Repository navigation
Release Notes
New features
- Upgraded support for Extrae 5.1.10.
- Upgraded support for Python 3.14.
- New Docker images with multi-architecture support (AMD64, ARM64 and RISCV64).
- New monitor interface (Graphana based) with real time data and interactive agent deployment support.
- New PyCOMPSs CLI: independent repository, graphical interface, COMPSs installation, environment management, and many more.
- New synchronization API function:
compss_wait_on_any. - New resource usage analytics tool from provenance:
compss_genprofiling.
Improvements
- Improved runtime and workers coordination (start and stop).
- Worker robustness: watchdog,slow-starting recovery and restarting.
- Graceful stop before SLURM walltime.
- CPU Affinity support unlimited.
- Extended context sharing in agents with Python.
- Unified runtime API methods: workflow vs runtime.
- Tasks no longer carry a language.
- Enabled intra-task checkpointing for
@MPI. - Event registration moved to the producers.
- Maven plugins removed from the framework.
- Improved NIO transfer resiliency.
- Improved resource usage capture using provenance: GPU usage captured and included in inspect, together with new specific plots.
- Workflow provenance includes all runtime logs by default in the crate, which is essential for debugging.
- Resource usage plots in provenance include an outlier analysis, highlighting any resource usage that differs from the rest of nodes in the computation.
- RCOMPSs has been ported to C++.
- Added support for
@constraintin RCOMPSs. - Added support for GPUs in RCOMPSs.
- Improved support for macOS ARM architectures.
- Configuration files for new HPC systems (Roquo) and updated (Fugaku).
- Several Bug fixes.
Deprecations:
- Dropped support for
@generated_jitfrom Numba. - Support for just-in-time functions within classes has been dropped.
- Dropped support for GAT.
- Dropped the CLI from the framework: moved to independent repository.
Known Limitations
- Known Extrae bug can make an execution fail with
tcacheerror. - Dynamic constrains are limited to task parameters declared as
INwhich are not future objects (generated by previous tasks). - Collections are not supported in
@httptasks. - macOS support is limited to Java and Python without CPU affinity (requires executing with
--cpu_affinity=disabled). Tracing is not available. - Reduce operations can consume more disk space than the manually programmed n-ary reduction.
- Objects used as task parameters must be serializable.
- Tasks that invoke Numpy and MKL may experience issues if a different MKL threads count is used in different tasks. This is due to the fact that MKL reuses threads in the different calls and it does not change the number of threads from one call to another. This can be also happen with other libraries implemented with OpenMP. To fix these issues use the DLB option for in the
cpu_affinityflag. - C++ Objects declared as arguments in coarse-grain tasks must be passed as object pointers in order to have proper dependency management.
- Master as worker feature is not working for executions with persistent worker in C++.
- Coherence and concurrent writing in parameters annotated with the
Concurrentdirection must be managed by the underlying distributed storage system. - Delete file calls for files used as input can produce a significant synchronization of the main code.
- Defining a parameter as
OUTis only allowed for files and collections of objects with a default constructor.
For further information, please refer to the COMPSs Documentation at: https://compss-doc.readthedocs.io/en/stable/
Please find more details about the COMP Superscalar framework at: http://compss.bsc.es/