Date: Wed, 05 Sep 2007 13:46:59 +0200 From: Tobias Ernst <tobi@casino.uni-stuttgart.de> To: freebsd-questions@freebsd.org Subject: Re: Booting to root on gmirror with disk failure, is it even possible? Message-ID: <46DE9733.8010402@casino.uni-stuttgart.de> In-Reply-To: <64c038660709041931i2f83ee9bv9a3e4276c750ff23@mail.gmail.com> References: <64c038660709041931i2f83ee9bv9a3e4276c750ff23@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Modulok schrieb: > Before I invest significantly more time into my current gmirror > issues, I have but two simple questions for anyone out there: > > 1. Has anyone used gmirror for the root partition and been able to > successfully boot with one failed (or un-plugged) disk? It's the > latter part of the question that is the real issue for me. I'm just > looking for a confirmed "it's possible". Yes, it is possible. IBM xSeries 346, FreeBSD 6.2-RELEASE, amd64. U360 hard drives. More specs are available from IBM. Using gmirror because we only have an Adaptec "HostRAID" (aka "FakeRAID") controller and not a real "ServerRaid", i.e. our SCSI controller basically has no useful RAID capabilities built in. My test case is to unplug any one disk while the system is running. (Don't do this with your system unless your hardware is specified for hot plugging!). FreeBSD detects a bus reset, marks the gmirror as degraded and continues operating normally, and I can also reboot the degraded gmirror without any problems. The more conservative test case is to power down the system, unplug any one disk, and restart the system. No problems with that either. In fact, the absolutely robust behaviour of gmirror was one of my key arguments for switching from Linux to FreeBSD :-). Of course there are a zillion ways to fail your hard disk, and there could be cases where one hard disk might start behaving erratically, and gmirror might not be able to detect all such cases and might try to continue using the failed disk. This could theoretically lead to some nasty data integrity issues in the worst case. But this is true for any RAID, even when implemented in hardware IMO. Regards Tobias -- Universität Stuttgart|Fakultät für Architektur und Stadtplanung|casinoIT 70174 Stuttgart Geschwister-Scholl-Straße 24D T +49 (0)711 121-4228 F +49 (0)711 121-4276 E office@casino.uni-stuttgart.de I http://www.casino.uni-stuttgart.de
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?46DE9733.8010402>