(Illustration by Gaich Muramatsu)
I think that databases must be kept as a single server. Files are stored on Coda, but only one DB server is running. If the machine fails, we first have to check the files on another machine, then start a new DB server, and redirect all queries to this new server. This isn't High Availability, but it's still good. Maybe there can be a problem when cheking the files on another machine, but Coda should take care of all that clearly. Nicolas Huillard -----Message d'origine----- De: Chris Snyder [SMTP:csnyder_at_mvpsoft.com] Date: mardi 9 mai 2000 22:04 Objet: Re: High Availability That seems like it would work, but what happens if there is a lot of data such as MySQL databases that is tough to combine, if two servers have different updates after a reconnect? ----- Original Message ----- From: "Jan Harkes" <jaharkes_at_cs.cmu.edu> To: <codalist_at_TELEMANN.coda.cs.cmu.edu> Sent: Tuesday, May 09, 2000 3:35 PM Subject: Re: High Availability > On Tue, May 09, 2000 at 06:46:44PM +0200, Nicolas Huillard wrote: > > I'm finishing reading Dr Le Blanc's presentation of Coda > > (http://www.mcc.ac.uk/~zlsiial/coda/text47.html) and also wonders > > if... (quote :) "Finally, I wonder whether Coda's success as > > high-availability software is impaired by the dependence of a cell on > > a single System Control Machine." I am really concerned about this > > issue, because my goal is to build High Availability... > > Does anyone have an idea ? > > The coda clients and servers do not rely on the existence of an SCM. The > only reason that machine is special is because of the way we propagate > updates to the user and volume databases. They are read/write on the > SCM, and the other servers check for updated files on the SCM once every > 30 seconds. > > If you lose the SCM, all servers can still rely on their local copy. But > in order to add a volume or user, you need to restore the old SCM. Or > designate one of the other servers (all of them should have all of the > required data) as the replacement SCM. To do this is is necessary to > update the /vice/db/scm file on all servers, and restart the > updateclnt/updatesrv processes. Not as good as a quorum based replicated > database server, but a lot simpler for the common case. > > Jan >Received on 2000-05-10 04:53:02