06.01.2011 15:56, KP Kirchdoerfer пишет: > Hi; > > this is from an old mail, but I tested today what happens if Bering-uClibc3 > runs out of memory and what happens on Bering-uClibc4 box. > > Am Samstag, 12. Juni 2010, 20:17:49 schrieb Andrew: >> Hi. >> I just finished migration from compressed minix initrd to initramfs that >> uses compressed as cpio.gz archive with files. >> I'm done this due to several reasons: >> 1) Support for multiple initrd files (so, it'll be possible to split >> initrd module packages); >> 2) It's much easier to repack .cpio archive, and minix driver can be >> excluded from kernels (both LEAF and host) >> >> Due to migration I omitted diverting of rootfs into separate ramdisk - >> IMHO it's really unuseful in that case; rootfs with it's default size >> will take up to half of available RAM during filling - so it'll be >> enough big, and it'll .never consume all available memory. (I check >> this, it's working) >> >> One of it's cons that I found - there is no initramfs usage statistics >> in output of 'df'. > I simulated running out of space with dd > dd if=/dev/zero of=output.file bs=1024 count=1024 > filling up the memory to the max. > > If a Bering-uClibc3 box runs out of memory the dd command fails with the > message "no disk on space" and refused to generate the last MB. > On a Bering-uClibc4 it starts to kill processes: > [36521.153494] Out of memory: kill process 2809 (mini_httpd) score 87 or a > child > [36521.160859] Killed process 2809 (mini_httpd) vsz:696kB, anon-rss:88kB, > file- > rss:180kB > [36521.190261] Out of memory: kill process 2228 (dnsmasq) score 85 or a child > [36521.197453] Killed process 2228 (dnsmasq) vsz:680kB, anon-rss:80kB, file- > rss:256kB > [36521.219629] Out of memory: kill process 2897 (aiccu) score 37 or a child > [36521.226586] Killed process 2897 (aiccu) vsz:4796kB, anon-rss:120kB, file- > rss:264kB > > This is a serious pb and it may happen that a remote box is unavailable even > for login via ssh, so you can't just reboot it. > > Something I've overlooked? > kp Hmm, it looks like initramfs behavior was changed from 2.6.32 kernel, and now it hasn't limit in half of available memory. We can switch back to rootfs on tmpfs, this will require some init scripts modification.
------------------------------------------------------------------------------ Learn how Oracle Real Application Clusters (RAC) One Node allows customers to consolidate database storage, standardize their database environment, and, should the need arise, upgrade to a full multi-node Oracle RAC database without downtime or disruption http://p.sf.net/sfu/oracle-sfdevnl _______________________________________________ leaf-devel mailing list leaf-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/leaf-devel
