| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-10-04 | |||
| 18:01:13 | dansmith | yeah | |
| 18:02:37 | clarkb | melwitt: on ubuntu (where uca is a thing) the deb packages should install to /usr and pip installed things to /usr/local. So if you get the order sorted out I expect it would just work | |
| 18:03:30 | mriedem | melwitt: weird, is there an ubuntu bug report? | |
| 18:03:36 | mriedem | we could ask jamespage about a fix | |
| 18:03:56 | mriedem | melwitt: alternatively we could workaround that maybe in the devstack-plugin-ceph repo by monkeying with the packages | |
| 18:08:34 | mriedem | heh, wtf, the compute rpcapi method for reboot_instance has a 'block_device_info' parameter that is always None from the API | |
| 18:13:36 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: move RP._set_traits() to module scope https://review.openstack.org/509028 | |
| 18:13:36 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: remove CRUD operations on Inventory class https://review.openstack.org/509029 | |
| 18:13:37 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: streamline InventoryList.get_all_by_rp_uuid() https://review.openstack.org/509030 | |
| 18:13:37 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: remove dead code in Allocation._create_in_db() https://review.openstack.org/509031 | |
| 18:13:38 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: remove ability to delete 1 allocation record https://review.openstack.org/509032 | |
| 18:13:38 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: fix up AllocList.get_by_resource_provider_uuid https://review.openstack.org/509033 | |
| 18:13:39 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: rework AllocList.get_all_by_consumer_id() https://review.openstack.org/509035 | |
| 18:13:39 | openstackgerrit | Jay Pipes proposed openstack/nova master: rp: remove _HasAResourceProvider mixin https://review.openstack.org/509036 | |
| 18:19:39 | cdent | jaypipes: is that orm-removal stuff basically ready for review, or are you still working through the kinks? i’m out of juice to review it now, but tomorrow morning might be good | |
| 18:20:49 | jaypipes | cdent: ready for review, thanks. | |
| 18:20:58 | cdent | coolio | |
| 18:21:06 | jaypipes | cdent: even doing three or four of the bottom ones would be useful. | |
| 18:21:13 | jaypipes | they're pretty smallish patches. | |
| 18:22:07 | jaypipes | cdent: fwiw, the HEAD of that series gets the resource_provider.py file to the point where I can do the root_provider_id/parent_provider_id fields (instead of the root_provider_uuid/parent_provider_uuid fields) that dansmith has asked for on the nested series. | |
| 18:22:33 | jaypipes | that dansmith has requested and that I agree with, btw | |
| 18:23:35 | mriedem | that reminds me, | |
| 18:23:47 | mriedem | andreykurilin: is there a devstack plugin for rally to make it easy to get an env setup? | |
| 18:24:01 | mriedem | oooo there is https://github.com/openstack/rally/tree/master/devstack | |
| 18:24:04 | mriedem | does it work? | |
| 18:24:05 | andreykurilin | mriedem: sure | |
| 18:24:43 | mriedem | what is the default backend when doing that? like do i need to also have ceilometer enabled? | |
| 18:24:44 | andreykurilin | mriedem: yes. it is used in all our gates | |
| 18:25:29 | andreykurilin | hm... ceilometer? rally has nothing related to ceilometer except test plugins for it. the default database backend is sqllite | |
| 18:25:57 | mriedem | ok yeah i was wondering about a db, i don't know much about rally, and i thought there was something to do with configuring it to send metrics collection to ceilometer | |
| 18:26:35 | mriedem | i'm looking for something simple to setup so i can compare test runs before and after some nova patch is applied | |
| 18:26:48 | mriedem | or, comparing scheduling times between ocata and pike | |
| 18:26:49 | mriedem | for example | |
| 18:28:19 | mriedem | i guess i should shut up and just read https://rally.readthedocs.io/en/latest/quick_start/tutorial.html | |
| 18:29:12 | andreykurilin | mriedem: so rally is quite simple and doesn't require any additional services to be installed. All the results are stored in the sql-dabase. One rally installation can check different clusters. The latest clients work perfectly with older releases of openstack. For example we have regular job which launch the scenarios at mitaka release | |
| 18:29:39 | mriedem | ok, i'm mostly interested in scheduling performance between ocata and pike, | |
| 18:30:01 | mriedem | i only have a single node devstack, so i'd have to get creative with mocking out computes | |
| 18:31:08 | mriedem | i know devstack has some flags for mocking out fake compute services when using the fake virt driver, but not sure if that works anymore with the tiered conductor stuff in devstack | |
| 18:31:16 | mriedem | i could run it in singleconductor mode to workaround that | |
| 18:36:04 | openstackgerrit | Merged openstack/nova master: Log consumer uuid when retrying claims in the scheduler https://review.openstack.org/507705 | |
| 18:36:38 | openstackgerrit | Merged openstack/nova master: [placement] gabbi tests for shared custom resource class https://review.openstack.org/485209 | |
| 18:38:38 | andreykurilin | mriedem: btw, I know some guys from huawei who are working on inner performance testing of openstack(as far as I know). I'll contact them about the hardware for such testing as you need | |
| 18:39:13 | mriedem | andreykurilin: are they in toronto? | |
| 18:39:17 | mriedem | hongbin's team? | |
| 18:39:44 | andreykurilin | mriedem: they are from China | |
| 18:39:48 | mriedem | oh | |
| 18:40:10 | mriedem | well, if you can get answers on internal openstack perf testing at huawei faster than i can, kudos to you sir | |
| 18:41:10 | andreykurilin | mriedem: heh. At least I'll try, but it will happen tomorrow, since they are slipping now | |
| 18:41:34 | mriedem | it's a chinese national holiday this week too | |
| 18:41:48 | openstackgerrit | priyaduggirala proposed openstack/nova master: Rename parameters in call() of nova/image/glance.py https://review.openstack.org/508533 | |
| 18:42:00 | melwitt | clarkb: sorry, was on a call. what did you mean by get the order sorted out between the uca package being installed and the pip package being installed? | |
| 18:43:03 | melwitt | mriedem: yeah, there's no ubuntu bug I don't think, because I think this is a unique problem to having devstack install PasteDeploy via pip and then the installation of ceph packages in the devstack plugin installing the uca packages | |
| 18:44:04 | melwitt | I thought the same, that I could probably do something like try uninstalling the uca python-pastedeploy package in the ceph devstack plugin and see if that would work | |
| 18:44:15 | clarkb | melwitt: basically make sure that apt-get happens first, then pip install, then keystone start | |
| 18:45:18 | melwitt | clarkb: ah, okay thanks. I'll see if that's possible in devstack plugin land, if there's a hook point for plugins that occurs before the pip installs that devstack does | |
| 18:46:04 | andreykurilin | mriedem: oh... yes, I forgot about it | |
| 18:46:09 | melwitt | because the devstack ceph plugin is doing the apt-get installs | |
| 18:48:17 | mriedem | melwitt: should be in this phase https://github.com/openstack/devstack-plugin-ceph/blob/master/devstack/plugin.sh#L12 | |
| 18:48:41 | mriedem | check the logs and see if that happens before or after the pip install happens | |
| 18:49:12 | melwitt | hm, yeah ... so it should be doing the right thing already. looking | |
| 18:54:07 | melwitt | says "Requirement already satisfied: PasteDeploy===1.5.2 in /usr/lib/python2.7/dist-packages" which means it skipped the pip install | |
| 18:54:13 | melwitt | (I think) | |
| 18:54:17 | melwitt | so how is keystone not finding it | |
| 18:54:24 | openstackgerrit | Merged openstack/nova master: fix nova accepting invalid availability zone name with ':' https://review.openstack.org/490722 | |
| 18:55:01 | melwitt | I had thought that because PasteDeploy was in the pip freeze output, that it had been pip installed too http://logs.openstack.org/32/508432/1/check/gate-tempest-dsvm-full-devstack-plugin-ceph-ubuntu-xenial/575e932/logs/pip2-freeze.txt.gz | |
| 18:55:17 | melwitt | but I guess that's not necessarily true | |
| 18:55:18 | mriedem | not necessarily | |
| 18:55:32 | mriedem | i remember yum installed packages show up in pip freeze output too | |
| 18:55:43 | mriedem | you just generally aren't using pip with rpms because things get wonky | |
| 18:56:03 | melwitt | yeah, huh. | |
| 18:57:25 | mriedem | right so it's setup via deb first http://logs.openstack.org/32/508432/1/check/gate-tempest-dsvm-full-devstack-plugin-ceph-ubuntu-xenial/575e932/logs/devstacklog.txt.gz | |
| 18:57:39 | mriedem | and then pip ignores it http://logs.openstack.org/32/508432/1/check/gate-tempest-dsvm-full-devstack-plugin-ceph-ubuntu-xenial/575e932/logs/devstacklog.txt.gz#_2017-10-04_01_50_59_811 | |
| 18:57:41 | clarkb | hrm pastedeploy 1.5.2 is latest | |
| 18:57:53 | clarkb | so is the ubuntu package just broken and not including all the bits? | |
| 18:57:58 | mriedem | oops first link is http://logs.openstack.org/32/508432/1/check/gate-tempest-dsvm-full-devstack-plugin-ceph-ubuntu-xenial/575e932/logs/devstacklog.txt.gz#_2017-10-04_01_49_51_453 | |
| 18:58:13 | melwitt | do you think it has anything to do with this? wsgi-file = /usr/local/bin/keystone-wsgi-public http://logs.openstack.org/32/508432/1/check/gate-tempest-dsvm-full-devstack-plugin-ceph-ubuntu-xenial/575e932/logs/etc/keystone/keystone-uwsgi-public.ini.txt.gz | |
| 18:58:16 | mriedem | we are using pike uca now | |
| 18:58:17 | mriedem | as of last week | |
| 18:58:41 | melwitt | is the /usr/local/ there causing keystone to only look under /usr/local for paste deploy? | |
| 18:59:51 | melwitt | clarkb: hm, maybe. I hadn't considered the package could be broken | |
| 19:00:10 | cfriesen | is there a way to specify running a single functional test class? | |
| 19:00:27 | clarkb | cfriesen: tox -e functional -- ClassNameHere ? | |
| 19:00:32 | melwitt | mriedem: yeah, the ceph-mgr started pulling in python-pastedeploy starting in the pike uca | |
| 19:00:35 | clarkb | that actually takes a regex (or should) | |
| 19:00:59 | cfriesen | clarkb: will try it...I had done that with the full dotted path to the class and it didn't like it. | |
| 19:01:15 | artom | cfriesen, clarkb's right, it should be a regex | |
| 19:01:22 | artom | So just ClassName should do it | |
| 19:01:30 | artom | Or even assName, because it matches as wlel | |
| 19:01:38 | mriedem | sorry make that monday https://review.openstack.org/#/c/505674/ | |
| 19:02:08 | cfriesen | artom: clarkb: nope. I ran "tox -e functional -- AggregateMigrationTestCase" and it started doing tests from nova.tests.functional.api.openstack.placement.test_placement_api | |
| 19:02:32 | melwitt | cfriesen: do you have an old checkout? there was some issues around the regex functionality that got fixed while we were at the PTG | |
| 19:02:44 | artom | cfriesen, hrmm - try without the --? | |
| 19:02:45 | mriedem | cfriesen: rebuild | |
| 19:02:48 | cfriesen | melwitt: entirely possible | |
| 19:02:49 | mriedem | tox -r -e | |
| 19:03:48 | melwitt | cfriesen: https://github.com/openstack/nova/commits/master/tox.ini I'd make sure you have the latest version of tox.ini | |
| 19:08:26 | cfriesen | combination of a fresh pull and tox -r seems to have done the trick | |
| 19:08:27 | cfriesen | thansk | |
| 19:14:40 | openstackgerrit | Matt Riedemann proposed openstack/nova master: use already loaded BDM in instance. |
|
| 19:14:41 | openstackgerrit | Matt Riedemann proposed openstack/nova master: use already loaded BDM in instance. |
|