Re: postgres win32 in FAT32 - Mailing list pgsql-hackers-win32

From Magnus Hagander
Subject Re: postgres win32 in FAT32
Date
Msg-id 6BCB9D8A16AC4241919521715F4D8BCE456A27@algol.sollentuna.se
Whole thread Raw
In response to postgres win32 in FAT32  (Albi Rebmann <albi@life.de>)
List pgsql-hackers-win32
> > Specifically, FAT32 does not support Directory Junctions,
> which means
> > tablespaces don't work. That's the only filesystem specific
> function
> > that is used directly. (security and reliability are of course used
> > indirectly)
> >
> > That said, you *can* run pg on FAT32, it's just not
> recommended. The
> > installer won't let you. But if you install just the files
> (you can do
> > this from the installer, just don't enable service or
> initdb) and run
> > it manually, it should work.
> >
> > Beware that no testing is done on FAT. And don't expect
> your data to
> > be around if the machine crashes. Right now I think only
> tablespaces
> > won't work, but if other features don't work, don't expect
> anybody to
> > put down a lot of effort to make a workaround for FAT.
> >
> > //Magnus
>
> OK, we understand that part of FAT32 not being reliable
> enough - but shouldn't that choice being made by the admin,
> not the installer? The rationale is clear, but sometimes
> people just have to use FAT32 (like in the cited case on top
> of this message), and can't do anything about it. An
> installer which works, but refuses to initdb is, er, not helpful.
> This whole 'FAT32 is bad, so we refuse it' issue seems like
> going too far - the admin is responsible for making sure the
> system is reliable enough.

There are other reasons for the installer not to do it as well. The
installer sets permissions and does several verifications on the
selected directories. All these operations fail miserably on FAT.
Workarounds could certainly be made, but it's not something that'll be
done until all other parts are finished.


> It's a bit like the 'psql can't be run by Administrator
> issue', but this is already covered nicely by the installer
> creating the service account. Can't the FAT issue be dealt
> with similarly?

To do that similarly, that would mean automatically converting the
filesystem. I doubt that's a good idea.


> Just pop up a warning dialog, saying something like
>
> Placing database data files on a FAT file system is not
> recommended because of possible data corruption. If possible,
> convert this filesystem to NTFS or use another drive.
> Conversion can be done using convert. Please look up convert
> in the windows help.
>
> This would have the additional benefit of making users aware.

Something like this would be how to put in the workaround eventually,
yes.

//Magnus


pgsql-hackers-win32 by date:

Previous
From: Joerg Hessdoerfer
Date:
Subject: Re: postgres win32 in FAT32
Next
From: Andrew Dunstan
Date:
Subject: pg_ctl strangeness under msys