[sles-beta] GEO Clusters, where to get basic information
Richard Brown
rbrown at suse.de
Sun May 25 02:01:05 MDT 2014
On 2014-05-25 09:15, Darren Thompson wrote:
> Team
>
> I have been using SLES HA for some time but have no exposure to SLES
> GEO Clusters.
>
> Where is a good source of "basic/orientation' information for someone
> familiar with HA but no exposure to GEO?
The manual is pretty good :)
https://www.suse.com/documentation/sle_ha/singlehtml/book_sleha/book_sleha.html#cha.ha.geo
> What is it?
GEO is an Extension to the SLE HA Extension, which enables you to build
'A Cluster of Clusters' spread across large Geographic distances
(Separate Cities/States/Countries/Continents) with services being
started/stopped on the separate cluster based on the network
availability (or lack thereof) of the other locations.
> What are it's use cases and where should it NOT bet used?
Its typical use case is 'Disaster Recovery' or 'Business Continuity'. A
cluster at Site A also has its HA resources and data replicated/mirrored
at Site B which is many miles away (eg. over 30km) with high latency
connection (eg over 15ms, too high for 'traditional' HA Local or Metro
Area clusters). In the event of Site A 'failing', all of the resources
that usually 'live' at Site A can be started at Site B, providing your
business with a way to continue operating despite the loss of Site A.
It probably shouldn't be used when you're clusters are on the same local
subnet and/or when latency is sufficiently low to make Metro clustering
viable - why complicate matters if a simple option is available?
> What is the basic principle on which it operates?
GEO introduces the concept of 'Tickets' - Tickets are tokens which can
be granted/revoked from a Cluster, and used as a dependency for the
operation of Cluster Resources. eg. a ticket called TicketA might be
required by all the Resources that are typically at Site A. When the
Ticket is revoked from Site A and granted to the Cluster at Site B, Site
B is then the only cluster able to operate the services that rely on
TicketA's presence. The distribution of Tickets is handled by a piece of
software called the 'booth'
Booth is an agent which runs on each cluster, and is configured to be
aware of the other sites involved in your Geo cluster. Using UDP, it's
designed to work in high latency environments. Booth usually has long
expiration times on Tickets, given significant protection against brief
connectivity issues. This comes at the potential expense of taking
longer before automatically failing over, but this isn't the sort of
thing you'd want the software to make a mistake about - more rapid
allocation of tickets can of course be done by sysadmins.
Arbitrators are special machines running Booth but not running HA
Clusters. This allows you to avoid split brain scenarios by having a 3rd
independent server arbitrate about the status of the other Sites.
Arbitrators are not necessary if you have an odd number of Clusters in
your Geo Cluster.
Taking my simple 2 site example from earlier, if Sites A and B can't
communicate with each other, there is no way Booth on either cluster can
automatically figure out which site is operational. An arbitrator
running elsewhere however should be able to provide enough information
to determine whether Site A or B is 'down' or that both sites are
actually 'up' and it's just a failure of the network link between Site A
and B.
> Are there configuration example/templates?
As the details depend totally on the structure of your network and the
local resources you want to enable for Geo clustering, templates are a
tricky thing to provide.
The manual has some very simple examples which should be enough to get
you started.
Best Regards,
- Richard
--
-------------------------------------------------------------------
Richard Brown, QA Engineer
Phone +4991174053-361, Fax +4991174053-483
SUSE LINUX Products GmbH, Maxfeldstr. 5, D-90409 Nuernberg
Geschaeftsfuehrer: Jeff Hawn, Jennifer Guild, Felix Imendoerffer,
HRB 16746 (AG Nuernberg)
-------------------------------------------------------------------
More information about the sles-beta
mailing list