Project

General

Profile

Bug #5374

MPI error

Added by John Freeman over 6 years ago. Updated over 3 years ago.

Status:
Rejected
Priority:
Normal
Assignee:
Category:
Known Issues
Target version:
Start date:
02/06/2014
Due date:
02/20/2014
% Done:

50%

Estimated time:
10.00 h
Co-Assignees:
Duration: 15

Description

On dsfr6, using artdaq-demo v2_00_04, if I switch from using the "standard" 2x2x2 configuration (2 BoardReaderMains, 2 EventBuilderMains, 2 AggregatorMains) and add an additional EventBuilderMain, then after running for a few thousand events (the number varies), the following error appears:

Thu Feb 06 15:53:40 -0600 2014: %MSG
Thu Feb 06 15:53:40 -0600 2014: Fatal error in MPI_Send: Other MPI error
Thu Feb 06 15:53:40 -0600 2014: Fatal error in MPI_Send: Other MPI error
Thu Feb 06 15:53:40 -0600 2014: Fatal error in MPI_Send: Other MPI error
Thu Feb 06 15:53:41 -0600 2014: EXITING:dsfr6:1:EventBuilderMain:5235
EventBuilderMain on dsfr6 is exiting.

History

#1 Updated by John Freeman over 6 years ago

  • % Done changed from 0 to 50

The "50% done" is my way of saying that I've discovered that as long as there's a 0.1s pause before a new event gets generated in simulation, this error no longer appears. This quantity hasn't been minimized, so it's not clear yet how brief a pause we can get away with before the error reappears.

#2 Updated by John Freeman over 6 years ago

  • Status changed from New to Assigned

#3 Updated by Eric Flumerfelt over 4 years ago

  • Target version set to 981

#4 Updated by Eric Flumerfelt over 3 years ago

  • Category set to Known Issues
  • Status changed from Assigned to Rejected
  • Target version deleted (981)

I'm closing this issue as the code path it refers to no longer exists. If there is a similar issue in MPITransfer, a new issue should be opened.

That being said, I think this may be related to some of the issue I was having during my integration testing of MPITransfer, notably the fact that the MPI routines did not appear to be as thread-safe as advertised.

#5 Updated by Eric Flumerfelt over 3 years ago

  • Target version set to artdaq-demo v2_09_00

Updating version number with the version where it became obsolete.



Also available in: Atom PDF