Dear MooseFS Users, Long-awaited MooseFS 3.0.107 is now stable!

This release is mainly a bugfix release (and includes a lot of fixes). We strongly recommend to upgrade to this latest version! This release also introduces some new features, including support for libfuse3.

We would like to thank all the contributors who helped us making this release, especially (alphabetically): @eleaner, @hradec, @itongqingbo, @njaard, @onlyjob, @steffenb7333. Thank you for all your help!

(If I forgot to mention somebody, I am very sorry and please let me know!)

Please find the list of fixes and new features that this release includes below:

• (master) added NOP command to changelog
• (master+supervisor) added time check
• (cs) fixed behaviour when job queue is full
• (cli+cgi) added time data to master servers
• (cs) fixed NOP packets ping-pong
• (client) fixed NOP flooding in write module
• (cs) fixed race condition in bgjobs module (jobhash)
• (cs+mount) added new status returned to mount when queue is full and support for it in client module
• (mount) added initial support for libfuse3 (related to issue #302)
• (mount) added support for readdirplus (related to issue #302)
• (mount) workaround for bug in fuse (wrong lock_owner in flush – issue #305)
• (mount) fixed deadlock in readdata.c (closes issue #306)
• (mount+master) fixed few small issues related to POSIX compliance (checked by pjd fstest)
• (all) changed preferred C standard from C99 to C11 and added tests for atomics
• (mount) move read/write structures initialization from open to read/write
• (mount,mfsio) added new inode->fleng map
• (configure) added option disable_all, added summary at the end
• (freebsd) detecting new fuse version in the kernel and adjusting behaviour
• (tools) better handling of unrecognized options
• (master) better hard links handling
• (master) added limit for number of hardlinks allowed for one object
• (master) added option for automatic removal of unused chunkservers (issue #247)
• (metadirinfo) fixed segfault when output filename is not specified
• (chunktool) added file name fixing option (issue #276)
• (all) changed chunk replication and deletion charts into two-color charts (successful/unsuccessful – issue #273)
• (all) added new charts with number of chunk creations and internal changes (also successful and unsuccessful)
• (master) added ignoring chunk with id zero sent from chunkserver
• (metadump) added option for dumping parts of metadata
• (master) added more logs to background data write subprocess (related to issue #280)
• (master) fixed LEADER/FOLLOWER desync on SETCHUNK command (pro only)
• (cli+cgi) fixed shebang in python scripts (preference for python3 in configure)
• (master) changed pipe in bgsaver to nonblocking mode
• (master) added check if the background saving process is alive
• (master) added lockfile for background saving process
• (master) added protection against time going back in the free inodes list
• (master) added protection in FOLLOWER against not keeping up with the LEADER (pro only)
• (master) changed condition for removing chunk from data structures (all chunkservers have to be connected)
• (master) added restore run mode (works similar to mfsmaster -a but without running daemon)
• (master) take write replication counters into account in server ordering for new chunk creations
• (master) changed calculating replication limits in chunk rebalance procedure (using actual usage difference)

Best regards, Piotr / MooseFS Team

MooseFS 3.0.105 is now stable! This release is mainly a bugfix release, however it also introduces a new feature of specifying subfolder for Block Devices on MooseFS.

We would like to thank all the contributors who helped us making this release, especially (alphabetically): @dragon2611, @eleaner, @jkiebzak, @onlyjob, @xandrus. Thank you for all your help!

(If I forgot to mention somebody, I am very sorry and please let me know!)

Please find the list of fixes this release includes below:

• (nbd) fixed segfault when filename not specified
• (client) fixed size of waiting threads counter (very rare deadlock in write module)
• (cli) fixed status returned in case of master name that can't be resolved (issue #238)
• (master) improved readability of some functions (related to issue #241)
• (nbd) added option for specifying subfolder on MFS for block devices (issue #252)
• (netdump) fixed using deprecated function pcap_lookupdev
• (cli) fixed 'free space' in plain mode

Best regards, Piotr / MooseFS Team

MooseFS 3.0.104 is now stable! This release includes both bugfixes and new features.

We would like to thank all the contributors who helped us making this release, especially (alphabetically): @borkd, @dragon2611, @nickb937, @njaard, @onlyjob. Thank you for all your help!

(If I forgot to mention somebody, I am very sorry and please let me know!)

Please find the list of fixes this release includes below:

• (master) fixed taking into account labels in classes 1-9
• (master) fixed recovering userperm and otherperm in ACL (in practice it doesn't matter – mainly cosmetics)
• (cli+cgi) fixed python3 compatibility issue (casting to int for %X format)
• (master) increased mmap allocation block size (Linux kernel issue – too many mmaps leads to memory not being properly released by the kernel)
• (client) fixed handling internal packets with size zero (client disconnection during chunkserver registration)
• (client) fixed symlink cache handling (potential race condition)
• (master) added global option for preventing using same ip or same rack for more than one chunk copy
• (client) added ignoring _netdev option – sometimes needed in case of mounting from fstab
• (master) added defining rackid as string paths in mfstopology.cfg (introduced optional higher distances between servers)
• (master) fixed desync on adding damaged chunk
• (master) added changelog save mode (background, async and sync)
• (all) fixed issues reported by cppcheck (fixes #179)
• (cli+cgi) fixed issues reported by pylint (fixes #180)
• (cs) added 'damaged' flag to chunk (prevents syslog flooding – issue #192)
• (metatools) fixed conflict with builtin function humanize_number (NetBSD)
• (master) better handling chunks with pending high-priority jobs
• (master) long loop detection (log info whenever time between polls is longer than 5 seconds)
• (master) fixed condition that checks if thare are all chunkservers and all chunks available (and using it for detecting ENOSPC condition – issue #201)
• (mount) added .params pseudofile with mfsmount working parameters (such as cache sizes, timeouts etc.)
• (cs) fixed removing chunks from data structures when marking disk as 'damaged' (issue #207)
• (cs) fixed handling 'damaged' disks with chunks on reload (issue #207)
• (cs) unlink duplicates week after detection (instead of removing them instantly – related to issue #207)
• (tools) added displaying number of blocks in mfsfileinfo
• (master) added error messages for 'fork' errors
• (cgi) fixed handling encodings (issue #206 and #208)
• (cgi) added timestamp (issue #197)
• (cgi) fixed chart.cgi url (issue #200)
• (cs) fixed crc values for non existing blocks
• (mount) fixed changing processname
• (supervisor) added reloading config parameters on SIGHUP (aka 'reload' – 'pro' version only)
• (cs) fixed exiting protocol
• (cs) added internal rebalance chart

Best regards, Piotr / MooseFS Team

Dear MooseFS Users,

MooseFS 3.0.103 is now stable! This release includes both bugfixes and new features.

Block Device

The biggest and most awaited feature introduced in this version is support for Block Storage which allows you to skip FUSE layer completely and give a block device instead of a file system layer to the OS on the Client using mbfsbdev instead of mfsmount.

See more in man mfsbdev.

This is kind of "feature release" of Block Device - we may improve it or its management in the future. You are very welcome to test it and report any issues by raising a GitHub issue. Your reports will help to make this feature and MooseFS better! 👍 :)

MooseFS Meta Dirinfo

The second new big feature is mfsmetadirinfo. It is so-called "offline" implementation of well-known mfsdirinfo -p and similarly allows you to count real space usage of a particular directory precisely, with taking snapshots into consideration. The tool is faster than mfsdirinfo -p and you can use it on your Master Server. The tool operates on a dumped metadata file (e.g. metadata.mfs.back or metadata.mfs).

See more in man mfsmetadirinfo.

Extremely fast Pseudo-random Generator

This release also includes extremely fast (>1GB/s) pseudo-random generator visible as a virtual file .random in root directory (not listed by ls). You can use it for testing purposes, similarly as you use e.g. /dev/urandom.

New Chunk testing routine in MB/s

MooseFS constantly checks Chunks' checksums in background in order to prevent silent data corruption. This version gives you ability to define speed in MB/s of this checking. It is especially important for big clusters with huge number of small files where definition in chunks per second could reflect in a very long time of whole cluster checking (like years).

Complete list of features and fixes in MooseFS 3.0.103

Please find the list of fixes this release includes below:

• (bdev) new module for Linux - dedicated block device
• (metatools) added new metatool mfsmetadirinfo - offline version of mfsdirinfo -p
• (mount) added extremely fast (>1GB/s) pseudo-random generator visible as a virtual file .random in root directory (not listed by ls)
• (cs) changed chunk testing routine (new definition is in MB/s per disk)
• (tools) changed data structure used by mfsdirinfo -p (faster, less memory usage)
• (cs) added optional high speed internal rebalancing
• (tools) network code refactoring (less code, reconnection to master on error)
• (master+client) added space reservation for superuser
• (daemons) added optional processname_set function
• (daemons) fixed waitpid condition
• (cs) added data precaching (posix_fadcvise(WILLNEED)) before reading chunk data
• (master) remove chunk copy from structures (instead of leaving as INVALID) when replication returned error
• (master) close listen sockets after fork (metadata save in background)
• (master) added subprocess for saving downloaded metadata in background (pro only)
• (master) fixed mfsrmsnapshot issue (SNAPSHOT flag not set in directories - intr. in 3.0.101)
• (mount) added option to load master password from given file
• (cs) proper disconnection from master during reload (maintenance mode during reload)

Best regards, Piotr / MooseFS Team

Dear MooseFS Users,

MooseFS 3.0.101 is stable now! Please find the list of fixes this release includes below:

• (master) make under goal replications respect topology (patch made by tclh123)
• (master) fix inforec undergoal (patch made by tclh123)
• (cgiserv) fixed handling connection close during data sending (leaving active pipes which caused leaving mfs.cgi subprocesses)
• (mount) decreased memory usage (fixed handling buffers allocated per thread)
• (all) setting correct errno when read returns 0 (in sockets.c)
• (mount) fixed double free of plock_data structure
• (master) fixed inconsistency in calculating realsize when storage class definition has changed
• (master+mount) fixed handling read-only filesystems (ro flag in exports and lookup-open lflags)
• (all) fixed differences in defaults between source, config and manpages
• (all) added nobreak macro to explicitly inform compiler that intentionally there is no break before following case
• (mount) added mfssuid and mfsdev options to mount (allows mounting with suid/dev options using system mount on Linux)
• (cs) added using posix_fadvise(...DONTNEED) for tested chunks
• (master) fixed HA bug (uninitialized variable might cause everlasting ELECT - PRO only)
• (cs) fixed metaid handling
• (master) added clearing SNAPHSOT eattr on file/directory modifications
• (master) added optional same rack preference for destination server during undergoal replications
• (mount+tools) increased masterproxy timeout from 1s to 20s

Best regards, Piotr / MooseFS Team

Dear MooseFS Users,

MooseFS 3.0.100 is stable now! Please find the list of fixes this release includes below:

• (cs) fixed rare segfault during chunk sending to master after disconnection
• (mount) added thread for cleaning released files with delay
• (mount) added assertion for lru cache (sustained open files)
• (master+mount) added chunks data cache invalidation after chunkserver disconnection and reconnection
• (mount) changed (lowered) number of connection retries (depends on I/O retry counter)

Best regards, Piotr / MooseFS Team

Dear MooseFS Users,

MooseFS 3.0.99 is stable now! Please find the list of fixes this release includes below:

• (master) disconnect all clients when exports has changed
• (mount) do not show message about 'negative travel time' when the measured time is zero (happens in virtual machines)
• (all) fixed get16bit function on big-endian machines

Best regards, Peter / MooseFS Team

Dear MooseFS Users,

MooseFS 3.0.98 is stable now! Please find the list of fixes and new features this release includes below:

• (mount) changed default cache mode on OS X from 'direct' to 'auto' due to problems with mmap
• (cgiserv) added seteuid to 'mfs' or 'nobody'
• (cs+tools) use readdir instead of readdir_r on glibc 2.24+
• (debian) added support for systemd in debian packages
• (all) fixed library dependencies (for libpcap,libz and libm)
• (mount) fixed race between defered 'open' and 'locks'
• (master) more chunk debugs and better handling chunk status change
• (cs) fixed handling mark for removal (after reload sometimes cs could send chunks marked for removal as normal chunks)
• (master) added file with metadata checksums (for future use)
• (master) fixed way of sending gids to changelog/follower
• (master) added inode checksum for some changes sent to changelog/follower
• (metadump) added '0x' prefix for fileds dumped as hex numbers
• (master) fixed ACL mask synchronization issues (restoring from changelog/leader->follower)

Best regards, Peter / MooseFS Team

Dear MooseFS Users,

MooseFS 3.0.97 is stable now! Please find the list of fixes this release includes below:

• (tools) fixed trash/sustained size desynchronization after file length changed by chunk creation
• (mount) fixed handling invalid arguments in posix lock command

Best regards, Peter / MooseFS Team

Dear MooseFS Users,

MooseFS 3.0.96 is stable now! Please find the list of fixes this release includes below:

• (tools) fixed segfault in mfsscadmin

Best regards, Peter / MooseFS Team