addprocs  : sh: 1: export:  usr bin julia bin julia: bad variable name  18587 Closed laurentheirendt opened this Issue on Sep 19 · 6 comments Projects None yet Labels parallel Milestone No milestone Assignees   amitmurthy amitmurthy 3 participants  laurentheirendt  amitmurthy  ViralBShah Notifications    laurentheirendt laurentheirendt commented on Sep 19 • edited      Relevant versions: Julia 0.5.0-rc4 & v0.4.6 & 0.4.5 || Linux 14.04 LTS     PATH is set to  usr bin julia bin  Adding processes using:  addprocs     first.last  server.com ,:auto  , tunnel true, topology :master_slave, dir `  `, sshflags `-p9000` ;  results in:  sh: 1: export:  usr bin julia bin julia: bad variable name  Is there anything that I am missing out here? This used to work ... On all the machines is the same version of Julia, freshly cloned.  amitmurthy amitmurthy was assigned by andreasnoack on Sep 19  ViralBShah ViralBShah added the parallel label on Sep 19  laurentheirendt laurentheirendt commented on Sep 20 • edited  Testing reveals that the issue is related to the function  launch_on_machine manager::SSHManager, machine, cnt, params, launched, launch_ntfy::Condition   in base managers-jl. The cmdstring is not properly built and $exename is called as part of the export command in $tval.  Following suggestions: 1  remove the ; after the ENV variable - gives an error when connecting      tval   haskey ENV,  JULIA_WORKER_TIMEOUT   ?         `export JULIA_WORKER_TIMEOUT $ ENV  JULIA_WORKER_TIMEOUT   ` : ``  2  change the cmdstring by adding && between $tval and $exename:      cmd   `cd $dir && $tval && $exename $exeflags`  Further testing is needed before submitting a PR.  laurentheirendt laurentheirendt referenced this issue on Sep 20 Closed addprocs   with SSHManager fails to load more than 1 process on remote node  18600  amitmurthy The Julia Language member amitmurthy commented on Sep 20  I cannot reproduce this locally on OSX.      This used to work ...  Which is the last version of Julia that it worked for you? Or did something else change when it stopped working?  Can you print the output of set | grep JULIA from your shell or the value of ENV  JULIA_WORKER_TIMEOUT   from the Julia REPL?  laurentheirendt laurentheirendt commented on Sep 20  I am actually trying to reinstall the old versions to figure out since when, but I am sure it still worked with v3.12.  Here is the output of the environment variable:  julia  ENV  JULIA_WORKER_TIMEOUT    1000    amitmurthy The Julia Language member amitmurthy commented on Sep 20  I just tested on an Ubuntu system locally and it worked fine.  Could you add a println cmd  after  cmd   `ssh -T -a -x -o ClearAllForwardings yes -n $sshflags $host $ shell_escape cmd  `  so that we can see how the cmd is being built in your environment?  laurentheirendt laurentheirendt commented on Sep 20  I upgraded all the Julia versions on all the nodes and on the host to v0.5.0. Most importantly, I set the environment explicitly in .bashrc as  export JULIA_WORKER_TIMEOUT 1000;  ... and it works now! It seems as if previously, the environment was set such that there was a whitespace after the number 1000:  julia  ENV  JULIA_WORKER_TIMEOUT    1000    Thanks for your help.  laurentheirendt laurentheirendt closed this on Sep 20  amitmurthy amitmurthy referenced this issue on Sep 20 Open Better handling and error reporting of incorrect JULIA_WORKER_TIMEOUT definitions  18602  amitmurthy The Julia Language member amitmurthy commented on Sep 20  Thanks. I have opened an issue to handle the error better -  18602