(the release notes also cover some changes from 0.34 to 0.35)
- simplify agent process tree, process naming
- improve session and agent termination
- several fixes and chages to the task state model (refer to
documentation!)
- fix POPEN state reporting
- split agent component into individual, relocatable processes
- improve and generalize agent bootstrapping
- add support for dynamic agent layout over compute nodes
- support for ORTE launch method on CRAY (and others)
- add a watcher thread for the ORTE DVM
- improves profiling support, expand to RP module
- add various profiling analysis tools
- add support for profile fetching from remote pilot sandbox
- synchronize and recombine profiles from different pilots
- add a simple tool to run a recorded session.
- add several utility classes: component, queue, pubsub
- clean configuration passing from module to agent.
- clean tunneling support
- support different data frame formats for profiling
- use agent infrastructure (LRMS, LM) for spawning sub-agents
- allow LM to specify env vars to be unset.
- allow agent on mom node to use tunnel.
- fix logging to avoid log leakage from lower layers
- avoid some file system bottlenecks
- several resource specific configuration fixes (mostly stampede,
archer, bw)
- backport stdout/stderr/log retrieval
- better logging of clone/drops, better error handling for configs
- fix, improve profiling of Task execution
- make profile an object
- use ZMQ pubsub and queues for agent/sub-agent communication
- decouple launch methods from scheduler for most LMs NOTE: RUNJOB
remains coupled!
- detect disappearing orte-dvm when exit code is zero
- perform node allocation for sub-agents
- introduce a barrier on agent startup
- fix some errors on shell spanwer (quoting, monotoring delays)
- make localhost layout configurable via cpn
- make setup.py report a decent error when being used with python3
- support nodename lookup on Cray
- only mkdir in input staging controller when we intent to stage data
- protect agent cb invokation by lock
- (re)add command line for profile fetching
- cleanup of data staging, with better support for different schemas
(incl. GlobusOnline)
- work toward better OSG support
- Use netifaces for ip address mangling.
- Use ORTE from the 2.x branch.
- remove Url class
--------------------------------------------------------------------------------