| *** mriedem_afk is now known as mriedem | 00:03 | |
| *** mriedem has quit IRC | 01:54 | |
| openstackgerrit | Merged openstack/watcher master: Add name for instance in Watcher datamodel https://review.opendev.org/660295 | 02:20 |
|---|---|---|
| openstackgerrit | Merged openstack/watcher master: improve the process of instance_created.end https://review.opendev.org/666795 | 02:29 |
| openstackgerrit | Merged openstack/watcher-dashboard master: Add py37 in tox.ini and setup.cfg file https://review.opendev.org/667515 | 02:29 |
| openstackgerrit | Merged openstack/watcher master: Configure nova notification format in non-grenade CI jobs https://review.opendev.org/663332 | 02:29 |
| openstackgerrit | Merged openstack/watcher master: Add uWSGI support https://review.opendev.org/666779 | 02:29 |
| *** zhurong has joined #openstack-watcher | 03:22 | |
| openstackgerrit | chenker proposed openstack/watcher master: Reduce the query time of the instances when call get_instance_list() https://review.opendev.org/668100 | 06:59 |
| *** licanwei has joined #openstack-watcher | 07:00 | |
| Dantalion | I will try the limited get_instance_list once I have confirmed in my environment what happens if I change back to find_instance | 07:04 |
| Dantalion | I think in the long run find_instance can be way more efficient because we can use futures to do many find_instance calls in parallel Watcher side | 07:04 |
| *** chenke has joined #openstack-watcher | 07:06 | |
| chenke | Hi,~ | 07:06 |
| chenke | https://review.opendev.org/#/c/668100/ | 07:07 |
| licanwei | yes, If find_instance can run in parallel, maybe it's fater. | 07:08 |
| openstackgerrit | pengyuesheng proposed openstack/watcher-dashboard master: Blacklist sphinx 2.1.0 (autodoc bug) https://review.opendev.org/668101 | 07:09 |
| licanwei | but it will have more api calls than one instance_list. | 07:10 |
| Dantalion | Maybe it should be user configurable | 07:11 |
| Dantalion | If you have one nova_api doing 15 find_instance calls might make nova unhappy | 07:11 |
| Dantalion | but for us parallesis find_instance will have very good performance since there are 15 nova-api instances | 07:11 |
| licanwei | yes, maybe user config | 07:13 |
| chenke | I think the reason why it takes a lot of time here is because of the limitation of limit. | 07:14 |
| chenke | Yes, it is very fast in parallel. | 07:15 |
| Dantalion | I will try the limit in get_instance_list today | 07:15 |
| licanwei | chenke just submit a commit to reduce the get_instance_list call | 07:16 |
| licanwei | https://review.opendev.org/#/c/668100/ | 07:16 |
| Dantalion | Exactly, that is what I meant with try the limit in get_instance_list | 07:16 |
| licanwei | you can try it | 07:17 |
| chenke | Okay. I think this patch will save about 300s. | 07:17 |
| Dantalion | I just tried find_instance with the single compute call | 07:20 |
| Dantalion | it reduces total time for medium cells from 300 to 260 :) | 07:20 |
| Dantalion | No I will try get_instance_list limit | 07:20 |
| chenke | Ok. wait for your result Email. | 07:26 |
| licanwei | Jun 27 09:57:57.130293 ubuntu-bionic-rax-dfw-0008422178 watcher-decision-engine[2170]: DEBUG keystoneauth.session [-] REQ: curl -g -i --cacert "/opt/stack/data/ca-bundle.pem" -X GET http://10.209.4.32/compute/v2.1/servers/detail?all_tenants=True&host=ubuntu-bionic-rax-dfw-0008422178&marker=dbb0828d-a42f-4582-b6c0-e72041067652 | 07:27 |
| licanwei | Jun 27 09:57:57.316596 ubuntu-bionic-rax-dfw-0008422178 watcher-decision-engine[2170]: DEBUG keystoneauth.session [-] RESP: [200] | 07:27 |
| Dantalion | The limit resolves all issues :) | 07:27 |
| Dantalion | all time low result 239 seconds | 07:27 |
| licanwei | http://logs.openstack.org/22/667822/1/check/watcher-tempest-basic_optim/17c2a7e/controller/logs/screen-watcher-decision-engine.txt.gz | 07:28 |
| chenke | why we can't see picture. | 07:28 |
| licanwei | Dantalion: I find it's faster than your cern. maybe the time depends on the backend database | 07:29 |
| Dantalion | Time is all relative ofcourse it depends on many things, Matt indicated that performance problems with get_instance_list typically only occur in multi-cell clouds | 07:30 |
| openstackgerrit | Dantali0n proposed openstack/watcher master: Releasenote for grafana datasource https://review.opendev.org/668110 | 08:00 |
| *** sum12 has joined #openstack-watcher | 08:17 | |
| *** sum12 has quit IRC | 09:00 | |
| *** sum12 has joined #openstack-watcher | 09:02 | |
| *** licanwei has quit IRC | 09:14 | |
| openstackgerrit | zhufl proposed openstack/watcher master: Fix invalid assert states https://review.opendev.org/668124 | 09:25 |
| *** chenke has quit IRC | 09:54 | |
| *** mriedem has joined #openstack-watcher | 13:24 | |
| *** mriedem is now known as mriedem_away | 16:18 | |
| Dantalion | I/18 | 16:57 |
| *** mriedem_away is now known as mriedem | 17:27 | |
| mriedem | Dantalion: what version of python-novaclient are you using in your stein testing? | 17:38 |
| mriedem | actually it probably doesn't matter | 17:42 |
| mriedem | i think i know how we can speed up add_instance_node | 17:42 |
| mriedem | ah nvm i see there is already a patch https://review.opendev.org/#/c/668100/ | 17:49 |
| Dantalion | mriedem: Yes chenker figured out that setting limit returns performance to normal | 18:55 |
| Dantalion | mriedem: new tests are already partialy done it looks really good :) | 18:56 |
| mriedem | awesome | 18:58 |
| mriedem | i'd be interested to know overall what the before / after comparisons are for stein/master with the 3 different optimization patches (add_instance_node and the other ones for the hypervisor APIs) | 18:58 |
| mriedem | but as long as you get back to or better than stein that's great | 18:59 |
| Dantalion | mriedem: it is better than stein now for sure :) so in the end the single API is faster but this issue with get_instance_list obscured the measurements | 19:01 |
| mriedem | that's great to hear | 19:01 |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!