| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-07-24 | |||
| 12:43:34 | jangutter | claudiub: where do you need to extract the SR-IOV capable PCI nics, is this for capabilities? | |
| 12:43:44 | openstackgerrit | Kashyap Chamarthy proposed openstack/nova master: libvirt: Post-migration, set cache value for Cinder volume(s) https://review.openstack.org/485752 | |
| 12:44:22 | jangutter | claudiub: I was confused with device_id for a bit, until I realised it's a synonym for product_id. | |
| 12:44:42 | claudiub | jangutter: nope, product_id, vendor_id, and device_id are distinct | |
| 12:45:03 | claudiub | you can have multiple devices with the same product_id, vendor_id, but not with the same device_id | |
| 12:45:22 | jangutter | claudiub: whoops, you're right! I forgot about that! | |
| 12:45:29 | claudiub | for example, 2 identical devices will have the same product_id and vendor_id, but different device_ids | |
| 12:46:15 | jangutter | claudiub: Is there something in lspci that indicates SR-IOV capability? | |
| 12:46:54 | claudiub | jangutter: i'm on windows / hyper-v. :) | |
| 12:46:59 | claudiub | jangutter: but yeah. | |
| 12:47:13 | jangutter | claudiub: you poor soul. | |
| 12:48:01 | jangutter | claudiub: "Capabilities: [200 v1] Single Root I/O Virtualization (SR-IOV)" | |
| 12:52:27 | claudiub | moshele: hm, I should be able to whitelist it using "devname" too, right? | |
| 12:52:45 | moshele | claudiub: yes | |
| 12:52:53 | claudiub | hm, let's see. | |
| 12:53:47 | claudiub | jangutter: hah. :)) | |
| 12:54:17 | claudiub | jangutter: what is that "Capabilities: [200 v1] Single Root I/O Virtualization (SR-IOV)" about? | |
| 12:57:16 | ralonsoh | claudiub: can I ask you what capabilities do you need? rso, tso, txvlan, etc? | |
| 13:00:39 | jangutter | claudiub: sorry, that's what I see with lspci. | |
| 13:03:01 | claudiub | jangutter: ah, i see. | |
| 13:03:33 | jangutter | claudiub: is there perhaps something like that in the Windows Device Manager registry? | |
| 13:05:38 | claudiub | jangutter: you mean for sr-iov? it can easily be checked through this: Get-VMHost (check if the host supports SR-IOV), and Get-NetAdapterSriov (check if the NICs support SR-IOV) | |
| 13:05:55 | claudiub | i'm good in both cases. :) | |
| 13:06:36 | claudiub | ralonsoh: any / all. :) currently looking into hyper-v sr-iov. will have to see what capabilities are available. | |
| 13:07:39 | ralonsoh | claudiub: hmmm, sorry, I didn't read that part (hyper-v). I'll take a look | |
| 13:15:06 | sdague | mriedem: when you get a chance this morning, I'd like to figure out what's needed to merge this - https://review.openstack.org/#/c/485602/ - I think the outstanding question is around using the existing config definition | |
| 13:15:29 | sdague | though that might be a reasonable follow on patch to pull that in for both nova-api and placement | |
| 13:17:13 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: Enable cold migration with target host(1/2) https://review.openstack.org/408955 | |
| 13:17:26 | openstackgerrit | Béla Vancsics proposed openstack/nova master: Transform rescue/unrescue instance notifications https://review.openstack.org/385644 | |
| 13:23:21 | mriedem | sdague: so the question is if you're using eventlet, wsgi_log_format config is how you tailor the log format however you want, but with this it's hard-coded so you lose the ability | |
| 13:23:37 | sdague | mriedem: yes | |
| 13:24:09 | sdague | I guess we should determine if that is valuable as config or not | |
| 13:24:22 | sdague | placement goes the hard coded route here as well | |
| 13:24:28 | sdague | so this is consistent with placement | |
| 13:24:30 | mriedem | %(request_line)s must come from eventlet? | |
| 13:24:49 | mriedem | and we're hard-coding: %(REQUEST_METHOD)s %(REQUEST_URI)s | |
| 13:25:34 | mriedem | this is the eventlet wsgi default | |
| 13:25:34 | mriedem | log_format='%(client_ip)s - - [%(date_time)s] "%(request_line)s" %(status_code)s %(body_length)s %(wall_seconds).6f' | |
| 13:26:24 | mriedem | seems odd that we define a default for our config option that's different from the eventlet wsgi default | |
| 13:26:32 | sdague | mriedem: https://github.com/openstack/oslo.service/blob/5a8a712c26cb053cbabf49044e6831ed98a2c2d5/oslo_service/wsgi.py#L184 | |
| 13:26:56 | sdague | mriedem: yeh, well, it's all leaky abstraction | |
| 13:27:02 | sdague | those are eventlet native stanzas | |
| 13:27:12 | mriedem | nova uses it here https://github.com/openstack/nova/blob/master/nova/wsgi.py#L184 | |
| 13:30:27 | sdague | mriedem: ah, I guess we're duping more of oslo.service than I realized, I thought that had been fully factored back out | |
| 13:30:41 | sdague | but, regardless, the current keying are eventlet server keywords | |
| 13:31:02 | mriedem | right, what i'm thinking is if we used the config, | |
| 13:31:14 | mriedem | it would be nice to have it default to None so we could tell if the operator actually cares about overriding it, | |
| 13:31:27 | mriedem | if it's None, and you're using eventlet wsgi, then you get eventlet wsgi defaults for the log format | |
| 13:31:30 | bauzas | mdbooth: do you remember some problems with volume migrations for their type ? https://bugs.launchpad.net/nova/+bug/1705700 | |
| 13:31:31 | sdague | mriedem: yeh, very much good point | |
| 13:31:31 | openstack | Launchpad bug 1705700 in OpenStack Compute (nova) "live migration does not work after volume migration" [Undecided,New] | |
| 13:31:40 | mriedem | if it's None and you're using apache, then you get our hard-coded line | |
| 13:32:01 | sdague | mriedem: we'd also have to normalize the keywords, because things like microversion aren't going to exist where eventlet does this | |
| 13:32:10 | mriedem | right, i'd munge that in with request_line | |
| 13:32:23 | bauzas | mdbooth: tl;dr cinder block volumes seem to be transformed as files | |
| 13:32:38 | mdbooth | bauzas: Looking | |
| 13:33:07 | sdague | https://github.com/eventlet/eventlet/blob/ce72c4378b176cdb0d8b6039023e3ad0026becde/eventlet/wsgi.py#L603-L611 - is where eventlet does it's thing | |
| 13:33:08 | mdbooth | bauzas: That does ring a bell, actually | |
| 13:33:14 | mriedem | sdague: fyi it's also a default value in oslo.service https://github.com/openstack/oslo.service/blob/master/oslo_service/_options.py#L59 | |
| 13:33:19 | sdague | mriedem: yep | |
| 13:33:26 | mdbooth | kashyap: You've fixed ^^^ recently, haven't you? | |
| 13:33:28 | bauzas | mdbooth: yup, I remember some discussion we had for that | |
| 13:33:53 | bauzas | kashyap: https://bugs.launchpad.net/nova/+bug/1705700 | |
| 13:33:54 | openstack | Launchpad bug 1705700 in OpenStack Compute (nova) "live migration does not work after volume migration" [Undecided,New] | |
| 13:34:08 | sdague | mriedem: so, honestly, my inclination is to just say this isn't customizable any more, and only bring that back if people really complain about it | |
| 13:34:16 | kashyap | mdbooth: It's the cache value being not set on the Cinder volume post-migration? | |
| 13:34:16 | mriedem | seems we could deprecate the nova specific wsgi_log_format, although if we want to default it to None we'd then end up relying on the config from oslo.service, which gets messy | |
| 13:34:21 | sdague | because the value in making that site specific seems really weird | |
| 13:34:26 | mdbooth | kashyap: No, different bug | |
| 13:34:33 | kashyap | If so -- https://review.openstack.org/#/c/485752/ -- libvirt: Post-migration, set cache value for Cinder volume(s) | |
| 13:34:38 | openstackgerrit | Gábor Antal proposed openstack/nova master: Transform instance.resize_revert notification https://review.openstack.org/396811 | |
| 13:34:46 | kashyap | mdbooth: Oh, is it the one bauzas pointed out just now? | |
| 13:34:53 | mdbooth | kashyap: Yes | |
| 13:35:12 | sdague | mriedem: because it feels like you could do a lot of effort here for 0 people ever doing this thing | |
| 13:35:14 | kashyap | mdbooth: Oh, this one -- I recall it | |
| 13:35:16 | mdbooth | https://bugs.launchpad.net/nova/+bug/1705700 | |
| 13:35:17 | openstack | Launchpad bug 1705700 in OpenStack Compute (nova) "live migration does not work after volume migration" [Undecided,New] | |
| 13:35:31 | mriedem | sdague: yeah i'm asking in the operators channel, but i don't care to bend over backwards for the config option | |
| 13:35:40 | bauzas | mdbooth: looked related but not the same problem | |
| 13:35:46 | bauzas | oops kashyap rather | |
| 13:36:15 | kashyap | mdbooth: Not me (Artom), but I did the triage of it with Eric Blake. The change was simple | |
| 13:36:19 | mdbooth | bauzas: Yeah, I've definitely seen a patch for the file='block device' thing | |
| 13:36:21 | bauzas | claudiub: I'm unclear whether https://bugs.launchpad.net/nova/+bug/1705683 is only for hyper-v driver | |
| 13:36:22 | openstack | Launchpad bug 1705683 in compute-hyperv "vif ports are not unplugged after cold migration" [Undecided,In progress] - Assigned to Alexandru Muresan (amuresan) | |
| 13:36:29 | mriedem | sdague: https://github.com/openstack/nova/commit/c01c852c808970ff649c26a0e52d34b3c2d80b0b | |
| 13:36:53 | kashyap | We need both the flags: VIR_DOMAIN_BLOCK_REBASE_COPY _and_ VIR_DOMAIN_BLOCK_REBASE_COPY_DEV | |
| 13:37:08 | mriedem | sdague: sounds like that was a bug fix at one point or a workaround for a limitation in eventlet | |
| 13:37:23 | mriedem | sdague: and then oslo.service copied it from nova | |
| 13:37:27 | mriedem | https://github.com/openstack/oslo.service/commit/6dd3ad50b6fee79b4cc08fe0830327803d4b2f5b | |
| 13:37:56 | bauzas | mdbooth: if you can find it, you'll get eternal gratitudes | |
| 13:38:17 | mdbooth | bauzas: I'll have a hunt in a bit. Just doing some fun, fun, fun bugzilla diving. | |
| 13:38:19 | bauzas | mdbooth: but I can lookup gerrit for libvirt/volume.py recent changes | |
| 13:38:55 | mriedem | sdague: so i'd say we keep them separate and actually deprecate the nova wsgi_log_format option since it's in oslo.service, and doesn't even really seem useful anymore | |
| 13:38:55 | bauzas | mdbooth: nah, I can try to find so | |
| 13:39:10 | kashyap | mdbooth: Whoops, I think I see the cause: "The issue is caused by the fact that dist type is changed from 'block' to 'file' after volume migration." | |
| 13:39:20 | bauzas | mdbooth: you'll just miss unlimited beer credits | |
| 13:39:22 | kashyap | It should _retain_ 'block' as the disk type post-migration, too. | |
| 13:39:27 | sdague | mriedem: ah, right, because we're going through the python logger the default logger in eventlet is also putting in timestamping | |
| 13:39:44 | bauzas | kashyap: that is the root cause as per the reporter indeed | |
| 13:40:07 | sdague | mriedem: ok, same patch for deprecation, or follow on? | |