| *** igordc has quit IRC | 00:02 | |
| *** sameo has quit IRC | 01:06 | |
| *** igordc has joined #kata-general | 03:30 | |
| *** igordc has quit IRC | 03:54 | |
| *** pvdp has joined #kata-general | 05:34 | |
| *** sgarzare has joined #kata-general | 07:55 | |
| *** sameo has joined #kata-general | 08:27 | |
| *** tmhoang has joined #kata-general | 08:36 | |
| *** gwhaley has joined #kata-general | 09:05 | |
| *** gwhaley has quit IRC | 12:08 | |
| *** pvdp has quit IRC | 12:56 | |
| *** gwhaley has joined #kata-general | 13:15 | |
| *** lcastell has joined #kata-general | 16:06 | |
| *** lcastell has quit IRC | 16:08 | |
| *** tmhoang has quit IRC | 16:41 | |
| *** sameo has quit IRC | 16:55 | |
| *** igordc has joined #kata-general | 17:13 | |
| *** igordc has quit IRC | 17:13 | |
| *** igordc has joined #kata-general | 17:13 | |
| *** sgarzare has quit IRC | 17:19 | |
| kata-irc-bot | <matt> To clarify the memory usage note in the docs, If I deploy a Kata container that is configured for 2G RAM, its going to immediately use 2G of host memory, there is no concept of overcommit. Correct? | 19:17 |
|---|---|---|
| kata-irc-bot | <graham.whaley> @manohar.r.castelino @xu ^^ can you comment authoritatively on this? | 19:23 |
| kata-irc-bot | <graham.whaley> I think, @matt, that the overcommit rules will be whatever your host is set up for. for instance, my hosts have always have overcommit disabled by default, so once you 'run out of allocated space', then KVM will refuse to launch another QEMU. | 19:24 |
| kata-irc-bot | <graham.whaley> I also think we have to be careful about the word 'use' maybe - I suspect there is effectively 'pre-allocate', and then 'use'. That is, I think it will pre-allocate 2G of RAM to the VM, but not actually use the physical memory to back that (unless you have enabled the pre-back option I think in the config....) until the container consumes it... | 19:25 |
| kata-irc-bot | <matt> From a basic POV, if I run pmap or what-not, I will see that 2G is allocated it seems | 19:27 |
| kata-irc-bot | <matt> From a using Kata POV, if I have 4 Kata containers allocating 2G each, running a busybox container, I could run **approximately** 4 of them on a host with 8G RAM (and no swap). But I could run more than 4 **plain** busybox docker containers because they wouldn't be allocating 2GB to spin up. -- Is that an accurate statement? | 19:29 |
| *** gwhaley has quit IRC | 19:31 | |
| kata-irc-bot | <eric.ernst> @graham.whaley - right -- that memory pre-allocated. | 21:37 |
| kata-irc-bot | <eric.ernst> So, because you run those 4 Kata Containers does not mean 8GB of physical memory is being used on the host. | 21:38 |
| kata-irc-bot | <matt> Right, 8GB allocated but probably not all used to just run busybox | 22:06 |
| kata-irc-bot | <matt> I guess the important question is would i be able to launch say, 8 Kata Containers in my scenario or would I effectively be limited to 4. | 22:07 |
| kata-irc-bot | <eric.ernst> more than 4 for sure | 22:15 |
| kata-irc-bot | <eric.ernst> we have some tests for measuring density that you can use. | 22:16 |
| kata-irc-bot | <matt> I'm going to restart my GCP instance and give it a quick test | 22:16 |
| kata-irc-bot | <eric.ernst> see https://github.com/kata-containers/tests/tree/master/metrics#density | 22:16 |
| kata-irc-bot | <matt> Should the delta in number of containers roughly be related to the VM overhead? In otherwords maybe the pre-allocation doesn't matter for density? | 22:17 |
| kata-irc-bot | <eric.ernst> I'd recommend just running the metrics tests | 22:17 |
| kata-irc-bot | <eric.ernst> yep. | 22:17 |
| kata-irc-bot | <eric.ernst> its just the active overhead | 22:17 |
| kata-irc-bot | <eric.ernst> better link: https://github.com/kata-containers/tests/tree/master/metrics/density | 22:17 |
| kata-irc-bot | <matt> disregarding the overhead, if I have a bunch of containers actively 2GB of RAM on an 8GB system, I will have issues with or without kata then | 22:18 |
| kata-irc-bot | <matt> thanks for talking this through | 22:19 |
| kata-irc-bot | <eric.ernst> np | 22:20 |
| kata-irc-bot | <eric.ernst> btw, make sure your gcp image has nested virt enabled | 22:20 |
| kata-irc-bot | <matt> yep it does | 22:20 |
| kata-irc-bot | <eric.ernst> https://github.com/kata-containers/documentation/blob/master/install/gce-installation-guide.md | 22:21 |
| kata-irc-bot | <eric.ernst> ok | 22:21 |
| kata-irc-bot | <matt> I've been playing around with some kata stuff on it but I power it off to save some cash | 22:21 |
| kata-irc-bot | <eric.ernst> :slightly_smiling_face: | 22:21 |
| kata-irc-bot | <matt> I think you could add a paragraph under `Memory` here: https://github.com/kata-containers/documentation/wiki/UserGuide - I would but I dont seem to have edit rights | 22:21 |
| kata-irc-bot | <matt> There's a memory allocation paragraph and probably should add a memory usage one also | 22:22 |
| *** eniac_petrov has joined #kata-general | 22:25 | |
| eniac_petrov | hey, anyone online? I'm curious about something | 22:25 |
| eniac_petrov | Kata is using namespaces inside the VM, right? | 22:25 |
| kata-irc-bot | <matt> @eric.ernst 16G host, I'm running 10 nginx kata containers and only really using 2.4G of memory. will also try the density stuff later | 22:27 |
| kata-irc-bot | <dangot> Hey all quick update on our JAVA containers not starting. We’ve fixed it by building our own ubuntu image and adding : `echo >/etc/security/limits.d/systemd.conf "* hard nofile 1048576"` | 23:33 |
| kata-irc-bot | <dangot> https://github.com/systemd/systemd/issues/10921 | 23:33 |
| *** eniac_petrov has quit IRC | 23:41 | |
| *** eniac_petrov has joined #kata-general | 23:42 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!