56 lines
2.8 KiB
Plaintext
56 lines
2.8 KiB
Plaintext
LizardFS is a highly scalable, fault-tolerant, POSIX-compatible, FUSE-based,
|
|
high performance distributed filesystem, licensed under GNU General Public
|
|
License version 3.
|
|
|
|
LizardFS is an implementation of GoogleFS, and a fork of the earlier project,
|
|
MooseFS. LizardFS supports writable snapshots (instant copies), undeleting
|
|
files, automatic data rebalancing, self-healing, data tiering, periodic data
|
|
patrols and many more.
|
|
|
|
LizardFS system consists of a master server, one or more metadata logging
|
|
servers (meta loggers), and many chunk servers, that store the data on their
|
|
locally-attached drives. Both meta loggers and chunk servers can be added and
|
|
removed without restarting the master server.
|
|
|
|
Filesystem metadata is stored on the master server (and constantly replicated
|
|
to meta loggers), whereas filesystem data is divided into chunks and spread as
|
|
files over chunk servers, according to pre-defined 'goals', which can be set
|
|
on file-, directory-, or filesystem level. A goal can be an n-way mirroring
|
|
goal, n+1 xor-ed goal (each chunk divided into n parts and xor-ed to calculate
|
|
one part of redundancy), or more sophisticated, erasure code based n+k
|
|
redundancy, where n parts of each chunk are backed by k parts of redundancy
|
|
data.
|
|
|
|
A set of administrative commands exists to support querying and setting
|
|
redundancy goals and trash preservation time. LizardFS is admin-friendly since
|
|
any missing chunks can be provided from any sort of backup to any running
|
|
chunk server.
|
|
|
|
This package contains all binaries needed to run LizardFS system: mfsmaster,
|
|
mfsmetalogger, mfschunkserver, as well as lizardfs-cgiserver (web-based
|
|
monitoring console).
|
|
|
|
You need an "mfs" user and group prior to building lizardfs. Something like
|
|
this will suffice for most systems:
|
|
groupadd -g 353 mfs
|
|
useradd -u 353 -g 353 -d /var/lib/mfs mfs
|
|
Feel free to use a different uid and gid if desired, but 353 is recommended to
|
|
avoid conflicts with other stuff from SlackBuilds.org.
|
|
|
|
It is also advisable to make name 'mfsmaster' pointing at your Master server
|
|
across your network. It is not strictly required, but it will make things much
|
|
easier. If you are unable to configure your DNS server, adding this line to
|
|
/etc/hosts on each master, metalogger, chunkserver, and client machines will
|
|
do:
|
|
|
|
a.b.c.d mfsmaster mfsmaster.my-domain.ext
|
|
|
|
where a.b.c.d is an IP address of your master server.
|
|
|
|
Then on each node add '/etc/rc.d/rc.lizardfs start' to /etc/rc.d/rc.local (or
|
|
wherever you find appropriate), and use '/etc/rc.d/rc.lizardfs setup' to
|
|
configure which services should run on the server. Since most installations
|
|
consists mostly of chunkservers, rc.lizardfs-chunkserver is marked executable
|
|
by default (but will not run until rc.lizardfs-chunkserver or rc.lizardfs is
|
|
added to rc.local, so no need to worry).
|