GitLab

- New filtering methods applies a filtering regular expression to all
node names in the node list of a job; the mode of the filtered node names
is then computed, and the job is processed only if this matches an
arbitrary "match" string
- Can be used to assign jobs to different islands in a HPC system based
on the allocation of nodes

- If a single unit of an operator fails computation, this does not
prevent anymore all other units from computing

- Hierarchical units are now supported across the framework in all plugins

- Added generateFromTemplate method
- Fixed handling of special cases (on-demand and root units)
- Added enforceTopics option to force a MQTT prefix for unit outputs
- printConfig in UnitTemplate now accounts for hierarchical units
- Updated README

- Fixed a bug that prevented pipelines in collectagent deployments
- Incremented the tolerance threshold of fuzzy queries to 1h
- Fixed the assignment of MQTT topics to job sensors

- Fixed a bug that prevented relaxed mode from working properly for
operators deployed on the collectagent
- Changed the function of the "delay" parameter: it now specifies a
fixed offset applied to the computation interval of each operator,
useful to tune the flow of data in operator pipelines

- Now using references instead of pointers
- Return type is now bool

- A regular expression (jobFilter) can be applied to the jobs' nodelists
to identify the jobs that need to be treated by each collectagent in a
cluster
- By default only the first node in the nodelist is considered, but this
behavior can be customized

- Integrated changes to the stop mechanism
- Also re-worked the interface to have the same execOn* methods as Sensor Groups

- Make clean might be required

- For plugins that use hierarchical units (e.g., job analyzers) a
"flatten" plugin option was added. If enabled, output sensors in the
sub-units of each unit will be exposed as well, and their output values
propagated
- Minor fixes here and there

- Fixed a fair amount of bugs related to job analyzers
- Added a "getJobsInIntervalRunning" method to JobDataStore, to get
the list of jobs that were running in a certain time interval
- Some code refactoring for the JobDataStore class to reduce duplication
- Job analyzers have been tested and are working properly as of now

- A jobData structure is now supplied alongside che corresponding unit
as input to the compute() method for job analyzers

- Compiles fine but completely untested for the moment
- Also addressed many bugs for job analyzer templates

- New method is lock-based and compatible with all features (e.g. REST API)
in the framework
- Performance impact is expected to be negligible as race conditions will
not be frequent

- The class is now derived from AnalyzerTemplate, reducing code
duplication

- Template for Job Analyzers, derived from AnalyzerInterface

- The "relaxed" option can now be used also to control on-demand
analyzers, allowing for pipelining when in this mode

- Update config files to reflect new topic structure
- Updated READMEs accordingly
- Added support for "relaxed" analyzers, whose units can be built
even if some sensors cannot be found in the system
- Added support for sub-units, so as to build hierarchical unit structures

- Fixed name of log files for collectagent
- Fixed a few compilation warnings under gcc 7
- README for dcdbpusher now correctly refers to dcdbpusher.conf files
- SensorNavigator is rebuilt after plugin reloads only if data analytics
are currently running

- Textual MQTT topics working in dcdbpusher and data analytics framework
- REST API still not working properly
- Testing on the collectagent not performed yet

- Added some more indentation to make the configurations more readable
- Fixed one more warning in dcdbquery

- Added a 'common' directory where all shared code has been placed
- Analytics package has been moved out of dcdbpusher, now has its own makefile
- Some renaming here and there for uniformity