Example Usage from the Command Line

In this page we briefly describe how to use JACAL in a DALiuGE graph. This assumes you already built JACAL.

Outline

In this example we will replicate one of the unit tests run in the GitLab CI pipeline, namely test_basic_imaging. This test performs basic imaging on an input MeasurementSet using the CalcNE and SolveNE JACAL components. The other unit tests work similarly, exercising different JACAL components in different modes of operation.

In DALiuGE a program is expressed as a graph, with nodes listing applications, and the data flowing through them. Graphs come in two flavours: logical, expressing the logical constructs used in the program (including loops, gather and scatter components), and physical, which is the fully-fledged version of a logical graph after expanding all the logical constructs.

This test is expressed as a logical graph. After translation into a physical graph it is submitted for execution to the DALiuGE managers, which need to be started beforehand. During execution one can monitor the progress of the program via a browser.

Preparing the Graph

This test needs a few inputs:

  • The logical graph.

  • A parset (parsets are text files containing configuration options, and are the configuration mechanism used throughout Yandasoft).

  • Some input data.

Put all three files above in a new directory, and then decompress the input data:

$> mkdir tmp
$> cd tmp
$> export TEST_WORKING_DIR=$PWD
$> wget https://gitlab.com/ska-telescope/ska-sdp-jacal/-/raw/master/jacal/test/daliuge/test_basic_imaging.json?inline=false
$> wget https://gitlab.com/ska-telescope/ska-sdp-jacal/-/raw/master/jacal/test/daliuge/test_basic_imaging.in?inline=false
$> wget https://gitlab.com/ska-telescope/ska-sdp-jacal/-/raw/master/data/chan_1.ms.tar.gz?inline=false
$> tar xf chan_1.ms.tar.gz
$> PARSET=$PWD/test_basic_imaging.in

Next, some adjustments will need to be made to the graph so that the JACAL shared library can be found, and the parset is correctly read at runtime:

$> sed -i "s|%JACAL_SO%|$PATH_TO_JACAL_SO|g; s|%PARSET%|$PARSET|g" test_basic_imaging.json

Starting DALiuGE

Firstly, one needs to start the DALiuGE managers, the runtime entities in charge of executing graphs. We will start two: the Node Manager (NM), in charge of executing the graph, and a Data Island Manager (DIM), in charge of managing one or more NMs. Note that starting the DIM is not strictly required, but is done for completeness.

Start the managers each on a different terminal so you can see their outputs independently. Also, to make the test simpler, start both in the same directory where the downloaded files are placed:

$> cd $TEST_WORKING_DIR
$> dlg nm -v
$> dlg dim -N 127.0.0.1 -v

Running

To execute a graph we submit it to one of the DALiuGE managers (in our case, the DIM). Also, because we are starting from a logical graph, we need to transform it into a physical graph that can be run on the deployed managers.

This can be done as follows:

$> cd $TEST_WORKING_DIR
$> cat test_basic_imaging.in \
   | dlg unroll-and-partition `# Logical -> Physical translation` \
   | dlg map                  `# Assign nodes to drops (i.e., schedule the graph)` \
   | dlg submit -w            `# Submit and wait until execution finishes`

Finally, connect to 127.0.0.1:8000 to see the graph running:

_images/dlg_runtime.png

Note that CalcNE now supports new gridders with more flexible data partitioning. This can be enabled with JACAL-specific parset parameter Cimager.gridder.dataaccess=datapartitions (set to yandasoft or leave unset to use the Yandasoft data iterators and gridders). The type of partitioning is set with JACAL-specific parset parameter Cimager.gridder.partitiontype:

GitHub: ICRAR/EAGLE-graph-repo (master): examples/jacal_CalcNElight.graph