Date: Fri, 3 Oct 1997 13:28:27 +0900 (JST) From: Michael Hancock <michaelh@cet.co.jp> To: Terry Lambert <tlambert@primenet.com> Cc: ccsanady@bob.scl.ameslab.gov, brandon@roguetrader.com, wilko@yedi.iaf.nl, freebsd-fs@FreeBSD.ORG Subject: Re: Known problems with async ufs? Message-ID: <Pine.SV4.3.95.971003132313.4232A-100000@parkplace.cet.co.jp> In-Reply-To: <199710020815.BAA24245@usr08.primenet.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 2 Oct 1997, Terry Lambert wrote: > > > It will also mean that there are no hooks to guarantee transactions > > > are idempotent (multiple atomic transactions considered as an > > > all-or-nothing unit) for something like a user accessible transaction > > > tracking system. > > > > I think you're looking for another term. Idempotent transactions are like > > reads where you can repeat the transaction without ill-effects. > > Non-idempotent transactions modify the state of the system such that they > > can't be repeated, say rmdir foo. > > > > Multiple atomic transactions that are atomic can be restated as an atomic > > transaction. > > Reads advance the file pointer, unless they are mread. Same for write vs. > mwrite. Maybe getattr would have been a better example. Though I've seen references where they separate VOP operations into idempotent and non-idempotent operations. It's probably more important in the context of NFS servers where things are stateless. A read here is packaged with file positioning so repeated receives of the same request doesn't hurt anything. Regards, Mike
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SV4.3.95.971003132313.4232A-100000>