hyperknot

openfreemap.org creator here. Yes, with vector tiles you are basically hosting static files, the server has nothing to do, except HTTPS encryption. Even gzipping is already done in the tiles.

It's also the first step in professional display calibration.

You are right, having two buckets is probably the easiest way to solve this. Then when loading the page it'd ask what to do with the autosave version.

Author here, I'll try to collect my thoughts into one comment here about why there is no auto-save.

If you implement auto-save, users will expect every click/keystroke to be saved, like how Google Docs does it. Simply adding a 1-minute timer will make people lose work in the last minute, as they switch tabs, go offline, etc. and forget about it for days, when the browser clears tab state.

The other reason is that there is no Undo-Redo implemented on MapHub. Implementing Undo properly in a web app is not trivial if you haven't designed the whole app around a time-traveling state management. It's actually a really difficult problem to solve, even though it looks simple on the surface. Combining it with real-time collaboration is even more complex. Short story: there is no Undo-Redo.

So if you don't have Undo, then Save button plays the role of making a checkpoint, trying out something, reverting if needed. With autosave this could be ruined.

Also, you write that you expect every web app to auto-save, but this is still not the universal case today and definitely wasn't the case a few years ago. I agree that most VC-backed startups with hundred-million-dollar valuations have autosave in their web apps. Again, the proper solution is to have a time-traveling diff system implemented, which can easily get really complicated with real-time multi-user collaboration. Have a look at Figma's technical blog post about this topic [1].

You might be right in saying that it's not a generation of users, but the same users being conditioned to web apps doing auto-save. I agree this is probably the case. The point of the article is this definitely wasn't the case all the time, and how I've experienced this on my side, during my 8+ years of running MapHub.

[1] https://www.figma.com/blog/how-figmas-multiplayer-technology...

Answered in detail here: https://news.ycombinator.com/item?id=42016231

The simple reason is that MapHub doesn't have a Google Docs like "time-traveling" version control nor Undo implemented. Undo is _really_ complicated if you don't start your app with that in mind. So the way to work is Save, try something and if you want to revert then you just reload the page.

Of course there is the "Previous versions" feature, but it still breaks my logic of Checkpoint-Experiment-Revert cycle. But I might be alone in this and users would be happier with autosave.

Update, answered in detail here: https://news.ycombinator.com/item?id=42016231

Yes, I agree, it might be a thing now.

Update, answered in detail here: https://news.ycombinator.com/item?id=42016231

The interesting bit is that for years this wasn't a problem. Like for 5 years I got 0 support requests about missing Save. Also web apps were not auto-saving traditionally, many of them still don't do it today.

Update, answered in detail here: https://news.ycombinator.com/item?id=42016231

The key is Undo-Redo information is lost. On Google Docs you can "time-travel" almost on a keystroke level, so that's when auto-save makes sense.

Update, answered in detail here: https://news.ycombinator.com/item?id=42016231

Hi, author here. I think it's not a good idea to auto save just at random intervals. I certainly don't want my any of my apps to auto save without me telling it to do so. Maybe I'm old fashioned in this regards.

Update, answered in detail here: https://news.ycombinator.com/item?id=42016231

Retested it, works brilliantly! I'll update the article accordingly.

Thanks for bringing it to the Free accounts, great outcome!

Great news, thanks for the amazing turnaround time!

I briefly mention that I don't go into L7 Load Balancing because it'd be cost prohibitive for my use case (millions of requests).

Also, the article is about DNS-RR, not the L7 solution.

Correct. I'm on macOS and I tried turning off DoH in Firefox and then it worked like Safari.

Thanks a lot for stating this. I'm happy to invest in AuthKit and recommend it if I know it will eventually support email change!

Blog post author. I think AuthKit is the best solution today!

But you really need to implement email change for users. It's a joke that it's not offered, that's a standard feature for every SaaS v0.1

Hi, blog post author here. This article is from Feb 2023, so quite dated at the moment.

I'll write an updated article for 2024 but as you've mentioned here, it's almost impossible to follow this landscape. So many solutions appeared during this time.

My recommendation right now would be to either:

- Use a reliable framework with a well-tested built-in solution like Rails/Django

- Choose WorkOS / AuthKit (1)

Some more thoughts:

- Clerk didn't turn out to what I thought when writing this article. Their offering is very much targeting the JS ecosystem, especially Next/Remix, it's not an universal solution.

- It's difficult to recommend hosted solutions when you see how much pricing changes even during 1.5 years. Like you decide on a solution then next year the company raises prices and you are stuck with it forever.

(1) Note for AuthKit: It's an amazing solution but it's a joke that users cannot change their emails. How can it be a production-ready solution, if it doesn't allow any app to offer email change for their users?

Yes, wanted to ask the same, what will happen to rclone? Is it unaffected?

Author here. That's a standard dialog built into the browser, which of course was implemented from Day 1. The problem is that people leave tabs open for indefinite time, the browser regularly saves memory by emptying old tabs and then it's gone.

I think he was referring to a freely available full planet OSM mbtiles dump, to be used with OpenMapTiles. I cannot find a reference to that in your blog post.

Yes, the key thing about Btrfs is how it handles inodes and how it can store data with the metadata. In the OpenFreeMap image, 60% of the files are stored with the metadata, essentially taking up no space.

Cloudflare Free is a good idea, but I'm afraid they'd be the one who limit me before. There is a comment in this thread who did 150 TB+ on Hetzner. I don't believe Cloudflare Free would allow 150 TB, not alone 300 TB in a month (I use two servers currently).

You are right in that I'd be afraid if this was hosted on AWS or on Cloudflare Workers. This whole idea is only possible because there are some hosts like Hetzner, which offer flat rate bandwidth.

Thanks a lot! Yes, I believe this dynamic pricing is making it impossible for small or non-profit projects to host maps. And you can see, "fun" maps basically disappeared from the internet - they were really common 10 years ago!

For POIs, they are present in the data, it's just that the styles might not have the matching icons for them. But you can always make them into a circle, etc.

Having said that, working on the styles is a big and never ending effort, so if someone would like to help with matching the POIs then contributors welcome.

Since I wrote it, one person made 1 commit related to the visual style. Before that, 1 visual commit in 2023 and then 2021 is when work kind of stopped.

Here the requests are really really small, on average 405 bytes each. I guess in your genomics work you are making larger requests, so probably it's not so much of an issue.

BTW, we are discussing latency with bdon in this issue, it seems to be specific to Cloudflare: https://github.com/hyperknot/openfreemap/issues/16

Thanks! It's not so bespoke, it's Fabric (https://www.fabfile.org/), it's been around at least since 2011, but probably much longer, since they already had 1.2.0 in 2011. It's a super reliable script for setting up servers over SSH. Yes, you run the same commands you'd type in a terminal, works perfectly.

I benchmarked and nginx can host 30 Gbps on cold cache, on actual real world load. (I played back a log of 1 million real world requests).

So using nginx in a 1 Gbps server is definitely not asking too much of it, or I might not have understood your comment clearly.

Why would Facebook decide to use my tile server? When a company's business depends on having reliable map tiles, they either self host or have an Enterprise deal with Mapbox, etc.

Thank you for sharing the story! I hope OpenFreeMap can contribute either with the MBTiles / Btrfs dumps, or with the public instance.

Yes, small files can fit in the metadata, which makes a super big difference when you have 300 million files of 405 bytes each. Also, inode handling is way better compared to ext4.