On Jun 20, 2007, at 9:43 PM, Murali Vilayannur wrote:
Sam,
Looks great!
A few minor comments
- pvfs2-touch.c: leaks io_servers and layout.server_list.servers.
That was just a quick example of how it would be used. I don't plan
on committing it and probably should have left it out of the patch.
Nice catch just the same!
- instead of sizeof(PVFS_BMI_addr_t), perhaps we can have sizeof
(*io_servers)?
Is the type going to change ever?
- sys-create.sm: seems to leak sm_p-
>u.create.layout.server_list.servers?
Also a nice catch. I've fixed this in my working version.
- I forget if these qhash_search(PINT_fsid_config_cache_table)
required a spinlock to serialize searches or if this a purely
client-side only code with no race conditions possible?
in pint_cached_config_*()
I think that one is ok.
- ANy thoughts on how client-core/vfs apps should make use of the
additional layout information? Another Extended attributes on parent
directory perhaps?
Maybe. Would a layout dir hint be something the VFS users would
use? I had thought that this would be for users of our interfaces only.
Perhaps we could start versioning system interface function
definitions
and have callers specify the version they are interested in?
something like callers have to #define PVFS_USE_VERSION before
including the pvfs.h header file.
#ifndef PVFS_USE_VERSION
#define PVFS_USE_VERSION CURRENT_VER
#endif
#if PVFS_USE_VERSION < CURRENT_VER
#define PVFS_sys_create(x,y,z) PVFS_sys_create_layout(x, y, z, NULL)
#endif
I am sure you guys would have thought of something along this, but
just wanted to doc it here in case :)
I'm not crazy about making users specify a version like that. I'd
rather just change the interfaces as little as possible, and where
necessary make the change and have users handle it themselves.
Thanks for the comments!
-sam
thanks,
Murali
On 6/20/07, Sam Lang <[EMAIL PROTECTED]> wrote:
On Jun 20, 2007, at 5:56 PM, Robert Latham wrote:
> On Wed, Jun 20, 2007 at 05:05:30PM -0500, Sam Lang wrote:
>> Let me know what you think.
>
> My only concern is keeping ROMIO in sync. But what's one more
> autoconf macro?
Its a PITA, is what it is. Rob and I talked about defining a new
PVFS_sys_create_layout function that would take a layout parameter,
leaving PVFS_sys_create untouched. If ROMIO is going to be one of
the users of the layout structure, that may not make a lot of sense
though, and it seems messy. With the latest PVFS FUSE port, maybe
there are more layers that have system interface deps, so we should
consider the potential impact there as well.
-sam
>
> ==rob
>
> --
> Rob Latham
> Mathematics and Computer Science Division A215 0178 EA2D B059
8CDF
> Argonne National Lab, IL USA B29D F333 664A 4280
315B
>
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
