多路徑環(huán)境下RHCS和GFS的timeout配置
適用環(huán)境:Cluster or GFS on RHEL4 and later
本文引用地址:http://www.ex-cimer.com/article/201610/306016.htm故障現(xiàn)象:日志報(bào)錯(cuò)
openais[3345]: [CMAN ] lost contact with quorum device
目前只要客戶(hù)有共享存儲(chǔ),在部署Cluster和GFS的時(shí)候,都建議配置quorum disk。而上面這個(gè)報(bào)錯(cuò)相信大家都不會(huì)陌生吧。這個(gè)問(wèn)題一般是因?yàn)閝disk進(jìn)程太長(zhǎng)時(shí)間沒(méi)有與cman/ais通信,超過(guò)了qdisk的poll投票時(shí)間,從而此節(jié)點(diǎn)被斷開(kāi)。特別是在配置了multipath、rdac等多路徑軟件環(huán)境中做鏈路失效切換測(cè)試時(shí),由于failover的時(shí)間可能比較長(zhǎng),造成鏈路切換之前qdisk就已經(jīng)丟失了,節(jié)點(diǎn)直接被reboot,而這當(dāng)然不是我們所期待的結(jié)果。那怎么解決這個(gè)問(wèn)題呢?
先來(lái)了解幾個(gè)基本概念:
① 集群要認(rèn)為一個(gè)節(jié)點(diǎn)健康,需要以下3要素
· CMAN認(rèn)為該節(jié)點(diǎn)online
· 該節(jié)點(diǎn)能足夠連續(xù)的讀寫(xiě)quorum disk
· 該節(jié)點(diǎn)heuristic有足夠的score
② qdisk包括兩個(gè)主要線程:主線程負(fù)責(zé)循環(huán)和進(jìn)行I/O操作;第二線程負(fù)責(zé)heuristic相關(guān)。
主線程另一個(gè)工作就是每隔一段時(shí)間告訴cman/ais自己還活著。如果qdisk超過(guò)quorum_dev_poll的時(shí)間而沒(méi)有和cman/ais通信,cman就會(huì)聲明說(shuō)此節(jié)點(diǎn)與quorum disk斷開(kāi)連接,此時(shí)日志便會(huì)有如上報(bào)錯(cuò)。默認(rèn)的cman.h里
#define DEFAULT_QUORUMDEV_POLL 10000
單位是ms,即10秒。修改quorum_dev_poll需要在cluster.conf文件里修改cman標(biāo)簽:
cman quorum_dev_poll=50000>/cman>
③我們平時(shí)指的qdisk timeout是指連續(xù)一段時(shí)間對(duì)quorum disk的讀寫(xiě)都是失敗。假如cluster.conf里
quorumd device=/dev/sdb1 interval=3 min_score=2 tko=13 votes=2>
其中
interval=3
This is the frequency of read/write cycles, in seconds.讀寫(xiě)quorum disk的頻率
tko=13
This is the number of cycles a node must miss in order to be declared dead.連續(xù)失敗多少次則判定此節(jié)點(diǎn)死掉
qdisk_timeout = interval x tko
④再來(lái)看看RHEL5里cman timeout是怎么去配置的,
token
This timeout specifies in milliseconds until a token loss is declared after not receiving a token. This is the time spent detecting a failure of a processor in the current configuration. Reforming a new configuration takes about 50 milliseconds in addition to this timeout. The default is 1000 milliseconds. 連續(xù)多長(zhǎng)時(shí)間沒(méi)有收到token就判定令牌丟失。默認(rèn)1秒,其中有50ms是生成一個(gè)新的配置的時(shí)間。
retransmits_before_loss
This value identifies how many token retransmits should be attempted before forming a new configuration. If this value is set, retransmit and hold will be automati- cally calculated from retransmits_before_loss and token. The default is 4 retransmissions. 連續(xù)丟失幾次token,才會(huì)生成新的cluster配置(將丟失token的節(jié)點(diǎn)踢出集群)。默認(rèn)4次。
token_retransmit
This timeout specifies in milliseconds after how long before receiving a token the token is retransmitted. This will be automatically calculated if token is modi- fied. It is not recommended to alter this value without guidance from the openais community. The default is 238 milliseconds. 重發(fā)token的時(shí)間間隔,這個(gè)值是由上面的token和token_retransmit自動(dòng)計(jì)算的。(1000-50)/4≈238ms
如果出現(xiàn)上面說(shuō)的丟失心跳token的時(shí)候,日志會(huì)出現(xiàn)如下報(bào)錯(cuò):
openais[3345]: [TOTEM] The token was lost in the OPERATIONAL state.
注意單位為毫秒。另外,也可以修改cman的標(biāo)簽:
注:RHEL4并未使用openais的架構(gòu),因此只能通過(guò)deadnode_timeout來(lái)修改。
好,有了前面的基礎(chǔ),不難想象到各個(gè)timeout值,用T(*)表示,應(yīng)有如下關(guān)系:
T(MPIO)
RH官方有如下建議:
T(qdisk) = T(MPIO) × 1.3
T(cman) = T(MPIO) × 2.7
參考文檔:
Red Hat Knowledgebase
、man page of
qdisk(5)
、
openais.conf(5)
評(píng)論