**** BEGIN LOGGING AT Mon May 03 02:59:56 2021 May 03 07:48:05 good morning May 03 14:43:28 denix: I think there already are 20.04 images? May 03 14:45:34 denix: We don't really have plans for other distros besides ubuntu; the idea is to use a sanity tested distro in the container and Ubuntu seems to be one of the best fits. Is there some specific reason you would want something else? May 03 16:01:24 JPEW: ok, "podman search garminpyrex" does show 20.04... I didn't see it earlier for some reason. Does pyrex default to 18.04 now? May 03 16:13:33 denix: Yes, it's defaulting to 18.04 May 03 16:13:45 I should update it now that 20.04 is valid May 03 16:18:12 JPEW: during last happy hour (btw, we missed you!) we were discussing containers and the fact that people are moving from docker to podman. I recently started using podmad on my new builder... May 03 16:20:43 JPEW: so, my new builder has ~2x faster CPU, ~3x faster disk IO and yet, it builds OE ~3x slower in a container than the old bare-metal builder - is that expected? I'd imagine a VM would even be faster... May 03 16:23:43 denix: Podman can have pretty slow I/O May 03 16:23:54 denix: I would be curious if you see the same with docker May 03 16:25:15 denix: https://github.com/garmin/pyrex/pull/61 May 03 16:25:20 JPEW: yeah, I was guessing the same and was wondering if there's a way to fine tune it... May 03 16:26:38 denix: Possibly. IIRC the problem is how it deals with --bind mounts... it might be emulating them with FUSE May 03 16:28:11 JPEW: I believe it does - it installed fuse, actually it replaced standard ubuntu fuse with its own from podman repo May 03 16:29:36 that seems odd vs just using mount binding? May 03 16:30:05 denix: Ya. I know it uses fuse-overlayfs for the main root file system in the container. That's probably slow, but ideally most of the file I/O is in the bind mounted workdir volume. I'm not sure how it deals with the bind mount May 03 16:30:40 smurray: Right; I'm not sure if it can actually do a bind mount. It's a little tricky because it's rootless May 03 16:31:24 Which is why the comparison to docker would be helpful since I *know* docker does the bind mount May 03 16:34:06 JPEW: ah, I'd figured there'd still be root involved in there somewhere May 03 16:35:53 smurray: you can "sudo podman" if you want (don't know if it'd change from fuse to bind then) but normally it's daemonless and rootless May 03 16:38:51 it is one of the selling points though - anyone can run podman and don't need root or a daemon and it's more secure, as compromised user account won't gain root May 03 16:39:51 I knew it was daemon-less, hadn't realized rootless was fully supported and not involving hackery May 03 16:47:27 * JaMa was using unprivileged LXC few years ago (before docker became so popular) and the performance was close to bare-metal May 03 16:54:18 JaMa: interesting. do you know what it was using on the backend? May 03 16:56:05 from my notes it looks like just bind (but don't know how it was implemented internally) lxc.mount.entry = /mnt/mirror mnt/mirror none bind,create=dir May 03 16:57:07 https://stgraber.org/2013/12/20/lxc-1-0-blog-post-series/ was what I was reading back then May 03 16:58:41 it was building images with dizzy in less than 2 hours :) May 03 16:58:58 good old days May 03 17:06:53 heh :) May 03 18:28:58 JPEW, smurray: came across this - https://news.ycombinator.com/item?id=25169423 May 03 18:30:53 denix: Ya May 03 18:31:51 I had to stop using podman because of the fd limit (although pyrex will automatically bump up the limit to the max hardlimit before running podman) May 03 18:33:16 Although sufficiently new fuse-overlayfs will do this also: https://github.com/containers/fuse-overlayfs/commit/d3729baa932bae444586ce9343bae59147ab2efb May 03 18:34:29 yeah, I haven't had fd limit issue May 03 18:34:53 but it is still slow and that post kind of explains the reasons **** ENDING LOGGING AT Tue May 04 02:59:57 2021