| mikal | noonedeadpunk / sean-k-mooney: I had some more thoughts about wsgi and wrote a reply to https://github.com/shakenfist/kerbside/issues/14, so I guess we should keep the conversation there at this point... | 09:11 |
|---|---|---|
| sean-k-mooney | mikal: ack. normally you woudl run this using two systemd unit files and 2 contianer in the kolla world and ues the fact that you can have depencies between the units/containers to enforce startup order. ovs in kolla for exampel has 2 contaiern one for the ovsdb and one for the ovsvswitchd | 10:24 |
| mikal | sean-k-mooney: that's true, and its what I think I should do here. Its more a case of that not having been how its already done, and me wanted to make sure I don't need to re-write the SPICE proxy before putting much more effort into the current implementation. | 10:31 |
| sean-k-mooney | mikal: yep i understand. im kind fo surpised that kolla are ok with the current approch, it was very contoverial when i added bifrost supprot with is the only example fo a multi service contaienr in kolla outside of how openstack service spawn a privesep service | 10:33 |
| mikal | sean-k-mooney: I don't think they know. Whilst the patches aren't secret I haven't gotten as far as proposing them in gerrit yet -- I needed the nova patches and client changes to merge first and that happened too late for 2025.1. I also want to sort out the plan for scaling first too. | 10:34 |
| sean-k-mooney | mikal: form a secity and oerational side i think it would make sense to supprot splitign kerbside long term but short term what you have now will work. i said the same in the pr | 10:34 |
| mikal | sean-k-mooney: I do think scaling the HTTP and SPICE aspects separately make sense, I just didn't occur to me a year or more ago when I wrote that code. | 10:35 |
| sean-k-mooney | mikal: i have not looked at the kerbside code in any detail but does the http side need to actully interact with the SPICE side directly via IPC | 10:38 |
| sean-k-mooney | if not or even if so it wprobaly woudl be easy to add flags to the helper to only run one or the other and use a unix socket for comunicatin | 10:38 |
| sean-k-mooney | that woudl allow you to supprot the all in one or split deplyment pretty simply | 10:39 |
| sean-k-mooney | and if someone didnt want to use the helper for running the http part that already possibel today | 10:39 |
| sean-k-mooney | they would just need to make sure to pass it the unix socket path to talk to the spice part assumign there is some IPC requireemnt | 10:40 |
| sean-k-mooney | if there isnt the its even simpler | 10:40 |
| sean-k-mooney | i assume there is since you spwaning it with a pipe today | 10:40 |
| mikal | All the interaction is via the MySQL database fortunately. The REST API / Nova Auth URL thingie create state (authorized tokens basically) that the SPICE side consumes. In return the SPICE side adds data to MySQL for the REST API to use (mostly information about currently connected sessions). | 10:41 |
| mikal | Its really a fairly simple app. The hard bit was the incompete state of the SPICE protocol documentation. | 10:41 |
| mikal | Especially the proprietary compression algorithms. That kept me out of mischief for a while. | 10:42 |
| sean-k-mooney | mikal: in that case it shoudl be trivial ot split the process and scale them indepently later. they just need to both have access to the db | 11:13 |
| opendevreview | Balazs Gibizer proposed openstack/nova master: Allow services to start with threading https://review.opendev.org/c/openstack/nova/+/948311 | 12:11 |
| opendevreview | Balazs Gibizer proposed openstack/nova master: Run nova-next with n-sch in threading mode https://review.opendev.org/c/openstack/nova/+/948450 | 12:11 |
| opendevreview | Stephen Finucane proposed openstack/nova master: api: Address issues with hypervisors APIs https://review.opendev.org/c/openstack/nova/+/950867 | 13:03 |
| stephenfin | sean-k-mooney: ^ | 13:03 |
| sean-k-mooney | ack so you are injectign that to the side fo the main seriese. i guess as long as we merge the 4 hypervior api paches at the same time its probaly fine | 13:23 |
| sean-k-mooney | gibi: i looked at a couple of your eventlet series and im mostly happy with the ones i looked at. i left a +1 with some comoens on 2? of them | 15:22 |
| sean-k-mooney | ill take a look at them again later in teh week | 15:22 |
| sean-k-mooney | but im going to switch to a diffent topic for a while | 15:22 |
| gibi | sean-k-mooney: ack, I will check you comments tomorrow | 17:17 |
| gibi | as the oslo.service patch is landed and released I bumped to that release in the series. Also I changed the env application order in the devstack patch as it was suggested on IRC | 17:18 |
| sean-k-mooney | yep i saw that update in my email (devstack order) | 17:21 |
| sean-k-mooney | i have not had time to test out yoru series locally but ill be deploying a multinode devstack to test some wather chagnes proably in the next week so once i have that up an running ill probaly check ot your schduelr branch | 17:22 |
| gibi | cool | 17:46 |
| gibi | thanks | 17:46 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!