Subject Re: Shadow files on a network share
Author wobbleian

Thanks for the quick reply, as always newsgroups beat "regular"
support organisations!

Brian L. Juergensmeyer wrote:

> From the IB 6.0 Beta Operations Guide (available at
> ):

> Shadowing has the following limitations:
> - Shadowing is not an implementation of replication....


> - Shadowing is useful only for recovery from hardware failures...

We are only looking for hardware recovery

> - ...Shadowing is an "all or nothing" recovery method.

We want it "all".

> - Shadowing can occur only to a local disk.

This is not our experience...!

> I hope you aren't planning on using a shadow for more than a backup
> to your nightly GBAK backups.

We intend to do all the usual things to maintain a good database no
matter what, including regular gbaks and frequent replication and not
just rely on shadowing.

> In any case, attempting to maintain a valid shadow over a network
> connection to a remote box is chancy at best. I don't know this
> certain, but I'd expect that it would be relatively easy to corrupt
> the shadow in case of network difficulties without receiving any
> errors back to the server. Then, something bad happens to the
> primary database, you try to flip over to the shadow, and discover
> that it is virtual gibberish.

This is not our experience from testing - with a "manual" shadow you
prevent additional connections if a shadow goes away. Obviously this
represents an additional point of failure, but out problem is not so
much 100% availability as 100% data recovery at any moment in time. So
far in testing all we seem to be able to lose are any transactions
active at the moment of failure.

Our problems would be solved if we could immediately activate a shadow
on a backup server in times of need. We have already looked at
replication but we need something more "real time", and have already
considered mirrored disks but will not necessarily have timely
physical access to the hardware.

Should we just concentrate on more elegant/code solutions to this
problem and make it go away, or is shadowing at least a potential stop
gap until we have coded the answer?