[MarxOk]

Sites like this often go down when they reach the HN first page. I've naively deployed stuff on AWS free tier with no scaling or anything that's handled thousands of concurrent requests out of the box. Is the HN kiss of death that bad, or is it just that a lot of people use weird/shared hosting providers?

[woudsma]

I've had the honour to build clickclickclick.click (frontend/backend), with the truly unique team at Studio Moniker in Amsterdam.

I was quite inexperienced - 1 year after graduating from art school - when I joined and started working on this project. The server went down 5x in a single day because of all the requests. We were using web sockets (simple node server with express + socketio and a react + rxjs frontend) and that put some strain on the server. And I made the rookie mistake of storing images directly into the database instead of in a S3 bucket. Also we chose CouchDB, which saves document revisions (I didn't know..) so at the end of the day the database took 100% disk space and I couldn't SSH into it anymore. There were something like 20M+ database writes within 24hrs of the launch because it went viral.

We span up larger DO droplets several times because of this issue which took a while to fix.

It was my first time setting up / working with a VPS and nginx + sockets as well, so I'm actually quite pleased with only 15-30mins downtime overall on launch day :)

I learned a lot during that time (it was made in 2016), especially with the help of HN. Projects are often developed by junior devs or people skilled in other areas. Some experience is useful when deploying scalable apps. Nowadays it's much more easy with services like Vercel/Cloudflare/Heroku/AWS/etc. We used SFTP to deploy the site. I think I moved it to a Docker container some time after that.

I'm still using DigitalOcean to this day for my personal stuff, only now I use Dokku + Cloudflare - which works like a charm.

Check out https://studiomoniker.com if you want to see more crazy projects! (They're my previous employer FYI).

[pontifier]

Fantastic projects there. I'm interested in "Red Follows Yellow Follows Blue Follows Red". It seems like it would be tons of fun for school kids. I can see a kit of headphones and capes being something a teacher could use to have a fun activity without too much work.

[pontifier]

Dang... I want to build a sand pen too!

[Lorin]

re Studio website: minor spook when the arm comes from the bottom of the screen to tap on the shop link unexpectedly

[PhillyG]

I'd read this comment and still had that horror movie jump scare feeling!

[pimlottc]

What’s frustrating is the lack of error handling, especially in an intentionally “mysterious” game like this. I spent a few minutes trying to “figure it out” before I realized, nope, it’s just broken (web socket connection failed).

[Honga]

We're (Moniker) doing palliative care for it at the moment. The technology is quite old.

Sorry about that! There may be an initiative to update it in the near future.

[c17r]

Doesn't seem to recognize the latest Chromium version of IE Edge

[mudlus]

Same here, had to switch off of Firefox

[Izkata]

Worked for me the second time I loaded it in Firefox

[codegeek]

It depends on what type of site it is. A dynamic site with lot of database calls/no caching would probably crash much quicker than a static HTML page with same amount of traffic. HN easily sends 100s of concurrent users if not many so it can crash a shared hosted dynamic site with no caching.

[brundolf]

My website is hosted on the cheapest non-free Heroku tier ($7), and served by a single Node process. I've had a couple of posts get to #1 on HN (one stayed there for 24+ hours if I remember correctly), with zero server issues.

I do statically-render everything, so not doing that is the only reason I can think of for why so many sites might be going down when they get up on the front page. Many of these are blogs so could probably do much better, though I did see someone mention that this one uses websockets for something, so it would definitely be doing some logic on the server-side and wouldn't be able to go fully static.

[woudsma]

You should look into Dokku!

I'm hosting 30+ apps (frontends / api's / cms's / other processes) on a single 5eu/mo DigitalOcean droplet. I believe Hertzner is the cheapest hosting provider with a 2,87eu/mo VPS.

Self-hosting is a lot of fun, and cheaper. Dokku has been around for several years, it's an open-source Heroku clone and I would recommend it to anyone looking into deploying web apps.

[brundolf]

I've heard of it; for me, messing with a system (even once) isn't really fun, it just gets in the way of spending time on my code. I don't have a ton of projects and Heroku has a free tier for things that don't need 100% uptime, so it's worth the extra couple bucks for me

Edit: I see DigitalOcean has it available as a 1-click install. Do you get totally automatic updates/system management and everything with that?

[ornornor]

How do you keep things secure and up to date? That’s always been my problem and when developing an app I don’t want to do system administration so I always end up going with heroku (or it’s cheaper sibling render.com lately) because I don’t have time for system administration, backing up, etc.

[syncsynchalt]

When I put up tls.ulfheim.net (a small-ish static site on a t2.micro) HN and Reddit were able to bring it down by maxing out the apache workers.

Some config changes fixed it right up, but my point is that it's not just the capabilities of the instance, the default http server configs might need some tweaking too.

[rkv]

Out of curiosity, what config changes did you use on apache?

[dopidop]

Same questions here. I’m not sure I would know where to start on those particular configs

[tibu]

Maxrequestworkers, server limit?

[syncsynchalt]

IIRC it was MaxRequestWorkers to 1500, and MaxConnectionsPerChild to 0 (unlimited)

[mceachen]

Here are some analytics from an HN hug a month an a half ago: https://forum.photostructure.com/t/front-page-of-hacker-news...

(Static sites should be fine, but many sites use dynamic CMS like wordpress, ghost or discourse, on very low-powered hosts, which really need a caching layer to hold up under non-trivial load)

[geek_at]

either badly coded website (heavily relying on some backend without static files) or weak hosters but not sure I have a blog that survived every single HN/reddit hug on a pretty weak VPS.

Blog written in PHP but without database interaction

[ddtaylor]

It appears to be hosted on Digital Ocean.

[jcpham2]

Surely not the first appearance?