Bug #23119
openMD5-checksum of the snapshot for rbd image in Ceph(as OpenStack-Glance backend Storage) doesNOT match that of the rbd image.
0%
Description
Ceph Version: 12.2.2 Luminous Stable
Problem description:
We use ceph as the backend storage for OpenStack Glance and Nova. There are four osd pools involved in our configuration, rep_glance, ec_glance, rep_nova, ec_nova. The ec_glance and ec_nova pools are backed on HDDs and configured in Erasure-Coding mode.
The rep_glance and rep_nova are backed by SSD acting as the cache tiering layer for ec_glance and ec_nova pools respectively.
When uploading a VM image into the OpenStack-Glance backend storage(Ceph cluster), a rbd image named with the UUID of the VM image such as 8fa76be3-4ade-a897-ca86123467c87 and the snapshot for the image named 8fa76be3-4ade-a897-ca86123467c87@snap are generated.
When the Ceph/OpenStack clusters run for about a month, we find that the rbd image is unavailable when launching new VMs, with error information "no bootable disk". Moreover, for VMs that have been launch before that, the filesystems in VMs also print error information, such as "XFS metadata lost...".
We export the rbd image and the snapshot for it from the rep_glance and ec_glance pools using the "rbd export reppool/8fa76be3-4ade-a897-ca86123467c87 rep-image" and "rbd export reppool/fa76be3-4ade-a897-ca86123467c87@snap rep-image-snap" commands, and compares the MD5 checksums of them. (MD5 checksums are generated with "md5sum -b rep-image" and "md5sum -b rep-image-snap" commands) We find that, the MD5 checksums are NOT matched, and the MD5 checksum of the rbd image 8fa76be3-4ade-a897-ca86123467c87 is consistent with that recorded in OpenStack-Glance, so we infer that the rbd image is not changed.
But why the snapshot of the rbd image is changed? Is it read-only?
Thanks very much!
Files