如何讓cluster reforming 變快?

[watermelonyu]

我有一組cluster and 一組package(folating ip用的)

我故意將一台shutdown

結果cluster reforming花了將近兩分鐘

這樣我的package (floating ip)也必須在cluster reforming 完成後才會跟著ready

有辦法將cluster reforming 時間變短嗎??

因為package 沒有ready 的話,我的oracle RAC根本也沒辦法提供服務

[watermelonyu]

建議你開cluster debug mode, 仔細觀察log是否能看出些端倪.

[watermelonyu]

教一下怎樣開操作debug???

[watermelonyu]

cmsetlog: /usr/contrib/bin/cmsetlog



enable and redirect log to /tmp/sg.log

# cmsetlog 5

# cmsetlog -f /tmp/sg.log



disable and redirect log to syslog.log

# cmsetlog -r

# cmsetlog -s



*Note: Log will grow up very fast.

[watermelonyu]

在正常狀況下你們遇到的狀況都是幾秒可以reforming finish

[watermelonyu]

我是關掉rac2

從log看從 RUNNING to RECONFIG 到RECONFIG to RUNNING已經花了兩分鐘了



May 19 05:49:02:2:CLM:1d: Action - Changed rac2's cm_state from RUNNING to RECONFIG

May 19 05:51:20:2:CLM:1c: Action - Changed rac1's cm_state from RECONFIG to RUNNING

May 19 05:51:20:2:CLM:1c: Action - Changed rac2's cm_state from RECONFIG to FAILED

[watermelonyu]

附上Log

[watermelonyu]

err ... 基本上不會真的拿馬錶去計算cluster reform需要幾秒. cluster最終的目的就是提供high availability. 慢有很多的可能性, 可能它就是要這麼久, 可能你的機器capacity不夠, 可能你的DB還有transaction .... 等等, 諸多的可能性.



btw, what's the value of your NODE_TIMEOUT and cluster heartbeat setting ?

[watermelonyu]

NODE_TIMEOUT 10000000

[watermelonyu]

HEARTBEAT_INTERVAL ?