Irun

• Released Spack package (https://github.com/leobago/fti-spack)
• Asynchronous HDF5 single file creation
• New installation folder structure (includes config file template, documentation, examples, etc)
• CI revision (bug fixes, new structure, compiler updates, added MPICH)
• Improved CPP interoperability (included into CI)
• Improved installation script (install.sh)
• Revised coding style constrains
• Minor patches (dcp, macros, testing)

Rabat

This release breaks compatibility with a few FTI functions for handling data types. Please check the documentation for details:
https://fault-tolerance-interface.readthedocs.io/en/latest/compatibility_notes.html

• Added support for defining non-opaque composite data types in Fortran when using HDF5;
• Added native and non-opaque bindings for Fortran Complex data type when using HDF5;
• Simplified the composite data type handling C API;
• Enhanced CMake configuration exports to include linked libraries and compiler definitions;
• Enhancement of the pre-commit hook including bug fixes.
• Implementation of Fast-Forward feature to allow checkpoints to be taken at sub-minute magnitude.
• Implementation of a checkpoint processor for reading and decoding FTI's checkpoints. Newer features are implemented including hdf5 write, N-dimentional variable support, last-checkpoint processing, and usage examples.
• New API function FTI_SetAttribute that allows adding descriptive attributes to the protected datasets

Florianopolis

• GetConfiguration feature: follow-up of issue #250
• Register a virtually unlimited number of protections with FTI_Protect
• Asynchronous postprocessing of shared HDF5 checkpoint file
• Enhancement of the ICR feature by separating variable recovery from the I/O tasks for all I/O modes
• IME I/O interface (IME native API)
• Refactoring of FTI-I/O interfaces (HDF5, SIONlib, MPI-I/O, FTI-FF)
• Addition of the FTI Integrated Test Framework (ITF), a tool to develop FTI Integration tests
• Complete refactor of local integration tests behavior into the new ITF format
• Removal of deprecated files in the local tests directory
• Integration of ITF into CMake test runner tool (CTest) and Jenkins
• Expanded CI tests with local tests for all I/O libraries
• Simplified runtime metadata handling in FTI (for FTI developers)
• Implementation of key-value storage for protected variables (for FTI developers)
• Documentation for how to use, develop tests and expand ITF added to the developer guide section
• Configuration of ReadTheDocs for an auto-build of the documentation from Doxygen's resources using Breathe and Sphinx

Heraklion

Release

This release includes the new version of differential checkpointing, a complete implementation of incremental checkpointing, full support for GPU checkpointing and full support for HDF5 checkpointing, including the option for checkpointing into a single file (N-1) and restarting with a different number of processes.

Changelog

• New major feature allowing users to checkpoint data allocated in the GPU device memory.
• New implementation of differential checkpointing that addresses performance issues for highly fragmented differential updates.
• New major feature allowing users to use incremental checkpointing for CPU and GPU data by adding one by one the variables to the checkpoint file.
• New major feature for DCPPosix allowing to recover from last non-corrupted checkpoint file.
• New examples in the examples/GPU directory that checkpoint GPU data.
• New major feature allowing to restart with a different number of processes using a shared HDF5 checkpoint file.
• New unitary tests for the new features.
• New configurable/flexible local test structure.
• Fixed Bug of RecoverVar.
• Fixed Bug on DCP recovery.
• Complete and full code documentation generated with Doxygen.

Cologne

• Fix for bug to find MPI with PGI compilers.
• New option to avoid killing head processes and let the user handle it.
• New major feature allowing users to use differential checkpointing including a new FTI file format.
• New major feature allowing users to keep ALL L4 checkpoints.
• New major feature allowing users to leverage FTI for general asynchronous I/O outside of checkpointing.
• New unitary tests for the new features and a few more with improved performance for the previous ones.
• Complete and full code documentation generated with Doxygen.
• Vastly improved User guide showcasing the new features of this release.

Coruna

• Fixed corner case bug when a specific failure type happens just after a restart.
• Fixed some unitary tests that were launched with the wrong arguments.
• Fixed bug on FTI File Format generating a segmentation fault.
• Implementation of checkpointing in HDF5 format.
• Support for HDF5 groups added to allow more flexibility.
• Additional unitary tests added to check HDF5 support.
• Extended wiki, documentation and user guide with HDF5 and FFF information.

Barcelona

• Switch to Jerasure 2.0

• Support for checkpoints that dynamically change size

• FTI_Realloc and FTI_GetStoredSize added to support evolving ckpt sizes

• Added support for partial local recovery with FTI_RecoverVar

• Feature to set sync time for applications with evolving iteration length

• Fix for examples bug with Intel compilers

• Fix for bug on Cray machines

• Added warning when using different compilers

• Add more unitary tests

• Wiki pages added on Github

• Fortran API added in the wiki and user guide

• Improved documentation

• Switch to BSD License

Poznan

• I/O interfaces for MPI-IO and SIONlib
• Checking checkpoint integrity with MD5 checksums
• Fixed bug for L2 and L3 checkpointing
• Fixed synchronization bugs
• Added automated testing with different compilers
• Expanded examples
• Fixed memory leaks and removed unused variables
• Local unitary tests in C and fortran
• Improved code formatting
• Updated developer documentation

Paris

• Fixed undefined behaviour while using snprintf function.
• Fixed bug when no checkpoints are taken if mean iteration time is bigger than 30 sec.
• Fixed while validating configuration for Level 2 and 3.
• Fixed bug when asked to keep last checkpoint if no checkpoint was taken during execution.
• Fixed bug of the example application crashes during recovery.
• Fixed bug when using of rint requires the math library at link.
• Added Cray Compiler names to FindMPI script
• Fixed bug for files larger than 2Gb (int -> long).
• Writing full datasets harder.
• Better documentation and NEW user guide!

Chicago

Here is the list of changes for this Release

• Fixed incorrect warnings for no-head ranks if using dedicated processes.
• Fixed problem with renaming/erasing local files if using dedicated processes.
• Fixed problem with creating files/directories which already exist and deleting files/directories which don't exist.
• Moved some global variables into static.
• Removed some unused variables or code.
• Fixed buffer overflow in jerasure library.
• Fixed some not null terminated string bugs.
• Fixed many unchecked or ignored return value form standard library functions.
• Fixed some uninitialized variable problems.
• Corrected some misleading warning/error messages.
• Fixed many resource leaks.
• Fixed many potential memory leaks.
• Fixed many TOCTOU problems.
• Added header file with declarations of all library functions.
• Added option to build examples.
• Added cmake files to build dependencies and examples.
• Cleaned library interface.