redhat cluster 问题

nntp

原帖由 "q1208c" 发表：


手工切换时当然没问题了.

可要是真的因为什么原因, 机器 hang 了. 你还能让它把 资源放出来么?

机器hung了，OS都不工作了，什么东西在占用资源？

所以要heart-beat+quorum牙，否则 lock manager干吗用的?

nntp

原帖由 q1208c 于 2005-8-29 18:34 发表


可不一定是 os 不工作.

所以, 重启了它就最好了.  

LOL.... if rebooting is the silver bullet, why do RAC getta to market? :"  

for most mission critical systems, rebooting means your boss is going to  kick your ass.

rebooting is the most stupid way to solve HA problem.

[ 本帖最后由 nntp 于 2005-12-1 00:58 编辑 ]

nntp

原帖由 q1208c 于 2005-8-30 17:30 发表
[quote]原帖由 "cnadl"]据我所知以前clustersuite是不重起的，是不是楼主quorum没配好？[/quote 发表：


会重启的.
当 hosta 不能访问 quorum 或是 tieroute 之类的就会因为 watchdog 而自 ...

This is because redhat cluster is a silly ha cluster system that was developped base on Kimberlite.
If you really familiar with HA system, Kimberlite and it's commerical edition Convolo Data Guard is the protype of RHCS.

RHCS utilize SCSI reservation to issue lock conflict when whole nodes are in brain-split status. To avoid 50%-50% election in a "dual node HA cluster ", trigger watchdog to reboot the system is the easiest way for those RHCS developer to make system rolling back.

Actually advanced HA software never use such mechanism for a Real "High Availability Production System".

nntp

Morning dude,

I was shocked by TruCluster When i started to help manufacture customers to setup high availability environment with Tru64 Unix on Alpha System.

I said, WoW ,  the cluster is the strongest HA cluster i've seen that can support real time sync data between nodes via special technology (we called it memory channel - A kind of HBA adapters on all nodes) . Ofcoz the whole system has sucessfully implemented SSI (single system image).And now days, people  know OCFS from Oracle, but a few years ago, Alpha already has it's VERY mature clusterwide filesystem by it's TruCluster.

Let's back to Oracle RAC:

- A not so mature ocfs , standalone clulster-aware(even not cluster-wide),
- A replacement solution but only stick on Oracle data management - ASM which support clusteraware storage too.
- A cache fusion core to support sync data between nodes, but limited to Oracle

anyway, Oracle learns faster,  they made it possible to  deploy a real time SSI Database Cluster with complete high availability on linux platform.

Try to access otn.oracle.com and check out linux section, there are 9i/10g RAC step by step guides for you . Furthermore, a guru who wrote series articles to instruct you setp by step installing  Oracle RAC on Linux with IEEE1394 system also can be searched. Very interesting? yep,that's made me happy for several weeks. ( but buddy you've to purchase Maxtor OneTouch Button IEEE1394 hardisk and enclosure to support two concurrent I/O block access from two nodes, suppose they are two standalone PC).

Metalink account(metalink.oracle.com) will booth you directly to getin know everytihing arround RAC.

I' don't want to explain more on how do these modern HA clusters to solve  the node dead/rebooting/hang situation because you can obtain those information easily from their official website. Some of them even provide evaluation copy of the cluster software as well.

To me, i'm an experienced HP MC/SG cluster guy, so i can explain your further question on this HA product if you really wanna give a shoot.

docs.hp.com --> high availability -> MC/SG for linux  Admin guide + toolkit guide witll be a good start point.

www.steeleye.com provide lifekeeper for linux, another pretty nice , well designed HA cluster for people who love to do so.

[ 本帖最后由 nntp 于 2005-12-6 05:12 编辑 ]