Tuesday, 2021-04-13

*** th0din has joined #kata-dev00:02
*** fuentess has quit IRC02:49
*** egernst has quit IRC05:08
*** ailan_ has joined #kata-dev06:20
*** fgiudici has joined #kata-dev06:33
*** snir_ has joined #kata-dev06:35
*** egernst has joined #kata-dev06:43
*** ailan_ has quit IRC06:44
*** egernst has quit IRC06:48
*** amorenoz_ has joined #kata-dev06:57
*** fgiudici has quit IRC06:59
*** pcaruana has quit IRC07:11
*** egernst has joined #kata-dev07:12
*** fgiudici has joined #kata-dev07:12
*** sgarzare has joined #kata-dev07:14
*** egernst has quit IRC07:18
*** jodh has joined #kata-dev07:24
*** pcaruana has joined #kata-dev07:40
*** dklyle has quit IRC07:50
*** davidgiluk has joined #kata-dev07:59
*** sameo has joined #kata-dev08:09
*** ailan_ has joined #kata-dev08:29
*** egernst has joined #kata-dev09:15
*** egernst has quit IRC09:19
*** snir_ has quit IRC11:05
*** snir has joined #kata-dev11:05
*** egernst has joined #kata-dev11:16
*** egernst has quit IRC11:22
*** fuentess has joined #kata-dev12:20
*** crobinso has joined #kata-dev12:23
*** devimc has joined #kata-dev12:35
*** jodh has quit IRC12:39
*** snir has quit IRC13:05
kata-irc-bot<fidencio> Friendly reminder that the Architecture Committee meeting is in *~90 minutes* from the time this message was sent, at 3PM UTC.13:29
*** egernst has joined #kata-dev13:42
*** egernst has quit IRC13:48
*** egernst has joined #kata-dev13:49
*** egernst has quit IRC14:13
*** egernst has joined #kata-dev14:25
*** ailan__ has joined #kata-dev14:43
*** ailan_ has quit IRC14:46
*** dklyle has joined #kata-dev15:09
*** snir has joined #kata-dev15:55
*** crobinso has quit IRC15:59
kata-irc-bot<eric.ernst> @christophe -- did you end up experimenting with the shared pages in same memory cgroup?16:12
kata-irc-bot<eric.ernst> ie, are they counted once if both processes which share the pages are within that cgroup?16:13
*** Yarboa has quit IRC16:14
*** Yarboa has joined #kata-dev16:17
*** fgiudici has quit IRC16:33
*** crobinso has joined #kata-dev16:43
*** crobinso has quit IRC17:07
*** sgarzare has quit IRC17:35
*** egernst_ has joined #kata-dev17:43
*** egernst has quit IRC17:46
*** devimc has quit IRC18:17
*** davidgiluk has quit IRC19:13
*** amorenoz__ has joined #kata-dev19:33
*** amorenoz_ has quit IRC19:36
kata-irc-bot<eric.ernst> virtio balloon :thread:. @jose.carlos.venegas.m I recall you had some fun playing with this before.20:29
kata-irc-bot<eric.ernst> What was your final lessons/learnings with this?20:29
kata-irc-bot<eric.ernst> Did you play with it in context of kata runtime itself, or was this just with the VMM directlyr.20:30
kata-irc-bot<jose.carlos.venegas.m> I have used with kata-runtime, tested with docker,  there was some concerns using it with VFIO devices for cases where the ballon driver zap pages used by devices20:32
kata-irc-bot<jose.carlos.venegas.m> @eric.ernst also some time ago got merged support for virtio-mem , I am not sure how well it work today20:34
kata-irc-bot<eric.ernst> I see; yeah with VFIO and page pinning, thing are a bit different, but I see those as being pretty different use cases likely.20:35
kata-irc-bot<eric.ernst> Did/do you have a branch from this work?20:35
kata-irc-bot<eric.ernst> Or any gist/docs summarizing your findings?20:35
kata-irc-bot<eric.ernst> I'm curious to try this again.20:35
kata-irc-bot<jose.carlos.venegas.m> I dont have any docs. Let me search the old  PR about it20:36
kata-irc-bot<jose.carlos.venegas.m> https://github.com/kata-containers/runtime/pull/79320:38
kata-irc-bot<jose.carlos.venegas.m> @eric.ernst any specific use case you are looking for? probably would worth try virtio-mem ?20:39
kata-irc-bot<eric.ernst> I'm just looking at long running pods, and reducing amount of memory used be caches20:40
kata-irc-bot<eric.ernst> in the guest.20:40
kata-irc-bot<eric.ernst> So, less about event when the container resources change, but more about putting pressure on the guest kernel to reduce memory usage20:41
kata-irc-bot<eric.ernst> Guest VM will use all the memory it can for a pod that is active on filesystem.20:41
kata-irc-bot<eric.ernst> since it'll cache every bit it can.20:41
kata-irc-bot<jose.carlos.venegas.m> and how would be the event to  reduce the guest pages?20:46
kata-irc-bot<eric.ernst> time?20:46
kata-irc-bot<jose.carlos.venegas.m> monitor by the agent if is idle and then ask to reduce memory ?20:46
kata-irc-bot<eric.ernst> tbh, I haven't played much with virtio-balloon.20:47
kata-irc-bot<jose.carlos.venegas.m> ok, the problem with ballon or virtio-mem is that there is a request to the VMM to reduce the memory the guest see,  not just mark pages to give back to the host20:49
kata-irc-bot<jose.carlos.venegas.m> so it is needed to ask to the VMM what is the new memory size that the guest can see20:49
kata-irc-bot<jose.carlos.venegas.m> I think better want to restrict the cache size of the guest kernel + mark pages to be freed by the guest  ?20:50
kata-irc-bot<jose.carlos.venegas.m> about cache, you are talking about cache pages from the filesystem ?20:52
kata-irc-bot<eric.ernst> Yeah, got it.20:53
kata-irc-bot<eric.ernst> Really, I want the guest to see what the pod limits are + a fixed amount that I expect the guest services to require (ie, agent)20:54
kata-irc-bot<eric.ernst> So, once the VMM starts and the containers are all setup, I'd want to ask the VMM to restrict the amount of memory it sees.20:54
kata-irc-bot<eric.ernst> Without this, the guest will keep its page cache, dentry cache until memory is exhausted.20:54
*** amorenoz__ has quit IRC21:20
*** sameo has quit IRC21:34
*** devimc has joined #kata-dev22:17
*** fuentess has quit IRC23:27
*** devimc has quit IRC23:29

Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!