[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] fence ' node1' failed if etho down



For a time I was restricted in the hardware I could use for a cluster, and the
hardware used in the two node cluster had the following fence devices:

node1 = power fencing through IPMI
node2 = manual fencing

Whenever node1 was fenced by node2, node1 would power down as expected.

Whenever node2 was fenced by node1, node1 would "manually fence" node2.

What basically happens from there is CMAN on node1 detects that the node2 is
still participating in the cluster, and because it can't remove node2 from the
cluster, node1 removes itself instead.  This is to stop damage to shared filesystems.

But the impact on your user base is that you've just lost both nodes on the
cluster until an admin manually intervenes - which serves little purpose when
your trying to achieve a clustered high availability.

Stewart



On Thu Feb 19 16:39 , ESGLinux  sent:

>Hello, 
>
>I can promisse you that the cluster runs with fencing at all, ;-) 
>
>But it runs in a way absolutelly impredectible. 
>
>I´m using xen because I´m just testing different scenarios before put the
cluster in a real production system, which is the final goal,
>
>
>I am going to read the faqs you send me (the other link I´ve already read and it
does not resolve my doubts, but I think because I didnt understand the
problematic, now I´ll see... )
>
>All of this about fencing makes me wonder. If all fails, and a node is complety
lost an its imposible to complete the fence process of it,
>
>what happens? In my actual situation without any fence, the cluster doesnt works
at all. 
>
>Greetings and thanks for the information, 
>
>ESG
>
>
>
>
>
>
>
>--
>
>Linux-cluster mailing list
>
>Linux-cluster redhat com
>
>https://www.redhat.com/mailman/listinfo/linux-cluster
>
>
>




[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]