When event indexing lags, the symptoms are obvious and painful: stale balances and missing transfers on user profiles, GraphQL endpoints returning incomplete timelines, production backfills that spike CPU and I/O and crush primary databases, and subtle correctness bugs caused by mishandled reorgs and duplicate events. You notice patterns: head processing keeps up for a while, historical queries choke the store, reorgs trigger mass rollbacks, and operational work escalates from a few minutes to overnight engineering sprints. Those symptoms tell you where the architecture needs to change: ingestion and storage, not just more RPC nodes.
Contents
Why latency and reliability are the product
When streaming wins and when batch beats streaming
Data modeling decisions: Postgres or ClickHouse for blockchain indexers?
