From: Kovacs Gabor Subject: NFS client freezes under heavy disk IO Date: Sun, 30 Dec 2007 01:13:49 +0100 Message-ID: <200712300113.49926.kgabor@bolyai.elte.hu> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Cc: gbakos-6Mj89AmeLNm+fmr0zi+kZQ@public.gmane.org To: linux-nfs@vger.kernel.org Return-path: Received: from mx2.mail.elte.hu ([157.181.151.9]:34635 "EHLO mx2.mail.elte.hu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752567AbXL3AOL (ORCPT ); Sat, 29 Dec 2007 19:14:11 -0500 Sender: linux-nfs-owner@vger.kernel.org List-ID: Hello All, We have some NFS problem on a big partition. NFS (v3) server runs kernel version 2.6.21.6, client runs kernel version 2.6.21.7. Exported file system is on XFS file system, 8.9T, typical file size is 1-10M. NFS tools are from Fedora 7: nfs-utils-1.1.0-3.fc7 On the client side, during heavy read-IO (a program walks on tens of thousands of files by opening them for reading and collects some data) after a couple of time (1-5 days) the user process freezes. Top command shows a "D" state for it, but many other commands trying to "touch" it, freezes, too. (ps, fuser /nfs_dir, w) making its own virtual console unusable. None of the frozen processes can be killed. Actually, the computer cannot be shut down cleanly any more, only the reset button helps. There was a kernel error: http://www.mediafire.com/imageview.php?quickkey=0to7xa0eowd I guess it is some kind of out of memory condition. Is this possible? How much memory required by kernel nfs (client) modules? Any idea to try? (System has 16G memory, swap is 32G, user program consumed approx. 12G.) A very similar deadlock was produced using NFSv4. Thanks Gabor Kovacs