comp.lang.ruby

Eric Hodel

1/22/2005 7:52:00 AM

On 20 Jan 2005, at 22:10, Dido Sevilla wrote:

> I'm in the process of designing a fault-tolerant distributed
> application using Ruby, and am looking at whether Rinda will be
> suitable for this purpose. I am wondering what strategies are
> available for recovering from the failure of the ring server, which
> seems to me like a critical single point of failure. It is possible to
> run the ring server/tuplespace daemon as part of a Linux-HA heartbeat
> cluster to guard against physical failure of the primary ring server,
> but this requires restarting the ring server on the secondary node.
> The new ring server instance running on the backup server is now
> ignorant of all services that were previously registered on the old
> ring server before the failure.

Yup.

> This is unacceptable for the
> distributed application. Is there a way for live services to
> automagically detect failure of the ring server, and automatically
> reregister themselves with it when it goes back up in that case, or
> some way for a primary and backup ring server to communicate with each
> other and share information about registered services transparently?

1) Run more than one RingServer, and have each cross-register the
other's services. (This service doesn't have to run on the RingServer
itself, actually...)

2) The RingServer removes services automatically when a service's
renewer fails to respond. Renewers are invoked after some timeout. On
the service side, if the service's renewer is not invoked within a
timeout, you could have the service re-register itself, something like
IRC's PING/PONG handshake.

--
Eric Hodel - drbrain@segment7.net - http://se...
FEC2 57F1 D465 EB15 5D6E 7C11 332A 551C 796C 9F04