From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932287AbWGLAQX (ORCPT ); Tue, 11 Jul 2006 20:16:23 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932288AbWGLAQX (ORCPT ); Tue, 11 Jul 2006 20:16:23 -0400 Received: from 58.105.229.78.optusnet.com.au ([58.105.229.78]:36522 "EHLO adsl-kenny.stuart.id.au") by vger.kernel.org with ESMTP id S932287AbWGLAQW (ORCPT ); Tue, 11 Jul 2006 20:16:22 -0400 Subject: Problems with oom killer From: Russell Stuart To: linux-kernel mailing list Content-Type: text/plain Date: Wed, 12 Jul 2006 10:15:12 +1000 Message-Id: <1152663312.4267.20.camel@ras.pc.brisbane.lube> Mime-Version: 1.0 X-Mailer: Evolution 2.0.4 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org The oom killer is being run, and I can't figure out why. As far as I can see there is plenty of memory available. Attached are some logs that show that state of the machine when the oom killer runs. Background info: The machine has 2Gb of RAM, and 2Gb of swap. The oom killer consistently strikes when the machine is doing a backup. The backup consists of mkfs'ing an ext3 partition, and then copying the files across. The oom killer does its stuff during the mkfs. It goes away ones the next ext3 partition is mounted and the backup starts. It is running the Debian stable kernel (2.6.8.1) with cfq on a dual core machine. Although it shouldn't be, it appears the machine is used a very high load at the time because the shell script attached is supposed to run every 5 seconds, yet at the time the oom condition happens there is approx 10 minute delay between runs. Two other odd things: there are many other machines that are identical software wise, as in installed from the same DVD image, and doing an identical backup. This is the only one with the issue. This box has just been replaced and a fresh install done. The previous box (completely different hardware) had the same issue. Any clues would be appreciated. -- Regards, Russell Stuart