On Thu, Feb 19, 2009 at 10:09 PM, Michael Schuster
<Michael.Schuster at sun.com> wrote:
> All,
>
> you may have seen (part of) the discussion concerning IPv6 addresses and
> kstats. This lead to a more fundamental question, and this is an effort to
> recap various elements of this discussion: (we also solve the kstats
> question btw)
>
> What is the "unit of currency" as far as individual servers are concerned?
>
> (ie, how do admins of ILB talk about back-end servers, and how does the
> system display information about them?)

Well, I use the following convention. For a site www.foo.com, the virtual server
tends to be called foo. The back-end servers are called something like foo-node1
and foo-node2, etc. (There are other models. For example, all the services for
functionality 'bar' may live on 'bar-server'.)

> Using IP addresses may be feasible for IPv4, but with IPv6 this gets
> unwieldy; therefore an abstraction is desired by which a (set of) server(s)
> can be identified and manipulated. Postulating the use of regular hostnames
> turns out not to be a valid option, as it requires name services to be
> maintained in the data center, something we're told cannot be relied on.

I use the unqualified hostnames (or something similar) as the names. It's then
absolutely obvious as you look at the configuration how to match
servers to hosts.
And it makes automated verification of configurations (and how the load-balancer
configuration is related to server configuration) much easier. Note
that I'm using a
convention that the names match - the IP address is a parameter and you never
rely on a nameservice to translate.

I do all configuration with names. IP addresses are transitory implementation
artefacts.

> Consensus was reached that we want to provide the capability of assigning
> names to servers; to avoid confusion with traditional hostnames, we call
> this a "server ID".
> This can be done in several ways; we selected to extend the "ilbadm
> create-servergroup" subcommand thus:
>
>        ilbadm create-servergroup ... [-n serverID=foo ] ..
>
> if the servernID option is not used, the ID defaults to the servergroup
> name. this name is then applied to every server specified, thus: In addition
> to the id specified, an index is appended to the name of every
> individual server. eg:
>
>        ilbadm create-servergroup -s server=192.168.1.1,192.168.1.2 group1
>
> will cause the servers to be named "group1.1" and "group1.2". (this is also
> done if only one server is used)
>
> In order to be able to associate different names with indivdual servers,
> "ilbadm add-server" will be extended to also a '-n serverID=' option.
>
> In all cases where a single server is specified, it will still be able to
> use a "traditional" hostname instead of the IP address; an option will be
> provided to use this hostname as the server ID (this name will remain, even
> if the naming DB changes after this server ID is created)
>
> "ilbadm remove-server", "enable-server" and "disable-server" will be
> modified to accept names instead of IP addresses; an option will be
> provided to indicate that either hostnames or IP addresses are used as
> arguments.
>
> this id will also be used as the name in per-server kstats (this is the
> "solving the IPv6 issue" part), both kstat(1M) and "ilbadm show-statistics",
> once implemented, will display this id for servers.
>
> Michael
> --
> Michael Schuster        http://blogs.sun.com/recursion
> Recursion, n.: see 'Recursion'
>
> _______________________________________________
> ilb-dev mailing list
> ilb-dev at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/ilb-dev
>



-- 
-Peter Tribble
http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/

Reply via email to