If you try to install on older hardware (something with only 32MB of ram) the 2.6.9 kernel kicks in OOM-Killer even if you have a large amount of swap and swappiness set to 100. This I think is a kernel itself problem (has happened since 2.6.8. 2.6.7 Works fine, would suggest people who have this problem use a 2004.2 livecd to install. Reproducible: Always Steps to Reproduce: 1. 2. 3.
You can't boot or what? If you can't bootstrap, then I would suggest GRP or a stage3 on that machine.
I don't see this as a bug. There are /many/ things that need more that 64MB of RAM to compile. Also, if you're low on physical RAM, (but frankly, not as low as this bug is talking about) then you should have a goodly sized swap file. I'd close this... Doesn't seem valid. I think this may also point out a flaw in our documentation. Do we have minimum system requirements? I would think anything less than 128MB of RAM (and a big swap partition) wouldn't be officially supported.
You can't really do anything to be honest - you cant compile any programs, and alot of programs can't be emerged, as they are compiling, the OOM-Killer kicks in. Moving to 2.6.7 "fixes" this. I had the problem, however, I still had a 2.4 kernel on my system that I booted to and used to compile 2.6.7. This was merely opened to notify that the problem exists, as 2004.3 doesn't have a 2.4 kernel (at least, thats what I was told.)
So do we add some "minimum system requirements" to the documentation? I could see it as a good thing. You must have at least a Pentium Pro and 128MB of RAM for support on compiling from stage1. If your machine has less CPU speed or less RAM than this, we recommend that you either use a stage3 install, or use GRP, if you plan on getting any support from Gentoo. Something like that?
Requirements are mentioned in handbook (2.a) Can you confirm 64Meg can be used if only GRP? Should we bump requirements to 128 for stage1/2 installs? Sounds reasonable to me. Why a pentiumPro? Are LiveCDs compiled for i686 and not x86? FYI, the smallest box I have is a Pentium 100 (i586) w/ 128Meg, stage1 w/1.4rc1. It does not boot CDs and I'm not rebooting it anyway (560 days uptime).
Actually, the only reason for the PPro is to keep complaints from machines that aren't i686 or better to a minimum. You can bootstrap from stage1 on a 486, provided you have enough RAM and swap.
What I am saying is that the kernel itself seems to be the culprit, when you have a small amount of ram, no matter what size your swap partition is, it doesn't swap. Swappiness can be set to 100, and it still doesn't swap. OOM-Killer kicks in on kernels 2.6.8+ and kills whatever the process it thinks is using the most ram that isn't necessary. It isn't just the install, I just recognized it because the person who brought it up was using a 2004.3 cd and it only had a 2.6 kernel available.
So this really is a kernel issue?
Does anyone have any idea if this was resolved in newer kernels? Perhaps by 2.6.11 or 2.6.12?
Yes, OOM killer stopped going mad around 2.6.10 or so
OK. I am marking this as FIXED as of 2005.0, if this is not the case, then *PLEASE* REOPEN this bug. Thanks
*** Bug 102130 has been marked as a duplicate of this bug. ***