*** lcastell has joined #kata-general | 00:51 | |
*** lcastell has quit IRC | 00:55 | |
*** igordc has quit IRC | 05:30 | |
kata-irc-bot | <dangot> Hey guys, what can we do to improve disk throughput for containers running in Kata ? ``` root@toolbox:/# sync; dd if=/dev/zero of=tempfile bs=1M count=8192; sync 8192+0 records in 8192+0 records out 8589934592 bytes (8.6 GB, 8.0 GiB) copied, 123.211 s, 69.7 MB/s ``` | 06:26 |
---|---|---|
kata-irc-bot | <dangot> On the same box running `runc` I get about 1GB … | 06:28 |
kata-irc-bot | <dangot> Looks like 9p problem ? | 06:28 |
*** sgarzare has joined #kata-general | 07:51 | |
*** tmhoang has joined #kata-general | 07:57 | |
*** gwhaley has joined #kata-general | 09:08 | |
*** sameo has joined #kata-general | 09:35 | |
kata-irc-bot | <graham.whaley> @dangot Hi. The normal first step is to set up your graph driver to be block based (devicemapper), and then kata can mount the block device directly instead of going over 9p. There is also a large effort going on right now to get the new virtio-fs enabled in kata (whilst we wait for all the bits to land in all the relevant upstreams like qemu and the kernel), which also gives much better performance than 9p. | 09:45 |
kata-irc-bot | <graham.whaley> @david532 - ah, is the joinin-existing-vm-network limitation a blocker in your use case? | 09:45 |
kata-irc-bot | <dangot> Thanks @graham.whaley - unfortunately containerd doesn't support devicemapper. We might have to go CRIO route. or wait for virtiofs, any rough ETA on that ? | 09:55 |
kata-irc-bot | <graham.whaley> @sebastien.boeuf @xu ^^ do you have a predicted ETA for landing virio-fs in a release? @dangot I think it is RSN - there is an active push right now to get it merged in... | 10:05 |
kata-irc-bot | <eric.ernst> FYI - we are working on a LVM/DM based snapshotter for containerd. There is a PR from folks at AWS to get this landed. I am not sure ETA for this being available, however. | 11:52 |
kata-irc-bot | <sebastien.boeuf> @graham.whaley the current target for the early support of virtio-fs with Kata is 1.7 (mid May) | 14:00 |
kata-irc-bot | <graham.whaley> @dangot ^^ fyi | 14:04 |
*** irclogbot_0 has quit IRC | 14:09 | |
*** irclogbot_0 has joined #kata-general | 14:11 | |
*** irclogbot_0 has quit IRC | 14:25 | |
*** irclogbot_0 has joined #kata-general | 14:27 | |
*** tmhoang has quit IRC | 14:33 | |
*** tmhoang has joined #kata-general | 15:13 | |
*** lpetrut has joined #kata-general | 15:14 | |
*** igordc has joined #kata-general | 15:23 | |
*** irclogbot_0 has quit IRC | 15:36 | |
*** irclogbot_0 has joined #kata-general | 15:38 | |
*** irclogbot_0 has quit IRC | 15:49 | |
*** irclogbot_0 has joined #kata-general | 15:50 | |
*** irclogbot_0 has quit IRC | 15:52 | |
*** irclogbot_0 has joined #kata-general | 15:55 | |
*** lpetrut has quit IRC | 16:44 | |
kata-irc-bot | <david532> @graham.whaley In this case, it does become a bit of a blocker due to my use of Traefik (in a non-Kube/non-Swarm) environment. My second theory in thinking is creating a sort of wrapper to query kata-networks for the interface IP and adding them "programatically" to Traefik's backend. | 16:58 |
kata-irc-bot | <david532> I suppose additionally, I could run Consul for internal DNS and resolve some of these issues. | 16:58 |
kata-irc-bot | <graham.whaley> @manohar.r.castelino @archana.m.shinde for any thoughts here ^^ | 17:00 |
*** lpetrut has joined #kata-general | 17:04 | |
kata-irc-bot | <manohar.r.castelino> Using consul for internal DNS should resolve the primary issue we face with custom networks | 17:08 |
kata-irc-bot | <manohar.r.castelino> but not with docker-shim | 17:11 |
kata-irc-bot | <david532> The networking joining is more for a convenience sake than it is a hard requirement, so I guess I'll plug away at it some more and see if I can't get my setup working. | 17:17 |
*** igordc has quit IRC | 17:19 | |
kata-irc-bot | <greg.bock> Re: virtio-fs, does anyone have a test setup on 1.5 working? I've ported the 1.4 patches to 1.5 but I've yet to get a container started under docker with it. | 17:26 |
kata-irc-bot | <graham.whaley> @sebastien.boeuf @stefanha ^^ re: virtio-fs | 17:27 |
kata-irc-bot | <sebastien.boeuf> nothing on my end. I'm using virtio-fs directly with QEMU, but I haven't applied to Kata yet | 17:29 |
*** lpetrut has quit IRC | 17:30 | |
kata-irc-bot | <greg.bock> no worries, I'll keep poking at it | 17:30 |
*** sgarzare has quit IRC | 17:41 | |
*** igordc has joined #kata-general | 17:58 | |
*** gwhaley has quit IRC | 18:04 | |
*** igordc has quit IRC | 18:08 | |
*** igordc has joined #kata-general | 18:13 | |
*** tmhoang_ has joined #kata-general | 20:22 | |
*** tmhoang_ has quit IRC | 20:23 | |
*** tmhoang has quit IRC | 20:24 | |
*** lpetrut has joined #kata-general | 20:27 | |
*** lpetrut has quit IRC | 20:31 | |
kata-irc-bot | <dangot> @greg.bock I'm keen on trying it if you can point me to that patch. | 20:48 |
kata-irc-bot | <dangot> Is the doc here https://virtio-fs.gitlab.io the one to follow ? | 20:49 |
*** lcastell has joined #kata-general | 21:24 | |
*** lcastell has quit IRC | 21:25 | |
kata-irc-bot | <greg.bock> yes, I can get it working with 1.4.0 | 21:26 |
kata-irc-bot | <greg.bock> https://github.com/greg-bock/runtime/tree/1.5.1-virtiofs https://github.com/greg-bock/agent/tree/1.5.1-virtiofs | 21:28 |
kata-irc-bot | <greg.bock> and small change to osbuilder | 21:28 |
kata-irc-bot | <greg.bock> | 21:28 |
kata-irc-bot | <greg.bock> I _think_ I have all the patches ported correctly | 21:29 |
kata-irc-bot | <greg.bock> | 21:36 |
*** tmhoang has joined #kata-general | 21:37 | |
*** sameo has quit IRC | 22:26 | |
kata-irc-bot | <raravena80> Yeah a few of us were there and there was also a booth | 22:33 |
*** tmhoang has quit IRC | 22:38 | |
kata-irc-bot | <raravena80> you can also use the consul traefik provider: https://docs.traefik.io/configuration/backends/consul/ | 22:42 |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!