Rulia

Skeleton projects for deploying R/Shiny apps that call Julia code via Docker. Two approaches are provided:

Approaches

sysimage/ — JuliaCall + Sysimage

Uses the JuliaCall R package with a precompiled sysimage for faster startup. Julia is installed in the Docker image at runtime.

• Pro: Simple R integration via julia_call()
• Con: Requires Julia installation in the Docker image (~1.5GB overhead)

cd sysimage
docker build -t myapp-sysimage .
docker run -d -p 4444:3838 myapp-sysimage

sharedlib/ — Compiled Shared Library

Compiles Julia code into a shared library (libmyjulia.so) via PackageCompiler.create_library. R calls Julia functions directly via .C() — no JuliaCall, no Julia installation in the runtime image.

• Pro: Smaller image, no runtime Julia dependency, faster cold start
• Con: Requires C shim for ABI bridging, more build setup

cd sharedlib
docker build -t myapp-sharedlib .
docker run -d -p 4444:3838 myapp-sharedlib

Access

http://localhost:4444/MyApp/

How It Works

Both approaches call the same Julia function (mymean) from R/Shiny:

	Sysimage	Shared Library
R calls Julia via	julia_call("myjulia.mymean", data)	.C("c_mymean", data, n, result)
Julia compiled as	Sysimage (.so loaded by JuliaCall)	Shared library (.so loaded by dyn.load)
Docker stages	1 (Julia installed in image)	2 (builder compiles, runtime is lean)
Needs JuliaCall	Yes	No
Needs Julia in runtime	Yes	No