| rinigus | vlagged: on my pc, qtwebengine compile would probably take about 4-6 hours when done for x86 (native). Qemu for aarch64 was easy more than 24h, maybe more than 48h. So it's rather expensive operation. But as cloud providers now do have arm64, it would make sense to use it | 04:21 |
|---|---|---|
| poetaster | rinigus, there's some discussion among the sailmates about provisioning an obs/build env in part to support you. Can you suggest what kind of a profile we'd need for a dedicated host? | 18:21 |
| poetaster | rinigus, I was thinking that storage would become an issue quickly and was high balling it with a https://www.hetzner.com/de/dedicated-rootserver/ax162-s/configurator/#/ | 18:23 |
| poetaster | rinigus slightly cheaper with the same storage but less oompf: https://www.hetzner.com/de/dedicated-rootserver/ex130-s/configurator/#/ | 18:24 |
| rinigus | poetaster: I would suggest to wait a bit and think what would be the best. getting x86 machines would not help us as we have issues while building on it and Jolla already has such OBS anyway. most of our work is in aarch64 and we can benefit from that. | 18:30 |
| rinigus | poetaster: also, I am not sure you need huge dedicated server. ideally, build system would scale with jobs and then scale down when compiles are done | 18:31 |
| rinigus | which build system supports such up/down scaling I don't know. I tried to look into OBS, but I didn't find such ability when quickly scanning through docs | 18:32 |
| rinigus | for OBS (or any other build system), we need someone who can admin it. I guess we can learn and do it as well in time | 18:34 |
| rinigus | poetaster: so, before we start thinking about renting a server, we should think what solution should we go for. and that requires some research | 18:35 |
| rinigus | maybe our current OBS maintainers know whether OBS can have workers switched on/off based loads? | 18:36 |
| b100dian[m] | So the storage part is because of the stored packages on the main server I would guess? | 19:46 |
| b100dian[m] | If so, hooking community workers under an "add_constraints" would be smarter | 19:49 |
| b100dian[m] | Wait, I had this suggestion in a community meeting back in the 5.15 days https://forum.sailfishos.org/t/community-meeting-on-irc-17th-august-2023/15962/4 | 19:51 |
| Keto | I am not aware of any autoscaling built-in in OBS | 19:52 |
| poetaster | rinigus, clear on the question of scaling on demand, but assuming that package space constraints will only grow. my experience with pay for time on cloud doesn't pan out in the end. | 20:18 |
| poetaster | rinigus, you had a single package building in 4 hours from what I understand? I fail to see how that could be anything like useful in the larger picture? | 20:20 |
| rinigus | Keto: thanks, maybe we should suggest it to OBS Devs. Would be rather useful | 20:22 |
| poetaster | b100dian[m], was there any feedback on that qt5.15 suggestion? | 20:22 |
| rinigus | poetaster: that's extreme package, qt webengine | 20:23 |
| poetaster | rinigus, which asks of us, what is the general case. obs sleeps most of the time, but when it chokes, it chokes everything. | 20:24 |
| poetaster | rinigus, cloud is certainly a way to go, but, if you do it to cover the bases, it's still expensive and at some point inconvinient. I've done bare metal commisioning of opencloud, so I have a bit of experience. | 20:26 |
| poetaster | rinigus, perhaps the focus, aarch64, so limiting the coverage is a key here? | 20:27 |
| b100dian[m] | poetaster: meeting notes are here https://irclogs.sailfishos.org/meetings/sailfishos-meeting/2023/sailfishos-meeting.2023-08-17-07.00.log.html but short answer is that they think about it | 20:28 |
| poetaster | b100dian[m], hmmm. not useful. Ok, drawing board. | 20:28 |
| b100dian[m] | I'm subscribed to get a qualcomm elite dev kit when I get the invitiation. unfortunately Elite has not finished Linux support as we speak. Any other powerful aarch64 you can get from home? Ampere is not an option :) | 20:30 |
| poetaster | b100dian[m], just step in and correct me vis. the sailmates. | 20:30 |
| b100dian[m] | *has no | 20:30 |
| rinigus | poetaster: aarch64 is surely an issue. But with these discussions we have to involve Keto and others from Jolla. They have large experience and know the admin side of it. But for me, I'll be off for tonight | 20:31 |
| poetaster | rinigus, all clear. I'll do the research. | 20:32 |
| poetaster | rinigus, I was speaking about parallel hardware which doesn't have anything to do with jolla directly, but I'd certainly try to coordinate. | 20:34 |
| Keto | our OBS is quite a bit behind the upstream, and I have not checked what has hapenned there, so I don't know if there is some hooks in the worker status that could be used in autoscaling | 21:36 |
| Keto | of course that kind of thing could also be done purely based on the worker load. OBS is quite fine with the workers coming and going, so the scaling can probably be done completely externaly | 21:39 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!