Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-01-04
15:25:50 kashyap mriedem: Still checking
15:25:57 hrw kashyap: thx
15:29:27 kashyap mriedem: Hmm, I just built the newest libvirt-python bindings, and don't see it any of the capabilities; let me ask one of the libvirt folks
15:30:16 kashyap mriedem: So no -- it isn't exposed via any capabilities currently (like `virsh (dom)capablities`)
15:30:39 kashyap mriedem: So Peter (who wrote that commit says):
15:31:16 kashyap 16:29 < pkrempa> kashyap: it is not exposed currently, since it's supposed to be transparent for the users
15:31:19 kashyap 16:30 < pkrempa> and if it's not transparent I'd suggest to complain to qemu
15:31:38 mriedem kashyap: ok so we're stuck with version checks
15:31:40 mriedem thanks for asking
15:31:56 mriedem we could, as noted, add a workaround config option to bypass the version checks if you've patched your packages
15:32:20 kashyap mriedem: But that'd be asking the user to be too awake, and alert and aware :P
15:33:01 kashyap More seriously, yeah - workaround config sounds good
15:33:09 mriedem it wouldn't be the end user, it'd be the deployer
15:33:12 mriedem but yes
15:33:17 mriedem depends on how much people want their multiattach
15:33:25 kashyap You mean in the [workarounds] section, right. Like that live snapshots thing (that now we removed, IIRC)
15:34:03 mriedem yes
15:34:04 openstackgerrit Marcin Juszkiewicz proposed openstack/nova master: libvirt: use 'host-passthrough' as default on AArch64 https://review.openstack.org/530965
15:34:19 mriedem a "this isn't tested upstream so it's not supported thing, but it's a backdoor if you need it"
15:34:27 kashyap Heh, yeah
15:35:08 mdbooth mriedem: lyarwood I also commented on the multi attach patch. There must be a better place to stash multiattach than connection_info. That's essentially adding pain to something we're effectively deprecating.
15:35:12 kashyap hrw: So, on the above patch, thanks for addressing that. Besides that, I don't have anything else
15:35:27 hrw kashyap: thanks for help
15:35:37 coreycb mriedem: will get back to you hopefully shortly. i'm checking with our libvirt maintainer.
15:36:14 mdbooth Incidentally, I think a 'can multiattach' flag on the BDM object would be appropriate.
15:36:24 ildikov mdbooth: what are we deprecating?
15:36:53 mdbooth ildikov: We're trying to get rid of the requirement for Nova to know anything about connection_info
15:37:01 kashyap hrw: Oh, hang on -- same update needs to be done in the rel note.
15:37:07 mdbooth Ideally we can just fetch it from cinder whenever we need it.
15:37:12 kashyap hrw: As that's more public facing; sorry, should've caught it earlier
15:37:15 mriedem mdbooth: with 2 weeks to feature freeze, i think that's a future improvement
15:37:18 hrw kashyap: will do
15:37:30 mriedem mdbooth: because i don't want to drag this out for queens with a schema and object migration and all that
15:37:44 mdbooth ildikov: We definitely don't want to be stashing more stuff in a foreign opaque dict, and then relying on it.
15:37:52 mriedem mdbooth: hell, we could just add a "multiattach" boolean flag to driver.attach_volume
15:38:08 mdbooth mriedem: Yeah, I wondered about that, but I think it needs to be persistent.
15:38:14 mriedem why?
15:38:29 mriedem we get the multiattach value from the volume
15:38:38 mdbooth mriedem: Admittedly I didn't stare at it for a long time, but it's used in volume_driver.get_config()
15:38:38 mriedem i don't really want to persist cinder state in nova's db
15:38:50 mriedem mdbooth: yeah i think that's called from driver.attach_volume
15:38:51 mdbooth I'm pretty sure we can call that outside of the context of attach()
15:38:54 mriedem so we can pass a boolean arg down
15:39:00 hrw kashyap: I just copy comment to release notes basically in next patch
15:39:10 mriedem mdbooth: hmm, for live migrate maybe
15:39:11 mriedem yeah
15:39:13 kashyap hrw: Yep
15:39:18 mdbooth mriedem: However, if that's not true (I didn't check), that would be great
15:39:32 ildikov mdbooth: mriedem: I'm open to suggestions it's just the first idea I had two years ago...
15:40:18 mdbooth Anyway, just my 2c, and I totally get that pragmatism might be required here.
15:40:46 mdbooth But it raised a flag for me. We'll need to unwind it eventually.
15:41:20 hrw also commit message got rewritten
15:41:21 openstackgerrit Marcin Juszkiewicz proposed openstack/nova master: libvirt: use 'host-passthrough' as default on AArch64 https://review.openstack.org/530965
15:41:26 mriedem yeah i think passing a boolean through attach_volume to get_config is easy, it's live migrate that i'm worried about
15:41:48 ildikov mdbooth: not the first time it came up, but we didn't manage to have this as a top priority problem as of yet to find a better way :(
15:42:33 mdbooth ildikov: At the very least we'll need to be able to find it in order to unwind it.
15:43:12 kashyap mriedem: Oh the previous point about capablities, libvirt upstream says, if I file a bug they could add it - as it shouldn't be too difficult
15:43:29 kashyap mriedem: That'd be cleaner for us (Nova), isn't it?
15:43:50 mdbooth ildikov: Incidentally, does service_uuid indicate a shared pool of volumes?
15:44:10 mdbooth ildikov: e.g. multiple volumes on the same NFS mount?
15:44:28 ildikov mdbooth: no, that's independent
15:44:40 mdbooth ildikov: Where can I read what it means?
15:44:55 mriedem kashyap: long-term that would be cleaner yes,
15:45:01 mriedem but not something that is going to help me in queens
15:45:37 kashyap Right. /me imagines: If you add the version check, and then the capability comes along later, no one will remember to swap that, until prompted by something
15:46:13 mriedem kashyap: feel free to file a bug if you want :)
15:46:17 kashyap ildikov: Can you point to the latest URL of the multi-attach specification, please?
15:46:27 openstackgerrit Matthew Booth proposed openstack/nova master: Use real block_device_info data in libvirt tests https://review.openstack.org/527916
15:46:27 openstackgerrit Matthew Booth proposed openstack/nova master: Pass DriverBlockDevice to driver.attach_volume https://review.openstack.org/528363
15:46:28 openstackgerrit Matthew Booth proposed openstack/nova master: Pass disk_info dict to libvirt_info https://review.openstack.org/529329
15:46:28 openstackgerrit Matthew Booth proposed openstack/nova master: Fix libvirt volume tests passing invalid disk_info https://review.openstack.org/529328
15:46:29 openstackgerrit Matthew Booth proposed openstack/nova master: Don't generate fake disk_info in swap_volume https://review.openstack.org/530787
15:46:29 openstackgerrit Matthew Booth proposed openstack/nova master: Expose volume host type and path independent of libvirt config https://review.openstack.org/530786
15:46:30 openstackgerrit Matthew Booth proposed openstack/nova master: Remove redundant swap_volume tests https://review.openstack.org/531179
15:46:30 openstackgerrit Matthew Booth proposed openstack/nova master: Local disk serial numbers for the libvirt driver https://review.openstack.org/529380
15:46:36 mriedem kashyap: https://specs.openstack.org/openstack/nova-specs/specs/queens/approved/cinder-volume-multi-attach.html
15:46:39 kashyap mriedem: Will do
15:46:46 kashyap Thanks
15:47:27 openstackgerrit Andreas Karis proposed openstack/nova master: Add debug output for selected page size https://review.openstack.org/530662
15:48:13 coreycb mriedem: this is being tracked in https://bugs.launchpad.net/ubuntu/+source/qemu/+bug/1716028
15:48:14 openstack Launchpad bug 1716028 in libvirt (Ubuntu) "qemu 2.10 locks images with no feature flag" [Medium,Triaged]
15:48:57 coreycb mriedem: cpaelzer says that after bionic is done he'll take a look at a potential SRU, although he currently has the SRU to artful(pike) as a low priority.
15:49:28 mriedem coreycb: ok thanks
15:51:09 kashyap It sucks hard that one cannot access the URLs (https://review.openstack.org/#/c/267587/78/nova/virt/libvirt/driver.py) without Gerrit account
15:51:31 kashyap Can't pass in-progress patch URLs to people aren't Gerrit users. Surely there must be a way
15:51:44 kashyap Without bothering people to ask to make an account (much like mailing lists).
15:52:35 mriedem kashyap: i'm not signed in and i can view https://review.openstack.org/#/c/267587/78/nova/virt/libvirt/driver.py
15:52:50 mriedem you just can't comment or vote
15:52:55 kashyap mriedem: Err, sorry. The person was complaining about typing in a comment.
15:54:00 kashyap mriedem: Anyway, the quick point that Peter wanted to add was: " sharing disk image is possible even with current qemu/libvirt if the image is 'raw' and <shareable/> is used"
15:55:36 mriedem we must be using qcow2 images
15:56:35 mriedem another thing i can try
15:57:03 ildikov mdbooth: the service_uuid field was added here: https://review.openstack.org/#/c/519025/
15:57:41 mdbooth ildikov: Just found it. I think it's orthogonal to multi-attach, tbh.
15:57:51 mriedem kashyap: although it's a bit confusing,
15:57:57 mriedem we dump the disk config before trying to attach the device
15:58:00 mriedem http://paste.openstack.org/show/638081/
15:58:00 mriedem and it says type="raw"
15:58:51 mdbooth ildikov: Still a good idea. I could also remove my NFS locking stuff in Nova if we had that, although my NFS locking is finer grained but significantly more complex.
15:59:09 ildikov mdbooth: it was added along with a shared_targets field so we can use a lock in case the target exported by the back end is shared among volumes/attachments

Earlier   Later