Having it locked in from bootstrap seems like a fair compromise to me, especially if this is well documented and in-line with other config attributes. Do feel free to reach out if you would like me to have a go at this. Thanks 🙂 ________________________________ From: Adam King <adking@xxxxxxxxxx> Sent: Monday, August 12, 2024 4:48 PM To: Alex Hussein-Kershaw (HE/HIM) <alexhus@xxxxxxxxxxxxx> Cc: ceph-users <ceph-users@xxxxxxx>; Joseph Silva <t-josilva@xxxxxxxxxxxxx> Subject: Re: [EXTERNAL] Re: Cephadm and the "--data-dir" Argument I think if it was locked in from bootstrap time it might not be that complicated. We'd just have to store the directory paths in some persistent location the module can access and make ceph cephadm mgr module use them when calling out to the binary for any further actions. This does have the slight issue that technically all the places cephadm uses for storing persistent settings can be modified by the user (config options and config-key store entries) although cephadm already has a number of other config-key store entries it doesn't expect users to modify so that might be fine. It gets more complicated if you allow users to change it post bootstrap as we'd have to implement some kind of migration of the existing dir on the hosts to the new location and make sure we don't make any calls using the new location until the migration is completed. That would take much more investigation. On Mon, Aug 12, 2024 at 11:13 AM Alex Hussein-Kershaw (HE/HIM) <alexhus@xxxxxxxxxxxxx<mailto:alexhus@xxxxxxxxxxxxx>> wrote: Thanks Adam - noted, I expect we can make something else work to meet our needs here. I don't know just how many monsters may be under the bed here - but if it's a fix that's appropriate for someone who doesn't know the Ceph codebase (me) I'd be happy to have a look at implementing a fix. Best Wishes, Alex ________________________________ From: Adam King <adking@xxxxxxxxxx<mailto:adking@xxxxxxxxxx>> Sent: Monday, August 12, 2024 4:05 PM To: Alex Hussein-Kershaw (HE/HIM) <alexhus@xxxxxxxxxxxxx<mailto:alexhus@xxxxxxxxxxxxx>> Cc: ceph-users <ceph-users@xxxxxxx<mailto:ceph-users@xxxxxxx>>; Joseph Silva <t-josilva@xxxxxxxxxxxxx<mailto:t-josilva@xxxxxxxxxxxxx>> Subject: [EXTERNAL] Re: Cephadm and the "--data-dir" Argument Looking through the code it doesn't seem like this will work currently. I found that the --data-dir arg to the cephadm binary was from the initial implementation of the cephadm binary (so early that it was actually called "ceph-daemon" at the time rather than "cephadm") but it doesn't look like that worked included anything to connect it to the cephadm mgr module. So, after bootstrapping the cluster, whenever the cephadm mgr module calls out to the binary to deploy any daemon, it sets the data dir back to the default, hence why you're seeing the unit files being overwritten. This seems to be the case fo all of the `--<thing>-dir` parameters (unit, log, sysctl, logrotate, and data). We are doing the planning session for the next release tomorrow. I might add this as a topic to look into. But for now, unfortunately, it simply won't work without a large amount of manual effort. On Mon, Aug 12, 2024 at 10:06 AM Alex Hussein-Kershaw (HE/HIM) <alexhus@xxxxxxxxxxxxx<mailto:alexhus@xxxxxxxxxxxxx>> wrote: Hi Folks, I'm trying to use the --data-dir argument of cephadm when bootstrapping a Storage Cluster. It looks like exactly what I need, where my use case is that I want to data files onto a persistent disk, such that I can below away my VMs while retaining the files. Everything looks good and the bootstrap command completes. For reference I am running this command: "sudo cephadm --image "ceph/squid:v19.1.0" --docker --data-dir /cephconfig/var/lib/ceph bootstrap --mon-ip 10.235.22.23 --ssh-user qs-admin --ssh-private-key /home/qs-admin/.ssh/id_rsa --ssh-public-key /home/qs-admin/.ssh/id_rsa.pub --output-dir /cephconfig/etc/ceph --skip-dashboard --skip-monitoring-stack --skip-pull --config my.conf" However, when I then try to continue with the deployment of my Storage Cluster, I find that I can't authenticate with the monitors. I run the suggested command to drop into a cephadm shell which then can't speak to the Storage Cluster. For example: $ ceph -s 2024-08-12T10:47:07.862+0000 7f998e59c640 -1 monclient(hunting): handle_auth_bad_method server allowed_methods [2] but i only support [2,1] [errno 13] RADOS permission denied (error connecting to the cluster) In the MON logs at the same time I can see: "cephx server client.admin: unexpected key: req.key=2c62e1471f111d12 expected_key=d18ce06d18f116b4" In the systemd unit files created I see: ... ExecStart=/bin/bash /var/lib/ceph/64415fba-58b0-11ef-9d27-005056014e4f/%i/unit.run ExecStop=-/bin/bash -c 'bash /var/lib/ceph/64415fba-58b0-11ef-9d27-005056014e4f/%i/unit.stop' ExecStopPost=-/bin/bash /var/lib/ceph/64415fba-58b0-11ef-9d27-005056014e4f/%i/unit.poststop ... Which does not contain my data directory. Looking at the source template it appears that it should: ceph/src/cephadm/cephadmlib/templates/ceph.service.j2 at 616fbc1b181ce15e49281553b35ca215d2aa1053 · ceph/ceph (github.com<http://github.com/>)<https://github.com/ceph/ceph/blob/616fbc1b181ce15e49281553b35ca215d2aa1053/src/cephadm/cephadmlib/templates/ceph.service.j2#L22> Manually modifying the unit file, reloading systemd and restarting the mon makes the authentication issue go away, although cephadm seems to be periodically rewriting my file and undoing the changes. Is there a templating bug in here? I note that there are no other variables being templated from the ctx in this jinja2 template so it seems likely it is broken. Many thanks, Alex _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx<mailto:ceph-users@xxxxxxx> To unsubscribe send an email to ceph-users-leave@xxxxxxx<mailto:ceph-users-leave@xxxxxxx> _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |