Date: Sat, 19 May 2012 13:13:28 +0300 From: Daniel Kalchev <daniel@digsys.bg> To: freebsd-fs@freebsd.org Subject: Re: Mirror of Raidz for data reliability Message-ID: <4FB77248.50709@digsys.bg> In-Reply-To: <CBDBEF7C.2B265%trent@snakebite.org> References: <CBDBEF7C.2B265%trent@snakebite.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 18.05.12 19:55, Trent Nelson wrote: > So, my thinking is=8A because both machines can see all disks, the=20 > master could import the zpool as normal, and the slave could import it = > read-only. (Or not import it at all...) The proper way of doing it is "not import it at all". ZFS is not an=20 shared filesystem. If you have the second host mount the zpool even if read-only, you only=20 guarantee that data on the pool will not be corrupted, but you cannot=20 avoid the second "read-only" host panic or otherwise crash if it tries=20 to access data which is no longer where it thinks it is, because the=20 second host doesn't have access to the primary host's in-memory metadata = about ZFS. Since ZFS is copy on write filesystem, chances are you will=20 be accessing data that is no longer valid. Refreshing the internal ZFS=20 state between two or more hosts is non-trivial (if it was, Sun would=20 have done this, as it suits their usage) and in any case performance=20 will suffer at least as much as an true networked filesystem does,=20 compared to "native" ZFS. Daniel
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4FB77248.50709>