Deployment

Both binaries are statically-linked Rust. They are 12-factor friendly: configuration is environment variables only, state lives in Postgres, and the processes are stateless across restarts (the cursor table is the only durable indexer state).

Indexer environment

Var	Purpose	Default
LAYERS_DB_URL	Postgres connection URL	required
LAYERS_JETSTREAM_URL	Jetstream subscribe URL	wss://jetstream2.us-east.bsky.network/subscribe?wantedCollections=pub.layers.*
LAYERS_ES_URL	Elasticsearch base URL (enables the ES sink when set)	unset
LAYERS_ES_AUTH	Authorization header value sent to ES	unset
LAYERS_NEO4J_URL	Neo4j base URL (enables the Neo4j sink when set)	unset
LAYERS_NEO4J_DB	Neo4j database name	neo4j
LAYERS_NEO4J_AUTH	Authorization header value sent to Neo4j	unset
LAYERS_SUBSCRIPTION	Cursor slot identifier	layers-default
LAYERS_LOG	Tracing filter	layers_indexer=info,idiolect_indexer=info

Orchestrator environment

Var	Purpose	Default
LAYERS_DB_URL	Postgres connection URL	required
LAYERS_BIND	<host>:<port> to listen on	0.0.0.0:8080
LAYERS_DID	Service DID used as the expected aud in service-auth	did:web:layers.pub
LAYERS_LOG	Tracing filter	layers_orchestrator=info

Migrations

Apply the SQL files under layers/migrations/ in order. The Rust binaries also call ensure_table() on startup so a fresh deployment boots without an explicit migration step, but a managed environment should drive migrations through a deliberate tool (e.g. sqlx migrate).

Failure recovery

• Lose the Postgres database: redeploy and let the indexer replay the Jetstream from cursor zero. Every record can be rebuilt.
• Lose the Elasticsearch cluster: redeploy and restart the indexer (no cursor logic touches ES; it re-indexes from the next live event). For a clean rebuild, drop the indexer cursor and replay.
• Lose Neo4j: same shape as Elasticsearch.
• Lose Redis: caches and rate-limit windows reset; nothing durable is lost.