MICROburst | holmanb: The apt: stuff substitutes the variable but offers only one line for source. So you can have only 'deb' or 'deb-src', using tricks like '\n' fail. As a result the stuff like here at rspamd (https://rspamd.com/downloads.html -- debian tab) has to be put in an extra file. Actually apt: is pretty useless. | 05:18 |
---|---|---|
SDes91 | does someone have some example cloud config examples on parititioning disks for QEMU? | 07:30 |
holmanb | MICROburst: I don't see anything there that demands multiline deb src support, so you'll have to be more specific about why you think it's necessary - it doesn't appear to be necessary to me. That all looks doable with apt: | 13:15 |
holmanb | ah, they left nvm | 13:18 |
cjp256 | falcojr: does my last comment on #1667 make more sense? | 19:37 |
falcojr | cjp256: it does. Me and (mostly) blackboxsw have been doing some digging to figure out exactly what's going on and the best way forward | 19:39 |
blackboxsw | cjp256: sorry we are actively discussing this :) | 19:40 |
blackboxsw | and want to hold 22.3 release on getting this included | 19:40 |
blackboxsw | ok so, per your question "can you think of a time when we update metadata but wouldn't want to write it back out?".... the only case I can think of is if/when we wanted to provide the ability to check via cloud-init CLI if metadata was updated.. Like a `cloud-init query --no-cache some_key` | 19:43 |
blackboxsw | falcojr: ^ | 19:43 |
blackboxsw | this was something we talked about a few years ago, but persisting the obj.pkl updates or not due to a Datasource.get_data() call could easily be a param/switch provided to the get_data method that opts to update the cache or not | 19:44 |
blackboxsw | "this was something we talked about" == the feature of potentially surfacing a cloud-init CLI query param to allow for direct datasource queries without using a cache.. so it implies some sort of ephemeral get_data() run without persisting the updated output to JSON or obj.pkl files | 19:45 |
blackboxsw | but both instance-data.json and obj.pkl output would fall into the same boat. you either want to persist all metadata updates to disk via get_data() if the data changes, or avoid it for all files | 19:46 |
falcojr | yeah, for now at least, I think it makes the most sense to keep the pickle writing with the update metadata | 19:47 |
falcojr | basically right below where we write the json | 19:47 |
falcojr | if we need to move it out later, we'll need to move the json writing out too, but I think it makes more sense to keep them together | 19:48 |
cjp256 | makes sense | 19:48 |
blackboxsw | yeah works for mw | 19:48 |
blackboxsw | yeah works for me | 19:48 |
blackboxsw | falcojr: I'm cobbling up a 2nd quick patch suggestion for this Azure obj.pkl cache handling. ~5 mins | 20:02 |
falcojr | blackboxsw: sounds good, thanks! | 20:03 |
blackboxsw | 5 mins, 500 mins what's the big diff. unraveling circular imports | 20:24 |
blackboxsw | ok PR up for review falcojr cjp256 holmanb I've migrated pkl_load/pkl_store into datsource/__init__ because it is really only applicable there. https://github.com/canonical/cloud-init/pull/1669 | 21:00 |
ubottu | Pull 1669 in canonical/cloud-init "sources: obj.pkl cache should be written anyime get_data is run" [Open] | 21:00 |
blackboxsw | I'm going through testing now on a live Azure system | 21:00 |
blackboxsw | while adding removing NICs | 21:00 |
cjp256 | thanks :) | 21:19 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!