Chapter 1
Introduction
Notice: there is one dependency to resolve: users’ computers need FUSE package to mount
MooseFS. It can be downloaded and installed from repositories.
1.1 Key differences between versions 1.6.2x and 2.0.x
1. Master host(s) configuration is done solely via DNS – it is no longer possible to list
master(s) IP address(es) in clients’ and chunkservers’ configuration; default name for
master domain is mfsmaster, it can be changed in configuration files;
2. In Pro version metaloggers become optional, they can be replaced by additional master
servers; in CE version it is still recommended to set up metaloggers.
3. Mfsmetarestore tool is no longer present in the system; instead, it is enough to start the
master process with -a switch;
4. Configuration files now sit in mfs subdirectory inside the etc directory (this change was
introduced in 1.6.27).
1.2 Many Master Servers – how does it work?
In previous MooseFS versions you had only one master process and any number of metaloggers.
In the event of master failure, system administrator was able to retrieve ”metadata” information
from the metalogger and start a new master (on a new machine, if necessary), so the file system
was up and running again. But this was always causing the system to be unavailable to clients
for a period of time and required manual work to bring it back up.
New Pro version introduces many master servers working together in multiple roles. One role
is ”leader”. The leader master is acting as it used to for the chunkservers and clients. There is
never more than one leader in any working system.
The other role is ”follower”. The follower master is doing what metaloggers used to do – it
downloads metadata from the leader master and keeps it. But unlike a metalogger, if a leader
3
