Re: Patch to improve reliability of postgresql on linux nfs

On Sep12, 2011, at 14:54 , ktm(at)rice(dot)edu wrote:
> Many, many, many other software packages expect I/O usage to be the same on
> an NFS volume and a local disk volume, including Oracle. Coding every application,
> or more likely mis-coding, to handle this gives every application another chance
> to get it wrong. If the OS does this, when it gets it right, all of the apps get
> it right. I think you should be surprised when other software actually deals with
> broken I/O semantics gracefully rather than concerned when one of a pantheon of
> programs does not. My two cents.

I don't buy that. People seem to be perfectly able to code correct networking
applications (correct from a read/write API POV at least), yet those applications
need to deal with partial reads and writes too.

Really, it's not *that* hard to put a retry loop around "read" and "write".

Also, non-interruptible IO primitives are by no means "right". At best, they're
a compromise between complexity and functionality for I/O devices with rather
short (and bounded) communication timeouts - because in that case, processes are
only blocked un-interruptibly for a short while.

best regards,
Florian Pflug