NFS lockups, default read/write size too big?
by notzed from LinuxQuestions.org on (#5MGHH)
I've had a persistent issue for a while where sometimes large copies or writes to a NFSv4 server would hang the client machine access to that disk (incl sync and reboot) and spam the network with retries. The logs on the server get filled with things like this:
Code:svc: svc_tcp_read_marker nfsd RPC fragment too large: 1048808Mount options are hard,nfsv4 and it never seems to give up and you have to power cycle or reset the client to recover.
Both client and server are using the same slackware64-current and although I use kernel.org kernels it's been a problem since I set it up so it's spanned many versions. It's currently nfs-utils-2.5.4 the server kernel is configured with only NFS version 4 support (no 4.x options).
I modified /etc/nfsmount.conf to set R|W|Bsize to 64K and it seems to have fixed it. I'm happy with that solution but I'm wondering if it either points to some package config issue that might be of interest to others or if it's just something local that doesn't matter (network, kernel config, etc).
Code:svc: svc_tcp_read_marker nfsd RPC fragment too large: 1048808Mount options are hard,nfsv4 and it never seems to give up and you have to power cycle or reset the client to recover.
Both client and server are using the same slackware64-current and although I use kernel.org kernels it's been a problem since I set it up so it's spanned many versions. It's currently nfs-utils-2.5.4 the server kernel is configured with only NFS version 4 support (no 4.x options).
I modified /etc/nfsmount.conf to set R|W|Bsize to 64K and it seems to have fixed it. I'm happy with that solution but I'm wondering if it either points to some package config issue that might be of interest to others or if it's just something local that doesn't matter (network, kernel config, etc).