From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ryusuke Konishi Subject: Re: Deadlocks! help, please! Date: Tue, 12 Feb 2008 21:57:57 +0900 Message-ID: <1202821077.9432.99.camel@localhost.localdomain> References: <20080121.140129.01311807.ryusuke@osrg.net> <20080122000052.215084db@vosztok> <1200970920.2844.59.camel@localhost.localdomain> <20080205.182829.44149266.ryusuke@osrg.net> <20080207110930.3df85a70@vosztok> Reply-To: NILFS Users mailing list Mime-Version: 1.0 Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: <20080207110930.3df85a70@vosztok> List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: users-bounces-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org Errors-To: users-bounces-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org Content-Type: text/plain; charset="iso-8859-1" To: NILFS Users mailing list Hi G=E1bor, On Thu, 2008-02-07 at 11:09 +0100, Gergely G=E1bor wrote: > I have installed it (with some other patches: the one for the mount opt= ion bug,=20 > and other for udivdi (that is needed to link the module properly at me)= and=20 > started a download in rtorrent for the night. It has not ferrzed, but i= have=20 > experianced massive memory leaks, either on the cleaner, or int he modu= le. > now i'm trying to reproduce the bug with fs running without cleaner.=20 > if it will not arise in a day, i'll try the cleaner as well.=20 > I have a question about the cleaner, that might help my find the bug: > can I send a signal to the cleaner to finish and commit unfinished cha= nges and > and can I manually start a cleaner (possibly with valgrind)? Yes, you can manually start and shutdown the cleaner. > Can I send signals to pause cleaning (for example I go on battery), and= to resume? No, the cleaner doesn't support pause/resume by signal, however=20 you can do the same thing by the manual shutdown and restart. > This leak error ate up all my memory, and forced all apps to swap, slow= ing the > system down so much, that it took almost an hour to exit rtorrent... > unfortunalety i forgot to vmstat after the kill of the cleaner, before=20 > umounting the fs. If I kill the fs will it not leave garbage (i mean ba= dly formed > entries in the log) after itself if it will, will those entries be clea= ned up?=20 Yes, they will. NILFS is designed not to leave garbage. However, the NILFS module holds memory for b-tree pages globally, so it may eat some memory even after umount. > after i umounted the partition, i had my ram back cleanly (i'd say that= this=20 > was possibly the cleaner, as the module was left in memory even after t= he umount, > still the ram wasn't wasted...) Is that possible, that the ram was not= wasted,=20 > but the cleaner/fs is bit inefficient, or can't be made simpler, and I = should=20 > buy more ram? it ate ~440Mb of memory. (i have 512Mb) OK, I will review the patch to find out an elemental leak bug. For now, the latest (unpublished) version seems to work properly=20 in my environment. We will soon release nilfs-2.0.0. I'd like to clean things up with the release. Since we don't want -testing release anymore, we may release it even if the problem ramains. If I can find the leak bug in time, I will include it in the release. > Otherwise the patch seems to work cleanly I haven't applied the > previous patch=20 > you proposed, should I have tried? (it gave errors) You don't have to apply the previous patch; the new patch includes the previous change. > Possibly if you set up a git/hg/other repo, life'd be more convinient. > they made my life simpler even in my small school projects. Sorry for inconvenience. We also have a plan to make a git (or other kind of) repository after it gets settled. So please wait a little longer. Best regards, --=20 Ryusuke Konishi NILFS team NTT http://www.nilfs.org/