Yes, only the ethernet interface connected to the gluster machines is down and everything else works. There's a "em0 watchdog timeout" line on the screen and log, and even if I restart the network subsystem the interface remains down. This is not a bug of glusterfs though, but glusterfs must have done something special at startup to trigger the bug. Also I'm wondering if it's possible to run only one instance of glusterfsd to serve both as a server and as a client. Thanks, - Wei Mark Mielke wrote: > On 09/26/2009 03:41 PM, Wei Dong wrote: >> Hi All, >> >> Our linux cluster uses a freebsd gateway, and it seems that starting >> a second glusterfs/glusterfsd instance (no matter as client or server >> listening on different port) on the linux machines tends to bring >> down the intel pro1000 NIC of the freebsd gateway. It doesn't happen >> every time on each particular machine, but it almost always happens >> when I try to do it on each of our 66 machines. I'm causing a lot of >> troubles to the cluster users by testing glusterfs and I'm wondering >> it someone on the list has an idea what's going on. >> >> Strictly speaking, it might not be a bug of glusterfs, but we've been >> running the cluster for half a year and glusterfs is the only program >> to trigger the bug. >> > > What does bringing down the NIC mean? Any kernel logs? Programs are > all working fine - no freezes - but just the network connectivity is out? > > Cheers, > mark >
