[Gluster-devel] glfs_resolve new file force lookup

[Rudra Siva]

The optimizations look good, I did find out that DHT is where the
information is for the specific volume so that is working well for me.

I'm testing for a 3 brick distributed volume at it gives the correct
sub volume to batch up the writes.

What gfapi is doing looks to be a little iffy to me (eg. glfs_creat) -
based on resolution does open/create - consider the case that 2
clients attempt to do a glfs_creat for the same file - the resolve may
indicate failure to both and they both attempt creation.

specifically the code in glfs_creat based on resolve:

if (ret == 0) {
        ret = syncop_open (subvol, &loc, flags, glfd->fd);
                DECODE_SYNCOP_ERR (ret);
    } else {
        ret = syncop_create (subvol, &loc, flags, mode, glfd->fd,
                     xattr_req, &iatt);
                DECODE_SYNCOP_ERR (ret);
    }

On Mon, Nov 24, 2014 at 9:55 AM, Shyam <srangana at redhat.com> wrote:
> On 11/24/2014 07:43 AM, Rudra Siva wrote:
>>
>> If I understand the structure correctly, the lookup itself generally
>> may not pose a problem - specific flags triggering the over-the-wire
>> request for me at this time I wish to suppress (the fop on the backend
>> is fully capable of creating the file if it does not exist) - the
>> lookup drives through the xlators specifically DHT.
>>
>> It looks like most of the stuff for the hashed subvolume is inside the
>> DHT - I will look into it.
>
>
> You maybe referring to the lookup everywhere that DHT triggers to be sure
> the file is not present anywhere else, and that layout changes caused the
> hashed to change.
>
> If so as an optimization for the same, this is where we are/should be
> headed,
> - http://review.gluster.org/#/c/7702/ (lookup-unhashed optimization)
> - http://review.gluster.org/#/c/8414/ (layout versions that should later
> lead to the same lookup-unhashed among other things)
>
> Shyam
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-devel



-- 
-Siva