| skraynev | hi. could you help with understanding difference between amphorav1/amphorav2? I mean, some doc about these differences. Also is it possible, that with amphorav2 I get error like "resource immutable" when send several create requests one by one (listener, pool, monitor, member). AND on amphorav1 I don't see such error. It looks like amphorav1 - synchronous, when amphorav2 - is asynchronous. | 06:55 |
|---|---|---|
| gthiemonge | skraynev: Hi, johnsom replied to this question yesterday, but basically there are both asynchronous, the only difference is internal: the type of data that are shared between the tasks (amphorav1 uses python objects, amphorav2 uses python dicts), with v2 it's easier to serialize the data and save it to a DB for HA | 07:33 |
| gthiemonge | skraynev: it would interesting to check your worker's logs, is there any backtraces/errors? | 07:34 |
| gthiemonge | skraynev: are those resources stuck in this immutable state? | 07:34 |
| skraynev | gthiemonge: oh, I missed reply from johnsom. sorry. | 07:38 |
| skraynev | gthiemonge: no. pool comes to ACTIVE state after some time. However the case is, that I send send several requests one by one and validate only http status code instead of provisioning status. for amphorav1 it worked without any error about immutable status, after switching on amphorav2 - I get immutable error in 90% situations, when try to repeat action. | 07:41 |
| skraynev | It potentially could be related to execution time, i.e. in amphorav1 resources came to ACTIVE state faster and I did not see same error. So now I try to investigate this behaviour | 07:43 |
| opendevreview | Tom Weininger proposed openstack/octavia master: Remove support for split listener configuration https://review.opendev.org/c/openstack/octavia/+/862599 | 08:05 |
| opendevreview | Tom Weininger proposed openstack/octavia master: Remove support for split listener configuration https://review.opendev.org/c/openstack/octavia/+/862599 | 08:10 |
| gthiemonge | skraynev: v2 might be a little bit slower ATM, it does more DB requests than v1, but I don't think it will more than 1 or 2 seconds to the duration of a request | 08:44 |
| skraynev | gthiemonge: I got it. Could it also be related to changes from ussuri to yoga? I mean, maybe some API changes here. | 08:46 |
| gthiemonge | skraynev: hmm, nop I don't think we made such impactful changes to the API | 08:49 |
| gthiemonge | but, maybe the amphora VM is slower because it's based on a different distribution | 08:49 |
| skraynev | gthiemonge: hm.. maybe.. I will take a look. thank you for the help :) | 08:50 |
| gthiemonge | skraynev: if DEBUG is enabled in the octavia-worker, you will see log message that may be helpful to get the duration of each request | 08:53 |
| gthiemonge | skraynev: for instance, https://paste.opendev.org/show/bUV3DGBcVYIaKTtlcK7T/ | 08:54 |
| gthiemonge | they indicate the start and the end of the creation of a LB | 08:54 |
| skraynev | gthiemonge: interesting. I found, that with amphorav1 octavia-create-pool-flow takes around 1-2 sec, when for amphorav2 it took 1-2 sec and time to time 5-6 sec .... | 09:56 |
| opendevreview | Tom Weininger proposed openstack/octavia master: Add test case for testing with unknown API version https://review.opendev.org/c/openstack/octavia/+/870075 | 10:51 |
| *** dkehn_ is now known as dkehn | 23:57 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!