On Fri, Sep 20, 2019 at 10:55 AM Sage Weil <sage@xxxxxxxxxxxx> wrote: > > On Thu, 19 Sep 2019, Alfredo Deza wrote: > > After deploying Ceph with ceph-deploy on Bionic, the latest luminous > > (12.2.12) has ceph-disk creating a file for a journal - something that > > is very surprising as I have never seen that functionality in > > ceph-disk, without specifying any flags that might indicate a file is > > needed. > > > > Using the same approach with ceph-ansible, the OSD would be created > > with a partition (again, via ceph-disk). Same arguments and all, > > similar to: > > > > ceph-disk -v prepare --cluster=ceph --filestore --dmcrypt /dev/sdX > > > > After going through all the ceph-disk output, this line got different > > results from the ceph-deploy cluster than the ceph-ansible one: > > > > /usr/bin/ceph-osd --check-allows-journal -i 0 --log-file > > /var/log/ceph/$cluster-osd-check.log --cluster ceph --setuser ceph > > --setgroup ceph > > This will always be true for filestore and always false for bluestore. > Perhaps this is a subtle change due to the default for osd_objectstore > having changed between versions? I think the "fix" is probably to pass > '--osd-objecstore bluestore' or '--osd-objecstore filestore' to this > command depending on which type of store was getting created? Ok, so that would explain the mistery... because ceph-disk asks for the OSD with an id of 0, in my case this OSD was created with bluestore, and then I am trying to create one with filestore! This would still be a bug in ceph-disk since I am passing --filestore in the invocation and it still goes out to check... but regardless I will try to clarify in the docs why and when this would happen. Thanks for clarifying Sage! > > sage > > > > > > The ceph-deploy cluster returns a 'no' the ceph-ansible one returns a 'yes'. > > > > The documentation doesn't seem to explain where or how to set/unset > > this. The references to the flag itself are minimal, just mentioning > > that the '--allows-journal' flag is to check if a journal is allowed > > or not. > > > > How does one tell a cluster that a journal is allowed (or not)? > > > > I am happy to go and expand on the documentation to explain this a bit > > even if it is for Luminous only since ceph-volume doesn't check this. > > > >
