| openstackgerrit | Eric K proposed openstack/congress: Use partitions in all DSE2 tests https://review.openstack.org/286911 | 00:05 |
|---|---|---|
| openstackgerrit | Eric K proposed openstack/congress: Use partitions in all DSE2 tests https://review.openstack.org/286911 | 00:08 |
| *** ekcs has joined #congress | 00:28 | |
| thinrichs | ekcs: you there? | 00:33 |
| ekcs | hey thinrichs | 00:33 |
| thinrichs | Just saw your new patch. | 00:33 |
| thinrichs | Was it the same problem we saw before? | 00:33 |
| ekcs | very similar. same cause. | 00:35 |
| ekcs | same solution. | 00:35 |
| thinrichs | I was hoping the other tests wouldn't be so sensitive. | 00:35 |
| thinrichs | And that we might only need to use partitions for a few. | 00:36 |
| thinrichs | But it seems that's not the case. | 00:36 |
| thinrichs | I never saw a failure locally—was it just in the gate? | 00:36 |
| ekcs | right. they’re all sensitive. every time I send an rpc call and there is another node with the same id still hanging around, the call can go to the wrong node. | 00:36 |
| ekcs | yes local failure happens when I delete the .tox dir to start clean. | 00:37 |
| ekcs | not totall sure why. | 00:37 |
| *** banix has joined #congress | 00:37 | |
| ekcs | as in not tottaly sure why and old .tox folder would cause the problem not to occur. | 00:38 |
| thinrichs | That .tox problem seems totally weird. | 00:38 |
| thinrichs | You think we could have also fixed it by stopping all the nodes at the end of every test? | 00:38 |
| thinrichs | I still think the partition approach is the right one, since you can fix 1 test before all the other tests are fixed. | 00:39 |
| ekcs | yes. making sure I stop all nodes is the first thing I did. that worked. | 00:40 |
| ekcs | but then I went partitions cuz I agree that’s the better approach. | 00:40 |
| thinrichs | With the node-stop approach, we're fixing test A by going around and fixing all of the other tests B. | 00:40 |
| thinrichs | Was that why you thought partitioning was the right answer? Or was it something else? (I'm always trying to re-evaluate those kinds of decisions after the fact.) | 00:41 |
| ekcs | well i still think either solution would be OK. Stopping in each test has the advantage that if someone forgets to do that in a test, it’ll likely fail on gate. On the other hand, the failure would be confusing because it’d show up all over the place in unrelated tests. | 00:44 |
| ekcs | partitioning has the disadvantage that if someone forgets to add partition in a new test, failure probably won’t show up right away (because everything else it’s in a diff part). On the other hand, when failures do eventually show up, it’ll be clear which tests are causing the failure because only the non-partitioned tests would fail. | 00:45 |
| ekcs | It’s not definitive, but I think partitioning makes sense. A new contributor would be thoroughly confused if s/he writes a new test and all of a sudden things fail all over the place. Let’s say it happens a year from now, diagnosis even by old-timers would depend on retaining this oral tradition of knowledge passed down about test interference. | 00:48 |
| thinrichs | Makes sense to me. Thinking long-term is definitely important here. | 00:49 |
| thinrichs | I pushed your change in by the way. Looks great. | 00:49 |
| ekcs | it’s a trade-off between fail-fast and fail-local. | 00:49 |
| ekcs | great thanks! | 00:50 |
| ekcs | diff topic, the plan now is to focus on testing right? | 00:50 |
| ekcs | haha wacky idea. we could add to the test suite a bunch of extraneous nodes sitting in the default partition and trying to interfere with things. that way anyone would forgets to use partitions would likely see failure right away. Don’t think we should do that now but if people forget to partition a lot, then maybe worth considering. | 00:53 |
| ekcs | then we get both fail-fast and fail-local. | 00:54 |
| ekcs | well actually a better way would be to let test base set a unittest flag. and then make DseNode throw an error every time no partition is supplied AND the unittest flag is set. | 00:59 |
| *** achanda has quit IRC | 01:03 | |
| thinrichs | Or maybe we should have the partition argument always be required and let Python handle it. | 01:06 |
| thinrichs | Yes. The idea is to focus on testing. I'm struggling to get a devstack set up. | 01:07 |
| thinrichs | That's the only real way to test the system. | 01:07 |
| *** banix has quit IRC | 01:07 | |
| thinrichs | (I always struggle to get devstack set up, by the way.) | 01:07 |
| ekcs | haha yea everyone has trouble with devstack. | 01:13 |
| thinrichs | ekcs: do you have a local.conf file that works? | 01:17 |
| ekcs | thinrichs: i think you can pull it out of the HOL image. | 01:23 |
| ekcs | that’s what I did before. will probably do that again. | 01:23 |
| thinrichs | I tried pulling it out of the devstack setup we use in the gate. That hasn't worked out. Maybe I'll see if the HOL is any different. | 01:24 |
| *** jasonsb has joined #congress | 01:25 | |
| openstackgerrit | Merged openstack/congress: Use partitions in all DSE2 tests https://review.openstack.org/286911 | 01:41 |
| *** fawadkhaliq has quit IRC | 01:42 | |
| openstackgerrit | Eric K proposed openstack/congress: Heartbeat update of subscribed tables https://review.openstack.org/281586 | 01:43 |
| *** achanda has joined #congress | 01:51 | |
| openstackgerrit | Eric K proposed openstack/congress: Heartbeat update of subscribed tables https://review.openstack.org/281586 | 01:59 |
| openstackgerrit | Eric K proposed openstack/congress: Heartbeat update of subscribed tables https://review.openstack.org/281586 | 02:00 |
| *** thinrichs has quit IRC | 02:18 | |
| *** banix has joined #congress | 02:34 | |
| openstackgerrit | Eric K proposed openstack/congress: Remove redundant dse2 partitioning code https://review.openstack.org/286939 | 02:42 |
| openstackgerrit | Eric K proposed openstack/congress: Require dse2 partitioning in unittesting https://review.openstack.org/286961 | 03:41 |
| *** fawadkhaliq has joined #congress | 03:43 | |
| *** ekcs has quit IRC | 03:56 | |
| *** banix has quit IRC | 04:20 | |
| openstackgerrit | Anusha Ramineni proposed openstack/congress: Migrate datasource api model https://review.openstack.org/280509 | 04:22 |
| openstackgerrit | Anusha Ramineni proposed openstack/congress: Migrate datasource api model https://review.openstack.org/280509 | 04:25 |
| *** fawadkhaliq has quit IRC | 04:58 | |
| *** banix has joined #congress | 05:06 | |
| *** banix has quit IRC | 05:09 | |
| *** achanda has quit IRC | 05:35 | |
| *** achanda has joined #congress | 05:39 | |
| *** achanda has quit IRC | 05:39 | |
| *** achanda has joined #congress | 06:00 | |
| *** masahito has joined #congress | 07:51 | |
| *** achanda has quit IRC | 08:03 | |
| *** achanda has joined #congress | 08:07 | |
| *** achanda has quit IRC | 08:18 | |
| *** jasonsb has quit IRC | 09:16 | |
| openstackgerrit | Béla Vancsics proposed openstack/congress: Use assert(Not)Equal/Less/GreaterEqual https://review.openstack.org/287060 | 09:39 |
| *** jasonsb has joined #congress | 09:43 | |
| *** masahito has quit IRC | 12:53 | |
| *** fawadkhaliq has joined #congress | 12:54 | |
| *** fawadkhaliq has quit IRC | 13:21 | |
| *** thumpba has joined #congress | 14:16 | |
| *** thinrichs has joined #congress | 15:48 | |
| *** thinrichs has quit IRC | 15:49 | |
| *** thinrichs has joined #congress | 15:50 | |
| *** fawadkhaliq has joined #congress | 16:06 | |
| *** banix has joined #congress | 16:12 | |
| *** fawadkhaliq has quit IRC | 16:14 | |
| *** bryan_att has quit IRC | 16:15 | |
| *** bryan_att has joined #congress | 16:20 | |
| openstackgerrit | Merged openstack/congress: Use assert(Not)Equal/Less/GreaterEqual https://review.openstack.org/287060 | 16:35 |
| openstackgerrit | Merged openstack/congress: Migrate datasource api model https://review.openstack.org/280509 | 17:11 |
| openstackgerrit | Merged openstack/congress: Remove redundant dse2 partitioning code https://review.openstack.org/286939 | 17:11 |
| *** achanda has joined #congress | 17:12 | |
| *** fawadkhaliq has joined #congress | 17:21 | |
| *** achanda has quit IRC | 17:27 | |
| *** jasonsb has quit IRC | 17:30 | |
| *** jasonsb has joined #congress | 17:32 | |
| *** jasonsb has quit IRC | 17:36 | |
| *** achanda has joined #congress | 17:45 | |
| *** absubram has joined #congress | 18:13 | |
| *** thinrichs has quit IRC | 18:16 | |
| *** thinrichs has joined #congress | 18:44 | |
| *** ekcs has joined #congress | 18:50 | |
| *** thinrichs has quit IRC | 19:46 | |
| *** thinrichs has joined #congress | 19:48 | |
| *** thinrichs has quit IRC | 20:30 | |
| *** thinrichs has joined #congress | 20:31 | |
| *** thinrichs has quit IRC | 20:41 | |
| *** thumpba has quit IRC | 20:44 | |
| *** tsandall has joined #congress | 21:18 | |
| *** achanda has quit IRC | 22:06 | |
| *** banix has quit IRC | 22:06 | |
| *** achanda has joined #congress | 22:13 | |
| *** achanda has quit IRC | 22:20 | |
| *** pballand has joined #congress | 22:38 | |
| openstackgerrit | Eric K proposed openstack/congress: Require dse2 partitioning in unittesting https://review.openstack.org/286961 | 22:48 |
| *** thinrichs has joined #congress | 23:04 | |
| *** achanda has joined #congress | 23:20 | |
| *** absubram has quit IRC | 23:25 | |
| *** achanda has quit IRC | 23:28 | |
| *** banix has joined #congress | 23:30 | |
| *** banix has quit IRC | 23:32 | |
| *** thinrichs has quit IRC | 23:38 | |
| *** thinrichs has joined #congress | 23:48 | |
| *** ramineni1 has joined #congress | 23:55 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!