The script reads the repos.txt file from the user-scripts and:

1. Compares the old version with the new one in all the repos, and
   if they new is smaller it stops execution
2. Changes all the versions in all the repos to the new version

Change-Id: I9ddff07ea935adb73e595a2e8d6a404b08e69a21

This patch implements all functionality required for dynamic, distributed
Nest interfaces that behave the same as single-process nest for local installs.

The existing MUSIC interfaces are deprecated and now subclass the adapters
and interfaces in the new hbp_distributed_nest package.

This patch does not support distributed, cluster launching, but this can
be adapted in the future from the existing MUSIC launchers.

Change-Id: I1aac3a59843d9fdfe051a9674566f3b3f5040ac1

Change-Id: I7b142d863c95d9597b290f3ab27c85637e56af2d

Change-Id: If63ddac9a6265b9ae6bc61e0350e087b0e1a29fe

Change-Id: I6b9ee3680630db3a3f2399a12164dba94b6a4c3e

This patch removes the hardcoded 20ms timestep value and instead
references the default value set in the CLE.

Change-Id: Iaa17d20a39b6c6b0b3fd271a19db3ddff42b6fe6

This patch adds support for setting a global RNG seed to be used
across all processes - the CLE and brain processes.

Change-Id: I8f4554e0684fa69bb801083e5f9dbecaa275f8f8

This patch resolves the distributed simulation performance issues by
ensuring all brain simulation processes are using the correct simulation/
integration timestep. Previously, the main CLE process would advance with
the correct timestep (default 20ms), but each distributed process was
advancing with the simulator timestep (default 0.1ms) - causing significant
overhead and 200x the calls on each remote process.

This patch improves performance to nearly realtime (small overhead for
multiple processes) rather than the prior hundreds of times slower. Further
optimizations can be made in the CLE as we add larger scale brains.

Change-Id: Id3974b31efc24508b3c48b9e7d1e250f44db8be8

Change-Id: Ica268ea6de36827b5c1dcf519176c12dbd3c1f6c

After the removal of the additional hbp-nrp dependencies to
support Bitbucket installation, our Jenkins builds now (again)
build one version of numpy with scipy and one separately as
specified by the CLE. This patch migrates the early installation of
numpy from the CLE to mitigate this.

All packages were updated to future proof dependencies, failure is
ignored and normal installation will proceed if there is no numpy
requirement specified.

Minor pylint cleanup also performed in the setup.py files.

Change-Id: I6b055439c80d931845937939261eb4b6b649c50a

Change-Id: Ice875bc5a3c36960370a83ac2d5713d167345b53

Change-Id: I5841d1f46fd9576402153e5ef8a7b3fcbccaf9b9

Change-Id: Ia51a19ea0aaf5b8d79c1c22a7b583bdfd078fd1f

Change-Id: Ib61b9b12602c1e614112df65212057a18c4c3ecf

Change-Id: Ie85485b9e5d42a0b66b8b73c8e0be71b2eb98e89

Change-Id: I56a07038278bd54bcab3b7206a8c73827b66dd36

Change-Id: Ica4c795a5c6da07c70627c6cd32e453915d54761

This patch changes the Lugano launch configuration to deploy the
nrp-variables script generated by Puppet to the remote hosts and
properly override NRP_VARIABLES_PATH to ensure Gazebo launch succeeds.

The previous version only set the _VERSION variables, but the file is
required to be accessible for the Gazebo launch to succeed in a staging
environment.

Change-Id: I06ec74420fdd35d3219ee311531aa8d57504936b

Change-Id: Ia73b4940cc7e8df13d2353821886c5f6fb004770

Change-Id: I728bdc39d3e510e63c39f72769a6bf05d29ce919

The distributed launch now creates a ROS node in the CLE to ensure
logs are properly redirected to stdout. This node and any future nodes
should be cleaned up as we do in the CLELauncher, but this needs to be
done after shutdown of all launched components.

Change-Id: I00733872978d13cf4c3f319c7e594feae866d930

The CLE no longer loads the brain module in terms of PyNN, but it
still needs to load the brain file source for the frontend to display.

Change-Id: I3cd7833387f6e542a05a29e651b38cf2031b6c54

This patch resolves the remaining issues with distributed Nest launching on the
Lugano vizcluster. It includes fixes to:

- pass the music working directory path as an environment variable rather than
  using MUSIC file configuration, as the CLE seems to change working directory
  during the launch of an expeirment and the proxy.xml file was inaccessible

- fix launch success/failure to wait for a non progress bar status message from
  the CLE, this is because the status bar messages are always marked as 'done'
  even in the event of a failure, so it was not a correct way to check success

- start gzbridge on the backend VM by waiting for Gazebo to be started on the
  remote nodes, this is because the vizcluster node cannot start gzbridge and
  would not be reachable by the frontend even if it could

Change-Id: Id182da74b6e01e748003bc8d7694fc83fe572336

Change-Id: I086344008eb04e9d6c17b4247fb39345c292f098

This patch addresses 3 broad issues:

[NRRPLT-3902]
- provides the ability to run simulations on vizcluster nodes
- parameterize absolute paths with appropriate NRP_ environment variable that can be
  properly expanded on the target host
- provides generic interfaces that can be extended for different hosts

[NRRPLT-4058]
- refactors the init/run mechanisms to more closely mirror the single process version
- this enables reporting of launch failures all the way through CLE initialization (which
  includes things like Gazebo launch failures, resource allocation failures, Exceptions)
- checks the runtime return code to determine if the mpirun command was abnormally
  terminated and properly reports the error
- propagates all errors appropriately back to the main pipeline

[NRRPLT-4840]
- increases test coverage of the new components back up to ~90%
- the reamining coverage is primarily for workaround/temporary code that is difficult to
  unit test without running the platform and will be removed in future commits

Change-Id: Ib33a1d35b26aeee6da10c95f30bb9be9243479dc

Change-Id: I186b69c74b1361a538958a242d5d5844a7944b35

This patch fixes a critical issue where the output of distributed simulations
did not match that of single process simulations.

The issue was due to static synapse creation for proxy connections that did
not consider the specific spike devices that were connected - each of which
have different default and user configurable values - and would produce
similar but incorrect output. Further, fixed port width meant that multiple
transfer functions would be connected to the same parrot neurons, resulting
in cumulative output to the actual brain neurons rather than properly
separated input from each device.

To resolve this issue, the following was implemented:
- faux large port widths to the brain allowing 5 x population connections
- dynamic configuration of these ports via MPI communication
- use of Nest freeze/thaw paradigm to try and reduce performance impact

Thus, the CLE notifies the remote brain processes of each TF's specific
configuration and it is replicated using the appropriate free proxy
neurons on each side through direct Nest manipulation. Once the CLE
notifies the processes that the transfer functions are loaded, they no
longer accept configuration messages and run the simulation.

This creates proper transfer function connectivity to workaround the lack
of dynamic port creation in MUSIC and produces the same results as the
single process version of the Husky experiment.

This patch should absolutely be removed and fully refactored when
MUSIC supports dynamic port creation, it is a workaround to achieve
functionality.

Change-Id: Ifee5e668c9644e679861dc7598dff9e9f5c6fa52

We are transitioning from OpenMPI to mvapich2, which has different
syntax for mpirun. This minor change updates the syntax and will ensure
that both local and deployed installed are using mvapich2.

The mpirun (mpiexec.hydra) is now a bit different than the OpenMPI
behavior. When executed, it sends a SIGSTOP to the parent process tree
and suspends everything (the bash terminal and our Python processes),
so it's necessary to now spawn it using the os.setssid parameter to
start a new session for the subprorcess that is separated from the
main process tree.

This patch also removes the subprocess.wait() call which according
to the documentation can cause buffering/memory issues. In it's place
poll() is used with live/faster output of the processes output to the
main log.

Change-Id: I3a9e255986ac6f39a01b7a1a801e8ab24a14abac

Change-Id: Ib773f68ab6cc0838c7aefe56c12d5c6bcb79fee9

Change-Id: I08139804e64c0f9cbc843c37665c447558f7f0ea

This patch updates the distributed NEST launcher to be in line with
the normal CLELauncher, which now requires a reservation parameter
to be passed into the init method.

Change-Id: Id6d52ea7b35b70205e35450e770cd45e9ccfc8a1

Change-Id: I17e42fd34b998a3b9b5c05527b7e8ccdc8e46183

Change-Id: Ibdfe7774f096795df4a51425a052d0eeb469bbb2

Change-Id: Ib5d1a2a63c39e3fd4f8caa205290ecf673a7c104

Change-Id: I6b5b94b9445896570067db08d88f6eef925a4217

Change-Id: I145b6cf2db4d0dcf945e92792adf9c8759c4c59f

Change-Id: Iddc4a9ae623a72adc6d17a9353773f6d4847eb6f