Run beacon's raw-WS indexer in production and retire the legacy firehose

[melvincarvalho]

What

beacon serves did:nostr documents from Mongo, but the data is still kept fresh by the legacy nostr-beacon indexers (firehose, followshose) running under pm2 on nostr.social. firehose is crash-prone (hundreds of pm2 restarts). To own the full pipeline, run beacon's own clean-room indexer (src/indexer.js — raw WebSocket, no nostr-tools) in production and retire firehose.

Scope

• Deploy src/indexer.js under pm2 (e.g. beacon-indexer), subscribing to kinds 0/3/10002 across the relay set, latest-wins upsert into beacon / follows / relay_lists.
• Validate it keeps the collections fresh (new profiles/follows show up), then retire firehose.

Open question: follows schema

The new indexer writes the raw kind-3 event ({ tags: [["p", hex], ...] }). The existing prod follows collection is the derived shape followshose produces: { pubkey, follows: [hex, ...], count, created_at }. The resolver (diddoc.js) already reads both, but writing two shapes into one collection is messy — decide the canonical going-forward shape (and whether the new indexer should emit the derived shape for consistency).

Watch-outs

• followshose may also populate the relay-health relays collection (uptime stats) that the new indexer does not produce — confirm before retiring it, or keep that piece.
• The indexer writes to the live prod Mongo (same DB the live resolver reads) — stage carefully.
• Preserve latest-wins dedupe; don't regress already-indexed newer events.

Acceptance

• beacon's indexer runs in prod, keeping beacon / follows / relay_lists fresh.
• firehose retired (and followshose retired or its relay-health role preserved).
• No resolver regression; the chosen follows schema is documented.