Nominatim on AWS ECS+RDS

[krishnajhadev]

Hi Team,
I want to set up reverse geo coding using nominatim using below desing :
Load balancer --> Nominatim service in ecs --> nominatim database in aws rds.
Can this design be possible, I checked online that this is possible but need to reconfirm?
Also is there any proper documentation for this type of set up in nominatim ?

[lonvia]

It is possible but not officially supported.

Latest Nominatim will happily work with a standard PostgreSQL/PostGIS installation, which AWS RDS should provide. Just make sure you have the permissions to creating functions and triggers. You might have to create the database manually and then use the unprivileged import.

If such a setup is a good idea, is another question. I doubt that it will have a very good performance or be particularly cost effective. Nominatim uses the database quite heavily and relies on having low-latency storage.

[krishnajhadev]

@mtmail for the above questions sir

[mtmail]

I'm not available for consulting contracts.

[krishnajhadev]

@mtmail sir, you can clarify the points I am confused on. This would be helpful

Right now we are using Nominatim 4.1.0 where Nominatim service (php based) and nominatim db is hosted in the same instance. Hence data base is hosted on machines where nominatim is hosted. Horizontal scaling would again make a new database + nominatim service set up.

Below things that needs confirmation :

• Why do we manually create the database ? We can provide the NOMINATIM_DATABASE_DSN before running any Nominatim set up command. This should set things up in aws rds while running commands in ecs (Nominatim ec2).
• Is connection pooling maintained by Nominatim service (for aws rds) via this NOMINATIM_API_POOL_SIZE ?
• Is nginx necessarily to be placed before Nominatim service ?
• Whats the port where nominatim should be running ?
• After doing nominatim import for new country, how can we make sure if import is done properly with all data inserted. is there any command to check if data is densely/properly populated?
• We have high throughput system and horizontal scaling with existing architecture is not feasible (Nominatim service + db) hosted on same machine. Hence we are looking forward for this set up (Load balancer --> Nominatim service in ecs --> nominatim database in aws rds). Your thoughts on this sir ?

[lonvia]

@krishnajhadev This is an issue tracker for the Nominatim software, not a free-as-in-beer support service for setting up your infrastructure. Harassing our volunteer contributors like that is not okay. Time to put an end to this discussion.

[mtmail]

• Why do we manually create the database? => That might be necessary if you don't have superuser permission for the database user. See https://nominatim.org/release-docs/latest/admin/Advanced-Installations/#importing-with-a-database-user-without-superuser-rights

• Is connection pooling maintained by Nominatim service? => Yes, that's the setting. https://nominatim.org/release-docs/latest/customize/Settings/#nominatim_api_pool_size I think it's unlikely you hit the limit.

• Is nginx necessarily to be placed before Nominatim service ? => No, you can use other webservers, or none even. In the past we had documentation for Apache2 and Caddy

• Whats the port where Nominatim should be running ? => You can choose any

• How can we make sure if import is done properly with all data inserted? => select count(*) from placex where country_code='...'; There's no feature to check for coverage, density, completeness.

• this set up (Load balancer --> Nominatim service in ecs --> nominatim database in aws rds) => Sure, that should work. I'd make sure only one Nominatim service is writing (running data updates) to the database.

• Also is there any proper documentation for this type of set up in nominatim ? => Not that I'm aware of.

In old github issues https://github.com/osm-search/Nominatim/issues?q=RDS you might see a 'nominatim.so' mentioned. Nominatim no longer requires that shared library file. It made installation on external hosted databases easier. We don't have much experience with AWS/Azure/GCP but users reported it works.