[Gluster-users] 3.8 granular heal questions

Ravishankar N ravishankar at redhat.com
Fri Jul 8 04:34:27 UTC 2016


On 07/08/2016 08:59 AM, Lindsay Mathieson wrote:
> Several question re the new granular heal feature in 3.8
> (cluster.granular-entry-heal: = true)
>
> - Whats the chunck size, i.e "granularity"
>
> - Can that size be configured?
There is no chunk size; granular-entry heal is for faster healing of 
directory entries, not healing specific portions within a given file.
The non-granular way of doing this was to do an 'expunge+impunge' of the 
parent directory.

i.e. i) for every file under a given parent directory of the bad brick, 
look it up on the good brick and if it is not present then delete it 
from the bad brick.
      ii)for every file under the same directory of the good good brick, 
look it up on the bad brick and if it is not present, create it.

As you can see, if a directory contains 1000 files to begin with, and 
you create just one new file in it when a brick was down, entry 
self-heal in the non-granular method required opendir, readdir of the 
parent dir + 2001 lookups. That is what is done away with granular 
entry-heal by storing only the name of the new file(s) to be *entry* 
healed and processing only that. It is the same thing for file deletions 
too. The data/metadata heal of a file (in case of creations) would be 
done after the entry heal is complete.

Hope this helps.
Ravi
>
> - Can we see how many chunks need to be healed? "heal info" just shows
> the files being healed, which gives you no idea as to heal progress.
>
> Thanks,
>



More information about the Gluster-users mailing list