[lustre-discuss] Odd behavior with Matlab.

[Phill Harvey-Smith]

On 03/06/2016 12:29, Patrick Farrell wrote:
Right,

I've now managed to get the LU-4185 patch to integrate into a copy of 
the lustre tree checked out on 31st may. I did have to make some minor 
changes (documented below), and could not get it to compile in a more 
recent checkout, but since I cannot get a more recent checkout to 
compile even without the patch I assume it's in a state of flux, and 
will be fixed shortly.

What I needed to do to get the patch to compile :

patched the source with patch -p1 < ../patch/9bb0fef5.diff
modified lustre/llite/lproc_llite.c as follows :

in function ll_create_no_open_opt_seq_write

changed val to be defined as :
         __s64   val;

changed the call
        rc = lprocfs_write_helper(buffer, count, &val);
to :
        rc = lprocfs_str_to_s64(buffer, count, &val);

Changed function ll_create_no_open_opt_seq_show
from:

static int ll_create_no_open_opt_seq_show(struct seq_file *m, void *v)
{
        return seq_printf(m, "%s\n",
                          ll_create_no_open_optimization ? "1" : "0");
}

to:
static int ll_create_no_open_opt_seq_show(struct seq_file *m, void *v)
{
         /*return */seq_printf(m, "%s\n",
                           ll_create_no_open_optimization ? "1" : "0");

         return 0;
}

Note I know the above patch to ll_create_no_open_opt_seq_show is 
probably not correct, I need to look and see how other instances of this 
check for errors, and change the return value to reflect this, but the 
above did at least allow me to compile the code and test the patch!

Lustre would then compile, build and install.

using :

lctl set_param llite.home-ffff88105b5e7000.create_no_open_optimization=0

Does indeed seem to allow matlab to run on this volume, so I can now 
proceed, as not being able to run Matlab would have been a show stopper 
for me.

Cheers.

Phill.