Downloading/installing JuliaPro on a headless machine


#1

Dear Julia Computing,

I find it very uncomfortable to install JuliaPro on a machine that does not have a GUI (headless). A typical scenario is an Ubuntu instance in the cloud (AWS, Azure) having only a terminal (SSH) connection available.

JuliaPro requires registrations and provides download links. Unfortunately, those links do not work with wget. Trying to download the file and copying the final download link does not work neither.
Hence, currently the only workaround for getting JuliaPro in headless environment running is to download JuliaPro installer (1GB) to my local machine and then upload it back to a virtual server in the cloud.

This problem could be easily solved if Julia Computing could provide links that are short-lived but can be pasted to wget and just work. Please note that installation is the first experience any newcomer has with Julia so I believe it should be just more friendly…

P.S.
There are also JuliaPro AMIs for AWS (https://juliacomputing.com/products/ami/juliapro-amis.html) but they do not have the latest JuliaPro version. On the other hans maybe not many people use this…


#2

Is there anything specific you need from JuliaPro that the official binaries + some packages cannot provide?

Specifically, if you include a Manifest.toml, then pkg"instantiate" should set the define package environment for you. This should be very convenient for a headless/cloud setup, as you can reproduce everything exactly.


#3

I second @Tamas_Papp. To me, the only real advantage of Julia Pro over usual julia + packages is that it ships with MKL. However, this can be easily solved by downloading the julia source code and compiling it with USE_MKL = 1.

Nonetheless, I guess it would still make sense to have better links available.

(Be aware though, that PyPlot, the wrapper to python’s matplotlib, won’t work with any Julia MKL installation up to this point because of some numpy incompatibility.)


#4

Using official binaries + packages or building Julia (with MKL, sometimes with LIBM as well) is what I do every time.

However, my post is not about myself, but about new people coming to the Julia ecosystem.

I know companies who have R and Python analytical groups (run on AWS or Azure via RStudio and Jupyter notebooks) and they want to try Julia (e.g. because I recommended them to try it out). Corporate world has many regulations and they rather want to use JuliaPro (e.g. they will need official support if they go into production). In my post I have just described the first experience those folks have with JuliaPro.

A very similar registration web page used to be present few years ago at the Python Anaconda website. But once one was registered you could copy the download link and continue the installation on different machine.

I think that for the first time (corporate world) users such small simple things that just do not work simply do not help convincing them to adopt Julia.


#5

Since it is the same language (Julia) and should be fully compatible (with a given package environment + MKL), I don’t see why they couldn’t just get support from Julia Computing or any other entity. If this is a concern, I would just contact JuliaPro directly for pricing (which they would need to do anyway for a SLA I assume).


#6

Also, please note that the packages are just catching up to 1.0, so while trying out Julia is fine, and one can start learning the language without any problems, using it in production takes quite a bit of dedication.


#7

other alternative - the JuliaPro - DockerHub version ( 0.6.3.1-ubuntu ; 0.6.3.1-centos )

but the latest JuliaPro Version : 0.6.4.1

$ docker images | grep juliapro
juliacomputing/juliapro    latest  e21e3188c45f    3 months ago    4.27GB

The Jupyter Notebook - can be used via “SSH tunnel”

An alternative - The “official” Julia Docker image is v1.0.1


#8

Thanks for collecting usefull links in one place.
Let’s hope that Julia 1.0 Pro will be simply downloadable with wget (after registration etc of course).