[Gluster-users] Broken after 3.7.8 upgrade from 3.7.6

[Alan Millar]

> The "unable to get index-dir on .." messages you saw in log are not 

> harmful in this scenario.
> A simple explanation : when you have 1 new node and 2 old nodes, the 
> self-heal-deamon and
> heal commands run on the new node are expecting that the index-dir 
> "<brickpath>/.glusterfs/xattrop/dirty" exists, to process 
> entries from it.
> But the mentioned directory doesn't exist on the old bricks. (it was 
> introduced as part of
> 3.7.7). As a result you see these logs.


OK.  Thank you for the explanation.   I see it in new bricks in <brickpath>/.glusterfs/indices/dirty, right next to .glusterfs/indices/xattrop   And my old bricks have indices/xattrop but not indices/dirty.  So that matches the warning.  Should I create the "dirty" directory on old bricks?

> But this doesn't explain why heal didn't happen on replacing the brick. 
> :-/
> After replacing, did you check volume heal info to know the status of heal?


Yes, I checked the heal info, and heal statistics, and both were reporting all zeroes, even after doing "heal" or "heal full". That is why I thought it was not working


However, now it looks like the data is healed, and I was just panicking :-/   


I think I was confused about server-side-healing versus client-side-healing.  After looking around some more, it looks to me like server-side-healing is reported in "heal info" and "heal statistics" and glustershd.log.  But client-side-healing is only reported in glfsheal-<volume>.log.  Does that sound correct?

Thanks for the help and information

- Alan