cerulea-relay

Realtime relay (1hr backfill window) for PDSes with fewer than 1000 repos.

The idea is that we can have much larger limits if we scale down the volume of the network.

• Large block sizes
• Large record size limit
• etcetcetc

todo

• metrics / tracing / observability shit
• keep track of currently subscribed PDSes in application state
• discover PDSes via crawling instead of hardcoding 1 (lol)
  • do not allow PDSes with more than 1000 repos
• robustness:
  • try to reconnect (tamed exponential backoff) if we lose connection to a PDS
  • allow catchup with cursor based on history data
• history:
  • timestamp instead of seq number as key
  • purge based on ttl
  • takedowns probably