| kata-irc-bot | <aprice> Speaking of QEMU 5.0, I ran into an issue recently with 9pfs and QEMU 5.0, likely to be fixed in 5.1 - https://github.com/kata-containers/runtime/issues/2978 | 00:06 |
|---|---|---|
| kata-irc-bot | <archana.m.shinde> @aprice Thats a strange error. We'll work on updating to 5.1 and see if that helps. | 01:12 |
| kata-irc-bot | <aprice> Thank you! | 01:12 |
| *** egernst has joined #kata-dev | 01:19 | |
| *** ajin has joined #kata-dev | 01:47 | |
| *** egernst has quit IRC | 01:54 | |
| *** egernst_ has joined #kata-dev | 01:56 | |
| *** egernst__ has joined #kata-dev | 02:00 | |
| *** egernst_ has quit IRC | 02:01 | |
| *** egernst__ has quit IRC | 02:05 | |
| *** egernst has joined #kata-dev | 02:35 | |
| *** egernst_ has joined #kata-dev | 02:38 | |
| *** egernst has quit IRC | 02:39 | |
| *** egernst_ has quit IRC | 02:43 | |
| *** egernst has joined #kata-dev | 03:15 | |
| *** sameo has joined #kata-dev | 04:30 | |
| *** sameo has quit IRC | 05:42 | |
| *** fgiudici has joined #kata-dev | 06:06 | |
| *** dklyle has quit IRC | 06:48 | |
| *** jodh has joined #kata-dev | 07:02 | |
| *** sameo has joined #kata-dev | 07:03 | |
| *** davidgiluk has joined #kata-dev | 08:05 | |
| *** pcaruana has quit IRC | 08:35 | |
| *** pcaruana has joined #kata-dev | 08:45 | |
| *** sgarzare has joined #kata-dev | 08:46 | |
| *** egernst has quit IRC | 09:05 | |
| *** devimc has joined #kata-dev | 12:25 | |
| *** crobinso has joined #kata-dev | 13:00 | |
| *** pcaruana has quit IRC | 13:16 | |
| *** fuentess has joined #kata-dev | 13:42 | |
| *** devimc has quit IRC | 14:33 | |
| *** devimc has joined #kata-dev | 14:36 | |
| *** dklyle has joined #kata-dev | 14:37 | |
| *** fuentess has quit IRC | 14:53 | |
| *** fuentess has joined #kata-dev | 14:59 | |
| *** devimc has quit IRC | 15:33 | |
| *** egernst has joined #kata-dev | 15:33 | |
| *** devimc has joined #kata-dev | 15:35 | |
| *** devimc has quit IRC | 15:39 | |
| *** devimc has joined #kata-dev | 15:39 | |
| kata-irc-bot | <eric.ernst> @gabriela.cervantes.te how's the metrics CI hunt going? | 15:45 |
| kata-irc-bot | <eric.ernst> We're looking @ boot-time, right? | 15:46 |
| devimc | mem footprint - and we blame you https://github.com/kata-containers/runtime/issues/2983#issuecomment-698418573 - haha :D | 15:47 |
| kata-irc-bot | <eric.ernst> Good find! | 15:49 |
| kata-irc-bot | <eric.ernst> I saw a lot of boot-time failures, though, too? | 15:49 |
| devimc | boot-time? nop | 15:51 |
| kata-irc-bot | <eric.ernst> Yeah, you're right. OK. Well... I don't think we're planning on removing virtio-fs? I think we're overdue on shrinking/address footprint at large, bu.... | 15:51 |
| kata-irc-bot | <eric.ernst> We either need to merge a PR which reduces are footprint, or update the metrics watermark :slightly_smiling_face: | 15:52 |
| davidgiluk | if your memory footprint is including the virtiofsd you can squash it down a bit by passing the flag to reduce the thread-pool-size | 16:32 |
| kata-irc-bot | <wmoschet> @eric.ernst somehow related...I'm wondering if there is a way to implement a waiver schema for situations like that. I mean, this metric job is blocking PRs from getting merged, PRs that won't affect the metric assessments at all, maybe we could have a way to temporarily disabling the job or disregard (waiver) the result | 16:33 |
| kata-irc-bot | <eric.ernst> The failing PR is just enabling the kernel config. | 16:42 |
| kata-irc-bot | <eric.ernst> I *think* we aren't actually using virtio-fs in that metrics test | 16:43 |
| kata-irc-bot | <eric.ernst> devimc, @gabriela.cervantes.te can you confirm? | 16:43 |
| devimc | afaik we are using it in the clh metrics ci | 16:44 |
| kata-irc-bot | <eric.ernst> which is being used for the CI? | 16:47 |
| *** sgarzare has quit IRC | 16:55 | |
| *** fgiudici has quit IRC | 16:59 | |
| kata-irc-bot | <archana.m.shinde> @eric.ernst The default kernel is built with the virtiofs configs now | 17:05 |
| kata-irc-bot | <archana.m.shinde> that increases the kernel size in memory | 17:06 |
| kata-irc-bot | <archana.m.shinde> remember those arent modules | 17:06 |
| kata-irc-bot | <eric.ernst> yep | 17:09 |
| kata-irc-bot | <eric.ernst> Yeah. IMO, virtio-fs should probably be default over virtio-9p, or at least this is the direction we are going in. | 17:10 |
| kata-irc-bot | <eric.ernst> Based on that, I don't think removing it makes a lot of sense. My point re: it being used in the test or not is whether the memory footprint is due to the thread-pool size (that shouldn't be allocated unless we are using the driver?) | 17:10 |
| *** jodh has quit IRC | 17:11 | |
| *** egernst has quit IRC | 17:28 | |
| *** egernst has joined #kata-dev | 17:30 | |
| *** egernst has quit IRC | 17:30 | |
| *** egernst has joined #kata-dev | 17:31 | |
| *** egernst has quit IRC | 17:35 | |
| *** egernst has joined #kata-dev | 17:45 | |
| *** snir_ has quit IRC | 17:47 | |
| *** snir_ has joined #kata-dev | 17:48 | |
| kata-irc-bot | <archana.m.shinde> Yes, agree it should not be removed @gabriela.cervantes.te is going to increase the memory limits to accomodate the change | 17:50 |
| *** egernst has quit IRC | 17:55 | |
| *** egernst has joined #kata-dev | 17:55 | |
| *** davidgiluk has quit IRC | 19:06 | |
| *** egernst_ has joined #kata-dev | 19:18 | |
| *** egernst__ has joined #kata-dev | 19:19 | |
| *** egernst has quit IRC | 19:21 | |
| *** egernst_ has quit IRC | 19:22 | |
| *** devimc has quit IRC | 19:26 | |
| *** egernst__ has quit IRC | 19:47 | |
| *** egernst has joined #kata-dev | 19:55 | |
| *** egernst has quit IRC | 20:00 | |
| kata-irc-bot | <gabriela.cervantes.te> I did that two weeks ago https://github.com/kata-containers/tests/pull/2840 | 20:02 |
| *** egernst has joined #kata-dev | 20:07 | |
| *** egernst has quit IRC | 20:09 | |
| *** egernst has joined #kata-dev | 20:09 | |
| kata-irc-bot | <eric.ernst> Thanks @gabriela.cervantes.te -- I wonder if we should increase he midvalue rather than the range of acceptable values? | 20:28 |
| kata-irc-bot | <eric.ernst> WDYT? | 20:28 |
| kata-irc-bot | <gabriela.cervantes.te> for that somebody will need to run the ci metrics multiple times and get the proper midvalue | 20:29 |
| kata-irc-bot | <gabriela.cervantes.te> I don't have the cycles for that but if some volunteers for me it is fine | 20:30 |
| *** devimc has joined #kata-dev | 20:31 | |
| kata-irc-bot | <eric.ernst> devimc: I noticed that if you specify default memory size for VM/SBS to < 256, it goes to the default 2048. | 20:33 |
| kata-irc-bot | <eric.ernst> Have you seen this? | 20:33 |
| devimc | @eric.ernst nop, but that could be a bug | 20:34 |
| kata-irc-bot | <eric.ernst> ``` // MinHypervisorMemory is the minimum memory required for a VM. MinHypervisorMemory = 256``` | 20:34 |
| kata-irc-bot | <eric.ernst> I (git) blame Archana. | 20:36 |
| kata-irc-bot | <eric.ernst> imo it should match ~PodOverhead value we specify, could be ~128MB | 20:37 |
| kata-irc-bot | <eric.ernst> assuming that kube sets an appropriate default memory size for each container/pod | 20:37 |
| kata-irc-bot | <eric.ernst> @archana.m.shinde wdyt? | 20:37 |
| kata-irc-bot | <archana.m.shinde> why blame me @eric.ernst :slightly_smiling_face: | 20:43 |
| kata-irc-bot | <archana.m.shinde> you are proposing to change from 2048 to 128mb? | 20:43 |
| kata-irc-bot | <archana.m.shinde> we also want to make sure that value is sufficient to run a container in a non-k8s context as well | 20:44 |
| kata-irc-bot | <eric.ernst> https://github.com/kata-containers/runtime/issues/2987 | 20:44 |
| kata-irc-bot | <eric.ernst> Default value is fine enough for me; I can change that. | 20:45 |
| kata-irc-bot | <eric.ernst> The problem is that there's an enforcement to make sure that the annotated value from toml >= 256MB | 20:45 |
| kata-irc-bot | <eric.ernst> I don't think we should check for this. Assume that users can set it appropriately (which, if pods /containers are well specified and overhead is utilized, it is feasible to make this ~120MB, or lower as we improve the footprint numbers) | 20:46 |
| kata-irc-bot | <archana.m.shinde> yeah, I am fine with removing it from the annotations | 20:47 |
| kata-irc-bot | <eric.ernst> @jose.carlos.venegas.m @chen.bo: question on CLH... :thread: | 20:47 |
| kata-irc-bot | <eric.ernst> Do we support a notion of memory prealloc? | 20:48 |
| kata-irc-bot | <eric.ernst> well, the check is just in the code. I'll remove it. | 20:48 |
| kata-irc-bot | <eric.ernst> Or, just move it to a much smaller number | 20:51 |
| kata-irc-bot | <eric.ernst> I guess it was 8MB before... | 20:51 |
| kata-irc-bot | <eric.ernst> Change/fix pushed. | 21:14 |
| kata-irc-bot | <eric.ernst> https://github.com/kata-containers/runtime/pull/2988 | 21:14 |
| *** devimc has quit IRC | 21:28 | |
| kata-irc-bot | <chen.bo> hi @eric.ernst, I was catching the context a bit with your message and links above. If I understand you correctly, the `memory prealloc` is the functionality that qemu has with `--mem-path` (`--mem-prealloc`)? | 21:31 |
| kata-irc-bot | <eric.ernst> For the memory object, setting prealloc=on | 21:34 |
| kata-irc-bot | <eric.ernst> ie, resulting in: -`object memory-backend-file,id=dimm1,size=128M,mem-path=/dev/shm,share=on,prealloc=on` | 21:34 |
| kata-irc-bot | <chen.bo> What's the effect of enabling `prealloc` ? | 21:49 |
| kata-irc-bot | <chen.bo> Sorry for the delay, I got into a meeting, and was talking a bit.. | 21:49 |
| kata-irc-bot | <chen.bo> ```--memory <memory> Memory parameters "size=<guest_memory_size>,mergeable=on|off,shared=on|off,hugepages=on|off,hotplug_method=acpi|virtio- mem,hotplug_size=<hotpluggable_memory_size>" [default: size=512M] --memory-zone <memory-zone> User defined memory zone parameters | 21:51 |
| kata-irc-bot | "size=<guest_memory_region_size>,file=<backing_file>,shared=on|off,hugepages=on|off,host_numa_node=<node_id>,id=<zone_identifier>"``` | 21:51 |
| kata-irc-bot | <chen.bo> This is what is supported in clh now. `--memory` is the common use case, where with `shared=on` the memory will be backup by filed in `/dev/shm` automatically. | 21:52 |
| kata-irc-bot | <chen.bo> `--memory-zone` is recently added as clh v0.10, and it allows users to specify `complex` memory configurations where each `memory zone` can be backed by different files or just normal guest memory. | 21:53 |
| kata-irc-bot | <eric.ernst> got it. I think zone --> numa topology stuff | 21:54 |
| kata-irc-bot | <chen.bo> Right. | 21:54 |
| kata-irc-bot | <eric.ernst> i'm not sure how strong of a use case it'll be, but i'm curious about mem prealloc | 21:55 |
| kata-irc-bot | <chen.bo> So, looks like `prealloc` requires special handling, which is not a part of clh (at least from its cli). | 21:55 |
| kata-irc-bot | <chen.bo> Wondering, what's the purpose/outcome of using `prealloc==on` ? | 21:56 |
| *** sameo has quit IRC | 22:55 | |
| *** sameo has joined #kata-dev | 22:58 | |
| *** sameo has quit IRC | 23:37 | |
| *** fuentess has quit IRC | 23:50 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!