Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-10-29
04:56:50 openstackgerrit Arthur Dayne proposed openstack/nova master: libvirt:volume:Disallow AIO=native when no 'O_DIRECT' is available https://review.opendev.org/682772
05:59:44 openstackgerrit Qiu Fossen proposed openstack/nova-specs master: support instance tag fuzzy query https://review.opendev.org/691651
06:12:51 openstackgerrit Qiu Fossen proposed openstack/nova-specs master: support instance tag fuzzy query https://review.opendev.org/691651
06:35:18 brinzhang stephenfin: alex_xu: https://review.opendev.org/#/c/682772/ this patach find the root issue and update the changes, change the conf.driver_io in the libvirt driver, not change the conf.driver_io in any disk_driver in https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L174-L200
06:36:16 brinzhang stephenfin: alex_xu: do you have time to review this, and give any suggestion?
07:02:25 openstackgerrit Daniel Pawlik proposed openstack/nova stable/train: Added openssh-client into bindep https://review.opendev.org/691808
07:02:37 openstackgerrit Daniel Pawlik proposed openstack/nova stable/stein: Added openssh-client into bindep https://review.opendev.org/691809
07:14:00 openstackgerrit Qiu Fossen proposed openstack/nova-specs master: support instance tag fuzzy query https://review.opendev.org/691651
07:49:24 openstackgerrit sunjiyun-inspur proposed openstack/nova master: On branch vm-query-notsameproject https://review.opendev.org/691629
07:51:34 openstackgerrit Arthur Dayne proposed openstack/nova master: libvirt:driver:Disallow AIO=native when no 'O_DIRECT' is available https://review.opendev.org/682772
08:03:30 openstackgerrit Qiu Fossen proposed openstack/nova-specs master: support instance tag fuzzy query https://review.opendev.org/691651
09:21:45 openstackgerrit Qiu Fossen proposed openstack/nova-specs master: Support fuzzy querying instance by tag https://review.opendev.org/691651
10:03:37 sean-k-mooney brinzhang: i guess that is one way to adress it. the other way is to say that its invalid to set the cache mode to 'writeback', 'writethrough'or 'unsafe'
10:04:15 sean-k-mooney e.g. use cache mode = node or direct_sysnc
10:07:33 brinzhang sean-k-mooney: yeah, it's the one way to resolve this issue.
10:07:50 sean-k-mooney this is also arguably a libvirt bug given we currently dont set teh driver io mode
10:08:21 sean-k-mooney im pretty sure i have hit that warning in the past and that is how i fixed it for the lvm driver
10:08:32 sean-k-mooney but using thread io is also accpetable
10:08:57 sean-k-mooney thread io is a little more expensive in terms of qemu cpu overhead
10:09:14 brinzhang I am not sure all of the volume driver can be set thread
10:09:36 sean-k-mooney well we dont want to default to threaded
10:09:47 sean-k-mooney that would be incorrect
10:10:07 brinzhang But if donot set driver_io to thread, the bug
10:10:34 brinzhang the bug is also exist, do you have another suggestion to fix it?
10:10:52 sean-k-mooney no what i mean is we dont want to defualt to therad in general outside of this edgecase
10:10:54 brinzhang I know set thread more expensive
10:11:19 sean-k-mooney by the way i dont think
10:11:20 sean-k-mooney if (hasattr(conf, 'driver_io') and
10:11:22 sean-k-mooney conf.driver_io == "native" and conf.driver_cache in
10:11:24 sean-k-mooney ['writeback', 'writethrough', 'unsafe']):
10:11:26 sean-k-mooney will work
10:11:42 sean-k-mooney do you know where conf.driver_io == "native" is set
10:12:20 sean-k-mooney oh its set in the volume plugins
10:12:55 brinzhang it's in the volume driver, such as iscsi https://github.com/openstack/nova/blob/master/nova/virt/libvirt/volume/iscsi.py#L57
10:13:06 sean-k-mooney yes im not sure its valid to override that
10:13:42 sean-k-mooney the code you added woudl basically change that to threaded on systems without direct io
10:14:40 sean-k-mooney brinzhang: ok i would want to see at least a release not to call out the behavior change
10:15:16 brinzhang Yeah, we test it in iscsi, change "native" to "thread" on my env, it's ok on the function,
10:15:28 sean-k-mooney and ideally i think we would also want a warning in init_host signally to the operator that we are falling back to threaded io
10:15:54 sean-k-mooney brinzhang: threaded i think will always be supported as it was teh legacy way fo doing io
10:16:14 sean-k-mooney so ya it should work with isci
10:16:46 brinzhang sean-k-mooney: thanks reply my doubt
10:17:44 sean-k-mooney brinzhang: ill leave some comment on the review but i think i would be ok with this as the direction for the fix. it will enable most cases to just work and as long as we warn the opertaotr of the behavior change i think its ok
10:18:45 brinzhang sean-k-mooney: yeah, thanks.
11:27:19 openstackgerrit Maciej Kucia proposed openstack/nova master: SR-IOV passthrough: Check PF only if VF is enabled https://review.opendev.org/476642
11:27:46 openstackgerrit sunjiyun-inspur proposed openstack/nova master: query diff user in same project https://review.opendev.org/691849
13:11:58 openstackgerrit Arthur Dayne proposed openstack/nova master: libvirt:driver:Disallow AIO=native when 'O_DIRECT' is not available https://review.opendev.org/682772
13:24:33 openstackgerrit Arthur Dayne proposed openstack/nova master: libvirt:driver:Disallow AIO=native when 'O_DIRECT' is not available https://review.opendev.org/682772
13:56:45 openstack bug 1835822 in OpenStack Compute (nova) "vms loose acess to config drive with CONF.force_config_drive=True after hard reboot" [Medium,Fix released] https://launchpad.net/bugs/1835822 - Assigned to sean mooney (sean-k-mooney)
13:56:45 openstackgerrit sean mooney proposed openstack/nova stable/stein: make config drives sticky bug 1835822 https://review.opendev.org/691864
14:02:33 mriedem huh, the changePassword server action api requires you specify adminPass but the value can be an empty string, in which case the server (nova-compute service) generates a password, but its not returned to the caller...
14:02:38 mriedem that's....helpful
14:03:42 sean-k-mooney very much so i can see why some people migt find that irrating
14:04:09 sean-k-mooney also i abandond ^ since i notice the other config dirve backports were abandoned too
14:05:01 efried mriedem: to what extent is it allowable to run osc version $later against a cloud at version $earlier ?
14:05:11 ganso has seen this problem before and it is an obvious one before I open the bug. Unfortunately I don't have access to the environment so it is hard for me to collect more data, and I couldn't reproduce in my lab. So, has anyone seen this weird behavior before?
14:05:11 ganso Hi folks. I am about to open a new nova bug that my customer is experiencing. They are running "openstack usage list --start 2019-08-01 --end 2019-08-31 --os-compute-api-version 2.40" and it stays in infinite loop. With --debug, I could see that the marker is never updated, therefore the server always sends the same page result over and over. The env is Queens, this was tested with OpenStack CLI 3.19 and 4.0. I decided to check here in case someone
14:05:22 mriedem so i guess the question is, would we consider it a bug that you can generate and change the admin password on a guest without getting the value back?
14:05:32 mriedem because if that's a bug, we might not require a microversion to change it
14:05:58 mriedem efried: osc is meant to run against old server versions
14:06:07 mriedem dtroyer is pretty adament about backward compat with old clouds
14:06:12 mriedem even for like nova-network support
14:06:14 efried how much older? The real question is: when can we drop nova-network support from osc?
14:06:16 efried d'oh
14:06:28 mriedem why are you asking?
14:06:44 mriedem we dropped nova-network support from novaclient and dean had to fix that all up in osc to continue supporting nova-net
14:07:02 efried because osc generates option helps based on whether it detects the 'network' (neutron) service or not.
14:07:04 sean-k-mooney can you tell form the api that it is nova networks?
14:07:11 efried which is fine when you're running in a cloud
14:07:22 efried but not so much when you're generating html docs to publish.
14:07:58 sean-k-mooney efried: well the sphinx docs for osc should generate for all intree options right
14:08:03 mriedem ganso: never heard of it but i wouldn't be surprised with markers, though it would be good to know more details about the marker, e.g. if it's a deleted server
14:08:09 sean-k-mooney and then plugins have there own docs
14:08:21 sean-k-mooney efried: do you wnat to skip the nova networks docs?
14:08:58 mriedem efried: option help or entire commands?
14:09:01 efried both neutron and nova-net options are in osc proper, non-plugin. Yes, I would love to be able to code that decision as: "If I'm not running in a cloud, assume I'm building docs, and since nova-network is dead, only generate the neutron options"
14:09:32 efried mriedem: uhm, not totally sure about that, would have to dig a bit more.
14:09:52 mriedem i'd say either drop nova-network stuff from published docs or you'd have to mark anything as nova-net as nova-net specific i guess? which probably sucks.
14:10:57 efried right
14:10:57 ganso mriedem: hmm that is an interesting detail. So, the marker I have is exactly the last instance returned from the first request. In the dict returned by the API, the instance is marked as "active". The CLI then uses this instance's ID for the next request URL, but then the API just returns the same page, with the same instance as last one. And over and over.
14:11:00 sean-k-mooney efried: given the last time i deployed with nova net osc was not a thing i have never seen the nova net commands/opetions do you have an example
14:11:58 ganso mriedem: Even though the instance is marked as "active" in the returned API, it is possible that the instance has been deleted but was active in the queried period?
14:12:34 efried sean-k-mooney: I'm not sure whether the docs have any of those today. Which would make my job easier, as what I'm doing will just produce the equivalent. Let me go dig some...
14:12:39 efried sean-k-mooney: fyi https://review.opendev.org/#/c/691767/1/openstackclient/network/common.py@174
14:13:57 mriedem ganso: i would expect that if the instance is deleted then it's vm_state is not active
14:14:18 mriedem or you're really unlucky in hitting a tight race window
14:14:57 mriedem i just asked about deleted markers because we've seen bugs with those so i wouldn't be surprised if that is the issue
14:15:21 ganso mriedem: hmm ok so the usage returns the current instance state. About races, the customer can consistently reproduce this, it happens every time
14:15:41 mriedem do you have this fix? https://review.opendev.org/#/c/632516/
14:15:53 mriedem that is in queens 17.0.10
14:16:06 ganso mriedem: yes. Customer is running 17.0.11
14:16:31 sean-k-mooney efried: for what its worth looking at https://docs.openstack.org/python-openstackclient/pike/cli/command-list.html and drilling down it port and network resouce i dont see nova network sepcific commands
14:17:00 mriedem https://docs.openstack.org/python-openstackclient/pike/cli/command-objects/server.html#server-ssh is nova-net specific
14:17:08 efried sean-k-mooney: one example is https://docs.openstack.org/python-openstackclient/latest/cli/command-objects/security-group.html#security-group-list
14:17:09 efried And now I'm noticing that they're tagged.
14:17:09 efried In this case, the nova-network options (just --all-projects) are a subset of the neutron options (which is what the doc has)
14:17:22 sean-k-mooney mriedem: you can use the ssh command with neutron
14:17:44 sean-k-mooney mriedem i use it all the time when im too lazy to lookup the ip
14:19:07 sean-k-mooney efried why is that nova networks specific. it should list the neutorn security groups
14:19:24 sean-k-mooney and --all-project shoudl still work in that context if you are an admin

Earlier   Later