Repository navigation
permission denied trying to mount /home on newly started nodes after accidental deletion #666
Description
It seems elasticluster can recover remarkably well when worker VMs are deleted by just starting again the cluster.
However as a consequence of the new IP address a VM gets assigned, the setup phase fails when it tries to mount /home from the frontend node: it seems that while the hosts file is properly updated, NFS caches the IP address it initially resolved from the exports file.
A manual restart of the NFS service on the frontend node fixes the situation and it could be suggested and/or automatically performed to be sure the running configuration is in sync with the configuration files.
Activity
riccardomurri commentedon Oct 31, 2019
Actually,
exportfs -ris unconditionally performed whenver thenfs-serverrole is run. What is happening, I think, is that, since the host names have not changed but only the IP addresses,exportfscan not detect any change in the configuration and exits without signallingrpc.mountdto reload the exports table.A fix would be to populate the exports table with IP addresses instead of host names, which would fix the issue at the cost of making the exports file more obscure.
Another one would be to restart only the
rpc.mountdservice, but on older SysV-based distros it does not have a separate startup script, we must restart the entire NFS subsystem.