Eric Leung Code, Data, Microbiome     About     Blog     Misc     Feed

Speed up Anaconda load on WSL

I use the Windows Subsystem for Linux on my work computer. Lately, the startup time for my Linux shell has taken too long for my taste and I set out to try and figure out why. I was able to figure out how to decrease my nearly 15 second wait (an eternity in programming) to nearly instantaneous. There is a slightly caveat to it but I don’t mind that extra inconvenience.

After a lot of searching around, I found out that my Anaconda/miniconda initialization was hogging all the time. This is because by default, I’ve set it up where conda activate base is called every time I create a shell. What the final solution does is remove this step and have you manually activate the environment whenever you need it.

Before I figured the eventual solution, I tried to blame the WSL shell itself. Looking around, I found there was an upgrade to WSL 2 available. This brought me to threads like this one.

One solution suggested to run

sudo apt update
sudo apt dist-upgrade

This gave me hope but it didn’t work. I eventually figured that because it was a work computer and I didn’t want to risk upgrading to a new system and everything breaking, I would abandon this potential solution.

This frustration then brought me to this thread. It sounds like I’m not the only one who has experienced this lag time. Even though the thread was from 2018, it seems relevant.

I gave their solutions a try. No luck.

The first thing I tried was change the absolute path to a relative one. I was skeptical this would work. And I was right in thinking so.

Scrolling down in the thread a bit more, I came across this comment. Near the bottom of the comment, it notes to comment out the code between # >>> conda initialize >>> and # <<< conda ini <<<. Then to just copy the inner if/else statements.

In my bash configuration (which should be somewhere either in .bashrc or .bash_profile), I have the following:

# >>> conda initialize >>>
# !! Contents within this block are managed by 'conda init' !!
__conda_setup="$('/home/leunge/miniconda/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
if [ $? -eq 0 ]; then
    eval "$__conda_setup"
else
    if [ -f "~/miniconda/etc/profile.d/conda.sh" ]; then
        . "~/miniconda/etc/profile.d/conda.sh"
    else
        export PATH="~/miniconda/bin:$PATH"
    fi
fi
unset __conda_setup
# <<< conda initialize <<<

I commented most of that out and copied out that inner if block.

if [ -f "~/miniconda/etc/profile.d/conda.sh" ]; then
    . "~/miniconda/etc/profile.d/conda.sh"
else
    export PATH="~/miniconda/bin:$PATH"
fi

Previously, my shell configuration essentially ran conda activate base with every new shell. With this new setup, I am no longer in an activated environment.

To double check that this was the issue, I timed it.

$ time conda activate base

real    0m15.461s
user    0m3.188s
sys     0m11.516s

Yep. That was the issue.

But can I still access conda and all of my tools? It turns out if I need to be in an Anaconda environment, I’ll have to remember to run conda activate base before doing anything. The export statement in the above code block ensures I still have access to conda and my Anaconda instance of Python.

This is a minor inconvenience I’m willing to take for the sake of time.

$ time bash ~/.bash_profile

real    0m0.131s
user    0m0.016s
sys     0m0.078s