[OpenAFS] Re: mixing AFS versions for db servers?

Andrew Deason adeason@sinenomine.net
Sat, 22 Oct 2011 13:30:56 -0500

On Sat, 22 Oct 2011 01:06:53 +0000
Coy Hile <coy.hile@coyhile.com> wrote:

> Out of curiousity (as I'm sure I'll come across it soonn enough after
> I rebuild my cell), if all dbservers must run the same version, what
> is the supported process for updating dbservers?  Certainly if you did
> "yum upgrade list-of-openafs-packages" on all three and rebooted them
> simultaneously, you'd have an outage -- however short it may be.

It depends on what you mean by 'outage'. If the dbserver processes do
not respond to requests for, say, 1 second while you switch over, that
doesn't mean clients will notice (even if they try to do something
within that 'outage' window). If clients did notice, then they'd notice
the same thing whenever the network dropped a packet.

There are many other operations for which data may be "unavailable" for
a short period of time, but I wouldn't really consider those an outage,
since from a client's perspective the only difference is that access
seems slow. For example, volumes are inaccessible for a (usually very
small) period of time during moves, releases, backups, etc. But if a
client tries to access the volume during that time, it will just wait
and retry the request. So from the level of applications accessing /afs,
there is no period of time where the data is inaccessible.

Andrew Deason