Opened 10 years ago

Closed 7 years ago

#480 closed defect (open)

bccd-syncdir essentially overwrites machines file

Reported by: fitz Owned by: skylar
Priority: major Milestone: Live CD
Component: Both Version:
Keywords: Cc:
Blocked By: Blocking:
Estimated Hours: 1 Total Hours: 2.00

Description (last modified by fitz)

  • bccd-syncdir copies via rsync, via ssh
  • ssh starts a login shell (bash)
  • bash sources/executes ~bccd/.bash_profile
  • bash_profile loads the openmpi module
  • loading the openmpi module regenerates the machines file

This has implications for using mpich2, namely that the machines file would need to be regenerated before running applications.

This goes beyond just bccd-syncdir as well. Other background tasks and even mpirun itself use SSH to get things done, so this is a much wider problem.

Change History (18)

comment:1 Changed 10 years ago by fitz

  • Owner set to fitz
  • Status changed from new to assigned

Proposed fix committed in r2287 to fitz-devel:

  • bash_profile looks for $RSYNC before loading modules
  • bccd-syncdir sets $RSYNC and passes through ssh
  • sshd set to allow the $RSYNC variable to be sent

comment:2 Changed 10 years ago by fitz

rm'ing the link .bashrc also fixes this issue.

comment:3 Changed 10 years ago by fitz

  • Component set to Both
  • Resolution set to fixed
  • Status changed from assigned to closed

Removing .bashrc link in r2299 fixes this issue.

comment:4 Changed 10 years ago by fitz

  • Resolution fixed deleted
  • Status changed from closed to reopened

Need to make sure terminals started with X are login shells.

comment:5 Changed 10 years ago by fitz

  • Resolution set to fixed
  • Status changed from reopened to closed

Starting aterm as login shells in r2302.

comment:6 Changed 8 years ago by fitz

  • Description modified (diff)
  • Resolution fixed deleted
  • Status changed from closed to reopened

comment:7 Changed 8 years ago by charliep

This has come back, it may be modules that's doing it and not re-generating it, or something similar. It seems to happen every time you run an MPI program.

comment:8 Changed 8 years ago by charliep

  • Owner changed from fitz to nobody
  • Status changed from reopened to new

comment:9 Changed 7 years ago by charliep

The fix here is to not run bccd-snarfhosts automagically but rather only manually. Also the output should be ~/machines-openmpi or ~/machines-mpich2 depending on the format that's generated.

comment:10 Changed 7 years ago by skylar

  • Owner changed from nobody to skylar
  • Status changed from new to assigned

comment:11 Changed 7 years ago by skylar

In 3773:

generating mpi implementation specific machines files, not auto-generating anymore (#480)

comment:12 Changed 7 years ago by skylar

In 3775:

make sure head node hostname only shows up once (#480)

comment:13 Changed 7 years ago by skylar

In 3776:

merging in new machines file generation (#480)

comment:14 Changed 7 years ago by skylar

  • Status changed from assigned to qa

comment:15 Changed 7 years ago by skylar

tested

comment:16 Changed 7 years ago by mmludin08

  • Priority changed from critical to major

comment:17 Changed 7 years ago by mmludin08

  • Type changed from defect to enhancement

comment:18 Changed 7 years ago by mmludin08

  • Resolution set to open
  • Status changed from qa to closed
  • Type changed from enhancement to defect
Note: See TracTickets for help on using tickets.