Date: Wed, 16 Dec 2015 09:43:21 +0200 From: Konstantin Belousov <kostikbel@gmail.com> To: Ian Lepore <ian@freebsd.org> Cc: Mark Felder <feld@FreeBSD.org>, bob prohaska <fbsd@www.zefox.net>, freebsd-arm@freebsd.org Subject: Re: No space left in lost+found Message-ID: <20151216074321.GO3625@kib.kiev.ua> In-Reply-To: <1450211241.25138.75.camel@freebsd.org> References: <20151215181047.GA29187@www.zefox.net> <1450204738.4176380.468287977.048387B9@webmail.messagingengine.com> <20151215191845.GB29187@www.zefox.net> <1450209274.4299.468358681.22757635@webmail.messagingengine.com> <1450211241.25138.75.camel@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Dec 15, 2015 at 01:27:21PM -0700, Ian Lepore wrote: > On Tue, 2015-12-15 at 13:54 -0600, Mark Felder wrote: > > > > On Tue, Dec 15, 2015, at 13:18, bob prohaska wrote: > > > Hi Mark, > > > > > > Here's what I get, da0p4 is /tmp > > > > > > root@www:/lost+found # tunefs -p /dev/da0p4 > > > tunefs: soft update journaling: (-j) enabled > > > > > root@www:/lost+found # tunefs -p /dev/mmcsd0s2a > > > tunefs: soft update journaling: (-j) enabled > > > > I would recommend turning off soft update journaling and see if your > > problem goes away. If that doesn't work, perhaps disable soft updates > > entirely. Anyone else have thoughts on this? > > > > I've always been anti-journaling, but only some of reasons are based inprovable facts. > > Over the years there have been a lot of complaints about it. Probably > some of them were genuine then, but have long since been fixed. Some > of them may have been user error or bad hardware. But all in all, it > has left me with a very negative opinion of journaling with ufs. > > But that's all emotion, not really hard facts. A few factual things... Factual thing about +J is that there are several unresolved deadlocks under high metadata i/o load. Also I saw several sporadic reports of fsck code having issues, but this mostly occured for the trashed journal and can be blamed to lack of proper sanity checks. Minor thing is that +J and snapshots do not live together. Returning to the OP problem, please show the output of ls -ld your/lost+found > > Journaling means doing a lot more writing and on an sdcard that's slow. > A lot of people say it's also bad in terms of wearing out the card, > but that doesn't worry me so much, it's a lot harder to kill an sdcard > than most people think. > > To me the strongest argument against it for most small-system users is > that the whole point of journaling is to take a small performance hit > on each write to avoid a long (sometimes hours-long) downtime doing > fsck after a crash. It doesn't improve reliability by storing extra > info that can make a better recovery than fsck alone, it's just a "pay > me now or pay me later" performance tradeoff. > > But on an sdcard the performance hit for extra writing isn't small, and > the time to do a full fsck after a crash isn't large. So in that sense > journaling adds nothing of value. > > IMO, soft updates (without journaling) is almost mandatory on an > sdcard. Without it, there is so much extra metadata IO that > performance is horrible.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20151216074321.GO3625>