=== chris14_ is now known as chris14 | ||
=== JanC_ is now known as JanC | ||
arighi | ricotz, I think it's a mistake, I don't even see nvidia-530 in lunar | 10:52 |
---|---|---|
arighi | I'll double check | 11:10 |
ricotz | arighi, https://launchpad.net/ubuntu/+source/nvidia-graphics-drivers-530 | 11:11 |
ricotz | I commented on https://bugs.launchpad.net/ubuntu/+source/nvidia-graphics-drivers-525/+bug/2015491 | 11:12 |
arighi | ricotz, yep, I see it in proposed for all the releases, but not in lunar (not yet at least) | 11:12 |
-ubottu:#ubuntu-kernel- Launchpad bug 2015491 in nvidia-graphics-drivers-525 (Ubuntu) "Introduce the 530 (UDA) NVIDIA driver series in Bionic, Focal, Jammy, Kinetic, Lunar" [High, In Progress] | 11:12 | |
arighi | oh ok sorry, you meant in general, I was only checking in lunar | 11:12 |
ricotz | yeah, those packages are broken | 11:13 |
arighi | ricotz, let me ask around :) | 11:13 |
ricotz | arighi, thank you | 11:13 |
tseliot | ricotz, hey, what's up with the driver? | 13:32 |
ricotz | tseliot, hi, see https://bugs.launchpad.net/ubuntu/+source/nvidia-graphics-drivers-525/+bug/2015491 | 13:33 |
-ubottu:#ubuntu-kernel- Launchpad bug 2015491 in nvidia-graphics-drivers-525 (Ubuntu) "Introduce the 530 (UDA) NVIDIA driver series in Bionic, Focal, Jammy, Kinetic, Lunar" [High, In Progress] | 13:33 | |
ricotz | tseliot, it seems bad to copy packages which ftbfs already in a PPA to the archive | 13:34 |
tseliot | Yes, I must've missed that. Only Lunar and bionic, apparently | 13:36 |
ricotz | tseliot, all of them, since i386 isn't picked up on the other series yet | 13:37 |
tseliot | Oh, I see. I thought you already had it in the PPA, and that 530 would be already whitelisted | 13:38 |
ricotz | I have, but it seems it only enabled lunar :\ | 13:38 |
tseliot | oh | 13:40 |
tseliot | Well, let me fix that build dep here | 13:40 |
ricotz | tseliot, there is another change for the PPA build, which seems to got fixed in debhelper | 13:41 |
ricotz | tseliot, https://paste.debian.net/plainh/b8c2b72c | 13:41 |
tseliot | So that it doesn't try it on all the packages? | 13:43 |
ricotz | yes, it failed a couple of days ago, apparently not on your side, so seems to got fixed | 13:44 |
ahasenack | rbasak: hi, is there a way to somehow highlight the "main" branch of development in a package? This list is very confusing to newcomers: https://code.launchpad.net/ubuntu/+source/nss | 14:05 |
ahasenack | 320 results! | 14:05 |
ahasenack | I suppose this needs some new launchpad feature? | 14:06 |
ahasenack | ideally we would have "main" (or "devel"), then one branch per release, and the rest is just details (foo-security, foo-proposed, applied and not applied, etc) | 14:07 |
ahasenack | rbasak: do you think we can let thunderbird be imported again? It's currently in the denylist, under "# uses up too much disk" | 14:23 |
ahasenack | and, hm, openssh is in the denylist too? Just saw it, a few lines below. But we are importing it. Are these lists still being used? | 14:24 |
ahasenack | but thunderbird is indeed not imported (or the default branch wasn't flipped): https://code.launchpad.net/ubuntu/+source/thunderbird | 14:24 |
tseliot | ricotz, err... zstd is not in main, at least in bionic and focal. xnox , is there anything we can do about this? | 14:25 |
ricotz | tseliot, it is a build-dep which fine | 14:25 |
tseliot | Yes, I know it's the actual dependencies that are not allowed, but still | 14:26 |
rbasak | ahasenack: interesting choice of channel? | 14:48 |
ahasenack | rbasak: bah | 14:48 |
ahasenack | off-by-one error | 14:48 |
ahasenack | ubuntu-devel is just above | 14:48 |
ghavil | Howdy, I opened https://bugs.launchpad.net/ubuntu/+source/linux-aws/+bug/2016186 yesterday and was wondering if anyone here had advice as to how I could debug further. I'm not sure how actual disk I/O would translate into these cgroup files, maybe I just need to go understand that side in more depth first. Curious if folks have ideas | 14:53 |
-ubottu:#ubuntu-kernel- Launchpad bug 2016186 in linux-aws (Ubuntu) "5.19 not reporting cgroups v1 blkio.throttle.io_serviced" [Undecided, New] | 14:53 | |
arighi | ghavil, hi, it really seems to be a cgroupv1 vs cgroupv2 issue, I think in kinetic with 5.19 we completely moved to cgroupv2, so you are not supposed to be able to use cgroupv1 anymore unless you tweak the kernel boot options/systems to enable the hybrid cgroupv1+cgroupv2 way | 15:23 |
ghavil | Hmmm oh boy that's what I'm worried about | 15:24 |
arighi | you can `grep cgroup /proc/mounts` to see the version of cgroup that is mounted | 15:24 |
arighi | if it says cgroup2, well, you're using version 2 | 15:24 |
ghavil | These are for sure cgroups v1 configured, I was trying to get us to v2 but it's breaking a lot of stuff along the way | 15:25 |
arighi | have you seen this? https://docs.kernel.org/admin-guide/cgroup-v2.html#io-interface-files | 15:25 |
arighi | v2 has a slightly different ABI for blkio, but you should be able to enforce the same IO limitations and get the same stats using cgroupv2 | 15:26 |
ghavil | Yep, our issue w/ cgroups v2 is much more nuanced at the moment. Moving to it isn't an option for now | 15:26 |
ghavil | Our kernels are configured with 'systemd.unified_cgroup_hierarchy=0 systemd.legacy_systemd_cgroup_controller=true' for reference | 15:27 |
ghavil | If 5.19 removes cgroups v1 support, I can work on pinning all of our stuff to 5.15 for now. Bummer though as this upgrade seems to be automatically happening for linux-aws packages (I don't think we opted into anything) and this is a Jammy node not Kinetic | 15:28 |
ghavil | i.e. https://packages.ubuntu.com/jammy-updates/linux-aws is showing the bump to 5.19.0.1022.23~22.04.6 and my expectation was that cgroups v1 would continue to work for 22.04 at least | 15:28 |
arighi | ghavil, if you boot with unified_cgroup_hierarchy you should still be able to use cgroupv1, the kernel still supports it | 15:33 |
arighi | I'm not sure if systemd is handling it properly.. I haven't done any test recently with cgroup1 | 15:33 |
arighi | let me do a quick test... | 15:35 |
ghavil | Okay sounds good, I'm been testing as well, I cannot get anything but '0's to appear in the /sys/fs/cgroup/blkio/blkio.throttle.io_serviced files on nodes w/ the cgroups v1 configuration & the 5.19.0.1022.23~22.04.6 kernel running. Downgrading to the 5.15 works and I unfortunately cannot change to cgroups v2 at this time. | 15:36 |
arighi | ghavil, which IO scheduler are you using? (grep . /sys/block/*/queue/scheduler) | 16:04 |
ghavil | '[none] mq-deadline' it looks like | 16:05 |
ghavil | Same for both of my testing 5.15 and 5.19 instances too | 16:06 |
arighi | ghavil, it looks like 'none' in 5.19 doesn't support cgroup blkio anymore (I'm trying to find the commit that changed this behavior) | 16:07 |
arighi | ghavil, if you switch to bfq for example (apt install linux-modules-extra and modprobe bfq) then you can see IO stats in the blkio.* in cgroupfs | 16:07 |
ghavil | Oh interesting | 16:08 |
ghavil | Lemme give that a whirl | 16:08 |
arighi | ghavil, BTW you also need to enable the bfq scheduler for your block devices, like `echo bfq > /sys/block/<dev>/queue/scheduler` | 16:10 |
ghavil | ack | 16:10 |
ghavil | arighi: Dumb question, should I be installing 'linux-modules-extra-aws' or is 'linux-modules-extra' good enough? | 16:26 |
ghavil | AH HA that does work! | 16:29 |
ghavil | I installed linux-modules-extra-aws for the record | 16:29 |
arighi | ghavil, yep modules-extra-aws is fine | 16:32 |
xnox | tseliot: i'm not sure what you are askinga bout zstd. what needs it in main? and if the request is legit we can promote it. | 16:33 |
tseliot | xnox, apparently, nvidia-530 needs it as a build-dep, only on i386. I suppose it should be fine for now. | 16:34 |
arighi | ghavil, however... I see the IO stats only in the blkio.bfq.* files, not in blkio.throttle.*, that means we are likely hitting a behavior change upstream (probably not many people noticed that, because it's cgroup1) | 16:34 |
ghavil | arighi: Ah good catch, yeah I see that too, bummer but this is better than nothing | 16:40 |
xnox | tseliot83: packages in main and restricted, can build-depend on packages in universe/multiverse, if they don't produce runtime Depends on them. | 17:04 |
xnox | tseliot83: i believe this has been true for long enough for you to use this fact everywhere. | 17:04 |
xnox | tseliot83: https://lists.ubuntu.com/archives/ubuntu-devel/2016-March/039274.html | 17:04 |
xnox | i love the title of this email, as it is so easy to find | 17:04 |
tseliot83 | xnox, right, it's good enough for me. Thanks! | 17:05 |
xnox | i think we did it 2016, so all releases you can add Build-Depends: zstd and it will work | 17:05 |
jchittum | ghavil: for pinning to the 5.15 kernel, you can use `linux-aws-lts-22.04` package: https://packages.ubuntu.com/jammy/linux-aws-lts-22.04 | 19:42 |
ghavil | Ah thanks for the tip! | 19:45 |
ghavil | jchittum: Is there a linux-gcp equivalent? I found https://packages.ubuntu.com/jammy/linux-azure-lts-22.04 for Azure | 19:46 |
jchittum | ghavil: based on history there will be. it looks like the gcp kernel hasn't rolled to 5.19 yet. I believe that the Kernel team makes the -lts-$VERSION around rolls. I see linux-gcp-lts-20.04 and linux-gcp-lts-18.04 | 21:35 |
=== halvors1 is now known as halvors |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!