756 lines
32 KiB
Plaintext
756 lines
32 KiB
Plaintext
# This is a config file for BeeGFS clients.
|
|
# http://www.beegfs.com
|
|
|
|
|
|
# --- [Table of Contents] ---
|
|
#
|
|
# 1) Settings
|
|
# 2) Mount Options
|
|
# 3) Basic Settings Documentation
|
|
# 4) Advanced Settings Documentation
|
|
|
|
|
|
#
|
|
# --- Section 1.1: [Basic Settings] ---
|
|
#
|
|
|
|
sysMgmtdHost =
|
|
|
|
|
|
#
|
|
# --- Section 1.2: [Advanced Settings] ---
|
|
#
|
|
|
|
connAuthFile = /etc/beegfs/conn.auth
|
|
connDisableAuthentication = false
|
|
connClientPort = 8004
|
|
connMgmtdPort = 8008
|
|
connPortShift = 0
|
|
|
|
connCommRetrySecs = 600
|
|
connFallbackExpirationSecs = 900
|
|
connInterfacesFile =
|
|
connRDMAInterfacesFile =
|
|
connMaxInternodeNum = 12
|
|
connMaxConcurrentAttempts = 0
|
|
connNetFilterFile =
|
|
|
|
connUseRDMA = true
|
|
connTCPFallbackEnabled = true
|
|
connTCPRcvBufSize = 0
|
|
connUDPRcvBufSize = 0
|
|
connRDMABufNum = 70
|
|
connRDMABufSize = 8192
|
|
connRDMAFragmentSize = page
|
|
connRDMATypeOfService = 0
|
|
connTcpOnlyFilterFile =
|
|
|
|
logClientID = false
|
|
logLevel = 3
|
|
|
|
quotaEnabled = false
|
|
|
|
sysCacheInvalidationVersion = true
|
|
sysCreateHardlinksAsSymlinks = false
|
|
sysMountSanityCheckMS = 11000
|
|
sysSessionCheckOnClose = false
|
|
sysSyncOnClose = false
|
|
sysTargetOfflineTimeoutSecs = 900
|
|
sysUpdateTargetStatesSecs = 30
|
|
sysXAttrsEnabled = false
|
|
|
|
tuneFileCacheType = buffered
|
|
tunePreferredMetaFile =
|
|
tunePreferredStorageFile =
|
|
tuneRemoteFSync = true
|
|
tuneUseGlobalAppendLocks = false
|
|
tuneUseGlobalFileLocks = false
|
|
|
|
|
|
#
|
|
# --- Section 1.3: [Enterprise Features] ---
|
|
#
|
|
# See end-user license agreement for definition and usage limitations of
|
|
# enterprise features.
|
|
#
|
|
|
|
sysACLsEnabled = false
|
|
|
|
|
|
#
|
|
# --- Section 2: [Mount Options] ---
|
|
#
|
|
|
|
# Valid mount options are:
|
|
# cfgFile, logLevel, connPortShift, connMgmtdPort,
|
|
# sysMgmtdHost, sysMountSanityCheckMS, connInterfacesList.
|
|
#
|
|
# Use the mount option "cfgFile" to specify a different config file
|
|
# for the beegfs client.
|
|
# Example:
|
|
# $ /bin/mount -t beegfs beegfs_nodev /beegfs -ocfgFile=/etc/anotherconfig.conf
|
|
#
|
|
# Use the mount option "connInterfacesList" to pass the list of interfaces names.
|
|
# These interfaces names should be space-separated.
|
|
# Example:
|
|
# $ /bin/mount -t beegfs beegfs_nodev /beegfs \
|
|
# -ocfgFile=/etc/anotherconfig.conf,connInterfacesList='ib0 eth0'
|
|
#
|
|
# Mount options override the corresponding config file values.
|
|
# Example:
|
|
# $ /bin/mount -ocfgFile=/etc/anotherconfig.conf,logLevel=3 ...
|
|
|
|
|
|
#
|
|
# --- Section 3: [Basic Settings Documentation] ---
|
|
#
|
|
|
|
# [sysMgmtdHost]
|
|
# Hostname (or IP) of the host running the management service.
|
|
# (See also "connMgmtdPort".)
|
|
# Default: <none>
|
|
|
|
|
|
#
|
|
# --- Section 4: [Advanced Settings Documentation] ---
|
|
#
|
|
|
|
#
|
|
# --- Section 4.1: [Connections & Communication] ---
|
|
#
|
|
|
|
# [connAuthFile]
|
|
# The path to a file that contains a shared secret for connection based
|
|
# authentication. Only peers that use the same shared secret will be able to
|
|
# connect.
|
|
# Default: <none>
|
|
|
|
# [connDisableAuthentication]
|
|
# If set to true, explicitly disables connection authentication and allow the
|
|
# service to run without a connAuthFile. Running BeeGFS without connection
|
|
# authentication is considered insecure and is not recommended.
|
|
# Default: false
|
|
|
|
# [connClientPort]
|
|
# The UDP port of the client.
|
|
# Default: 8004
|
|
|
|
# [connMgmtdPort]
|
|
# The UDP and TCP port of the management node.
|
|
# Default: 8008
|
|
|
|
# [connPortShift]
|
|
# Shifts all following UDP and TCP ports according to the specified value.
|
|
# Intended to make port configuration easier in case you do not want to
|
|
# configure each port individually.
|
|
# Default: 0
|
|
|
|
# [connCommRetrySecs]
|
|
# The time (in seconds) for retries in case a network communication fails
|
|
# (e.g. if a server is down). After this time, the I/O operation will fail
|
|
# and the calling process will receive an error.
|
|
# Note: Set this value to 0 for infinite retries. In this case, a process
|
|
# accessing the file system will block until the corresponding server
|
|
# becomes available (or until it is interrupted by a signal).
|
|
# Default: 600
|
|
|
|
# [connFallbackExpirationSecs]
|
|
# The time in seconds after which a connection to a fallback interface expires.
|
|
# When a fallback connection expires, the system will try to establish a new
|
|
# connection to the other hosts primary interface (falling back to another
|
|
# interface again if necessary).
|
|
# Note: The priority of node interfaces can be configured using the
|
|
# "connInterfacesFile" parameter.
|
|
# Default: 900
|
|
|
|
# [connInterfacesFile]
|
|
# The path to a text file that specifies the names of the interfaces, which
|
|
# may be used for communication by other nodes. One interface per line. The
|
|
# line number also defines the priority of an interface.
|
|
# Example: "ib0" in the first line, "eth0" in the second line.
|
|
# Values: This setting is optional. If unspecified, all available interfaces
|
|
# will be published and priorities will be assigned automatically.
|
|
# Note: This has no influence on outgoing connections. The information is sent
|
|
# to other hosts to inform them about possible communication paths.
|
|
# Default: <none>
|
|
|
|
# [connInterfacesList]
|
|
# Comma-separated list of interface names. Performs the same function as
|
|
# connInterfacesFile.
|
|
# If use as the mount option "connInterfacesList" to pass the list of interfaces
|
|
# names then it override the corresponding config file/list values..
|
|
# The interfaces names should be space-separated.
|
|
# Example:
|
|
# $ /bin/mount -t beegfs beegfs_nodev /beegfs \
|
|
# -ocfgFile=/etc/anotherconfig.conf,connInterfacesList='ib0 eth0'
|
|
#
|
|
# Default: <none>
|
|
|
|
# [connRDMAInterfacesFile]
|
|
# The path to a text file that specifies the names of IPoIB interfaces, which
|
|
# may be used for outbound RDMA communication with other nodes. One interface
|
|
# per line. These interfaces must be RDMA-capable NICs.
|
|
#
|
|
# All storage and metadata servers must be IP-reachable from each specified
|
|
# interface.
|
|
#
|
|
# Specifying interfaces in this file limits which RDMA NICs are used for outbound
|
|
# RDMA. Specifying multiple interfaces allows the client to use multiple RDMA
|
|
# interfaces for outbound communication.
|
|
#
|
|
# Example: "ib0" in the first line, "ib1" in the second line.
|
|
# Values: This setting is optional. When none are specified, the client will use
|
|
# the first client host interface that can reach the remote node via IPoIB,
|
|
# as decided by rdma_cm. When multiple interfaces are specified, the client
|
|
# round-robins creation of outbound RDMA connections across the specified
|
|
# interfaces.
|
|
# Default: <none>
|
|
|
|
# [connMaxInternodeNum]
|
|
# The maximum number of simultaneous connections to the same node.
|
|
# Default: 12
|
|
|
|
# [connMaxConcurrentAttempts]
|
|
# The maximum number of simultaneous connection attempts. This may help in case
|
|
# establishing new connections keeps failing and produces fallbacks.
|
|
# It may happen particularly when using RDMA in an Omni-Path setup. If you
|
|
# don't have failing connection attempts, tuning this option might still lead
|
|
# to a faster connection process. This option is experimental, so there is no
|
|
# experience with different values. Setting it to 0 disables it, which means
|
|
# concurrent connection attempts are not limited.
|
|
# Default: 0
|
|
|
|
# [connNetFilterFile]
|
|
# The path to a text file that specifies allowed IP subnets, which may be used
|
|
# for outgoing communication. One subnet per line in classless notation (IP
|
|
# address and number of significant bits).
|
|
# Example: "192.168.10.0/24" in the first line, "192.168.20.0/24" in the second
|
|
# line.
|
|
# This value is optional. If unspecified, all addresses are allowed for
|
|
# outgoing communication.
|
|
# Default: <none>
|
|
|
|
# [connUseRDMA]
|
|
# Enables the use of Remote Direct Memory Access (RDMA) for InfiniBand or RoCE.
|
|
# For this setting to be effective, OFED ibverbs support also has to be enabled
|
|
# at compile time of the beegfs client modules (typically via
|
|
# beegfs-client-autobuild.conf).
|
|
# Default: true
|
|
|
|
# [connTCPFallbackEnabled]
|
|
# Enables fallback from RDMA to TCP sockets when there is a problem connecting
|
|
# via RDMA to a storage or meta node.
|
|
# Default: true
|
|
|
|
# [connTCPRcvBufSize], [connUDPRcvBufSize]
|
|
# Sets the size for TCP and UDP socket receive buffers (SO_RCVBUF). The maximum
|
|
# allowed value is determined by sysctl net.core.rmem_max. This value is ignored
|
|
# if it is less than the default value determined by net.core.rmem_default.
|
|
# For legacy reasons, the default value 0 indicates that the buffer size is set
|
|
# to connRDMABufNum * connRDMABufSize.
|
|
# -1 indicates that the buffer size should be left at the system default.
|
|
# Default: 0
|
|
|
|
# [connRDMABufNum], [connRDMABufSize], [connRDMAFragmentSize]
|
|
# InfiniBand RDMA buffer settings.
|
|
# connRDMABufSize is the maximum size of a buffer (in bytes) that will be sent
|
|
# over the network; connRDMABufNum is the number of available buffers that can
|
|
# be in flight for a single connection. These client settings are also applied
|
|
# on the server side for each connection. Ideally, the largest, commonly used
|
|
# filesytem chunksize should be < connRDMABufNum * connRDMABufSize to achieve
|
|
# the best performance.
|
|
#
|
|
# The minimum usable value for connRDMABufNum is 3, which is required by the
|
|
# BeeGFS RDMA protocol. Lower values will immediately result in communication
|
|
# failures.
|
|
#
|
|
# connRDMAFragmentSize determines how contiguous memory is allocated per
|
|
# buffer. If connRDMAFragmentSize=4096 and connRDMABufSize=8192, each buffer
|
|
# is allocated in 2 regions of 4096 contiguous bytes. Less fragmentation
|
|
# improves performance. The value "none" indicates that buffers will not be
|
|
# fragmented, resulting in allocation of contiguous regions of
|
|
# connRDMABufSize. The value "page" uses the Linux PAGE_SIZE as the
|
|
# fragmentation value.
|
|
# The reason for using fragmentation is that large allocations
|
|
# are more likely to fail if there is a shortage of heap memory. The minimum
|
|
# fragmentation value is PAGE_SIZE. Using larger values (or 0) should improve
|
|
# performance and allows for larger values of connRDMABufSize.
|
|
#
|
|
# Note: RAM usage per connection is connRDMABufSize x connRDMABufNum x 2. Keep
|
|
# resulting RAM usage (x connMaxInternodeNum x number_of_clients) on the
|
|
# server in mind when increasing these values.
|
|
# Default: 70, 8192, page
|
|
|
|
# [connRDMAMetaBufNum], [connRDMAMetaBufSize], [connRDMAMetaFragmentSize]
|
|
# InfiniBand RDMA buffer settings for connections to beegfs-meta.
|
|
# These settings behave in the same way as connRDMABufNum, connRDMABufSize
|
|
# and connRDMAFragmentize except that they are used for connections to
|
|
# beegfs-meta.
|
|
# Metadata messages are usuallly small and do not require the large amount
|
|
# of buffer space that is typically configured for connections to
|
|
# beegfs-storage. One exception to this would be if large extended attributes
|
|
# are added to files.
|
|
# connRDMAMetaBufNum = "default" indicates that connRDMABufNum should be used.
|
|
# connRDMAMetaBufSize = "default" indicates that connRDMABufSize should be used.
|
|
# connRDMAMetaFragmentSize = "default" indicates that connRDMAFragmentSize
|
|
# should be used.
|
|
# The minimum value for connRDMAMetaBufNum is 3.
|
|
# Default: default, default, default
|
|
|
|
# [connRDMATypeOfService]
|
|
# InfiniBand provides the option to set a type of service for an application.
|
|
# This type of service can be used by your subnet manager to provide Quality of
|
|
# Service functionality (e.g. setting different service levels).
|
|
# In openSM the service type will be mapped to the parameter qos-class, which
|
|
# can be handled in your QoS configuration.
|
|
# See
|
|
# www.openfabrics.org/downloads/OFED/ofed-1.4/OFED-1.4-docs/
|
|
# QoS_management_in_OpenSM.txt
|
|
# for more information on how to configure openSM for QoS.
|
|
# This parameter sets the type of service for all outgoing connections of this
|
|
# daemon.
|
|
# Default: 0 (Max: 255)
|
|
|
|
# [connRDMAKeyType]
|
|
# In RDMA, an "rkey" is used to provide an access token for a peer to access
|
|
# local memory regions that are registered for RDMA. Historically,
|
|
# BeeGFS used either a "DMA key" or an "unsafe global rkey" depending upon
|
|
# whether or not "unsafe global rkey" is supported by the operating system.
|
|
# This is now selectable. "DMA key" is not supported on kernel >= 4.9
|
|
# unless MOFED is installed. If an unsupported option is specified, there
|
|
# will be warnings in syslog and RDMA connections will not be established.
|
|
# Use of "unsafe global rkey" is preferred, but generates a syslog message
|
|
# every time an RDMA connection is established: "enabling unsafe global rkey".
|
|
# Neither option is considered "safe" because they both provide access
|
|
# to all DMA mapped memory for a given connection. This technique is
|
|
# used to provide better performance for small I/O requests.
|
|
# "register" uses a memory registration per connection to provide an rkey.
|
|
# "register" is not compatible with NVIDIA GPUDirect Storage.
|
|
# Specify "dma" or "register" to squelch the syslog warning.
|
|
# Values: "global" (unsafe global rkey), "dma" (DMA key), "register"
|
|
# (memory registration)
|
|
# Default: "global"
|
|
|
|
# [connTcpOnlyFilterFile]
|
|
# The path to a text file that specifies IP address ranges to which no RDMA
|
|
# connection should be established. This is useful e.g. for environments where
|
|
# all hosts support RDMA, but some hosts cannot connect via RDMA to some other
|
|
# hosts.
|
|
# Example: "192.168.10.0/24" in the first line, "192.168.20.0/24" in the second
|
|
# line.
|
|
# Values: This setting is optional.
|
|
# Default: <none>
|
|
|
|
# [connMessagingTimeouts]
|
|
# These constants are used to set some of the connection timeouts for sending
|
|
# and receiving data between services in the cluster. They used to be hard-coded
|
|
# (CONN_LONG_TIMEOUT, CONN_MEDIUM_TIMEOUT and CONN_SHORT_TIMEOUT) but are now
|
|
# made configurable for experimentation purposes.
|
|
# This option takes three integer values of milliseconds, separated by a comma
|
|
# in the order long, medium, short.
|
|
# WARNING: This is an EXPERIMENTAL configuration option that should not be
|
|
# changed in production environments unless properly tested and validated.
|
|
# Some configurations can lead to service lockups and other subtle issues.
|
|
# Please make sure that you know exactly what you are doing and properly
|
|
# test any changes you make.
|
|
# Default: 600000,90000,30000
|
|
|
|
# [connRDMATimeouts]
|
|
# These constants are used to set some of the timeouts for sending and receiving
|
|
# data between services in the cluster via RDMA. They used to be
|
|
# hard-coded IBVSOCKET_CONN_TIMEOUT_MS, IBVSOCKET_COMPLETION_TIMEOUT_MS,
|
|
# IBVSOCKET_FLOWCONTROL_ONSEND_TIMEOUT_MS,
|
|
# IBVSOCKET_FLOWCONTROL_ONRECV_TIMEOUT_MS and a 10000 literal for poll timeout
|
|
# but are now made configurable for experimentation purposes.
|
|
# This option takes five integer values of milliseconds, separated by a comma
|
|
# in the order connectMS, completionMS, flowSendMS, flowRecvMS and pollMS.
|
|
# WARNING: This is an EXPERIMENTAL configuration option that should not be
|
|
# changed in production environments unless properly tested and validated.
|
|
# Some configurations can lead to service lockups and other subtle issues.
|
|
# Please make sure that you know exactly what you are doing and properly
|
|
# test any changes you make.
|
|
# Default: 5000,300000,180000,180000,10000
|
|
|
|
|
|
# --- Section 4.2: [Logging] ---
|
|
#
|
|
|
|
# [logClientID]
|
|
# Defines whether the ClientID should appear in each log line.
|
|
# This is mainly helpful if BeeGFS is mounted multiple times on this machine.
|
|
# Default: false
|
|
|
|
# [logLevel]
|
|
# Defines the amount of log messages. The higher this level, the more detailed
|
|
# the log messages will be.
|
|
# Level 3 will print connection messages, level 4 will print syscall messages,
|
|
# level 5 will print debug messages.
|
|
# Note: Levels above 3 might decrease performance.
|
|
# Default: 3 (Max: 5)
|
|
|
|
#
|
|
# --- Section 4.3: [Quota Settings] ---
|
|
#
|
|
|
|
# [quotaEnabled]
|
|
# Enables user and group quota support by transferring extra user data to the
|
|
# servers. This uses quota information of the underlying file systems on the
|
|
# storage servers, which needs to be enabled by the server administrator.
|
|
# Note: In the first implementation, only quota monitoring is available.
|
|
# Note: Get quota information with "beegfs-ctl --getquota".
|
|
# Note: If this option is true, performance might be slightly decreased due to
|
|
# extra information tracking.
|
|
# Default: false
|
|
|
|
|
|
#
|
|
# --- Section 4.4: [System Settings] ---
|
|
#
|
|
|
|
# [sysCacheInvalidationVersion]
|
|
#
|
|
# Enable the client to invalidate its cache and reload the inode of a file when
|
|
# the version parameter of the file on metadata changes due to internal metadata
|
|
# operations such as stripe pattern change. This is done by comparing the
|
|
# version parameters on client and metadata on the first lookup after internal
|
|
# metadata changes. If the versions differ, the client invalidates the cache and
|
|
# reloads the inode.
|
|
# Note: If this option is set to true, performance may be decreased.
|
|
# Default: true
|
|
|
|
# [sysCreateHardlinksAsSymlinks]
|
|
# Create a symlink when an application tries to create a hardlink for files in
|
|
# different directories.
|
|
# Default: false
|
|
|
|
# [sysMountSanityCheckMS]
|
|
# Perform some basic checks during mount (e.g. whether the client helper daemon
|
|
# and storage servers are reachable). Mounting will fail if a problem is
|
|
# detected.
|
|
# Values: Set the time (in ms) you want to spend waiting for the servers
|
|
# (especially the management daemon) to respond. Use 0 to disable all checks
|
|
# and allow mounting even if no servers are reachable.
|
|
# Default: 11000
|
|
|
|
# [sysSessionCheckOnClose]
|
|
# Checks for a valid session on the storage servers when a file is closed. If
|
|
# this option is set to true, a potential cache loss from a crash of a storage
|
|
# server can be detected. This will be reported to the user application as a
|
|
# close() error code.
|
|
# Note: There is also a session check included in all read/write/fsync messages,
|
|
# which is independent of this setting.
|
|
# Note: If this option is set to true, more network messages are required on
|
|
# close(), so performance will decreased.
|
|
# Default: false
|
|
|
|
# [sysSessionChecksEnabled]
|
|
# Enable session checks in read/write/fsync operations to be able to detect
|
|
# server crashes that could have caused a loss of server side caches. Disabling
|
|
# these checks is useful in certain system configurations to be able to cleanly
|
|
# resume I/O after a server crash/unclean failover.
|
|
# WARNING: Disabling session checks can lead to undetected cache loss and
|
|
# therefore silent data corruption on the storage servers. Only disable the
|
|
# checks if absolutely necessary and if there are measures in place to prevent
|
|
# cache loss (synchronous mounts, battery backed caches) on the storage servers.
|
|
# Default: true
|
|
|
|
# [sysSyncOnClose]
|
|
# Sync file contents on close. If this option is set to true, the storage
|
|
# servers will flush the write cache of a file to disk immediately when it is
|
|
# closed by the application. If this option is set to false, the write cache
|
|
# will be flushed to disk asynchronously after a few seconds.
|
|
# Note: If this option is true, performance will be decreased.
|
|
# Default: false
|
|
|
|
# [sysTargetOfflineTimeoutSecs]
|
|
# Timeout until all storage targets and metadata nodes are considered offline
|
|
# when no target state updates can be fetched from the management server.
|
|
# If this value is 0, targets will never be set to offline due to an
|
|
# unreachable management node and will stay in state probably-offline.
|
|
# Note: This must be at least twice as large as the value of
|
|
# sysTargetOfflineTimoutSecs in the server config files.
|
|
# Values: time in seconds
|
|
# Default: 900
|
|
|
|
# [sysUpdateTargetStatesSecs]
|
|
# Interval in which the states of the storage targets are checked.
|
|
# Note: This must be significantly shorter than the sysTargetOfflineTimeoutSecs
|
|
# value set in the server (recommendation: maximum 1/3 of it).
|
|
# Values: time in seconds
|
|
# Default: 30
|
|
|
|
# [sysXAttrsEnabled]
|
|
# Enable extended attributes (also know as EAs/xattrs).
|
|
# Default: false
|
|
|
|
# [sysXAttrsCheckCapabilities]
|
|
# Check inodes for existing "security.capability" extended attribute and
|
|
# optionally cache to reduce metadata requests and increase write performance.
|
|
# The Linux kernel uses a security mechanism that automatically removes
|
|
# setuid/setgid bits and capabilities from files when they are changed. This is
|
|
# done to prevent users from executing binaries with elevated privileges that
|
|
# were changed after the privileges were originally set. That mechanism requires
|
|
# that, by default, the kernel has to check each file for existing capabilities
|
|
# on every write which leads to a large overhead in metadata RPCs to fetch the
|
|
# "security.capability" extended attribute. To optimize this, Linux allows file
|
|
# systems to set a flag (S_NOSEC) on the file, which short-circuits these
|
|
# checks.
|
|
# This configuration option configures the file system mount to allow the
|
|
# client to either always check, check once and cache that flag after a first
|
|
# lookup of the extended attribute returns an empty result, or set the flag on
|
|
# inode creation and never check. The flag will automatically be cleared when
|
|
# capabilities are modified on this client. It will, however, currently not
|
|
# be cleared when a different client modifies capabilities or sets setuid/setgid
|
|
# bits, which can lead to capabilities not being cleared, even after the file is
|
|
# written to. If this is a concern, this option should be set to "always".
|
|
# As long as BeeGFS is mounted using the "nosuid" mount option (which is
|
|
# recommended and the default setting), elevating privileges via setuid/setgid
|
|
# bits and capabilities is disabled and it is safe to set this option to "never".
|
|
# Possible values:
|
|
# always (always check for security xattr, never cache the result)
|
|
# cache (check for security xattr once, then cache)
|
|
# never (mark new inodes immediately, never check security xattr)
|
|
# Default: never
|
|
|
|
# [sysBypassFileAccessCheckOnMeta]
|
|
# Allow this client to bypass file access restrictions enforced by metadata
|
|
# servers. When enabled, the client is permitted to open files even if access
|
|
# restrictions are currently in place (e.g., file marked read-only,
|
|
# write-locked, or fully restricted). This setting is primarily intended for
|
|
# specialized clients in controlled environments, such as HSM (Hierarchical
|
|
# Storage Management) systems that need to restore file data, or backup/recovery
|
|
# tools that require access to otherwise locked files.
|
|
# Note: This setting only affects metadata-level access checks and has no effect
|
|
# on other security or permission mechanisms.
|
|
# Default: false
|
|
|
|
|
|
# [sysACLsEnabled]
|
|
# Allow the creation and enforcement of Access Control Lists (ACLs).
|
|
# Note: Only works if sysXAttrsEnabled=true.
|
|
# Note: Requires at least Linux kernel version 3.2.
|
|
# Note: Enabling this setting can affect metadata performance.
|
|
# Default: false
|
|
|
|
# [sysFileEventLogMask]
|
|
# Specifies which file system events shall be logged by the metadata servers. If
|
|
# unset, this client doesn't send log events at all. This can either be "none"
|
|
# or a comma separated list of event types to log. The following event types
|
|
# (and any comma-separated combination) are possible:
|
|
# "flush" (explicit data flushes on files), "close" (close writable file),
|
|
# "trunc" (file truncation), "setattr" (set attributes),
|
|
# "link-op" (create,mkdir,mknod,create symlink,create hardlink,rmdir,unlink, rename),
|
|
# "read" (deprecated, see notes) or "open-read" (file opened in read-only mode),
|
|
# "open-write" (file opened in write only mode),
|
|
# "open-readwrite" (file opened for both read and write),
|
|
# Note: If a client opens a file multiple times, "close" will only generate an
|
|
# event if the last fd is closed. If events for each fd shall be generated,
|
|
# "flush" needs to be used. However, "flush" might have a small performance
|
|
# impact. Also note that "read" is deprecated for file opening in read-only mode,
|
|
# but it is still allowed for backward compatibility. It is recommended to use
|
|
# "open-read" instead of "read" for clarity and consistency.
|
|
# Example: sysFileEventLogMask = close,trunc,setattr,link-op,open-read
|
|
# Default: none
|
|
|
|
# [sysRenameEbusyAsXdev]
|
|
# Changes the semantics of rename() to return an EXDEV error if a file could not
|
|
# be moved because it is in use (instead of the default EBUSY). Applications and
|
|
# tools like mv can handle EXDEV and fall back to copy/unlink for the files.
|
|
# This is mostly useful for NFS exports, where files may not be closed by the
|
|
# server until after the last open file handle has been closed by clients. This
|
|
# can cause spurious EBUSY errors in clients that close a file and rename it
|
|
# immediately afterwards.
|
|
# Default: false
|
|
|
|
|
|
#
|
|
# --- Section 4.5: [Tuning] ---
|
|
#
|
|
|
|
# [tuneFileCacheType]
|
|
# Sets the file read/write cache type.
|
|
# Values: "none" (disable client caching), "buffered" (use a pool of small
|
|
# static buffers for write-back and read-ahead), "native" (use the kernel
|
|
# pagecache), "paged" (experimental, deprecated).
|
|
# Note: The cache protocols are currently non-coherent (but caches are
|
|
# automatically flushed when a file is closed).
|
|
# Note: When client and servers are running on the same machine, "paged" mode
|
|
# contains the typical potential for memory allocation deadlocks (also known
|
|
# from local NFS server mounts). So do not use "paged" mode for clients that
|
|
# run on a metadata or storage server machine.
|
|
# Default: buffered
|
|
|
|
# [tunePreferredMetaFile], [tunePreferredStorageFile]
|
|
# Path to a text file that contains the numeric IDs of preferred storage targets
|
|
# and metadata servers. These will be preferred when the client creates new file
|
|
# system entries. This is useful e.g. to take advantage of data locality in the
|
|
# case of multiple data centers. If unspecified, all available targets and
|
|
# servers will be used equally.
|
|
# Usage: One targetID per line for storage servers, one nodeID per line for
|
|
# metadata servers.
|
|
# Note: TargetIDs and nodeIDs can be queried with the beegfs-ctl tool.
|
|
# Default: <none>
|
|
|
|
# [tuneRemoteFSync]
|
|
# Controls whether fsync() syscalls from a user application should only be
|
|
# executed on the client to transfer data from the client cache to server
|
|
# cache (=false); or also on the servers to flush the server's cached file
|
|
# data to the disks (=true).
|
|
# Default: true
|
|
|
|
# [tuneUseGlobalAppendLocks]
|
|
# Controls whether files opened in append mode should be protected by locks on
|
|
# the local machine only (=false) or globally on the servers (=true).
|
|
# Default: false
|
|
|
|
# [tuneUseGlobalFileLocks]
|
|
# Controls whether application advisory file locks via flock() and fcntl()
|
|
# should be checked for conflicts on the local machine only (=false) or
|
|
# globally on the servers (=true).
|
|
# Default: false
|
|
|
|
# [tuneCoherentBuffers]
|
|
# Enables or disables coherence between the buffers used by
|
|
# tuneFileCacheType=buffered and the page cache.
|
|
# If a file is concurrently accessed via mmap() regions and read()/write()
|
|
# system calls, the buffers used by tuneFileCacheType=buffered and the page
|
|
# cache used by mmap() may go out of sync - changes made in an mmap()ed region
|
|
# may not be visible to read() calls immediately, or changes made by write()
|
|
# calls may not be immediately reflected in mmap()ed regions.
|
|
# Many programs that use both methods of accessing a file assume that
|
|
# read()/write() and mmap() present the same view of the file, if this is not
|
|
# the case, those programs may not work correctly. Programs that have been
|
|
# observed to misbehave with non-coherent buffers are, for example, git and
|
|
# some in-memory database applications.
|
|
# When this option is enabled, files that are currently mmap()ed will behave as
|
|
# though they had been opened with tuneFileCacheType=none
|
|
# Default: true
|
|
|
|
|
|
#
|
|
# --- Section 5: [Expert options]
|
|
#
|
|
|
|
# [connUnmountRetries]
|
|
# Retry communications during unmount.
|
|
# If this option is set to `true` and a communication error occurs during
|
|
# unmont, for example due to a transient network fault, the unsuccessful
|
|
# communications will be retried normally. When set to `false` they will not be
|
|
# retried; this leads to a quicker unmount, but resources allocated to current
|
|
# client will not be freed for a few hours.
|
|
# Default: true
|
|
|
|
# [tuneFileCacheBufSize]
|
|
# When using buffered mode: maximum size of the (contiguous) data cache for an
|
|
# open file.
|
|
# When using native mode: threshold for direct operations. If a read() or
|
|
# write() passes a buffer size larger than tuneFileCacheBufSize the client will
|
|
# bypass the page cache and send/receive the data directly to/from the storage
|
|
# servers.
|
|
# Default: 524288 (512KiB)
|
|
|
|
# [tuneFileCacheBufNum]
|
|
# When using buffered mode: maximum number of file caches to preallocate
|
|
# for the mount. When a file is opened a cache is allocated, up to this number.
|
|
# If the maximum number of caches is reached no cache is allocated and all
|
|
# read/write operations for the file go to the storage servers directly.
|
|
# Default: 4*(number of CPUs)
|
|
|
|
# [tunePageCacheValidityMS]
|
|
# Maximum lifetime of cached data in the page cache.
|
|
# In buffered mode the page cache is used for mmap(), in native mode the page
|
|
# cache is used for all data. Data in the page cache that was not yet written
|
|
# to the storage server is written after at most this time, data that was read
|
|
# but not modified is discarded.
|
|
# Default: 2000000000 (approx. 23 days)
|
|
|
|
# [tuneDirSubentryCacheValidityMS]
|
|
# Validity time of directory attribute data, in milliseconds.
|
|
# Attributes of directories (eg stat() data) that have been loaded from the
|
|
# metadata servers are assumed to be valid for this amount of time without
|
|
# requiring a refresh. Once the time has passed the next access will cause a
|
|
# refresh.
|
|
# Default: 1000
|
|
|
|
# [tuneFileSubentryCacheValidityMS]
|
|
# Validity time of file attribute data, in milliseconds.
|
|
# Attributes of files (eg stat() data) that have been loaded from the metadata
|
|
# servers are assumed to be valid for this amount of time without requiring a
|
|
# refresh. Once the time has passed the next access will cause a refresh.
|
|
# Default: 0
|
|
|
|
# [tuneENOENTCacheValidityMS]
|
|
# Validity time of the non-existing file in milliseconds.
|
|
# A negative result of a stat call indicating "No such file or directory"
|
|
# (ENOENT) is assumed to be valid for this amount of time without requiring a
|
|
# new request to the meta server. Once the time has passed the next access will
|
|
# cause a refresh.
|
|
# Default: 0
|
|
|
|
# [tunePathBufSize]
|
|
# Size of buffers used for constructing paths.
|
|
# Whenever a full path must be constructed (eg for log messages) a preallocated
|
|
# buffer of this size will be used.
|
|
# Default: 4096
|
|
|
|
# [tunePathBufNum]
|
|
# Number of path buffers for path construction.
|
|
# Determines how many path buffers are preallocated during mount. If no buffers
|
|
# are available for an operation the operation must wait for another thread to
|
|
# free enough buffers.
|
|
# Default: 8
|
|
|
|
# [tuneMsgBufSize]
|
|
# Size of buffers used for messaging.
|
|
# Messages sent and received by the client (except logging messages) use
|
|
# buffers preallocated at mount time. Buffers are allocated with the size
|
|
# given here.
|
|
# Default: 65536
|
|
|
|
# [tuneMsgBufNum]
|
|
# Number of message buffers.
|
|
# During mount this many message buffers are preallocated. If an operation
|
|
# requires communication with a server but all buffers are used, the operation
|
|
# must wait until a buffer is released.
|
|
# Default: 4*(number of CPUs) + 1
|
|
|
|
# [tuneRefreshOnGetAttr]
|
|
# If set to `true`, file attributes will be loaded from the server on each call
|
|
# to fstat(). When set to `false` a call to fstat() may return stale
|
|
# information for files that are not currently open; this can happen mainly
|
|
# when NFS exports are used.
|
|
# Default: false
|
|
|
|
# [tuneInodeBlockBits]
|
|
# Sets the block size of file on the mountpoint to 2**tuneInodeBlockBits.
|
|
# Default: 19 (512KiB)
|
|
|
|
# [tuneEarlyCloseResponse]
|
|
# Request close responses from the metadata server before the file is fully closed.
|
|
# This may improve close() performance, but closed files may be accounted as
|
|
# open for a short time after close() has returned. Files accounted as open
|
|
# cannot be moved.
|
|
# Default: false
|
|
|
|
# [tuneUseBufferedAppend]
|
|
# Used only buffered mode. If set, writes to files opened with O_APPEND will be
|
|
# cached. Ignored unless tuneUseGlobalAppendLocks is also set.
|
|
# Default: true
|
|
|
|
# [tuneStatFsCacheSecs]
|
|
# Validity time of statfs() results, in seconds.
|
|
# Results of statfs(), once queried from the storage servers, will be cached
|
|
# for this amount of time.
|
|
# Default: 10
|
|
|
|
# [sysInodeIDStyle]
|
|
# Sets the hash function used to compute inode numbers from metadata IDs.
|
|
# The *32 options produce 32 bit inodes numbers, the *64 variants produce 64
|
|
# bit inode numbers.
|
|
# Possible values:
|
|
# hash32
|
|
# hash64
|
|
# md4hash32
|
|
# md4hash64
|
|
# Default: md4hash64
|