Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-06-26
20:27:48 mriedem sec
20:28:06 mriedem https://review.openstack.org/#/c/539035/
20:28:14 mriedem i assume ^ is related
20:28:59 dansmith mriedem: that's part of their workaround
20:29:08 dansmith mriedem: that isn't needed if we have this, regardless of the oslo_db version
20:33:18 dansmith so, that bug says, I think, that oslo.msg does a split(',') on the netloc part of the url, yeah?
20:33:39 dansmith that'll break my extraction of the username and password,at least, and probably the hostname
20:36:33 dansmith ugh, yeah
20:36:53 dansmith I dunno why that was done instead of just multiple full urls..
20:56:13 mriedem dansmith: ok comments inline
20:56:40 mriedem reading mike's comment now, i suppose i should have read that first
20:57:17 melwitt does anyone know off the top of their head how instances can end up with no InstanceMapping?
20:58:11 melwitt chatting with someone from rdo cloud about it, it's happened to them a couple of times
20:58:52 mriedem no instance mapping or no cell mapping in the instance mapping?
20:58:53 melwitt it seems like something that could only happen during an upgrade, if instance(s) were somehow missed upon nova-manage cell_v2 map_instances
21:01:16 melwitt I'm not sure yet, will be looking at the details tomorrow (they're in EU time zone). but just wanted to ask in case this was a common thing other people have seen
21:01:36 openstackgerrit Merged openstack/nova master: Remove support for /os-fixed-ips REST API https://review.openstack.org/568516
21:02:04 melwitt we know CERN has hit the no cell mapping in the instance mapping, and I still don't understand how that occurs
21:07:27 mriedem i do
21:07:38 mriedem well i know one way you can hit it
21:07:42 mriedem https://review.openstack.org/#/c/576161/
21:11:21 melwitt I won't get to see what the situation is in their db until tomorrow. I doubt they're doing the ip filter though
21:13:41 mriedem well i also started https://review.openstack.org/#/c/575556/ to try and debug when this does happen
21:14:11 mriedem only need to determine if i should lookup the build request to see if we somehow missed the build request and went straight to listing instances with a marker that is on an instance not yet mapped to a cell
21:14:54 mriedem there is a window of time between when we create the instance in a cell and when we actually update the instance mapping in conductor (and then we delete the build request)
21:16:09 melwitt I think last time this happened, it was that there were a few instances in the cell0 db but they had no instance mappings, so they showed up in a 'server list' but could not be deleted because no instance mapping or no cell mapping in instance mapping
21:16:13 mriedem dansmith: yeah so if i create my cell mapping with multiple rabbit hosts using , it would mess up that urlparse - or transport_url for that matter right?
21:16:28 mriedem and transport_url is a stropt which is comma-delimited
21:16:29 dansmith mriedem: yes, but I've got a solution
21:16:38 mriedem [oslo_messaging_rabbit]/rabbit_hosts is a listopt
21:16:40 dansmith the weird chars in db thing is more annoying
21:16:48 mriedem yeah...
21:16:51 openstackgerrit Dan Smith proposed openstack/nova master: Allow templated cell_mapping URLs https://review.openstack.org/578163
21:16:52 dansmith mriedem: checkityo ^
21:23:37 dansmith mriedem: I guess I'm missing something
21:23:57 dansmith surely ? in the url anywhere other than before the query string is not valid right?
21:25:16 melwitt it could be part of the password also, right?
21:25:45 dansmith that's the assertion, but I'm just surprised most url parsing implementations would allow it
21:25:50 melwitt unless that doesn't work in general, connecting to mysql db separately
21:25:59 dansmith clearly python's does not
21:29:07 dansmith yeah, password should be url encoded
21:29:49 dansmith https://bugs.python.org/issue18140
21:29:54 dansmith so I call BS on that
21:30:48 dansmith I would take the three years of no answer to that last question as "nobody has a good reason"
21:31:06 melwitt hm, yeah. I was thinking this same question, "Where do such unencoded URLs come from?"
21:31:30 dansmith same would go for having "@" or ":" in the password.. you'd *have* to urlencode it for sanity
21:31:38 dansmith or / or any of the other reserved characters
21:31:42 melwitt yeah
21:35:24 mriedem while on the subject, i saw this today https://bugs.launchpad.net/nova/+bug/1778515
21:35:24 openstack Launchpad bug 1778515 in OpenStack Compute (nova) "nova-manage list_cells doesn't work is some special characters are in the passwords" [Undecided,New] - Assigned to Surya Seetharaman (tssurya)
21:37:25 melwitt oh yeah, similar deal
21:37:29 mriedem so apparently CERN has [ or ] but not both, or in the wrong order, in their db/mq passwords
21:37:44 mriedem urlparse thinks they are invalid ipv6 urls
21:39:28 mriedem this is the thing we use in map_cell0 https://github.com/zzzeek/sqlalchemy/blob/15ea75981305fdad8286f6803671b864ccda13f2/lib/sqlalchemy/engine/url.py#L203
21:53:25 openstackgerrit Dan Smith proposed openstack/nova master: Allow templated cell_mapping URLs https://review.openstack.org/578163
21:53:36 dansmith I think I got all the actionable feedback ^ but I gotta run off for a bit now
21:57:21 mriedem just noticed something weird, but surely i can't be the first...
21:57:29 mriedem we move instance groups and instance group members from the cell to api db yeah
21:57:47 mriedem and when we delete an instance, the instance_destroy db api code used to also remove the instance group member table entry for the deleted instance
21:57:59 melwitt yeah instance group stuff moved to api db
21:58:03 mriedem we don't have anything like that with deleting an instance, from what i can tell, when deleting an instance and the members are now in the api db
21:58:37 mriedem so if i create an instance in a group, then delete the instance, and show members on the group, will it show the deleted instance uuid?
21:58:57 melwitt hm
21:59:13 mriedem ah _get_not_deleted
21:59:36 mriedem that seems pretty inefficient
22:00:05 mriedem but,
22:00:20 mriedem if you delete the group membership in the api when deleting the instance, if we fail to delete the instance in the cell then we're hosed
22:00:28 mriedem similar to deleting request specs and instance mappings
22:00:45 mriedem seems we should clean up old instance group members during archive like we do for mappings and request specs
22:01:15 melwitt ah, yeah
22:01:45 mriedem although _get_not_deleted also relies on instance mappings, so if we're correctly deleting instance mappings for deleted / purged instances, maybe it's not a problem
22:01:57 melwitt I was trying to think why we didn't go ahead and delete the group member at instance_destroy time but it must be like you said
22:04:04 mriedem looks like we'd add something here if we wanted to https://github.com/openstack/nova/blob/master/nova/cmd/manage.py#L553
22:05:20 mriedem this is where the api does the filtering https://github.com/openstack/nova/blob/master/nova/api/openstack/compute/server_groups.py#L48
22:05:54 mriedem so, we could potentially avoid passing more uuids into the cell db query https://github.com/openstack/nova/blob/master/nova/api/openstack/compute/server_groups.py#L75
22:09:40 mriedem ah well, can track it as low priority https://bugs.launchpad.net/nova/+bug/1778804
22:09:40 openstack Launchpad bug 1778804 in OpenStack Compute (nova) "instance_group_member records not purged from nova_api DB for deleted instances" [Low,Triaged]
22:21:48 openstackgerrit Matt Riedemann proposed openstack/nova master: Remove unused DB API instance_group_delete method https://review.openstack.org/578223
#openstack-nova - 2018-06-27
00:08:18 openstackgerrit karim proposed openstack/nova master: Handle rebuild of instances with image traits https://review.openstack.org/569498
00:39:45 openstackgerrit karim proposed openstack/nova master: Update scheduler to use image-traits https://review.openstack.org/576054
02:12:09 openstackgerrit Zhenyu Zheng proposed openstack/nova master: Use ThreadPoolExecutor for max_concurrent_live_migrations https://review.openstack.org/563505
02:55:32 openstackgerrit Merged openstack/nova master: [placement] Add test demonstrating bug 1778743 https://review.openstack.org/578126
02:55:33 openstack bug 1778743 in OpenStack Compute (nova) "When POSTing to /allocations with multiple consumers it is possible violate inventory capacity constraints" [Medium,In progress] https://launchpad.net/bugs/1778743 - Assigned to Chris Dent (cdent)
03:01:00 openstackgerrit Zhenyu Zheng proposed openstack/nova master: Compute: add support to abort queued live migration https://review.openstack.org/568542
03:08:36 deepak_mourya efried: Hi, what should be the good string to use here according to you https://review.openstack.org/#/c/577726/6/nova/api/openstack/placement/handlers/resource_provider.py
03:09:11 efried deepak_mourya: Yeah, sorry, I didn't leave a lot of detail on that. Are you familiar with the framework for defining the error codes?
03:10:23 deepak_mourya efried: It's ok, any example for the same in the nova before ?
03:10:32 efried deepak_mourya: You want to define it in nova/api/openstack/placement/errors.py, and then look at how the other consts therein are being used.
03:10:46 efried deepak_mourya: But you can do that in a followon patch. The -1 is just for the failing functional test.
03:16:43 deepak_mourya efried: ok, i didn't understand much from here nova/api/openstack/placement/errors.py
03:16:43 deepak_mourya ok so what exactly these things are
03:16:43 deepak_mourya DEFAULT = 'placement.undefined_code'
03:16:43 deepak_mourya INVENTORY_INUSE = 'placement.inventory.inuse'
03:16:43 deepak_mourya CONCURRENT_UPDATE = 'placement.concurrent_update'
03:17:02 deepak_mourya Do i need to create another consts for the same?
03:17:40 efried deepak_mourya: Yes. It's explained in this spec: http://specs.openstack.org/openstack/nova-specs/specs/rocky/approved/placement-api-error-handling.html
03:18:01 efried deepak_mourya: But again, don't try to do it in this same patch. Submit a separate one on top of this one.
03:19:17 deepak_mourya efried: ok i will go through with it first, thanks for the reply and help
03:19:54 efried deepak_mourya: Thanks for the fix!
03:29:32 openstackgerrit xulei proposed openstack/nova master: Disable limits if force_hosts or force_nodes is set https://review.openstack.org/576693

Earlier   Later