| johnsom | That was my question too on the bug | 00:06 |
|---|---|---|
| eandersson | I just spotted that :D | 00:07 |
| eandersson | https://bugs.launchpad.net/designate/+bug/1940976 | 00:07 |
| eandersson | We should probably update the documentation tbh | 00:07 |
| eandersson | Because outside of a DEV env you cannot run Designate without a coordinator | 00:08 |
| eandersson | https://github.com/openstack/designate/blob/05343d4226822da8b9776201ea18e000d366573d/designate/coordination.py#L72 | 00:08 |
| johnsom | Yeah, I was thinking about that today actually. I'm guessing it needs to go in the install guides? | 00:08 |
| eandersson | Yea | 00:08 |
| eandersson | I believe Zookeeper, Redis and Memcached are the valid backends | 00:09 |
| eandersson | But wouldn't put memcached in PROD lol but still better than without any coordinator | 00:09 |
| eandersson | https://docs.openstack.org/tooz/latest/user/compatibility.html | 00:11 |
| johnsom | Yeah, the tooz docs says consul too, but... | 00:11 |
| johnsom | It is this list I think: https://docs.openstack.org/tooz/latest/user/compatibility.html#grouping | 00:11 |
| eandersson | The problem is that we use more than just locking | 00:11 |
| eandersson | Yea | 00:11 |
| eandersson | I think that is right, we use Locking + Grouping | 00:12 |
| johnsom | eandersson While you are here... https://review.opendev.org/c/openstack/designate/+/813176 | 00:13 |
| johnsom | I also commented on https://review.opendev.org/c/openstack/designate/+/813151 that I think we should just remove the SOAP akamai backend as it's EOL. | 00:14 |
| eandersson | Yep was reviewing it earlier today, but.. got distracted lol | 00:14 |
| johnsom | I had some days off, so had a heck of a stack of emails and updates to go through today. | 00:15 |
| eandersson | We need this as well btw https://review.opendev.org/c/openstack/designate-tempest-plugin/+/813299/ | 00:15 |
| johnsom | Ah, missed that | 00:15 |
| johnsom | That is more concerning actually. hmm | 00:16 |
| johnsom | Though, secondary zones were broken with the old dnspython. | 00:16 |
| eandersson | I would like more information | 00:16 |
| eandersson | btw for the other patch I am fine with just the hotfix and then a follow up patch to remove the actual driver | 00:17 |
| eandersson | I will +2 it and let you pick the path (both are fine to me at least) | 00:17 |
| eandersson | I am fine with disabling the test assuming it's something we intend to fix | 00:18 |
| eandersson | I marked all patches with +2 and will let you decide if we should merge them. | 00:21 |
| eandersson | I'll rebuild my local dev env sometime this week and do some testing | 00:21 |
| johnsom | Yeah, I an spinning mine back up now too. | 00:22 |
| johnsom | It's just a bad test | 00:36 |
| eandersson | Good to hear | 00:52 |
| eandersson | Worth considering removing it, or do you think it is salvagable? | 00:52 |
| johnsom | It's salvageable. I will poke at it tomorrow. | 00:53 |
| opendevreview | Merged openstack/designate-tempest-plugin master: Skip test broken by dnspython2 https://review.opendev.org/c/openstack/designate-tempest-plugin/+/813299 | 01:41 |
| opendevreview | Merged openstack/designate master: Fix test failures with dnspython>=2 https://review.opendev.org/c/openstack/designate/+/813176 | 04:10 |
| crohmann | eandersson: johnsom: Sorry I missed your responses and discussion last night (I am in Germany). Looking at the code references I now realize that there is no synchronization within a single instance of the Designate components even and it fully relies on a DLM? And no, I am NOT running a coordinator yet. Certainly any serious setup will likely use multiple instances - but having code is not | 07:02 |
| crohmann | "thread-safe" even for a single instance is really unfortunate. Also the requirement for grouping in the DLM really reduces the choice and likely pushes people to install software they might not need otherwise or even are unfamiliar with. Do you guys see any chance to get this whole problem a little more lightweight? The change nicolasbock proposed at least synchronizes the serial generation / | 07:02 |
| crohmann | incrementation via the storage backend which would, so my theory allow a DEV or single instance Designate setup to run without DLM, right? | 07:02 |
| opendevreview | Arkady Shtempler proposed openstack/designate-tempest-plugin master: Add "cleanup" for created recordsets + delete zone test https://review.opendev.org/c/openstack/designate-tempest-plugin/+/796469 | 09:27 |
| opendevreview | Gunju Kim proposed openstack/designate master: Remove akamai(v1) backend as API no longer exists at Akamai https://review.opendev.org/c/openstack/designate/+/813380 | 14:46 |
| opendevreview | Gunju Kim proposed openstack/designate master: Remove akamai(v1) backend as API no longer exists at Akamai https://review.opendev.org/c/openstack/designate/+/813380 | 14:56 |
| opendevreview | Gunju Kim proposed openstack/designate master: Remove akamai(v1) backend as API no longer exists at Akamai https://review.opendev.org/c/openstack/designate/+/813380 | 14:58 |
| opendevreview | Gunju Kim proposed openstack/designate master: Remove akamai(v1) backend as API no longer exists at Akamai https://review.opendev.org/c/openstack/designate/+/813380 | 14:58 |
| opendevreview | Michael Johnson proposed openstack/designate master: Remove akamai(v1) backend as API no longer exists at Akamai https://review.opendev.org/c/openstack/designate/+/813380 | 15:53 |
| opendevreview | Michael Johnson proposed openstack/designate master: Remove Akamai (eDNS SOAP API) backend https://review.opendev.org/c/openstack/designate/+/813380 | 16:36 |
| eandersson | crohmann: We want a lightweight implementation, but central needs a lot of work. | 17:02 |
| eandersson | I think nicolasbock had a few ideas, but not sure how far along he got | 17:04 |
| opendevreview | Michael Johnson proposed openstack/designate-tempest-plugin master: DNM: Testing dnspython 2.1 tempest issue. https://review.opendev.org/c/openstack/designate-tempest-plugin/+/813685 | 18:24 |
| frickler | johnsom: I did quite some digging on ^^ already. it boils down to dns.query.udp() requiring an IP address now | 18:28 |
| frickler | but we hand it the DNS name of the nameserver. it used to work with dnspython1, though unexpectedly | 18:29 |
| frickler | it also works when running under eventlet, because the greendns wrapper reintroduces doing a resolve step on the nameserver | 18:30 |
| frickler | https://github.com/rthalley/dnspython/issues/552 essentially | 18:32 |
| frickler | and since we use ns1.devstack.org as nameserver entry, for which there isn't an A or AAAA record, this is always going to fail anyway | 18:32 |
| frickler | so IMHO this whole test is bogus in it's current state | 18:33 |
| opendevreview | Merged openstack/designate master: Remove Akamai (eDNS SOAP API) backend https://review.opendev.org/c/openstack/designate/+/813380 | 18:33 |
| frickler | to work properly, it would need to setup a zone with the local devstack IP as nameserver and then disable answering anything | 18:34 |
| johnsom | frickler Thanks for the details, it didn't reproduce for me local so I was trying to gather some more info. | 18:42 |
| eandersson | Indeed! Thanks for the details | 18:44 |
| eandersson | frickler does dns.query.tcp also require an ip? | 18:45 |
| eandersson | https://github.com/openstack/designate/blob/897fc7925bd7c821dcc13df577c0ec75974f5458/designate/mdns/notify.py#L302 | 18:47 |
| eandersson | https://github.com/openstack/designate/blob/05343d4226822da8b9776201ea18e000d366573d/designate/worker/utils.py#L68 | 18:47 |
| frickler | eandersson: yes | 18:47 |
| eandersson | I wonder if these two are meant to be different | 18:47 |
| eandersson | or if that is just an oversight | 18:48 |
| eandersson | The notify implementation wraps host with socket.gethostbyname(host), | 18:48 |
| eandersson | while the worker one does not | 18:49 |
| frickler | I think both need to be the same https://dnspython.readthedocs.io/en/stable/query.html | 18:51 |
| frickler | when reproducing locally I was also first tricked by the eventlet wrapper. needed a held node and debugging in-place to find that out | 18:52 |
| frickler | eandersson: actually the wrapped solution doesn't sound correct to me either. it only does ipv4 and I fear it will also fail when host already is an IP address instead of a name https://docs.python.org/3.8/library/socket.html#socket.gethostbyname | 19:36 |
| frickler | seems we also need to do some proper IPv6 testing | 19:36 |
| crohmann | eandersson: Thanks for the info. In the meantime I shall be hooking up an DLM to our designate then. | 20:27 |
| eandersson | frickler that could be why the new impl does not have it | 20:37 |
| eandersson | crohmann: let me know if you run into any issues | 20:38 |
| eandersson | worst case you can just throw memcached in there :D it's what we use for our CI | 20:38 |
| eandersson | frickler if you want I can take a stab at implementing / fixing that tonight | 21:18 |
| johnsom | eandersson I think I agree with frickler on this "to work properly, it would need to setup a zone with the local devstack IP as nameserver and then disable answering anything" really, this test is going to require a lot more setup. Otherwise it is just always going to skip. | 21:52 |
| johnsom | This test was passing for bogus reasons before. | 21:56 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!