Hi Bryan, On Fri, Jul 26, 2019 at 04:02:10PM +0900, Bryan Linton wrote: | On 2019-07-25 13:01:28, Mike Larkin <mlar...@nested.page> wrote: | > On Thu, Jul 25, 2019 at 09:54:22PM +0200, Paul de Weerd wrote: | > > A little more follow-up on this vmd-memory-leak issue. | > > | > > Comparing the two VMs I have running, I started to stress parts where | > > these two hosts differ. The testvm hardly does any traffic, while the | > > undeadly vm sees quite a few visitors on a daily basis, so networking | > > may be part of the leak. | > > | > > Running tcpbench against this machine (averaging at ~250Mbit/s) | > > results in vmd growing by about 80MB to 100MB per minute. Running | > > tcpbench against the testvm has similar results. | > > | > > The undeadly VM also has a second disk configured (on slower storage), | > > but putting load on that didn't significantly change the memory | > > consumption (above the 'expected' growth that I've been seeing). | > > | > > The growth during daily(8) runs still confuses me, as that doesn't do | > > anything network-related... | > > | > > Paul | > > | > | > I'll try to look for leaks in that area then. Thanks for the report. | > | > -ml | > | | To Paul, is it related to disk activity on the VM? I.e. Does | doing lots of I/O on the system cause memory usage to increase?
It does not, I specifically tried that (also because the VM with this issue has two disks, one of them on slower media) but I didn't notice a difference. | Can you test it with a snapshot/kernel dated May 7th or earlier? Not easily on the undeadly production environment. But I'm building a test environment at home that I may be able to use for this. | I CCed both of you into a bug report I just submitted to bugs@ | because I thought it may possibly be related. | | https://marc.info/?l=openbsd-bugs&m=156412299418191&w=2 Yep, saw that. I can definitely try reverting that diff on my test environment later today. | In brief, I'm seeing large amounts of memory being consumed | followed by a system hang when files are copied to a vnd(4) | device. | | I don't see it with regular disk I/O, only with vnds. | | If the VMs are using/accessing memory in a similar way to the | method that vnd(4) does, it might explain why the daily(8) runs | are causing the memory usage to increase. I can't really comment on that as I'm not familiar with the implementation of these parts of the kernel. | If this is an unrelated issue, then I apologize for the noise. I | figured it better to CC both of you in so you could evaluate it on | your own rather than for me to do nothing. Cheers, Paul -- >++++++++[<++++++++++>-]<+++++++.>+++[<------>-]<.>+++[<+ +++++++++++>-]<.>++[<------------>-]<+.--------------.[-] http://www.weirdnet.nl/
