[Gluster-devel] 3.6.1 issue

[Vijay Bellur]

On 12/20/2014 12:09 PM, David F. Robinson wrote:
> Seems to work with "-xPf".  I obviously couldn't check all of the files,
> but the two specific ones that I noted in my original email do not show
> any problems when using -P...
>

This is related to the way tar extracts symbolic links by default & its 
interaction with GlusterFS. In a nutshell the following steps are 
involved in creation of symbolic links on the destination:

a) Create an empty regular placeholder file with permission bits set to 
0 and the name being that of the symlink source file.

b) Record the device, inode numbers and the mtime of the placeholder 
file through stat.

c) After the first pass of extraction is complete, there is a second 
pass involved to set right symbolic links. In this phase a stat is 
performed on the placeholder file. If all attributes recorded in b) are 
in sync with the latest information from stat buf, only then the 
placeholder is unlinked and a new symbolic link is created. If any 
attribute is out of sync, the unlink and creation of symbolic link do 
not happen.

In the case of replicated GlusterFS volumes, the mtimes can vary across 
nodes during the creation of placeholder files. If the stat calls in 
steps b) and c) land on different nodes, then there is a very good 
likelihood that tar would skip creation of symbolic links and leave 
behind the placeholder files.

A little more detail about this particular implementation behavior of 
symlinks for tar can be found at [1].

To overcome this behavior, we can make use of the P switch with tar 
command during extraction which will create the link file directly and 
not go ahead with the above set of steps.

Keeping timestamps in sync across the cluster will help to an extent in 
preventing this situation. There are ongoing refinements in replicate's 
selection of read-child which will help in addressing this problem.

-Vijay

[1] http://lists.debian.org/debian-user/2003/03/msg03249.html