Resolwe

===================

Fixed
-----
- Account for file system overhead when processing Data objects with large
inputs

Changed
-------
- Improve storage manager to only process applicable storage locations instead
of iterating through all of them
- Skip hash computation when connector itself provides data integrity check
- Remove ``job`` prefix from kubernetes job name
- Make error messages in Python processes more useful

Added
-----
- Add label ``job_type`` to Kubernetes job to separate interactive jobs from
batch jobs


===================

===================

Fixed
-----
- Fixed progress reporting in Python processes
- Do not override content-type of S3 object when storing hashes
- Support upload of files larger than 80G to AWS S3

Changed
-------
- Download input data in init container
- Storage objects are sent to the listener over socket instead of using files
on the shared filesystem
- Make it possible to run the platform without shared filesystem. All inputs
for processed data object are prepared in input container and all outputs are
uploaded to the chosen storage connector when they are referenced.
- Overcommit CPU in kubertenes processing container by 20%
- Move docker images from Docker Hub to Amazon ECR

Added
-----
- Make automatic removal of Docker containers configurable
- Terminate processing immediately when data object is deleted
- Make default processing image configurable


===================

===================

Changed
-------
- **BACKWARD INCOMPATIBLE:** Remove ``resolwe/upload-tab-file`` and
``resolwe/archiver`` Docker images
- **BACKWARD INCOMPATIBLE:** Remove obsolete processes: ``archiver``,
``upload-tab-file`` and ``upload-image-file``
- **BACKWARD INCOMPATIBLE:** Python process syntax has changed: all the
attributes of Data object are now available in Python process and therefore
accessing outputs using syntax ``data_object.output_name`` is no longer
valid since ``output_name`` could be the name of the attribute. The new
syntax is ``data_object.output.output_name``.
- **BACKWARD INCOMPATIBLE:** Communication between the processing script
and listener has changed from printing to stdout to sending messages over
sockets. Messages printed to stdout or sent using old version of the
``resolwe-runtime-utils`` (YAML processes) are no longer processed. YAML
processes need new version of ``resolwe-runtime-utils`` while Python
processes require a rewrite to the new syntax and Python version 3.4 or
higher in the container (``resolwe-runtime-utils`` package is no longer
needed).
- Use Github Actions to run the tests
- Listener communicates with containers through ZeroMQ instead of Redis
- Start two containers for each process instead of one: the second one is
used to communicate with the listener
- Move settings for Python processes from files to environmental variables

Added
-----
- Add Kubernetes workload connector
- Support running process instant termination
- Support registering custom command handlers in listener and exposing data
objects (possibly defined in other Django applications) to Python processes
- Support Django-like syntax in Python processes to create, filter or access
attributes of the exposed data objects
- Support creating new base classes for Python processes


===================

===================

Fixed
-----
- Allow retrieval of Storage object that was linked to more than one ``Data``
object

Changed
-------
- Migrate docker images to Fedora 33 and Ubuntu 20.04


===================

===================

Added
-----
- Support Python processes in Sphinx ``autoprocess*::`` directive


===================

===================

Added
-----
- **BACKWARD INCOMPATIBLE:** Only copy parent relations when duplicating
``Data`` objects
- Add duplicate data dependency to indicate from which object the ``Data``
object was duplicated
- Support accessing Data name in Python processes through ``self.name``
- Add ``permission`` filter to ``collection``, ``entity`` and ``data`` that
returns only objects on which current user has given permission

Changed
-------
- Make relations in collection visible to public user if he has view
permissions on the collection


===================