v0.12

New features:

• libmfu: improve support when copying files to HPSSFS-FUSE by calling ioctl to declare file size to select the appropriate HPSS Class of Service (COS). Enable with new -DENABLE_HPSS=ON CMake option. #548 (thanks hawartens)
• dcmp, dcp, ddup, dsync, dtar: added --open-noatime option to open files with O_NOATIME to avoid modifying file last access time when reading source files. #561
• dcp: added --uid and --gid options to set effective user and group id of running process #540 (thanks bdevcich)
• dfilemaker: promote dfilemaker and retire dfilemaker1. #604 #607 #609 #610 #613 #614 #615 #617
• Added dsync tests. #593 #596 #617 #619 #620
• dsync/dcmp: Added support for symlinks targets #618
• DAOS: DAOS-9902 API: To support empty container lable #531

Bug Fixes:

• dsync: Fix for issue dsync not updating times correctly #560
• libmfu: Function is_mtime_nsec_used now returned true on stat function call failure. #622
• doc/dwalk: add --no-atime in manpages while walking file system without stat. #590
• dsync: Showing usage and nothing else when failure or error occurs. #612
• libmfu: fix bufsize typo in mfu_copy_xattrs and DCOPY_copy_xattrs #601
• .gitignore: ignore tags file created by ctags #606
• libmfu: Fix compile warnings #605
• libmfu: mfu_flist_walk.c: set WALK_RESULT = -1 #600
• dsync: disable --delete on src walk error #599
• libmfu: wrap xattr IO functions with DCOPY_USE_XATTRS #583
• dsync: We are verifying the destination is writable, so we don't perform the source walk and then fail. #595
• build: build.rst shell script: stop if dependency download fails #592
• DAOS: DAOS-14547: Source pool/cont only needs read permission #568
• doc: add -DENABLE_HPSS #551
• libmfu: Add in initial support for HPSSFS-FUSE #546
• DAOS: use proper permissions for obj_open. #543
• libmfu: mfu_param_path: fix cutting root component #536
• DAOS: remove --daos-prefix handling #539
• DAOS: daos-9200 daos-gen: Fix compatible code for API update. #529
• libmfu: fix: hang if file size shrinks during a copy #526
• doc: fix libarchive url #524
• doc: readme: Remove google group #523
• DAOS: DAOS-9871 API: fix more daos_cont_open #522
• DAOS: Daos api update API: daos_cont_open only supports uuid string. #521
• libmfu: properly check for CAP_CHOWN and CAP_FOWNER when changing group and permissions on files #562
• libmfu: update to libdtcmp-v1.1.5 to avoid segfault when sorting large file counts #552
• ddup: add newline when printing warning about changing file size #559
• dcp: when giving up privileges, call setgroups before setuid to avoid permission leaks#569
• dcp: Restore UID/GID after setting with --uid/--gid #586
• rpm: use %cmake macros to build with distribution default compiler and linker flags (thanks rezib) #564

Recommended:

• liblwgrp v1.0.6
• libdtcmp v1.1.5
• libcircle v0.3
• libarchive v3.7.7

v0.11.1

New features:

• Release tarball to package mpiFileUtils with appropriate versions of LWGRP, DTCMP, and libcircle to simplify builds. See https://mpifileutils.readthedocs.io/en/latest/build.html#cmake
• dcp, dsync: added --xattrs={none, all, non-lustre, libattr} option for better control when copying extended attributes. Lustre extended attributes are no longer copied by default. Copying of extended attributes is now independent of the --preserve option to copy owner, group, permissions, and timestamps. #503

New DAOS features:

• New daos-serialize and daos-deserialize tools for moving containers to/from HDF5 format.
• Support for paths formatted as daos://<pool>/<cont>/[path]

Bug Fixes:

• libmfu: incorrect item count in mfu_flist_copy progress message #497
• libmfu: segfault in strmap #501

Recommended:

• liblwgrp v1.0.4
• libdtcmp v1.1.4
• libcircle v0.3
• libarchive v3.5.1

v0.11

New features:

• libmfu: mfu_flist_archive function to create and extract tar (tape archive) files
• libmfu: updated default I/O buffer and chunk sizes from 1MB to 4MB
• libmfu: define major, minor, patch versioning on libmfu.so
• dcmp, dcp, dsync: renamed --synchronous to --direct as the option to enable O_DIRECT
• dcmp, dsync: added --chunksize and --blocksize options to size work units when slicing files
• dcp: log more errors and return 1 from main on error
• dcp, dsync, dwalk: added --dereference option for dereferencing symbolic links
• dtar: promoted from experimental to production tool

New DAOS features:

• dcmp, dcp, dsync: added DAOS support for DAOS POSIX containers
• dcp: added DAOS support for generic DAOS containers
• see DAOS-Support.md for documentation on DAOS usage

Bug Fixes:

• libmfu: improved O_DIRECT buffer and file offset alignments when copying files
• libmfu: compile bug to define SYS_getdents on aarch64 systems

Recommended:

• liblwgrp v1.0.3
• libdtcmp v1.1.1
• libarchive v3.5.1

v0.10.1

This is primarily a bug fix release of v0.10. Among other improvements, it includes fixes to support newer Lustre, Open MPI, and GCC versions.

New features:

• dchmod: check for CAP_FOWNER and CAP_CHOWN to reduce need for --force option

Bug fixes:

• libmfu: drop dead Lustre-specific code that prevented compilation with Lustre v2.13 and newer
• libmfu: convert deprecated MPI functions to use newer MPI_Comm_create_keyval, MPI_Comm_set_attr, and MPI_Comm_get_attr
• libmfu: support empty lists in mfu_flist_sort
• libmfu: switch from "external32" to "native" MPI I/O data representation to better support Open MPI v4.0.3, and add error checking around MPI I/O calls
• libmfu: pad short path names with null when writing binary cache files
• libmfu: avoid multiple definitions of global variables to allow easier builds with the GCC v10.1 compiler
• libmfu: update file offset alignment, buffer alignment, and transfer size for improved O_DIRECT support in mfu_flist_copy
• dfilemaker: correct usage message
• dfind: support file names containing spaces in --exec {} substitutions and avoid seg fault when missing terminating ;

v0.10

New features:

• libmfu: reduced verbosity of debug messages when copying GPFS ACLs
• libmfu: tweaked lite walk progress to count directories after reading them, rather than discovering them
• libmfu: include item rate in walk progress messages
• libmfu: added mfu_progress_start/update/complete functions to periodically print progress information
• progress messages enabled in dchmod, dcmp, dcp, dreln, drm, dstripe, dsync, dwalk, see new --progress option for more details
• dchmod: enable --user and --group options to accept numeric user id and group id values in addition to names
• dchmod: added algorithm to avoid stat on walk when stat info is not needed
• dchmod: skip chown/chmod calls on items that do not need to be changed
• dchmod: added --force option to always call chown/chmod
• dchmod: added --silent option to suppress EPERM error messages
• dcp, dcp1: added --chunksize and --blocksize options to control slicing of large files during copy
• drm: added --stat, --text, and --output options to record list of files drm attempts to remove
• dsync: added --link-dest option to create hardlinks to conserve storage space and inodes during incremental backups
• dsync: added --sparse option to write sparse files

Bug Fixes:

• libmfu: fix call to segmented scan leading to false positives and false negatives when detecting file content differences in dcmp and dsync --contents
• dfind: mask file mode with S_IFMT instead of file type to avoid collisions between different file types, previously -type f would also return symlinks

Requires:

• libcircle v0.3 or higher

v0.9.1

This is primarily a bug fix release for v0.9, but we also promote dbz2 to be a released tool.

New features:

• libmfu: added functions for parallel compression / decompression of dbz2 files
• dbz2: compress and decompress a large file in parallel. Thanks to Ahana Roy Choudhury for contributing the original dbz2 implementation using libcircle for parallel compression and decompression and for proposing the original dbz2 file format that facilitates parallel decompression.
• dwalk: --file_histogram option prints a default size histogram
• dwalk: --text option is now documented. Use this option with --output to create a text file.
• per common user request, tools now verbose by default, use new --quiet option to silence
• simplified informational messages to be more concise
• many tools have had their usage and output text updated

Bug fixes:

• cmake: rpaths now supported in both build and install directories
• cmake: tests added to enable Lustre APIs
• cmake: add path to find FindGPFS.cmake when using -DENABLE_GPFS=ON
• libmfu: fixed bug in cache file format that was introduced in v0.9
• dbcast: fixed to no longer create target directory if it already exists
• dfind: output file was mistakenly hardcoded to write to text format
• dwalk: --distribution option led to a hang when using incorrect syntax

v0.9

We've officially converted to CMake!
Instructions for building are here: https://mpifileutils.readthedocs.io/en/v0.9/build.html

New Features:

• dcmp: include nanoseconds when comparing timestamps
• dcmp: new --lite option to compare files based on file type, file size, and modification time rather than file content
• drm: new --aggressive option to delete files while walking
• dsync: default behavior no longer deletes files at the destination, deleting now requires new --delete option
• dsync: optionally copies in batches with --batch-files option as form of self-checkpointing long running dsync jobs
• drm: fix segfault when deleting a large number of files
• libmfu: avoid problematic MPI I/O external32 for more consistent file format
• libmfu: support for GPFS ACLs

New Tools:

• dfind: filters file list based on different criteria
• dreln: update symlinks whose targets use absolute paths, useful after dsync

v0.8.1

Bug fixes:

• fix lseek positioning when overwriting an existing file with data from a newer source file when using dsync --contents

md5sum: acbd5b5c15919a67392509614bb7871e

v0.8

New features:

• dchmod: added --owner option to change user on files
• dchmod: fixed bug when using 'a' option in symbolic notation
• dcmp: added expressions to compare permissions and ACL on files
• dcp: fix bug that prevented some subdirectories from being created
• dcp1: original dcp from LANL, may be faster than dcp for directory trees with lots of small files, a similar algorithm to be merged into dcp in future release
• dfilemaker: updated to create multiple directory levels, files of different sizes with random content, and symlinks
• dfilemaker1: dfilemaker from v0.7, to be merged into dfilemaker in future release
• dstripe: added support for Lustre 2.5
• dsync: new tool to synchronize one directory tree with another (good for backups or completing partial copies)
• libmfu: added API calls to define new list elements and set their properties
• libmfu: added function to write file list to text file

New experimental tools (work in progress):

• dbz2: compress a single file with bz2
• dfind: filter file lists with find-like tests
• dgrep: parallel grep
• dtar: parallel tar (incomplete)

Known bugs:

• dtar uses libarchive that assumes a single process is writing the tar archive. It likely will generate corrupt tar archives if run with more than one process. This will be fixed in a future release.
• The binary format for reading and writing filelists to files uses MPI I/O external32 data representation. This has proved to be buggy across MPI implementations. In a future release, mpiFileUtils will be changed to read/write these files with POSIX I/O. If a work around is needed, one may change "external32" to "native" in src/common/mfu_flist_io.c.
• dsync --contents has a bug when computing its offset during lseek for overwriting an existing file (fixed in v0.8.1)

Updated dependency:

• Requires update of DTCMP from v1.0.3 to v1.1.0 for new DTCMP_Segmented_scanv/exscanv calls.

md5sum: 1082600e7ac4e6b2c13d91bbec40cffb

v0.7

• dbcast: now creates destination directory before broadcasting file
• dbcast: added --size option to set file segment size
• dchmod: process umask if user provides no ugoa letter in symbolic notation
• dcmp: added --sync option to synchronize source and target directory trees
• dsh: added support for wildcard filters, limit ls output to 100 items by default
• dsh: added --output and --file options to save modified flist on exit
• dstripe: now recursively processes files under a directory
• dstripe: added --report to print file striping info
• dstripe: added --count and --size options to set stripe parameters
• dstripe: added --minsize option to only process files above a certain size
• libmfu: moved copy logic from dcp to new mfu_flist_copy routine
• libmfu: write flist cache files using multiple stripes, one per MPI rank
• configure: fixed --enable-experimental builds

Known bugs:

• using an input list with dcp is broken in v0.7, but this is patched on the main branch

md5sum: c081f7f72c4521dddccdcf9e087c5a2b