Groups


195 of 99+  



julia-users ›
Package Directory of Remote Workers
2 posts by 1 author  



Young-Jun Ko 	

Jan 26


Hello,

I am trying to use the parallel processing capabilities of julia in the following way.

I have two machine m1, m2 with the following characteristics:
- They do not share a file system. 
- User names are different
- Julia and packages are installed in different locations.

I connect m1 and m2 using master_slave topology, i.e. from m1 I run:

addprocs( [("m1", 1)], tunnel=true, topology=:master_slave, dir=<...>, exename=<...> )

Now, I have some data in an HDF5 file that I would like to load on m2.
HDF5 is installed on both machines and works when imported from the respective REPLs.
But when I run

addprocs( as above )
import HDF5

I get the following output:

julia> workers()
1-element Array{Int64,1}:
 2

julia> import HDF5
ERROR: On worker 2:
InitError: error compiling __init__: could not load library "<Pkg Dir on m1>/julia/v0.4/Blosc/src/../deps/libblosc"
/home/90days/ko/julia/v0.4/Blosc/src/../deps/libblosc: cannot open shared object file: No such file or directory

It seems that m2 is somehow using the configuration of m1.

However, if I check Pkg.dir() on m2, it returns the correct Pkg directory.

julia> d = @spawnat 2 Pkg.dir()
RemoteRef{Channel{Any}}(2,1,4)

julia> fetch(d)
"$HOME/.julia/v0.4" 

Does anyone have an idea how to debug whats going on?

Thanks!




 


Young-Jun Ko 	

Jan 26


What seems to work here is to put the import statements into the .juliarc.jl file on the remote machine.
This seems a bit clumsy though.
- show quoted text -