[prev] [thread] [next] [lurker] [Date index for 2002/10/11]
On Fri, Oct 11, 2002 at 03:46:49PM +0100, Richard Clamp wrote: > On Friday, Oct 11, 2002, at 13:24 Europe/London, Simon Wistow wrote: > > > On Thu, Oct 10, 2002 at 05:45:08PM +0100, Nicholas Clark said: > >> However, is there any sort of sanity locking - presumably a race > >> condition > >> can exist where (say) two e-mails both requesting configuration > >> changes > >> can arrive, execute simultaneously: > > Last change wins. I agree > I think most of the by email changes be gated by ack > messages anyhow, so that I can't surreptitiously upset you by spoofing > an envelope and asking for extra header munging. But I meant that if the stuff calling the Storage system knows that it is "doing stuff" where it reads data, modifies it and then re-writes data, then it needs to implement a form transactions to be robust. (And certainly whatever implements the Storage system needs to be told to do locking, else there's the non-zero probability of corrupting DBM files or the like on disk. Rare, yes. Impossible no. Therefore it will happen to someone at some point) That's all I meant to say. (And I don't give a fuck for long term "well, they shouldn't do two things at once" arguments. Short term, yes, valid. Long term, it will allow lusers to shag systems. And admins will get pissed off and use something else.) Nicholas ClarkThere's stuff above here
Generated at 13:56 on 01 Jul 2004 by mariachi 0.52