Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-06-21
11:39:19 kosal openstack-nova, I have query .... What is the maximum number of host aggregate can be created in Availability Zone ?
11:39:56 openstackgerrit Maciej Kucia proposed openstack/nova master: Add JSON schema validation to PCI whitelist https://review.openstack.org/476098
11:45:04 openstackgerrit Maciej Kucia proposed openstack/nova master: Add JSON schema validation to PCI whitelist https://review.openstack.org/476098
11:48:38 openstackgerrit Chris Dent proposed openstack/nova master: Optional separate database for placement API https://review.openstack.org/362766
12:21:08 ebbex Other than max_concurrent_builds, what can I tune to speed up the scheduling/building of instances? (Stress-testing systems, by creating 3-400 instances at a time.)
12:45:17 alex_xu nova api meeting is 15 mins at #openstack-meeting-4
12:45:36 alex_xu s/is 15 mins/is in 15 mins/
12:48:31 openstackgerrit Luis Pigueiras proposed openstack/nova master: Handle keypair not found from metadata server using cells https://review.openstack.org/476122
12:49:50 kosal openstack-nova, I have query .... What is the maximum number of host aggregate can be created in Availability Zone ?
12:50:31 kosal openstack-nova, somebody can help on this query?
12:51:05 openstackgerrit Artom Lifshitz proposed openstack/nova master: API support for tagged device attachment https://review.openstack.org/376021
13:04:43 openstackgerrit Zhenyu Zheng proposed openstack/nova master: Don't delete neutron port when attach failed https://review.openstack.org/403568
13:08:10 mriedem jaypipes: let me start your morning by mentioning when we claim in the scheduler we'll also have to do that for unshelve :)
13:13:32 gibi_ mriedem: hi! I answered you concerns in the soft_delete transformation https://review.openstack.org/#/c/460029/
13:24:16 mriedem jaypipes: would it be possible to put hypervisor driver capabilities on compute node resource providers as traits, so the scheduler could use those for things like scheduling and picking hosts during unshelve?
13:24:46 mriedem then unshelve would inject a required trait into the request spec saying i need a CN RP that can fulfill this thing that is needed for unshelve
13:40:24 sfinucan alex_xu, gmann: comments left on [1] - both easy fixes though, IMO https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:master+topic:bug/1693168
13:41:36 gmann_ sfinucan: thanks, ll check and update in morning.
13:41:50 sfinucan (y)
14:02:48 ftersin mdbooth: hi. did you make a decision for my review?
14:07:50 edleafe p!spy
14:11:41 stvnoyes mriedem: good morning matt. When you get some time, I'd appreciate it if you could review the cinder v3 migrate change - https://review.openstack.org/#/c/463987/ thanks...
14:16:03 mriedem stvnoyes: yup, it's in the list
14:17:07 stvnoyes mriedem: ok thx
14:23:10 openstackgerrit Stephen Finucane proposed openstack/nova master: Monkey patch the blockdiag extension https://review.openstack.org/476159
14:23:10 openstackgerrit Stephen Finucane proposed openstack/nova master: docs: Document the scheduler workflow https://review.openstack.org/475810
14:23:30 tbachman sfinucan: it’s a hook they added to the channel to be able to see when they’ve been named in channel
14:23:32 tbachman sfinucan: try it
14:23:36 tbachman p!spy
14:23:51 sfinucan melwitt, cfriesen_: Think the latter of those resolves your word-wrapping concerns with that activity diagram thing ^
14:23:55 sfinucan p!spy
14:25:31 sfinucan tbachman: Aha! Cheers :) I use xpyhighlight with ZNC for something similar. Seems like less work though, heh
14:25:41 tbachman sfinucan: np!
14:40:53 edleafe sfinucan: I use highlightwin for irssi, but I'm on the road and didn't have access to my bouncer. So it's a good backup.
14:44:50 mriedem jaypipes: nevermind, i guess we never delete allocations on shelve, so we don't claim on unshelve
14:45:00 mriedem we retain the quota while the instance is shelved offloaded
14:48:10 edleafe sfinucan: more info: https://anticdent.org/purple-irc-bot.html
14:49:06 sfinucan edleafe: I think I might have actually read that once upon a time 🤔 Ta for the link
15:02:57 mriedem dansmith: melwitt: to your great sadness, i will be unable to attend the cells v2 meeting today
15:03:12 dansmith mriedem: egads
15:03:35 dansmith probably not much point in having it then, and nothing else burning on the agenda that I know of
15:03:49 mriedem nothing a little penicilin can't fix
15:08:24 dansmith heh
15:10:29 jaypipes mriedem: I'd be amenable to hypervisor-specific traits, sure. I'd have to see what some examples of them would be, though
15:12:07 mriedem jaypipes: ok easy ones came up in the api meeting,
15:12:09 mriedem e.g.
15:12:44 mriedem we have the "supports_device_tagging" capability in the virt drivers
15:12:52 mriedem only libvirt and hyperv support that capability today,
15:13:12 mriedem if you try to create a server with a tagged nic/block device and land on a compute that has supports_device_tagging=False, your instance goes to ERROR state and the build is aborted
15:13:59 mriedem so the thinking is, the compute node resource provider has a trait for that capability flag, and if the server create request has tagged nics/bdms, we put a required trait in the request spec for the scheduler to use fo filtering hosts
15:14:03 jaypipes mriedem: that seems like a fine one to me, sure
15:14:21 mriedem ok coll
15:14:23 mriedem *cool even
15:32:07 efried sfinucan Looks like actdiag will generate SVG with the -T option. Guessing if it's not there already, this would have to be enabled via the docs builder code.
15:32:37 sfinucan efried: I'm using SVG atm (which is why you can copy paste the text)
15:33:01 efried sfinucan Oh, okay. I thought I saw an email that said you were frustrated by that.
15:33:29 sfinucan aha - no, sorry, I should have said I'd managed to figure that one out
15:33:37 sfinucan but not the other one. My mistake :)
15:33:50 sfinucan Though I have figured out the other one now, give or take
15:34:07 efried sfinucan Oh, was about to propose a patch for that.
15:34:32 efried My current crude workaround is jamming \n in the strings by trial/error.
15:34:40 sfinucan efried: Do anyway. Mine is kind of hacky (monkey patching stuff)
15:34:47 sfinucan ...though maybe not as hacky as yours :D
15:35:09 efried I'm looking to see if there's block size and/or autowrap options in the tooling. But I assume you've already done that.
15:46:44 openstackgerrit Markus Zoeller (markus_z) proposed openstack/nova master: docs: Explain the flow of the "serial console" feature https://review.openstack.org/476188
15:48:30 openstackgerrit Matt Riedemann proposed openstack/nova master: Explain why API services are filtered out of GET /os-services https://review.openstack.org/476189
15:49:37 mgagne mriedem: what kind of exception do you suggest it raises? https://review.openstack.org/#/c/454322/18/nova/virt/libvirt/driver.py@1420
15:51:56 efried sfinucan Looks like the best we can do (without hacking the source) is specify box height/width and sprinkle in newlines.
15:53:26 mriedem mgagne: the volume driver's default implementation for the extend_volume method could be just to raise NotImplementedError
15:53:38 sfinucan efried: Hmm, I wonder if hacking the source is such a bad thing though?
15:53:48 sfinucan I mean, IMO the current implementation is broken
15:54:00 mriedem mgagne: the libvirt driver code would have to handle that and translate it to some NovaException, like ExtendVolumeNotSupported or something
15:54:03 mgagne mriedem: so just reraise the NotImplementedError exception?
15:54:12 mgagne ok, that was my question
15:54:16 melwitt dansmith: +1 to skipping the cells meeting today
15:54:51 mgagne there are 2 things: virt might not support extend volume. volume driver within virt driver might not support extend volume. we need to handle both cases. first can be handled by capabilities. I'm not sure about the second one.
15:55:05 efried sfinucan I don't disagree. But where are you planning to hack?
15:55:19 mgagne so you suggest we reuse ExtendVolumeNotSupported if volume driver does not support extend volume?
15:55:21 mriedem mgagne: related comment here https://review.openstack.org/#/c/454322/18/nova/compute/manager.py@6891
15:55:31 sfinucan efried: https://review.openstack.org/476159
15:56:05 mgagne yes, handled that part, I'm not sure about the volume driver as it was not explicitly mentionned. will use the same exception then
15:56:43 efried sfinucan Yeah, that'd do it... partway. What about automatically adjusting the box size, though? With just this, you're going to get cut off.
15:57:01 mriedem mgagne: we could use different exceptions if we want, like you could raise ExtendVolumeNotSupported for the capabilities check, and subclass that with an ExtendVolumeNotSupportedForVolumeType exception if the volume driver doesn't support it - with that one having the type in the exception message
15:57:10 mriedem mgagne: i don't think it makes a ton of difference though
15:57:22 mgagne ok, will use ExtendVolumeNotSupported as suggested
15:57:29 mriedem the difference in error message that goes into the fault could be useful for an operator
15:57:56 mriedem but you tell me :) you're the one deploying this
15:58:16 sfinucan efried: You mean make all boxes bigger?
15:58:27 sfinucan I was trying to avoid that to stop stuff going off screen
15:58:30 mgagne haha, I have a totally different implementation and we fully support the feature so this code path is never used =)
15:58:45 efried sfinucan Well, not necessarily, though that would be easiest (changing the default box height).
15:59:09 efried sfinucan I was talking about scaling the box size dynamically according to font face/size + number of lines after your autowrap.
16:00:38 efried sfinucan ...which sounds pretty difficult to me.
16:01:02 sfinucan efried: That depends on how that tool is implemented
16:01:14 sfinucan efried: If they wrap text, then generate the box, it should be fine
16:01:41 sfinucan If the other way around, it might be tough/impossible as you'd need to resize an existing 'rect' and move anything below it too
16:01:50 efried sfinucan Still not sure how you would account for font face
16:02:27 efried sfinucan Oh, I guess it has a "measuring" facility built in.
16:02:29 efried Which makes sense.
16:02:46 sfinucan Yeah, it does that for us

Earlier   Later