Bring back fabric map

[bertday]

Is your feature request related to a problem? Please describe.
It would be nice to be able to preview individual address points on a map (i.e. a point "fabric"). From what I understand this used to be a feature of the OA website, but stopped working at some point (see #346).

Describe the solution you'd like
A new slippy map with vector tiles of discrete OA addresses.

Describe alternatives you've considered
None so far.

Additional context
n/a

[ingalls]

@bertday We had a dot map of very very early data that we had on the website, but as far as I know it hasn't been auto-refreshing in years and the cost to host it was prohibitive to keep our kindly donated map usage free so it was removed.

I've sketched out a per-source live view with PMTiles. I attempted a "fabric" with PMTiles but got a lot of errors in generation and haven't gotten time to work through them.

@iandees would probably know more about the OG map on openaddresses.io and data frequency.

[bertday]

I've been using PMTiles + S3 on a few projects and it's been great so far. Sure beats hosting and wrangling thousands of individual Z/X/Y tiles.

Something else I'm using at the moment which maybe be of interest:

https://github.com/tobinbradley/dirt-simple-postgis-http-api

It's basically a zero-config Node server you point any Postgres DB and it serves up vector tiles on the fly. I've found it works well for real-time stuff you wouldn't want to tile every 15 minutes 😅 I run it on a DigitalOcean droplet for $6/month. No issues so far.

[iandees]

I think it'd be great to pipe the global collection through tippecanoe to generate a pmtiles after we finish building the collections. Put that on R2 and it'd be a good way to showcase the data we have available.

[ingalls]

@bertday Yeah I've done a lot with the internal PostGIS ST_AsMVT but we don't currently load the Address data into a database at any point due to the sheer volume of it, so tiling has to happen async. It is something we could explore and would certainly be the most straightforward option but would be fairly pricey to get a database to be performant with this much data + we would still have the loading/unloading cost.

The PMTiles+Tippecanoe combo is what I was testing out but was getting Leaf errors on creation.
The current code exists here: https://github.com/openaddresses/batch/blob/master/task/fabric.js
On my list to pick back up exploration of this again but if someone beats me to it, 🎉