PDA

View Full Version : Cluster resource stop on 3 node cluster



lywwing
15-Nov-2016, 08:53
hi Guys,

I have 3 node cluster server with resource running as below.

https://s17.postimg.org/7g9n7f9fj/Screenshot_11_15_2016_1_52_41_PM.jpg

Then I shutdown node 1 server, it become below.

https://s4.postimg.org/hz23dpbtp/Screenshot_11_15_2016_1_53_24_PM.jpg

Then I shutdown node 2 server, it become below.

https://s18.postimg.org/cik9lyzu1/Screenshot_11_15_2016_1_54_10_PM.jpg

With only one server online, but there have nothing resource can running, any idea?

There have no "cli-ban" constraints and any constraints rule here.

Regards
Thanks...KEN

ab
15-Nov-2016, 12:28
On 11/15/2016 12:54 AM, lywwing wrote:
>
> Then I shutdown node 2 server, it become below.
>
> [image:
> https://s18.postimg.org/cik9lyzu1/Screenshot_11_15_2016_1_54_10_PM.jpg]
>
> With only one server online, but there have nothing resource can
> running, any idea?

Is it safe to assume that node2 was running the resources when it was shut
down? It probably does not matter, because at the end of the day you lost
quorum and thus the only node left could not establish if it was in a
split-brain scenario, and thus it did not know if it could run resources,
so it did not run them.

Think about this from another perspective, that of the other nodes that
could have been running, but just could not reach this third node. In
that scenario, the other two boxes would have had a quorum (two out of
three) and would have tried, legitimately, to run resources. This node,
being on its own, would have appeared exactly as you presented it, and
should NOT have run any resources, because some resources simply cannot
run in multiple places at time (active/passive vs. active/active
resources), so it should have stopped, which it did.

If you want more opportunity for failure, add more nodes (preferably
keeping the numbers odd, so three, five, seven, etc.) so that quorum can
be potentially-maintained longer.

I believe you can also configure the cluster how it should act in the case
of a loss of quorum, and this is often done for invalid-by-definition
two-node clusters (thank-you for having a three-node cluster) and while
you could try that, you should should not because you already have a valid
cluster.

> There have no "cli-ban" constraints and any constraints rule here.

No, there need not be any.

--
Good luck.

If you find this post helpful and are logged into the web interface,
show your appreciation and click on the star below...