Ponder

Goal: By the end of 2025, even the largest Ponder app backfills will be fast enough to run in CI. Seconds or minutes, not hours or days.

It should feel like a build or reload, not a "backfill".

Building ponder.sh. EVM, TypeScript, open-source enjoyer.

Two main tricks
1) An architecture decision to decouple "sync" (download RPC data) from "indexing" (run user code and write to the database)
2) A in-memory cache for database rows to minimize blocking queries

🦀 /foundry @paradigm | founding engineer /ithaca

Two main tricks
1) An architecture decision to decouple "sync" (download RPC data) from "indexing" (run user code and write to the database)
2) A in-memory cache for database rows to minimize blocking queries

http://ithare.com/oltp-db-optimizations-102-100-coherent-app-level-cache-for-single-writing-db-connection/

What are the tricks that make Ponder fast and what makes Subgraph slow? Never fetching more than strictly necessary? Efficient batching? Bloom filters?

this would be amazing! Thanks for the hard work. truly appreciate it.