.TH mfsmaster.cfg "5" "August 2021" "MooseFS 3.0.116-1" "This is part of MooseFS" .SH NAME mfsmaster.cfg \- main configuration file for \fBmfsmaster\fP .SH DESCRIPTION The file \fBmfsmaster.cfg\fP contains configuration of MooseFS master process. .SH SYNTAX .PP Syntax is: .TP .IB OPTION " = " VALUE .PP Lines starting with \fB#\fP character are ignored as comments. .SH OPTIONS Configuration options: .TP .B WORKING_USER user to run daemon as .TP .B WORKING_GROUP group to run daemon as; optional value - if empty then default user group will be used .TP .B SYSLOG_IDENT name of process to place in syslog messages; default is mfsmaster .TP .B LOCK_MEMORY whether to perform mlockall() to avoid swapping out mfsmaster process; default is 0, i.e. no .TP .B LIMIT_GLIBC_MALLOC_ARENAS limit malloc arenas to given value - prevents server from using huge amount of virtual memory (Linux onty, default is 4) .TP .B DISABLE_OOM_KILLER disable out of memory killer (Linux only, default is 1) .TP .B NICE_LEVEL 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 .TP .B FILE_UMASK set default umask for group and others (user has always 0); default is 027 - block write for group and block all for others .TP .B DATA_PATH where to store metadata files and lock file .TP .B EXPORTS_FILENAME alternate location/name of \fBmfsexports.cfg\fP file .TP .B TOPOLOGY_FILENAME alternate location/name of \fBmfstopology.cfg\fP file .TP .B LICENCE_FILENAME alternate location/name of \fBmfslicence.bin\fP file (pro version only) .TP .B BACK_LOGS number of metadata change log files (default is 50) .TP .B METADATA_SAVE_FREQ how often (in hours) master will store metadata (default is 1) .TP .B METADATA_DOWNLOAD_FREQ how often (in hours) leader will download metadata from followers (pro version only ; default is 24) .TP .B METADATA_DOWNLOAD_LIMIT metadata downloading limit in mebibytes per second (pro version only ; 0 means no limit, default is 50 - 50MiB/s = ~52.4MB/s) .TP .B BACK_META_KEEP_PREVIOUS number of previous metadata files to be kept (default is 1) .TP .B CHANGELOG_PRESERVE_SECONDS 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) .TP .B CHANGELOG_SAVE_MODE Changelog save mode (default is 0) .br 0 - write in background by different process (less safe, but doesn't make master stop in case of heavy hdd load) .br 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) .br 2 - write in foreground with fsync after each write (very safe, but may make your master very slow unless you have very sophisticated hardware) .TP .B MISSING_LOG_CAPACITY 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) .TP .B MATOML_LISTEN_HOST IP address to listen on for metalogger, masters and supervisors connections (* means any) .TP .B MATOML_LISTEN_PORT port to listen on for metalogger, masters and supervisors connections .TP .B MASTER_HOST MooseFS master host, IP is allowed only in single-master installations (pro version only ; default is mfsmaster) .TP .B MASTER_RECONNECTION_DELAY delay in seconds before next try to reconnect to master-leader if not connected (pro version only ; default is 5) .TP .B MASTER_TIMEOUT timeout in seconds for master-leader connections (pro version only ; default is 10) .TP .B BIND_HOST local address to use for connecting with master-leader (pro version only ; default is *, i.e. default local address) .TP .B MATOCS_LISTEN_HOST IP address to listen on for chunkserver connections (\fB*\fP means any) .TP .B MATOCS_LISTEN_PORT port to listen on for chunkserver connections .TP .B MATOCS_TIMEOUT default timeout in seconds for master-chunkserver connection (default is 10) .TP .B AUTH_CODE 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. .TP .BR REMAP_BITS ", " REMAP_SOURCE_IP_CLASS ", " REMAP_DESTINATION_IP_CLASS 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) .TP .B REPLICATIONS_DELAY_INIT initial delay in seconds before starting replications (default is 60) .TP .B REPLICATIONS_RESPECT_TOPOLOGY whether to make undergoal replications respect topology (default is 0) .br 0 - do not respect topology .br 1 - pick destination server at random, but then choose best source server .br 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) .TP .B CREATIONS_RESPECT_TOPOLOGY whether new chunks should be recorded with respect to topology (default is 0) .br 0 - do not respect topology .br 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 .TP .B CHUNKS_UNIQUE_MODE avoid using same ip/rack for different chunk copies (default is 0) .br 0 - ignore ip and rackid (standard behaviour) .br 1 - avoid storing more than one copy on chunkservers using same IP number .br 2 - avoid storing more than one copy on chunkservers using IP number from the same rack id .TP .B CHUNKS_LOOP_MAX_CPS Chunks loop shouldn't check more chunks per seconds than given number (default is 100000) .TP .B CHUNKS_LOOP_MIN_TIME Chunks loop shouldn't be done in less seconds than given number (default is 300) .TP .B CHUNKS_SOFT_DEL_LIMIT Soft maximum number of chunks to delete on one chunkserver (default is 10) .TP .B CHUNKS_HARD_DEL_LIMIT Hard maximum number of chunks to delete on one chunkserver (default is 25) .TP .B CHUNKS_WRITE_REP_LIMIT Maximum number of chunks to replicate to one chunkserver (default is 2,1,1,4 - see NOTES) .TP .B CHUNKS_READ_REP_LIMIT Maximum number of chunks to replicate from one chunkserver (default is 10,5,2,5 - see NOTES) .TP .B CS_HEAVY_LOAD_THRESHOLD Threshold for chunkserver load. (default is 150 - see NOTES) .TP .B CS_HEAVY_LOAD_RATIO_THRESHOLD Threshold ratio for chunkserver load (default is 3.0 - see NOTES) .TP .B CS_HEAVY_LOAD_GRACE_PERIOD Defines how long chunkservers will remain in 'grace' mode (default is 900 - see NOTES) .TP .B CS_MAINTENANCE_MODE_TIMEOUT Maximum number of seconds server can be in maintenance mode (default value is 0 - which means 'forever') .TP .B CS_TEMP_MAINTENANCE_MODE_TIMEOUT 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) .TP .B CS_DAYS_TO_REMOVE_UNUSED How many days unused (disconnected) chunkserver should be kept in master data structures (valid values: 0 - 365 ; 0 means indefinitely ; default value: 7) .TP .B ACCEPTABLE_DIFFERENCE Maximum difference between space usage of chunkservers (deprecated, use \fBACCEPTABLE_PERCENTAGE_DIFFERENCE\fP instead) .TP .B ACCEPTABLE_PERCENTAGE_DIFFERENCE Maximum percentage difference between space usage of chunkservers (default is 1 = 1%) .TP .B PRIORITY_QUEUES_LENGTH Length of priority queues (for endangered, undergoal etc. chunks - chunks that should be processed first - default is 1000000) .TP .B MATOCL_LISTEN_HOST IP address to listen on for client (mount) connections (\fB*\fP means any) .TP .B MATOCL_LISTEN_PORT port to listen on for client (mount) connections .TP .B SESSION_SUSTAIN_TIME How long to sustain a disconnected client session (in seconds; default is 86400 = 1 day) .TP .B QUOTA_TIME_LIMIT Grace period in secods for soft quota (deprecated, use \fBQUOTA_DEFAULT_GRACE_PERIOD\fP instead for default value or specify it individually) .TP .B QUOTA_DEFAULT_GRACE_PERIOD Default grace period in seconds for soft quota (default is 604800 = 7 days) .TP .B ATIME_MODE Set atime modification mode (default is 0 = always modify atime - see NOTES) .TP .B RESERVE_SPACE Set amount of space reserved for superuser (default is 0 = do not reserve space for superuser - see NOTES) .TP .B MAX_ALLOWED_HARD_LINKS Define limit for number of hardlinks allowed for one object (default is 32767; possible values are from 8 to 65000) .TP .B INODE_REUSE_DELAY 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: \fBmfsattrcacheto\fP, \fBmfsxattrcacheto\fP, \fBmfsentrycacheto\fP, \fBmfsdirentrycacheto\fP, \fBmfsnegentrycacheto\fP, \fBmfssymlinkcacheto\fP. (default is 86400; possible values are from 300 to 3000000) .SH NOTES .PP Chunks in master are tested in a loop. Speed (or frequency) is regulated by two options \fBCHUNKS_LOOP_MIN_TIME\fP and \fBCHUNKS_LOOP_MAX_CPS\fP. 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 \fBCHUNK_LOOP_MIN_TIME\fP, but when number of chunks beccomes bigger then time of loop can increase according to \fBCHUNKS_LOOP_MAX_CPS\fP. .PP Example: \fBCHUNKS_LOOP_MIN_TIME\fP is set to 300, \fBCHUNKS_LOOP_MAX_CPS\fP 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. .PP 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. .PP Replication limits are divided into four cases: .IP \[bu] 2 first limit is for endangered chunks (chunks with only one copy) .IP \[bu] 2 second limit is for undergoal chunks (chunks with number of copies lower than specified goal) .IP \[bu] 2 third limit is for rebalance between servers with space usage around arithmetic mean .IP \[bu] 2 fourth limit is for rebalance between other servers (very low or very high space usage) .PP 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). .PP Whenever chunkserver load is higher than \fBCS_HEAVY_LOAD_THRESHOLD\fP and \fBCS_HEAVY_LOAD_RATIO_THRESHOLD\fP times higher than average load, then chunkserver is switched into 'grace' mode. Chunkserver stays in grace mode for \fBCS_HEAVY_LOAD_GRACE_PERIOD\fP seconds. .PP There are five values for \fBATIME_MODE\fP (all other values are treated as 0): .IP \[bu] 2 \fB0\fP = Always modify atime for files, folders and symlinks. .IP \[bu] 2 \fB1\fP = Always modify atime but only in case of files (do not modify atime in case of folders and symlinks). .IP \[bu] 2 \fB2\fP = 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). .IP \[bu] 2 \fB3\fP = Same as above but only in case of files. In case of folders and symlinks do not modify atime. .IP \[bu] 2 \fB4\fP = Never modify atime during access (like "noatime" option). .PP You can reserve space for superuser using \fBRESERVE_SPACE\fP option. You can define it as number of bytes, percent of total space, capacity of biggest chunkserver, etc. .IP \[bu] 2 \fI#\fP or \fI#\fP\fBB\fP = number of bytes reserved for superuser. Standard metric prefixes can be used - SI and IEC (k,K,M,Mi,G,Gi etc.) .IP \[bu] 2 \fI#\fP\fB%\fP or \fI#\fP\fB.\fP\fI#\fP\fB%\fP = percent of total capacity of MooseFS instance .IP \[bu] 2 \fI#\fP\fBU\fP or \fI#\fP\fB.\fP\fI#\fP\fBU\fP = multiplies of "U" value; U is defined as maximum number of bytes currently used by a single chunkserver .IP \[bu] 2 \fI#\fP\fBC\fP or \fI#\fP\fB.\fP\fI#\fP\fBC\fP = multiplies of "C" value; C is defined as maximum total capacity of a single chunkserver .PP 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. .SH COPYRIGHT 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 http://www.gnu.org/licenses/gpl-2.0.html .SH "SEE ALSO" .BR mfsmaster (8), .BR mfsexports.cfg (5) .BR mfstopology.cfg (5)