Date: Thu, 24 May 2012 01:38:53 +0400 From: Lev Serebryakov <lev@FreeBSD.org> To: Konstantin Belousov <kostikbel@gmail.com> Cc: "Bjoern A. Zeeb" <bzeeb-lists@lists.zabbadoz.net>, freebsd-current FreeBSD <freebsd-current@freebsd.org>, Kirk McKusick <mckusick@freebsd.org> Subject: Re: UFS+J panics on HEAD Message-ID: <12410676034.20120524013853@serebryakov.spb.ru> In-Reply-To: <20120523131046.GC2358@deviant.kiev.zoral.com.ua> References: <38A5BC8F-A8FB-4371-AB1D-9548F5957254@lists.zabbadoz.net> <20120523131046.GC2358@deviant.kiev.zoral.com.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
Hello, Konstantin. You wrote 23 =D0=BC=D0=B0=D1=8F 2012 =D0=B3., 17:10:46: KB> This panic is another protective panic caused by on-disk inconsistent KB> structures. The bitmap indicated that an inode was free, but actual ino= de KB> context suggested that the inode is in use. KB> I would not worry much about ffs code until known hardware problem on KB> the machine are fixed. Konstantin, it is very sad, that official position of one of FFS maintainers (according to mailing list activity), is to blame hardware on every FFS/SU/SUJ inconsistency and do nothing with code. According to my experience, we all live in real world, where HDDs could lost write cache on power failure or kernel panic unrelated to FFS, UPSes and PSUs fails (which leads to power failures) and HDDs with turned off write cache are unusable -- because they become slow like hell without writing cache. You could name it "broken hardware," but let face it -- all non-top-server hardware, everything, but HBAs with battery installed in double-PSU-equipped cases, are "broken" now in this sense. My home server with almost-read-only load crashes due to burned out PSU twice in last 2 years (and I buy "good" desktop PSUs, in $150-$200 price range, not china boxes for $30) and I've got one memory problem in this time period (DIMM was detected and replaced, but I've got two or three panics before I become sure, that it is memory problem, because memtest86+ doesn;t find any problems in 12+ hours run). It is good desktop hardware, with good cooling system, not something low-end, but not server-grade one, of course. And after EVERY of such crashes my main storage area (95% read, 5% write) had dozens of "unexpected SU inconsistences," background fsck fails to create snapshot and I was need to run foreground fsck for many hours. It seems, that "async" mount without SU will not be worse that SU solution! And, if you read through mailing lists, you cold find dozens such reports. And answer almost always is "broken hardware". Yes, Ok, it is broken hardware, all right. But we haven't other one! We need to live with what we have! What I want to say: FFS/SU become almost unusable on this hardware. Protective panic, my ass! Every solution (link this inode to lost+found and mark as used, mark it as free, etc) is better than protective panic. One mismatched inode is not the end of the world, it is even not end of cylinder group, not to say about whole FS, system could (and must) complain about it, but not panic! Did you hear term ``self-healing''? It seems, that modern hardware needed better solution, that "just panic and blame hardware." Or should we call FFS officially dead and promote ZFS as only usable FS on modern FreeBSD now? --=20 // Black Lion AKA Lev Serebryakov <lev@FreeBSD.org>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?12410676034.20120524013853>