Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-06-09
14:02:01 artom mriedem, btw, continuing the transient/persistent domain discussion from last night
14:02:47 artom mriedem, you may be interested in what kashyap told me this morning
14:02:48 artom http://eavesdrop.openstack.org/irclogs/%23openstack-nova/latest.log.html
14:03:00 artom (Sorry I can't link to a specific line, but it's near the end)
14:05:07 mriedem ok
14:05:08 mriedem read it
14:05:08 melwitt here's a link to a line http://eavesdrop.openstack.org/irclogs/%23openstack-nova/latest.log.html#t2017-06-09T12:59:43
14:05:30 mriedem artom: i'm out today in about 20 minutes,
14:05:35 mriedem this is my debug patch https://review.openstack.org/#/c/471494/
14:05:43 mriedem in neutron which depends on a nova change that dumps debug info when we fail
14:05:54 mriedem interestingly it's survived a few rechecks without hitting hte issue
14:05:55 mriedem of course
14:31:56 bauwser vladikr: do we have some documentation that I can point to for NIC device tagging in order to mark https://bugs.launchpad.net/nova/+bug/1696664 as invalid ?
14:31:56 openstack Launchpad bug 1696664 in OpenStack Compute (nova) "Order of the interfaces is not honored when using virt and SR-IOV interfaces" [Low,Confirmed]
14:32:08 bauwser vladikr: I can point to the spec but it's not super handy
14:32:16 bauwser superdan: in case you know too ^
14:32:37 superdan bauwser: not sure, ask artom
14:32:45 vladikr bauwser, I'm not aware of anything besides the spec..
14:32:51 superdan I'm guessing maybe just the spec and the api-ref
14:33:31 bauwser fair enough, the api-ref is a valid document
14:33:45 artom bauwser, in the end, regardless of what magic we're able to achieve across the entire stack, the guest OS can still do whatever it wants
14:34:07 artom So even if we magically can preserve whatever order the user expects, the guest can say "lulz no, have it my way"
14:34:13 bauwser artom: that's the reason why you proposed tags, right?
14:34:37 artom bauwser, it's the reason tags came about, yeah
14:34:46 bauwser artom: I just want to explain to the bug reporter that he can lookup those tags instead of awaiting specific ordering
14:35:13 bauwser and FWIW, I'd even be in favor of removing the codepath that does the ordering now that we have that feature
14:35:26 bauwser because it makes false promises
14:36:13 artom bauwser, I'm not aware of any doc that promises preserved ordering
14:37:20 bauwser artom: that's worst, that's in code
14:37:39 artom bauwser, it's technically better, because it's easier for us to say "it's not supported"
14:37:46 bauwser which would mean that someone looking at the code could add some documentation
14:37:54 bauwser without getting full knowledge
14:38:10 bauwser artom: see my original comment on the bug
14:38:15 bauwser I was saying that
14:38:23 bauwser but mriedem pointed me the code
14:42:34 artom I'm pretty sure mriedem reads the code as bedtime stories :)
14:50:39 kashyap artom: melwitt: Another way to look at the transient / persistent is:
14:50:43 kashyap If you are going to manage the guest yourself (e.g. via `virt-manager`), you want persistent domains by default, so that even when the guest powers down, libvirt still remembers all the details, and can use those details to reboot the guest.
14:50:58 kashyap However, if you are using a higher-level application to manage guests (like Nova), you probably prefer transient guests, since all the smarts for managing the guests are tied to the cluster as managed by Nova, while libvirt management of persistent guests is local only to one host.
14:51:07 kashyap And lastly...
14:51:40 kashyap A change to a persistent guest is only made on the host where the guest is currently living -- but in a cluster setup, you want to make sure the change lives on, no matter which host in the cluster is currently running the guest; which is why Nova takes care of the configuration, and uses transient guests instead of letting libvirt store (potentially stale) information.
14:51:47 kashyap </EOM>
14:52:36 artom kashyap, aha!
14:52:43 artom That's very helpful, thanks!
14:53:26 kashyap Np; /me wonders if I should paste the above in a .txt file and put it on the web somewhere. I'm sure it's not the first time this came up
14:53:57 artom kashyap, the libvirt docs are pretty sparse on the "why"
14:54:03 kashyap Yeah, I hear you
14:54:14 artom They explain "what", but without the context you just gave, that's not very useful
14:55:13 kashyap I agree, it can be improved
14:55:42 kashyap artom: In other 'news', for upstream QEMU, I began writing a long document for its block operations. An HTML version - https://kashyapc.fedorapeople.org/virt/qemu/live-block-operations.html
14:56:08 kashyap It took a couple of months to slowly chip away at it
14:56:15 kashyap Still in review on upstream qemu-devel list
14:56:53 openstackgerrit Hironori Shiina proposed openstack/nova-specs master: Ironic: Cold migration support https://review.openstack.org/449155
14:57:13 openstackgerrit Gábor Antal proposed openstack/nova master: Transform instance.resize_prep notification https://review.openstack.org/465081
14:59:01 artom bauwser, your message to close the bug looks fine to me.
14:59:26 artom But perhaps you're right, we need to be clearer in the documentation that we don't make guarantees about nic ordering
15:02:02 bauwser artom: I can amend api-ref
15:02:14 bauwser well, if gerrit allows me...
15:04:23 bauwser gibi: would love your insights on https://bugs.launchpad.net/nova/+bug/1696759
15:04:23 openstack Launchpad bug 1696759 in OpenStack Compute (nova) "Inconsistent RabbitMQ messages after subnet update" [Undecided,New]
15:05:59 bauwser gibi: and also, I was looking at a Launchpad tag for that one, can't find it, so will use 'notifications'
15:17:30 openstackgerrit Sylvain Bauza proposed openstack/nova master: Amend api-ref for multiple networks request https://review.openstack.org/472733
15:22:58 bauwser artom: see https://review.openstack.org/472733
15:23:03 danpawlik Hi. Why when I have different password for placement user as nova api has, nova-status upgrade check is raising an error? Is this normal?
15:25:38 artom bauwser, thanks for proposing that, left a suggestion :)
15:26:20 artom bauwser, if you want I can just write the thing myself
15:30:41 stvnoyes mriedem, johnthetubaguy, I finally got migrate working with the cinder v3. The review is here - https://review.openstack.org/#/c/463987/ , please take a look when you get a chance. thx.
15:30:55 bauwser artom: absolutely, flood is yours
15:31:29 bauwser artom: you're Canadian, at least living there, so your english skills are way better than mines
15:31:54 bauwser even if you're living in a very specific area where people speak way more French than we do
15:32:13 bauwser and related, I'll go on 'fin de semaine' soon
15:33:15 artom bauwser, bonne fin de semaine alors :)
15:33:46 bauwser ooooh, and *floor*, not flood
15:34:13 artom bauwser, I know, but we did get flooding in the periphery of Montreal not long ago, so you were kinda correct
15:34:32 artom If you'd said that to vladikr you'd have been 100% right ;)
15:34:37 bauwser orly?
15:34:38 artom Happily, his house was spared
15:34:53 bauwser oh yeah, I think I heard that
15:35:04 vladikr :)
15:37:17 openstackgerrit Stephen Finucane proposed openstack/nova master: objects: Remove custom comparison methods https://review.openstack.org/472285
15:48:42 openstackgerrit Artom Lifshitz proposed openstack/nova master: Amend api-ref for multiple networks request https://review.openstack.org/472733
15:52:28 openstackgerrit Jay Pipes proposed openstack/nova-specs master: Return alternate allocation requests to scheduler https://review.openstack.org/471927
16:10:29 openstackgerrit Jay Pipes proposed openstack/nova-specs master: Return alternate allocation requests to scheduler https://review.openstack.org/471927
16:11:30 openstackgerrit Merged openstack/nova master: Add earliest-version tags for stable branch renos https://review.openstack.org/469969
16:16:05 breton hi
16:17:22 ajay i am seeing a failure in nova-compute creating resources providers for placement on ocata
16:17:24 breton what happens if notification https://github.com/openstack/nova/blob/5d95cb9dbca403790db4e9680919e6716fa5cb76/nova/compute/manager.py#L2092-L2094 fails due to rabbit failure for example? Will the instance status be set to "Failed"?
16:17:31 ajay Failed to create resource provider record in placement API for UUID eff80c6e-a087-458e-8821-8ae4a9f64387. Got 300: {"choices": [{"status": "SUPPORTED", "media-types": [{"base": "application/json", "type": "application/vnd.openstack.compute+json;version=2"}], "id": "v2.0", "links": [{"href": "http://15.0.0.42:8778/v2/resource_providers", "rel": "self"}]}, {"status": "CURRENT", "media-types": [{"base": "application/json", "type":
16:17:50 ajay it seems like it is posting to version=2 even though all my nova endpoints are 2.1
16:19:24 ajay anyone have an idea why this api incompatibility with placement
16:24:40 cfriesen is there a novaclient command that does a GET on /servers/list ?
16:27:23 cfriesen whoops, I mean a GET on /servers
16:30:23 cfriesen answered my own question..." nova list --minimal" will do it.
16:39:45 superdan ajay: placement is a different api versioning scheme
16:40:06 superdan ajay: if placement is running, I would suspect a problem with your service catalog
16:40:17 superdan ajay: do you see any logs from placement that indicate the computes are even hitting it?
16:43:19 openstackgerrit Ildiko Vancsa proposed openstack/nova master: Remove check_detach https://review.openstack.org/446671
16:54:14 ajay no i dont see any logs on placement
16:54:19 ajay saying compute is hitting it
16:56:10 superdan ajay: I expect that means you've got your service catalog pointing to the wrong thing

Earlier   Later