Firehose is explicitly positioned as files-based and streaming-first blockchain-data infrastructure, not as a generic node RPC product.
The most important design move is decomposition of ingest into separate roles: instrumented nodes, readers, relayers, mergers, stream servers, and downstream Substreams processing.
The docs frame Firehose as a response to brittle JSON-RPC patterns and slow reprocessing, which makes it a strong comparison point for anyone claiming to index chain data faster without exposing their ingest architecture.
Cursor and ForkDB handling are central: Firehose treats resumable, fork-aware stream state as part of the product rather than as an afterthought.
Low-latency behavior is not only a performance anecdote; the docs explicitly describe racing readers fronted by relayers that deduplicate and forward the first-arriving block.
The Graph’s own docs say Firehose is intended to replace The Graph’s original extraction layer, so it matters as a sublayer inside a broader indexing and data-services ecosystem.
The firehose-core README reveals a second useful layer: governance and maintenance flow through a shared upstream core that chain-specific integrations are expected to track.
Specific source signals worth retaining
Firehose overview: strongest concise articulation of why the system exists and how it differs from RPC-style access.
Firehose components page: best short source for the explicit system decomposition into readers, relayers, mergers, Firehose, and Substreams.
The Graph Firehose docs page: strongest signal that Firehose matters inside The Graph stack rather than only as a StreamingFast-side product.
firehose-core README: best source for the shared-maintenance and chain-integration model that would be easy to miss in higher-level docs.
Why this entry belongs in the corpus
It adds a lower-layer ingestion and stream-delivery comparison point beneath The Graph, Substreams, and generic blockchain-data infrastructure labels.
It makes blockchain-data access legible as node instrumentation, file-based history custody, relay topology, cursor semantics, and chain-integration maintenance rather than one flat indexing bucket.
The durable insight is that faster and more reliable data access often comes from restructuring the extraction pipeline itself, and that whoever controls those extraction, storage, relay, and core-maintenance layers can become a real chokepoint.