viktorvsk / saltivka Goto Github PK

It seems that rate limiting won't work on DigitalOcean AppPlatform currently due to:

• https://www.digitalocean.com/community/questions/how-to-bind-server-to-ipv4-only-connections-in-app-platform
• https://ideas.digitalocean.com/app-platform/p/ability-to-change-listening-to-ipv4ipv6-clients

Cloud Flare seems to be able to handle this (https://developers.cloudflare.com/support/network/understanding-and-configuring-cloudflares-ipv6-support/#h_877db671-916a-4085-9676-8eb27eaa2a91) but it was not tested/adjusted to WAF deployments

Currently single Redis backend is used for all operations. However, workload may be split among different instances (each potentially having different settings and topology):

• Relay Redis
• Sidekiq Redis
• Cache Redis (in future when caching functionality is added)
• Misc Redis (i.e. rate limiting, HyperLogLog to count events and other stats, settings (i.e. maintenance)

Currently if RELAY_CONFIG.max_limit is set to 100 and REQ command is sent by a client with 5 different filters (i.e. ["REQ", "subscription", {...}, {...}, {...}, {...}, {...}] the response in theory will contain 500 records maximum (100 per filter). Also, in case filters have #authors filter, NIP-26 delegation will trigger another UNION to each subquery.

We could limit resulting set, but it could be not clear for clients, so the best and the most obvious solution should be defined and enforced.

According to NIP-01 e tags must have valid Event#sha256 in the second value and this value should be searchable by predicate and p tags must have valid Author#pubkey as the second value.

Currently it works in the same way NIP-12 is supported — all single letter tags are added as SearchableTag. However, it means those values are stored as is, having duplicates and occupying a lot of space. SearchableTags on e and p tags require the most memory among all indexes currently.

It would be great to store references to other tables instead.

It is pretty straightforward for p tag since we can just create an Author record before saving an event if it does not exist and leave this record in database even if Event is not valid (given that pubkey is valid).

However for e tag it's not as straightforward. Because it may happen e tag refers to the Event we don't know. And currently events table expects valid events and no drafts.

• Create Redis structure to hold latest (or the most popular?) events, say, for 1-3 days
• On REQ command find if any of recent/popular cached events match
• Consider how to handle LIMIT in this case (can we avoid database request at all?)
• TBD

Implement some handy features for relay operators who may want to sell names on exclusive domains:

• ENV variable to set one of 3 levels of configuration for users: free, `pay-with-day', 'subscribers-only'
• Enforce those levels
• ENV variable to set price in days for registering a name
• Define how to release registered names in case payment strategy changed
• Cache /.well-known/nostr.json endpoint aggressively since it may experience huge amount of requests
• Improve UI of handling errors on /profile page
• Make NIP-05 UI more user-friendly and add details on how and why to use it