[Date Prev][Date Next] [Chronological] [Thread] [Top]

RE: syncrepl questions



> -----Original Message-----
> From: Hallvard B Furuseth [mailto:h.b.furuseth@usit.uio.no]

> Howard Chu writes:
> > I had considered this design once before; it would be pretty easy to
> > implement using back-ldap in an overlay in serial form. Concurrently
> > updating multiple targets would be a bit more effort. This simple
> > approach using back-ldap would have no resilience; if any target
> > servers are unavailable then you lose...
>
> No good.  While load balancing will be one reason for
> multiple servers,
> another is to have fail-over when one of the servers goes down.

Understood. I didn't pursue the design very far...

> > Though the basic "update everyone" concept is simple enough, we'd
> > still need LDAP transactions to make it clean, which would allow for
> > consistent multimastering.
>
> Well, we'll probably write it so that one daemon on one host
> will do all
> the updates.  So that daemon will handle consistency issues.

That's what Kurt's reply alluded to - without two-phase commit your clients
will get an inconsistent view. In the time it takes for your daemon to send
an update to each slave, some client queries will be directed to slaves with
fresh data, others will go to slaves with stale data.

  -- Howard Chu
  Chief Architect, Symas Corp.       Director, Highland Sun
  http://www.symas.com               http://highlandsun.com/hyc
  Symas: Premier OpenSource Development and Support