Why only support odd number monitors in ceph cluster?

Why only support odd number monitors in ceph cluster?

[Eric_YH_Chen]

Hi, all:

     I am curious about why ceph cluster only support odd number monitors.

     If we lose 1 monitor (becomes odd number), 

      would it cause any problem if we do not handle this situation in short time? 

     Thanks!

--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Why only support odd number monitors in ceph cluster?

[Wido den Hollander]

Hi,

On 12/23/2011 08:21 AM, Eric_YH_Chen@wistron.com wrote:
> Hi, all:
>
>       I am curious about why ceph cluster only support odd number monitors.

It doesn't only support a odd number. A even number of monitors works in 
theory, but it's recommended to use a odd number to get a majority.

>
>       If we lose 1 monitor (becomes odd number),

You shouldn't look to the number of online monitors, but to the number 
of monitors in your monmap. That should be a odd number.

Wido

>
>        would it cause any problem if we do not handle this situation in short time?
>
>       Thanks!
>
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Why only support odd number monitors in ceph cluster?

[Sage Weil]

On Fri, 23 Dec 2011, Wido den Hollander wrote:
> Hi,
> 
> On 12/23/2011 08:21 AM, Eric_YH_Chen@wistron.com wrote:
> > Hi, all:
> > 
> >       I am curious about why ceph cluster only support odd number monitors.
> 
> It doesn't only support a odd number. A even number of monitors works in
> theory, but it's recommended to use a odd number to get a majority.

Right.  A bit more detail:

Running an even number of monitors means a slightly higher risk of failure 
than running one less monitor.  For example, if you run 3 monitors, you 
can tolerate 1 failure (2 is a strict majority).  And you have 3 machines 
that can fail.  

On the other hand, if you run four, you can still only tolerate 1 failure 
(3 is a majority but 2 is not).  But then you have 4 machines that can 
potentially fail, and your risk of 2 dying is slightly higher.

The difference is risk is probably pretty small for most failure rates, 
but, generally speaking, odd numbers are slightly better.

sage