On Fri, Sep 21, 2018 at 4:06 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Thomas Munro <thomas.munro@enterprisedb.com> writes:
> > ... There may be ways to fix the dsm_resize() path
> > based on the observation that you don't need to fallocate() if you
> > made the mapping smaller, and if you made it bigger then you could
> > always undo that on error (or not) and you haven't thrown away any
> > data. Hmm... I note that there are actually no callers of
> > dsm_resize(), and it's not implemented on Windows or SystemV.
Erm, actually you probably only need to do ftruncate() *or*
posix_fallocate(), depending on the direction of the resize. Doing
both is redundant and introduces this theoretical hazard (in practice
I'd be surprised if fallocate() really can fail after you shrank a
file that was already fully allocated).
> Why would we fix it rather than just removing it?
I assumed we wouldn't remove an extern C function extension code
somewhere might use. Though admittedly I'd be surprised if anyone
used this one.
--
Thomas Munro
http://www.enterprisedb.com