Feature #47274
open
cephadm: make the container_image setting available to the cephadm binary independent of any deployed daemons (maybe ceph.conf?)
Added by Sebastian Wagner over 3 years ago.
Updated about 2 years ago.
Description
The problem is:
If someone calls
cephadm shell -- ceph -s
and no daemon was ever deployed on this host, cephadm will pull the default image which is hardcoded
in cephadm, even if this was overwritten by in the config store.
In order to not depend on downstream tools to pre-pull the image on every host,
mgr/cephadm should add
[client.cephadm]
container_image = registry/image:tag
to the host's /etc/ceph/ceph.conf and also letting cephadm use the container_image as written to the
ceph.conf. Thus making sure we pull the correct image.
wdyt?
- Related to Bug #46561: cephadm: monitoring services adoption doesn't honor the container image added
- Related to Feature #45111: cephadm: choose distribution specific images based on etc/os-releaes added
- Related to Bug #50502: cephadm pull doesn't get latest image added
Seems reasonable. So what happens during a "cephadm pull"? I imagine:
- determine the new version
- set it in the config
- then do the pull and upgrade
...and then unwind the mess if something fails, of course.
Jeff Layton wrote:
Seems reasonable. So what happens during a "cephadm pull"? I imagine:
- determine the new version
- set it in the config
- then do the pull and upgrade
...and then unwind the mess if something fails, of course.
I wouldn't make cephadm pull that complicated. I'd simply pull the image that is configured in the config. Cause that is the image that is going to be used by future deployment on that host. Pulling a different image by default would be a bit strange.
If we want to fetch a new image, we could make ceph orch upgrade ... more elaborate.
I'm not sure this is a good idea since container_image now references a specific digest. That would mean that 'cephadm pull' would be a no-op after the initial pull. Also, we don't currently updated ceph.conf if the container image changes (e.g. upgrade) so that would be another round of updates (though we could do this).
My sense is that it is pretty harmless now because cephadm in each stable series will pull the latest for that series, and having 'cephadm shell' use the latest pacific (or whatever) seems ok?
- Subject changed from cephadm: put the container_image setting into the deployed ceph.conf to cephadm: make the container_image setting available to the cephadm binary independent of any deployed daemons (maybe ceph.conf?)
default image now is still hard-coded by it points to the 'quay.ceph.io/ceph-ci/ceph:master'
Also available in: Atom
PDF