Hello, As a follow-up of the thread "RBD migration between 2 EC pools : very slow". I'm running Octopus 15.2.13. RBD migration seems really fragile. I started a migration to change the data pool (from an EC 3+2 to an EC 8+2) : - rbd migration prepare - rbd migration execute => 4% after 6h, and no progress 12h later - rbd migration abort => does not return After that, the state of the migration on destination image is "unknown'. rbd info on source (in trash) and destination image shows migrating in the features. Debbugging another abord command shows that it tries to take/put a lock, and there's one already, due to either the cacncelled execute or the subsequent abort. rbd lock delete does not work, I have a strange message about a read only filesystem... So, I'm stuck. My production on the source image is stopped so far, because I use krbd and I must terminate the migration (either commit or abort) before being able to map and mount it. What I think I'm just understanding, is that abort is meant to be used only when the migration execution is finished, like commit ? Am I wrong ? If so, How can we stop an ongoinfg migration, and also, How do we recover the source image ? Do I need to : - delete the destination image - restore the source image from the trash What about the migrating feature that forbid krbd to map the image ? _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |