Description of problem: When using an nvme for both a journal device for an OSD and an OSD for backing the index pool an issue is encountered when creating a docker container with ceph+nvme partitions, the bash scripts doing the OSD prepare inside the container fails because it expects a full drive to be provided and not a partition. Because it's code inside the container, there's no workaround, we need to fix the image and build another one. Version-Release number of selected component (if applicable): RHCS 3.0 How reproducible: Constant Steps to Reproduce: 1. Use nvme device as journal for a ceph OSD 2. Use same nvme device as data/journal partition for an OSD. This forces us to not be able to specify the entire block device and we must create the partitions for the data and journal. 3. Deploy the container with partitons instead of the full block device. The docker image during the osd prepare phase fails because it expect a partition in a partition (nvme0n1p14 is the partition we use, it expects nvme0n1p14p1). Actual results: Unable to deploy the container as we expect to see full block device path Expected results: Should creater the container on the partitioned device.
Unfortunately, this is somewhat expected. However, if you pass the full device, ceph-disk (inside the container) will find the next available partition and will create it with the proper size (if enough space). Can you try this? Also, can we get the log from the container prepare sequence? Thanks.
Closing this BZ as duplicate to: https://bugzilla.redhat.com/show_bug.cgi?id=1541415 *** This bug has been marked as a duplicate of bug 1541415 ***