amusecode / amuse-distributed Goto Github PK

Workers are expected to start just about immediately. We need a timeout on the startup, to detect failures

If a worker is starting using the SocketChannel, and MPI is used, this only works properly if the rank-0 mpi process is running on the same machine as the AMUSE script.

The library needs unit tests. The test framework is already in place.

In debug mode, we should save the output of codes somewhere, regardless of output settings

We need to add some tests to the amuse/test set for distributed amuse

If for some reason the connection between the pilots and the central amuse script is severed, we should make sure to shutdown any running jobs. One way to achieve this is a watchdog timer.

see amusecode/amuse#149 and xenon-middleware/xenon#462 !

To see what is going on, we need a simply monitoring web app build into the library. Probably easiest to use Jetty to generate some HTML tables

Note: it would help if we had a NLeSC css file for the layout/colors/fonts/etc

If a user submits a job which matches no pilot, it will never be run. We should consider this to be an error.

We need a function in Distributed Amuse to set and get the port/address the webinterface is running on.

When running a distributed simulation, we should make sure we use an identical or at least equivalent AMUSE, java, python, etc.

We probably should add some checks for this.

We should also support python workers, where the python code is copied to the remote side dynamically (like the script jobs)

By default, hubs are started on the frontend of a cluster. We should also support starting the hub on a node using the scheduler.

We need some sort of user guide. Preferably inside the AMUSE documentation.

When finished, the code needs to be moved from the amuse sandbox to the amuse source tree

We need some examples for running amuse applications using the distributed code.

We need to add javadoc where applicable, so future developers of this code have somewhere to start

the current source in AMUSE is out of sync and more up to date then the NLeSc repo - the current situation with both (in tree and seperate repo) is not so nice, so either we should
1 drop the NleSc repo, make the code part of AMUSE only (ie delete the update target)
2 drop the AMUSE code, have it a downloadable code from the NLeSc repo
3 combine 1+2: drop the AMUSE code, have it a downloadable code but move the code to a seperate repo as part of the amuse project (so under amusecode)