout, notconn, err)
end
self.__net = _connect(host, onconnect, onclose, 30)--30秒逾時
end
function Sentinel:onConnect(net)
printi("Sentinel:onConnect", net:closed())
if net:closed() then
return
end
self.net = net
delayf("Sentinel.keeplive", self.keeplive, 60, self)
end
function Sentinel:onClose(net, timeout, notconn, err)
self.net = nil
end
Sentinel.lastcommand = {}
function Sentinel:command(cmd, ...)
Sentinel.lastcommand.cmd = cmd
Sentinel.lastcommand.args = {...}
print("-------------->Sentinel:command", cmd)
if not self.net or self.net:closed() then
print("INFO", "Sentinel:command net is nil", cmd, self.net, self.net and self.net:closed())
return
end
local config = Sentinel.commands[cmd]
if not config then return end
local r, parse = Sentinel.multibulk
if type(config) == "table" then
if config.request then
r = config.request
end
if config.response then
parse = config.response
end
end
self.net:send(r(cmd, ...))
local reply = Sentinel.response(self.net)
if type(reply) == 'table' and reply.queued then
reply.parser = parse
return reply
else
if parse then
return parse(reply, cmd, ...)
end
return reply
end
end
function Sentinel.hash_multi_request_builder(builder_callback)
return function(command, ...)
local args, arguments = {...}, {}
if #args == 2 then
table.insert(arguments, args[1])
for k, v in pairs(args[2]) do
builder_callback(arguments, k, v)
end
else
arguments = args
end
return Sentinel.multibulk(command, arguments)
end
end
function Sentinel.multibulk(cmd, ...)
local args = {...}
local argsn = #args
local buffer = {true, true}
if argsn == 1 and type(args[1]) == 'table' then
argsn, args = #args[1], args[1]
end
buffer[1] = '*' .. tostring(argsn + 1) .. "\r\n"
buffer[2] = '$' .. #cmd .. "\r\n" .. cmd .. "\r\n"
local table_insert = table.insert
for i = 1, argsn do
local s_argument = tostring(args[i] or '')
table_insert(buffer, '$' .. #s_argument .. "\r\n" .. s_argument .. "\r\n")
end
return table.concat(buffer)
end
Sentinel.commands = {
-- Sentinel = {-- >= 2.0
-- request = Sentinel.hash_multi_request_builder(function(args, k, v)
-- print('=====kkk', k, v)
-- table.insert(args, k)
-- table.insert(args, v)
-- end),
-- },
Sentinel = true, -- >= 2.0
}
function Sentinel.finalerr(net, msg)
net:close()
print("ERROR", "Sentinel lastcommand", Sentinel.lastcommand.cmd or 'nocmd', serialize(Sentinel.lastcommand.args))
error(msg)
end
function Sentinel.getdata(net)
local res, payload = pcall(net.receiving, net, "\r\n", 32, 10)--10秒逾時
if not res then
Sentinel.finalerr(net, "Sentinel err, receiving something wrong.")
end
if not payload or not payload.tostr then
Sentinel.finalerr(net, "redis err, receiving payload wrong, "..type(payload))
end
local ret = payload:tostr()
return ret:sub(1, -#ret), ret:sub(2)
end
function Sentinel.response(net)
local prefix, data = Sentinel.getdata(net)
print("------------------>Sentinel.response", prefix, data)
if not prefix then return end
return Sentinel.response_data(net, prefix, data)
end
function Sentinel.response_data(net, prefix, data)
-- status reply
if prefix == '+' then
if data == 'OK' then
return true
elseif data == 'QUEUED' then
return {queued = true}
else
return data
end
-- error reply
elseif prefix == '-' then
return Sentinel.finalerr(net, "Sentinel err, -" .. data)
-- integer reply
elseif prefix == ':' then
local number = tonumber(data)
if not number then
if data == 'nil' then
return nil
end
return Sentinel.finalerr(net, "Sentinel err, isnot number")
end
return number
-- bulk reply
elseif prefix == '$' then
local length = tonumber(data)
if not length then
return Sentinel.finalerr(net, "redis err, cannot parse " .. length .. " as data length")
end
if length == -1 then
return nil
end
local res, nextchunk = pcall(net.receiving, net, length + 2, 2)
if not res then
return Sentinel.finalerr(net, "redis err, something wrong.")
end
return nextchunk:tostr():sub(1, -3)
-- multibulk reply
elseif prefix == '*' then
local count = tonumber(data)
if count == -1 then
return nil
end
local list = {}
if count > 0 then
for i = 1, count do
list[i] = Redis.response(net)
end
end
return list
-- unknown type of reply
else
error('Sentinel error: ' .. 'unknown response prefix: ' .. prefix)
end
end
function Sentinel.getMasterHost()
local info = sentinel:command('Sentinel', 'get-master-addr-by-name', 'mymaster')
if info then
return string.format('%s:%s', info[1], info[2])
else
error('Sentinel error: ' .. 'unknown errorr no master')
end
end
--應該周遊 選擇一個可用的
_G.sentinel = new(Sentinel, '10.1.67.28:26379')
-----------------------------------------------------------------------------------
![資料遷移方法資料遷移原則資料遷移之雙寫方案資料遷移之級聯同步方案[圖]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)