name: titlepage class: middle, centre # Julia for scripting Fredrik Ekre, JuliaCon 2020 Slides @ [fredrikekre.se/juliacon](https://fredrikekre.se/juliacon/) --- # What is "scripting"? Executing a script from the command line, e.g. ```bash $ cat script.jl println("Hello JuliaCon!") ``` -- ```bash $ julia script.jl Hello JuliaCon! ``` --- name: problem # What are the problems? -- * Julia startup time ```bash $ time julia script.jl Hello JuliaCon! real 0m0.170s user 0m0.111s sys 0m0.045s $ time python script.py Hello JuliaCon! real 0m0.011s user 0m0.004s sys 0m0.007s ``` -- * Package load time -- * Compile time -- *This is continuously beeing worked on and improved with each Julia release.* --- # Is it a problem? -- * Yes, for short workloads (runtime ~ startup + compile time), e.g. quick tasks for interactive use at the command line -- * No, for longer workloads where time spent on compilation is negligible compared to compute time, e.g. cluster computations --- name: solutions # Possible remedies Fewer Julia startups: * keep the same session running for a long time * pay for startup once, pay for compile time once * very useful workflow together with e.g. Revise -- Now we are not scripting... -- <hr /> .daemon[ *[DaemonMode.jl](https://github.com/dmolina/DaemonMode.jl) by Daniel Molina (@dmolina):* * *Server: Julia session that is running in the background* * *Client: Wrapper script that passes everything to the server* ] --- name: solutions # Possible remedies Save compiled code for reuse: * enhance the Julia sysimg by appending compiled code corresponding to your workload * particularly useful to decrease package load times * specify sysimg on startup: ```bash $ julia --sysimage custom-img.so script.jl ``` -- *See Kristoffer Carlsson's talk on [PackageCompiler.jl](https://github.com/JuliaLang/PackageCompiler.jl) from JuliaCon 2020.* --- name: solutions # Possible remedies Spend less time compiling: * decrease/skip compiler optimizations (see the `-O`/`--optimize` flag) * utilize Julias interpreter (see the `--compile` flag) --- name: example # Example [`jlpkg`](https://github.com/fredrikekre/jlpkg): a command line interface to Julia's package manager * `--compile=min` to minimize compilation * `--optimize=0` to minimize optimizations * benefits from not using any external packages --- name: example # Example [`jlpkg`](https://github.com/fredrikekre/jlpkg): a command line interface to Julia's package manager * `--compile=min` to minimize compilation * `--optimize=0` to minimize optimizations * benefits from not using any external packages ```bash $ time jlpkg status real 0m0.210s user 0m0.253s sys 0m0.080s $ time julia -e 'using Pkg; Pkg.status()' real 0m0.444s user 0m0.442s sys 0m0.129s ``` --- name: example # Example [`jlpkg`](https://github.com/fredrikekre/jlpkg): a command line interface to Julia's package manager * `--compile=min` to minimize compilation * `--optimize=0` to minimize optimizations * benefits from not using any external packages ```bash $ time jlpkg add DataFrames real 0m0.864s user 0m0.929s sys 0m0.125s $ time julia -e 'using Pkg; Pkg.add("DataFrames")' real 0m1.958s user 0m1.644s sys 0m0.185s ``` --- name: example # Example [`jlpkg`](https://github.com/fredrikekre/jlpkg): a command line interface to Julia's package manager * `--compile=min` to minimize compilation * `--optimize=0` to minimize optimizations * benefits from not using any external packages ```bash $ time jlpkg rm DataFrames real 0m0.209s user 0m0.257s sys 0m0.145s $ time julia -e 'using Pkg; Pkg.rm("DataFrames")' real 0m0.701s user 0m0.691s sys 0m0.200s ``` --- class: center, middle # Thanks for listening!