[Gluster-devel] upstream: Symbolic link not getting healed

Vijay Bellur vbellur at redhat.com
Sat Dec 21 17:41:57 UTC 2013


On 12/21/2013 12:32 AM, Harshavardhana wrote:
>
> On Fri, Dec 20, 2013 at 4:36 AM, Vijay Bellur <vbellur at redhat.com
> <mailto:vbellur at redhat.com>> wrote:
>
>     On 12/19/2013 02:28 PM, Harshavardhana wrote:
>
>         GFAPI observes ENOENT with glfs_stat() - so the fix is necessary.
>
>
>     I agree that the fix is necessary. We will address it for
>     release-3.5 and master now. Getting this into release-3.4 at this
>     point in time is dicey as we are planning to release 3.4.2 on
>     Monday. Given that the libgfapi problem has existed in 3.4.1 and is
>     not a new regression in 3.4.2, we can target the complete fix for
>     3.4.3. At the moment, I am inclined to revert that fix for getting
>     3.4.2 out.
>
>     -Vijay
>
>
> There is a business case for Bluedatainc
> (http://www.bluedata.com/) which needs that fix who are a consumer of
> GFAPI - they are using the community version 3.4.1qa1 with this fix.
>
> Hadoop jobs "TestDFS_IO" fails without this fix.
>


Sure, there could be a custom 3.4.2 build on which they can continue to 
operate till the problem gets fixed and tested thoroughly in 3.4.3. As I 
mentioned in my OP, the current fix causes self-healing of symbolic 
links to fail and this can potentially cause data loss. This has far 
more serious consequences than a particular workload failing through 
libgfapi.

We have delayed 3.4.2 by a fair while and I don't think we can afford to 
hold it back further. We can consider releasing 3.4.3 soon after the 
errno issue gets fixed properly and we have adequate test coverage on 
that. For now the only option is to revert this patch.

Regards,
Vijay





More information about the Gluster-devel mailing list