Configure komoot photon server

[davecraig]

For our geo-search during development we have been using the demo photon server. For deployment we'll need our own server. The home page of the project is https://github.com/komoot/photon and there are some simple instructions on installation here.

The summary is that there are two pieces that need installed:

1. The application itself from https://github.com/komoot/photon/releases
2. The world wide index from https://download1.graphhopper.com/public/

Original postings on platform slack:

It's trivial to run the photon server locally (see https://nominatim.org/2020/10/21/photon-country-extracts.html) and it just involves downloading a release, downloading the latest photon index (~76GB for the world, or ~3GB for just the UK) and running it. However, there's not a lot of information about the CPU/RAM required to run photon, all I found was komoot/photon#726.

I did run some artillery.io testing against a local install and also posted the results

It took me 20 minutes to get started and run a test against a local photon server using "made up street names" from one of their plugins https://ngneat.github.io/falso/docs/address#randstreetname
I was only running a UK photon index and the memory usage on the server was pretty much just the size of the unzipped index (5.7GB).
https://app.artillery.io/share/sh_20965116282a81feae7b5b1f7ba011ac31884fc9eebb94b30d9f1123830eafda
If I limit the memory that I give the Java process running photon to 1GB then it runs a fair amount slower at peak, but still serving up ~120 requests per seconds which is still way more than I expected. I'm going to download the world index overnight and see how low I can trim the memory when load testing that.

I haven't got as far as load testing the worldwide index, so that still needs looking at.