[Gluster-users] Return previously broken server to gluster cluster

[Strahil Nikolov]

>Do I do the following:
>gluster peer detach urd-gds-021
>gluster peer probe urd-gds-021
>gluster volume replace-brick gds-home urd-gds-021:/brick urd-gds-021/brick


>I just want to be sure before I enter any commands so I do not destroy
instead if repairing.



I saw recently in the mails that the most appropriate way would be to reduce the replica count (remove-brick) and then increase the replica count (add-brick). I guess something like:
- gluster volume remove-brick  gds-home replica 1 <previously-failed-host>:/brick <arbiter-node>:/brick force
- gluster peer detach <previously-failed-host>
- gluster peer probe <newly-reinstalled-host>

On the arbiter:
umount /brick
mkfs.xfs -f -i size=512 /brick
mount /brick
#Avoid using bricks that are actually a mount point:
mkdir /brick/brick

gluster volume add-brick gds-home replica 3 arbiter 1 <newly-reinstalled-host>:/brick/brick <arbiter-node>:/brick/brick


Then trigger a full heal:
gluster volume heal gds-hom full

P.S.: The approach you have described is also valid, so stick with whatever you feel comfortable with.
Don't forget to test you changes before pushing them to production.


Best Regards,
Strahil Nikolov