Re: [BUGS] postgresql 8.0b1 Win32 observations

> > >> What I would like
> > >> to see is eiter 1. An installer that lets u install on fat.
> > >> OR 2. a zipped binary distro of the windows version so i
> > can happily
> > >> run on fat, however that seems to have stopped which means
> > I have to
> > >> compile it myself which is a hassle :/
> > >
> > >NTFS checking is now coded into the installer which will
> > only let you
> > >install the data directory on an NTFS partition. The rest
> of the app
> > >can be installed anywhere. I will commit a minor change in
> a minute
> > >that will allow you to specify a non-NTFS partition, but
> only if you
> > >don't automatically initdb.
> >
> > I think we should emit a warning if it's installed with *any*
> > component on FAT. Permit doing it, but tell people we recommend
> > strongly against it.
>
> Why? I know FAT is evil, but the aim is to ensure that
> peoples data is safe and that just requires the data
> directory be on a parition with reasonable resiliency. If
> they start losing static stuff like entire .exe's then the
> chances are there are bigger problems than any filesystem can prevent!

For one, there is no security on the files. We take explicit steps to
ensure the security of the files against the pg service account on NTFS
(deny-write permissions on everything except the PGDATA directory). This
step cannot be done on FAT, obviously.

Oh, and losing entire .EXE's *does* happen on FAT. I've seen directories
that themselves were touched only by read-only operations lose files on
FAT (not during normal operation, but in case of a crash something can
apparantly hapepn). But the security part is the main reason for me.

//Magnus