Thread: Run-time posix_fallocate failures

Buildfarm member "rover_firefly" has been failing like this since
269e780 went in:

creating template1 database in /export/home/postgres/HEAD/pgsql.47842/src/test/regress/./tmp_check/data/base/1 ...
FATAL: could not allocate space for file "pg_xlog/xlogtemp.60072" using posix_fallocate: Invalid argument
 

Apparently OmniOS has a version of posix_fallocate that doesn't actually
work, or at least doesn't work on the specific filesystem rover_firefly
is using.  The Single Unix Spec says that that is the correct error to
return if the filesystem doesn't provide support:
[EINVAL]The len argument is less than zero, or the offset argument is less than zero, or the underlying file system
doesnot support this operation.
 

I think you'd better rejigger that patch so that it falls through to the
old implementation if posix_fallocate() fails.
        regards, tom lane

On Sat, 2013-07-06 at 10:33 -0400, Tom Lane wrote:
> Apparently OmniOS has a version of posix_fallocate that doesn't actually
> work, or at least doesn't work on the specific filesystem rover_firefly
> is using.  The Single Unix Spec says that that is the correct error to
> return if the filesystem doesn't provide support:

Thank you.

> I think you'd better rejigger that patch so that it falls through to the
> old implementation if posix_fallocate() fails.

Do you mean "fails at all" or "fails with EINVAL"?

Regards,Jeff Davis

Jeff Davis <pgsql@j-davis.com> writes:
> On Sat, 2013-07-06 at 10:33 -0400, Tom Lane wrote:
>> I think you'd better rejigger that patch so that it falls through to the
>> old implementation if posix_fallocate() fails.

> Do you mean "fails at all" or "fails with EINVAL"?

I'd be inclined to ignore the specific error code and just do it the
hard way for any failure.  Less code, and I see no particular advantage
to being pickier than that.
        regards, tom lane