Summary
- New Token Extractor usage identification using the payload setup string rather than relying on the release version, allowing the new mechanism using positional event numbers with release 19.2.5.3. The old pilot version would start the Token Extractor for all releases older than 20.3.3.
- Support for server instructions ‘softkill’ (letting AthenaMP finish the downloaded event ranges, but won’t download any further event ranges; OS stage-outs are allowed to finish) and ’tobekilled’ (abort AthenaMP immediately; existing OS stage-out queue is allowed to finish).
- Protection against problematic import of python argparse module (only available from python v 2.7); problem seen with one HPC in China.
Updates related to event service on HPC:s
- Implementation of one rank Yoda
- Fix for a merge problem
Version info
General fixes
- Protected against too old python versions while importing argparse. Problem seen on Chinese HPC site (EventStager, MVEventStager)
Event service:
- Updated setUseTokenExtractor() to use payload setup string to decide whether the Token Extractor should be used or not, instead of hardcoded release \
string (RunJobEvent)
- Sending runCommandList[0] to setUseTokenExtractor() instead of release string, from __main__() (RunJobEvent)
- Now sending jobId to updateEventRange() from asynchronousOutputStager(), listener() (RunJobEvent)
- Added jobId argument to updateEventRange() (EventRanges)
- Now supporting softkill in asynchronousOutputStager() (RunJobEvent)
- Created shouldBeAborted(), setAbort(), shouldBeKilled(), setToBeKilled() - used in asynchronousOutputStager(), __main__() (RunJobEvent)
- Created new error code 1237, ERR_ESKILLEDBYSERVER, 'Event Service job killed by server'
Updates from Wen Guan
- Fixed ATLAS site information
- Implementation of one rank yoda
- Fixed merge problem in site information
No comments:
Post a Comment