| opendevreview | Merged openstack/puppet-manila master: Remove redundant order enforcement https://review.opendev.org/c/openstack/puppet-manila/+/897514 | 06:37 |
|---|---|---|
| opendevreview | Merged openstack/puppet-manila master: Add support for share manger options https://review.opendev.org/c/openstack/puppet-manila/+/897511 | 06:48 |
| opendevreview | Merged openstack/puppet-manila master: PowerMax: Fix wrong nfs client package name in Ubuntu/Debian https://review.opendev.org/c/openstack/puppet-manila/+/897560 | 07:06 |
| eliaswimmer | Hi! Recently I stumbled upon an issue with cephfs driver in manila, when using a filesystem named other than cephfs. | 09:02 |
| eliaswimmer | With native cephfs. The export location does not include the actual filesystem in use, thus users have to know from somewhere else and in Kubernetes manila CSI driver needs an extra kernel parameter set. | 09:06 |
| eliaswimmer | As far as I know from reading the code, this export information is rather freestyle. | 09:07 |
| eliaswimmer | For native cephfs we could switch to the new mounting syntax (since Pacific?) or add additional fields providing more information. | 09:08 |
| eliaswimmer | The manila CSI driver for Kubernetes, relies on the current format. There the new format would give troubles cause it does not include the mon addresses. | 09:10 |
| eliaswimmer | So either way is not perfect... | 09:10 |
| eliaswimmer | In my opinion, it's best to add more fields returned. | 09:12 |
| eliaswimmer | This would not break APIs? | 09:13 |
| eliaswimmer | Or maybe as metainfo? | 09:24 |
| opendevreview | kiran pawar proposed openstack/manila master: Delete share network subnet during network deletion https://review.opendev.org/c/openstack/manila/+/896541 | 13:08 |
| -opendevstatus- NOTICE: The Gerrit service on review.opendev.org will be offline momentarily while we restart it for a combined runtime and platform upgrade | 16:33 | |
| carloss | o/ eliaswimmer - I believe gouthamr might have some context on this :) | 20:47 |
| eliaswimmer | carlos: true! I hope he will answer me. Maybe I will discuss this at the next PTG | 20:55 |
| * gouthamr o/ lost scrollback, but will consult the logs | 21:36 | |
| gouthamr | eliaswimmer: yes, i think i need some more context as to what relies on this filessytem name data - we've never encoded the filesystem name in the export paths; in the past (xena and prior) releases, we used to allow prefixing the paths with a directory name. that's gone now.. | 21:40 |
| gouthamr | this thread may be relating to the same issue: https://lists.openstack.org/pipermail/openstack-discuss/2023-September/035098.html | 21:40 |
| gouthamr | the background for this comes from the change in the API that manila uses to create shares with cephfs.. we're now using the ceph mgr volumes (and subvolumes) API in the driver... this API has been supported since Ceph Nautilus | 21:41 |
| gouthamr | in Pacific, Ceph killed the older python API module manila used in Xena and earlier releases | 21:42 |
| gouthamr | eliaswimmer: i'd like to understand the usecase a bit better, maybe export location metadata can be a good vehicle for funneling the filesystem information since the driver has it.. if you're available to talk about this at the PTG, please do add an item here: https://etherpad.opendev.org/p/caracal-ptg-manila-planning | 21:44 |
| eliaswimmer | gouthamr: Is there already a time for the manila meetings? If it's before midnight I am in :) | 21:49 |
| eliaswimmer | My usecase was a new openstack setup which uses the same ceph cluster. Thus I wanted to use a different fs for manila. | 21:50 |
| eliaswimmer | This worked as expected. But the export location is lacking the details that this is a different fs than "cephfs" | 21:51 |
| gouthamr | eliaswimmer: yes, slots for the PTG aren't scheduled yet.. carloss was looking to do that soon, the time slots we'd pick though would be between 1300-1700 on Mon, Wed-Fri.. | 21:52 |
| gouthamr | eliaswimmer: ah i see; and kubernetes/CSI consumes from both openstack setups? | 21:53 |
| eliaswimmer | Also manila csi needs this information supplied via storgeclass | 21:53 |
| eliaswimmer | No thankfully not! | 21:53 |
| eliaswimmer | But users need to know! | 21:53 |
| eliaswimmer | It's hidden to the users | 21:54 |
| eliaswimmer | 13:00-17:00 UTC? | 21:54 |
| gouthamr | manila csi would only need the share type info in the storage class today; couldn't you encode this information through that? | 21:54 |
| eliaswimmer | sure, but it's not always me in control | 21:55 |
| gouthamr | (not as transparent, but, the share type can specify the cephfs_filesystem_name) | 21:55 |
| eliaswimmer | yes | 21:55 |
| gouthamr | yes 1300-1700 UTC | 21:57 |
| gouthamr | but not all days; please feel free to indicate a preferred time if you'd like on the etherpad.. if its in the window, carloss can use that to pencil in a spot | 21:58 |
| eliaswimmer | great | 21:58 |
| eliaswimmer | but a 'openstack share type show' does not show the fs | 21:59 |
| eliaswimmer | Anyway I will discuss that in two weeks | 22:00 |
| gouthamr | ++ eliaswimmer | 22:02 |
| gouthamr | eliaswimmer: yes; it'd be a "hidden" extra-spec as far as end users are concerned.. | 22:02 |
| eliaswimmer | It's after midnight now! I wish you nice evening. Will go to bed now :) | 22:08 |
| opendevreview | Takashi Kajinami proposed openstack/puppet-manila master: DNM: Testing ... https://review.opendev.org/c/openstack/puppet-manila/+/897741 | 22:24 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!