On Thu, Mar 30, 2006 at 03:22:39PM +0200, JaniD++ wrote: > Hello, list, > > Can comebody write me a little patch to the client, to newer give up and can > resume the connection on some packet lost? > I mean exactly the same as nfs "hard" option, to avoid these messages: > > nbd3: Attempted send on closed socket > end_request: I/O error, dev nbd3, sector 66961896 > I/O error in filesystem ("md31") meta-data dev md31 block 0xff707b0 > ("xfs_trans_read_buf") error 5 buf count 8192 > > My 8TB raid is too much often crashes, and any type of script is too slow to > resume (reconnect) the nbd-client automatically... I've just committed a modified nbd-client to the subversion trunk. With that nbd-client running in a qemu machine and an nbd-server on the host, I was running 'mkfs -t ext2 /dev/nb0' when I killed the nbd-server that was serving for that box. It reconnected immediately, and mkfs was not interrupted (at least, it did not seem to be). There were /some/ read errors, but not a lot of them. This is probably the fastest it's going to be without kernel patches. If you want to try it out, get the code from subversion as explained on the sourceforge.net website. Note, only use the client from subversion trunk -- the server is pretty broken at the moment. You need to run it with the -persist option for the automatic reconnect to succeed; and if you want to disconnect, you may have to run 'killall nbd-client' _before_ doing the nbd-client -d (will depend on your kernel). -- Fun will now commence -- Seven Of Nine, "Ashes to Ashes", stardate 53679.4
Attachment:
signature.asc
Description: Digital signature