Initial structure for out of the application workflow orchestration #5
Conversation
* Add wrapper to start flux with proxy * Add infrastructure of connecting to and receiving from RMQ * Add initial daemon implementation
|
|
||
| import flux | ||
| import json | ||
| from flux.security import SecurityContext |
There was a problem hiding this comment.
@milroy I get an import error. I am guessing because of running on an old flux version. Can you make the appropriate changes to support both versions?
There was a problem hiding this comment.
@milroy This is a general 'driver' of the code we are going to run outside of compute nodes. I re-used and restructured your previous commit and tried to abstract out some of the hard-coded paths.
| import argparse | ||
|
|
||
|
|
||
| def main(): |
There was a problem hiding this comment.
@milroy I have not tested any of this code. Take a look and let me know of any considerations. Effectively I bundle both the AMSDaemon and the FluxDaemon at the same python entry point.
src/AMSWorkflow/ams/rmq.py
Outdated
| import json | ||
|
|
||
|
|
||
| class RMQClient: |
There was a problem hiding this comment.
@lpottier This is a first re-implementation of the RMQClient we had on our previous hands-on meeting. Connecting with the server needs to happen with a context manager python API. The same applies for enabling/pulling messages from channels. This is inspired from the orchestrator requirements, but use it as a baseline to support the database client. Ideally, we would like to have a single API-point of entry to communicating messages across AMS components.
…messages to a RabbitMQ queue Signed-off-by: Loic Pottier <[email protected]>
* Added initial script to bootstrap Flux on CORAL/IBM machine * Added support Slurm based system, tested with flux-core 0.49 on Ruby/Lassen * Added script to launch AMS miniapp with Flux * Reverted script to support older version of flux, Lassen bootstrap only works wityh Flux<= 0.45 (tested with 0.45) * Added scripts to add secrets on OC * Added new scripts to launch the entire AMS workflow * Upgrade all scripts, they are now fully functional (main script communicates with AMS daemon via RMQ) --------- Signed-off-by: Loic Pottier <[email protected]>
Signed-off-by: Loic Pottier <[email protected]>
Signed-off-by: Loic Pottier <[email protected]>
* This commit addresses multiple problems in the broker part of AMS - we are not sending input/output as encoded string anymore, we send binary blobs - base64 has been removed - a bug has been fixed with (very) old libevent version (<= 2.0.21-stable) - offloading inputs/outputs to the thread managing RMQ is now much faster * Moved to ResourceManager, created AMSMessage structures, moved to smart pointers. * Complete re-design of the RabbitMQ backend * Removed EventBuffer, removed pthread and signals. Big cleanup of the code. * Added documentation and new AMSMsgHeader class + moved from memcpy to ResourceManager::copy
No description provided.