Deep-dive documentation¶
These documents walk through major subsystems end-to-end. They are maintained against the main branch of the HyperbyteDB repository; when in doubt, treat the cited src/... paths as the source of truth.
| Document | Topic | Key source files to cross-check |
|---|---|---|
| Write path | HTTP → WAL → flush → Parquet | adapters/http/write.rs, application/ingestion_service.rs, application/flush_service.rs, adapters/storage/parquet_writer.rs |
| Read path | InfluxQL → CH SQL → chDB | adapters/http/query.rs, application/query_service.rs, influxql/, adapters/chdb/ |
| Compaction | Merge, compact_all, verified compaction | application/compaction_service.rs, application/compaction_merge.rs |
| Self-repair | Bucket hash, sync, membership repair | application/compaction_service.rs, cluster/ sync + peer paths |
| Clustering | Raft, replication, anti-entropy | cluster/raft/, cluster/peer_client.rs, cluster/sync.rs |
Validation notes (spot-checks)¶
- Parquet writer (
parquet_writer.rs): ZSTD level 1, max row group 65,536, page statistics — matches Write path §9. - Replication backoff (
peer_client.rs): exponential delay starting at 1s, capped at 30s, up toreplication_max_retriesattempts per peer send — matches Write path §13 narrative. - Proxy (optional front tier): not part of the database binary; see hyperbytedb-proxy.
If you find a mismatch, prefer updating the deep-dive text to match src/ (or open an issue if the code is wrong).