Re: softupdates recommendation

[Gustavo Vieira Gonçalves Coelho Rios <gustavo.rios@terra.com.br>]

BTW:
I heard it is not recommend using softupdates when you need to known the
file status, ffor a transaction need. For intance, in qmail, it states
not to use softupdates. But postfix, do suggest softupdates. the
question is: is it really rational using fsync(fd) system call on a file
system mounted with softupdates enabled ?

Jesper Louis Andersen wrote:
> 
> Quoting Marco Peereboom (marco@peereboom.us):
> > I use softupdates just about everywhere. Sure there are corner cases
> > where one can loose some data however I (usually) prefer the
> > performace boost over the potential to loose data. Keywords here are
> > UPS, stability and easily recreated/restored data.
> 
> I might be wrong here but:
> 
> 1. If you cannot afford to loose some data your program should use
> fsync(2) to flush the in-core state down to the disk. If the semantics
> in OpenBSD is not that way, I would like to know how I can do it safely.
> At least that is the original semantics of fsync(2) as far as I know.
> 
> FreeBSD seems to support the thing written above at least.
> 
> 2. By using softupdates you trade memory for speed. I do not know if
> this issue is real though as I do not know how much memory the
> dependency graph uses memory.
> 
> --
> j.