Hi all, I have a setup with 2 nodes mirrored, if I simulate a disk crash (take a server node out, clear the data and restart it) on the first stat of files on the client, it appears to be self healing but during this time (it's syncing about 2G) the mount point becomes unusable..... is this expected behaviour? Josh.
On Thu, 9 Jul 2009 15:16:49 +0100 "Hiren Joshi" <josh at moonfruit.com> wrote:> Hi all, > > I have a setup with 2 nodes mirrored, if I simulate a disk crash (take a > server node out, clear the data and restart it) on the first stat of > files on the client, it appears to be self healing but during this time > (it's syncing about 2G) the mount point becomes unusable..... is this > expected behaviour?If you mean that the healing more or less crashes the mount and leaves it unusable and does never complete, well, then you see exactly what I see ... I have not managed to complete a healing process with everything called 2.0.3rc. I have not checked 2.0.3 so far, will do tomorrow. -- Regards, Stephan
----- "Hiren Joshi" <josh at moonfruit.com> wrote:> Hi all, > > I have a setup with 2 nodes mirrored, if I simulate a disk crash (take > a > server node out, clear the data and restart it) on the first stat of > files on the client, it appears to be self healing but during this > time > (it's syncing about 2G) the mount point becomes unusable..... is this > expected behaviour?Self heal happens whenever a directory or a file is accessed for the first time. While a file is being self-healed, all other operations on it will be blocked. So if a 2GB file is being healed, then accesses on that file will be blocked until self-heal completes. You might see that "ls" done on the mountpoint is blocked until self heal of _all_ files in the directory completes. This is because in most cases ls will be aliased to 'ls --color', which causes ls to do a 'stat' on each file in the directory, thus triggering their self-heal. Self-healing in the background is in the roadmap.