Scroll to navigation

mfsmaster.cfg(5) This is part of MooseFS mfsmaster.cfg(5)


mfsmaster.cfg - main configuration file for mfsmaster


The file mfsmaster.cfg contains configuration of MooseFS master process.


Syntax is:

Lines starting with # character are ignored as comments.


Configuration options:

user to run daemon as
group to run daemon as; optional value - if empty then default user group will be used
name of process to place in syslog messages; default is mfsmaster
whether to perform mlockall() to avoid swapping out mfsmaster process; default is 0, i.e. no
limit malloc arenas to given value - prevents server from using huge amount of virtual memory (Linux onty, default is 4)
disable out of memory killer (Linux only, default is 1)
nice level to run daemon with; default is -19; note: process must be started as root to increase priority, if setting of priority fails, process retains the nice level it started with
set default umask for group and others (user has always 0); default is 027 - block write for group and block all for others
where to store metadata files and lock file
alternate location/name of mfsexports.cfg file
alternate location/name of mfstopology.cfg file
alternate location/name of mfslicence.bin file (pro version only)
number of metadata change log files (default is 50)
how often (in hours) master will store metadata (default is 1)
how often (in hours) leader will download metadata from followers (pro version only ; default is 24)
metadata downloading limit in mebibytes per second (pro version only ; 0 means no limit, default is 50 - 50MiB/s = ~52.4MB/s)
number of previous metadata files to be kept (default is 1)
how many seconds of change logs have to be preserved in memory (default is 1800; this sets the minimum, actual number may be a bit bigger due to logs being kept in 5k blocks; zero disables extra logs storage)
Changelog save mode (default is 0)
0 - write in background by different process (less safe, but doesn't make master stop in case of heavy hdd load)
1 - write in foreground without syncing data (master waits for every changelog to be saved to hdd, but without syncing - a little more safe than the background option, but may cause master to stop and wait for flushing hdd buffers)
2 - write in foreground with fsync after each write (very safe, but may make your master very slow unless you have very sophisticated hardware)
how many missing chunks will be stored in master (up to 100*MISSING_LOG_CAPACITY bytes of memory will be allocated ; default value is 100000)
IP address to listen on for metalogger, masters and supervisors connections (* means any)
port to listen on for metalogger, masters and supervisors connections
MooseFS master host, IP is allowed only in single-master installations (pro version only ; default is mfsmaster)
delay in seconds before next try to reconnect to master-leader if not connected (pro version only ; default is 5)
timeout in seconds for master-leader connections (pro version only ; default is 10)
local address to use for connecting with master-leader (pro version only ; default is *, i.e. default local address)
IP address to listen on for chunkserver connections (* means any)
port to listen on for chunkserver connections
default timeout in seconds for master-chunkserver connection (default is 10)
Optional authentication string. When defined - then only chunkservers with the same AUTH_CODE are allowed to connect to this master. When not defined (default) - then all chunkservers are allowed. If you want to switch on chunkserver authentication, then first define AUTH_CODE in all your chunkservers (and reload/restart them) then define this option and master and reload/restart it. Remember that after reload currently connected chunkservers are NOT disconnected. New AUTH_CODE will be used only when chunkservers will make new connection.
Optional IP class remapping. Remap chunkserver IP addresses with first REMAP_BITS of IP equal to first REMAP_BITS of REMAP_SOURCE_IP_CLASS. During remapping system will set first REMAP_BITS of IP address to first REMAP_BITS of REMAP_DESTINATION_IP_CLASS. (defaults are zeros which means no remap)
initial delay in seconds before starting replications (default is 60)
whether to make undergoal replications respect topology (default is 0)
0 - do not respect topology
1 - pick destination server at random, but then choose best source server
2 - try to find destination server in the same rack as one of the existing copies and then replicate chunk locally (in the same rack)
whether new chunks should be recorded with respect to topology (default is 0)
0 - do not respect topology
N (N>0) - first try to create new chunks on servers with topological distance LOWER than N from the client; if not possible, for example because of storage class, chunk servers being busy or lacking space, then try servers with distance greater or equal to N
avoid using same ip/rack for different chunk copies (default is 0)
0 - ignore ip and rackid (standard behaviour)
1 - avoid storing more than one copy on chunkservers using same IP number
2 - avoid storing more than one copy on chunkservers using IP number from the same rack id
Chunks loop shouldn't check more chunks per seconds than given number (default is 100000)
Chunks loop shouldn't be done in less seconds than given number (default is 300)
Soft maximum number of chunks to delete on one chunkserver (default is 10)
Hard maximum number of chunks to delete on one chunkserver (default is 25)
Maximum number of chunks to replicate to one chunkserver (default is 2,1,1,4 - see NOTES)
Maximum number of chunks to replicate from one chunkserver (default is 10,5,2,5 - see NOTES)
Threshold for chunkserver load. (default is 150 - see NOTES)
Threshold ratio for chunkserver load (default is 3.0 - see NOTES)
Defines how long chunkservers will remain in 'grace' mode (default is 900 - see NOTES)
Maximum number of seconds server can be in maintenance mode (default value is 0 - which means 'forever')
Maximum number of seconds server can be in "temporary" maintenance mode (server is switched to this mode whenever is stopped gracefully, after reconnection server is switched back to normal mode automatically ; default value: 1800)
How many days unused (disconnected) chunkserver should be kept in master data structures (valid values: 0 - 365 ; 0 means indefinitely ; default value: 7)
Maximum difference between space usage of chunkservers (deprecated, use ACCEPTABLE_PERCENTAGE_DIFFERENCE instead)
Maximum percentage difference between space usage of chunkservers (default is 1 = 1%)
Length of priority queues (for endangered, undergoal etc. chunks - chunks that should be processed first - default is 1000000)
IP address to listen on for client (mount) connections (* means any)
port to listen on for client (mount) connections
How long to sustain a disconnected client session (in seconds; default is 86400 = 1 day)
Grace period in secods for soft quota (deprecated, use QUOTA_DEFAULT_GRACE_PERIOD instead for default value or specify it individually)
Default grace period in seconds for soft quota (default is 604800 = 7 days)
Set atime modification mode (default is 0 = always modify atime - see NOTES)
Set amount of space reserved for superuser (default is 0 = do not reserve space for superuser - see NOTES)
Define limit for number of hardlinks allowed for one object (default is 32767; possible values are from 8 to 65000)
Delay time in seconds after which inodes of deleted objects will be reused. BE AWARE if you change this value below 86400 (1 day) you MUST ensure that this value is higher than any of the following timeouts in all clients: mfsattrcacheto, mfsxattrcacheto, mfsentrycacheto, mfsdirentrycacheto, mfsnegentrycacheto, mfssymlinkcacheto. (default is 86400; possible values are from 300 to 3000000)


Chunks in master are tested in a loop. Speed (or frequency) is regulated by two options CHUNKS_LOOP_MIN_TIME and CHUNKS_LOOP_MAX_CPS. First defines minimal time between iterations of the loop and second defines maximal number of chunk tests per second. Typically at the beginning, when number of chunks is small, time is constant, regulated by CHUNK_LOOP_MIN_TIME, but when number of chunks beccomes bigger then time of loop can increase according to CHUNKS_LOOP_MAX_CPS.

Example: CHUNKS_LOOP_MIN_TIME is set to 300, CHUNKS_LOOP_MAX_CPS is set to 100000 and there is 1000000 (one million) chunks in the system. 1000000/100000 = 10, which is less than 300, so one loop iteration will take 300 seconds. With 1000000000 (one billion) chunks the system needs 10000 seconds for one iteration of the loop.

Deletion limits are defined as 'soft' and 'hard' limit. When number of chunks to delete increases from loop to loop, current limit can be temporary increased above soft limit, but never above hard limit.

Replication limits are divided into four cases:

first limit is for endangered chunks (chunks with only one copy)
second limit is for undergoal chunks (chunks with number of copies lower than specified goal)
third limit is for rebalance between servers with space usage around arithmetic mean
fourth limit is for rebalance between other servers (very low or very high space usage)

Usually first number should be grater than or equal to second, second greater than or equal to third, and fourth greater than or equal to third ( 1st >= 2nd >= 3rd <= 4th ). If one number is given, then all limits are set to this number (for backward compatibility).

Whenever chunkserver load is higher than CS_HEAVY_LOAD_THRESHOLD and CS_HEAVY_LOAD_RATIO_THRESHOLD times higher than average load, then chunkserver is switched into 'grace' mode. Chunkserver stays in grace mode for CS_HEAVY_LOAD_GRACE_PERIOD seconds.

There are five values for ATIME_MODE (all other values are treated as 0):

0 = Always modify atime for files, folders and symlinks.
1 = Always modify atime but only in case of files (do not modify atime in case of folders and symlinks).
2 = Modify atime only when it is lower than ctime or mtime and when current time is higher than ctime or mtime respectively, also modify atime when current atime is older than 24h. Do it for all objects during access (like "relatime" option in Linux).
3 = Same as above but only in case of files. In case of folders and symlinks do not modify atime.
4 = Never modify atime during access (like "noatime" option).

You can reserve space for superuser using RESERVE_SPACE option. You can define it as number of bytes, percent of total space, capacity of biggest chunkserver, etc.

# or #B = number of bytes reserved for superuser. Standard metric prefixes can be used - SI and IEC (k,K,M,Mi,G,Gi etc.)
#% or #.#% = percent of total capacity of MooseFS instance
#U or #.#U = multiplies of "U" value; U is defined as maximum number of bytes currently used by a single chunkserver
#C or #.#C = multiplies of "C" value; C is defined as maximum total capacity of a single chunkserver

When your network has two (or more) IP classes you may want to use one network for standard communication between MFS modules and separate network only for I/O. It can be done by setting REMAP_BITS, REMAP_SOURCE_IP_CLASS and REMAP_DESTINATION_IP_CLASS. When you set these options then master will change internally IP addresses of chunkservers and will send them as chunk locations, so clients will make connections with chunkservers using new (destination) IP for all I/O, but still communicate with master using original (source) IP. Also chunkservers will use original IP to communicate with master, but they will use new IP's to communicate between themselves during replication. Beware that all clients and chunkservers must have access to both networks, but masters, metaloggers etc. will need only access to the source network.


Copyright (C) 2021 Jakub Kruszona-Zawadzki, Core Technology Sp. z o.o.

This file is part of MooseFS.

MooseFS is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2 (only).

MooseFS is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with MooseFS; if not, write to the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02111-1301, USA or visit


mfsmaster(8), mfsexports.cfg(5) mfstopology.cfg(5)

August 2021 MooseFS 3.0.116-1