Setting up CTDB on OCFS2 and VMs ...

[ronnie sahlberg]

Steve, please. You are confused on what CTDB is.


What you are describing and you use is basically a primitive
Active/Passive failover pair where you have to manually trigger
failover by disabling/enabling one of the two CTDB nodes.
For such Active/Passive failover pairs there are heaps of really good
solutions. Use one of those solutions, not CTDB, if all you want is a
two-node active/passive failover pair.
CTDB is not a good solution for your use case.


What Richard and Rowland are aiming for is a proper multi-node cluster
with CTDB and OCFS2 for the use case that CTDB is designed for, i.e.
an All Active cluster where ALL nodes are active simultaneously and
serving data.
The whole point of ctdb is to NOT have active/passive and do failover
but instead a cluster where ALL nodes are active simultaneously.
In such a cluster there is no longer any failovers in the traditional
sense but merely a shuffle of ip addresses when nodes enter/leave the
active cluster.

For proper CTDB use in a production environment DRDB is not a viable solution.


>Not sure of your use case but it's not let us down all term and we've really thrown everything we could at it. It's serving 80 real boxes from real hardware too.

That is not exactly enterprise. People using CTDB use clusters usually
ranging from 4 to 16 nodes, serving tens of thousands of simultaneous
clients.
I would say that while you CAN build a 2 node CTDB cluster, that is
probably not a great idea for a production environment. I think that 4
nodes is probably the smallest cluster you should use.
Feel free to disagree.


>We've had only 1 sb (after a power cut) and drbd messaged me. It's then just a case of choosing your node.

That is not how CTDB is supposed to be used. If a node fails, CTDB is
supposed to automatically redistribute all traffic across the
surviging nodes. No human intervention required.
Seems you have a single point of failure since you need to manually intervene.




Please do not misrepresent what ctdb is or how it should be used. It
is a disservice to the mailing list readers that come here for
information and may form a broken idea of how/what ctdb is.


regards
ronnie sahlberg

On Sat, Dec 6, 2014 at 10:05 AM, steve <steve at steve-ss.com> wrote:
> On 06/12/14 18:40, Rowland Penny wrote:
>
>>>
>>>
>>
>> Yes, I know, but I am trying to do it without DRBD.
>>
>> Rowland
>>
>
> Not sure of your use case but it's not let us down all term and we've really
> thrown everything we could at it. It's serving 80 real boxes from real
> hardware too.
>
> We've had only 1 sb (after a power cut) and drbd messaged me. It's then just
> a case of choosing your node.
>