# redesign: i rewrote this site's server code from scratch

this site used to serve all the posts in plaintext. but now it serves them all in html so it's convenient to read the content from any device. this is because i rewrote the whole server from scratch with a little bit different goals this time.

# server

i wrote the old server in c. it was quite simple but i found it a bit hard to modify after a while. in the past year i was working with go and found it a nice language so i decided to rewrite the whole thing in go.

the old c code's main loop was this:

• accept connection
• read what the request is for (thanks to TCP_DEFER_ACCEPT this was already ready after accept)
• look up the post in memory (all posts were served from memory, disk wasn't touched in the serving path)
• write the contents (i've set the tcp buffer to be at least 100 KiB and ensured all my posts are shorter so my tcp writes never blocked)
• close the connection and repeat the loop

it was a dumb server but super reliable. i didn't even need any state tracking and dynamic memory allocation. the server was single threaded so it couldn't become overloaded at all. the kernel itself could still become overloaded but at least that's not the server's problem.

with go things became a little bit more complex with lots of goroutines and dynamic memory allocations. and i had to be a little bit more careful about ensuring it doesn't fall over for some obvious attacks. but in exchange the code was much simpler to write because go is a really convenient language. now it will be much easier for me to add more features if i ever want to (e.g. comments). my little rpi might break sooner if i expose a more complicated server on it to the wild internet. but hey, live and learn.

# backup

an important thing i want from my blog is easy consumability. i want a single megapage that contains all the posts one page that people can simply scroll through. i call this "the backup" page and i want it to be available even if my domain or myself is unavailable. i host this on github (https://ypsu.github.io/blog/) so that it doesn't rely on my tiny webserver.

with the old solution i always simply regenerated the full backup site whenever i created a new post. but over time this became quite slow on my old raspberry pi as the number of posts and the length of the full backup page grew. this was getting more and more annoying as time went on.

in my new solution i use an automated github action for this. apparently github lets you run arbitrary code on specific events completely free for anyone on the internet. i guess i got to try it while it lasts. so it's quite neat now: i push a new post, and the backup page is automatically updated in the background after a few minutes. and even if free github actions are gone, i can now do this easily as a background task on my rpi without disrupting my usual post creation workflow.

# html

i was in a little bit of a pickle when it came to plaintext vs html. i really like the simplicity of plaintext but having the flexibility of html is sometimes awesome. i've decided it's time to go mainstream and have things in html. especially given in the last year i've created a few html demos. maybe this way i'll be tempted to do more of that type of stuff.

but i'll still write everything in plaintext files, it's just the presentation that changes.

the migration to html was quite easy because i wrote all my posts with markdown in mind. the backup page from before was already rendered as html with markdown so i was pretty much prepared for this since the beginning.

# summary

rewriting the site was fun and should make reading the dumb brainpoop i write here much easier. if i ever write something worth sharing now i hope the plaintext-ness of the site won't discourage people from reading it.

published on 2021-09-05, last modified on 2023-09-04

to the frontpage