On Sun, Feb 04, 2007 at 08:45:15PM +0100, Wouter Verhelst wrote: > On Sat, Feb 03, 2007 at 11:54:49AM -0800, Vagrant Cascadian wrote: > > when using nbd-server with the "-a timeout" option, it will often > > disconnect running nbd-client instances that have merely been inactive > > for a while. it would be much, much more useful if it implemented some > > sort of "ping" to determine if the nbd-client instance is still there, > > just not very active. > > > > the workaround is just to have the machine running an nbd-client make > > nbd accesses periodically just to keep the connection alive... but > > that's kind of ugly. > > Or just do not use it at all. nbd-server does set the SO_KEEPALIVE > socket option, which will do such keepalive requests every once in a > while. However, for that to kick in, it takes two hours (by default), > and then you still need a number of probes before it is actually > disconnected and the server killed.
hmmm... i'm not sure i understand you here. are you saying that nbd-server instances disconnect if there is no nbd-client process connecting to it after some period of time? i've seen nbd-server instances run for days without an nbd-client connected to them. i would like for some sort of idle timeout mechanism, so that an nbd-server instance doesn't indefinitely consume resources to no purpose. if there's a way to get behavior like that without using the "-a timeout", please let me know. > Anyhow, if you want to write a patch that makes the -a option more > useful, feel free... but that would probably also require a kernel > patch, which I don't maintain. and well over my head... thanks for your work and time. live well, vagrant -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
