What happened.
Hydra PR #2767 changes its persisted event store from JSON to CBOR. On a real 3-node benchmark database, the reported size fell from 27.4 MB to 10.5 MB, or 2.6x smaller. The same record reports migration plus a full replay in 5.3 seconds.
The change includes an automatic, one-shot migration from schema v1 to schema v2. Existing event rows are re-encoded when the database is first opened. Hydra’s client-visible JSON output is described as byte-unchanged, so the storage change is intended to stay behind the existing client-facing format.
Why it matters.
For builders and stake pool operators running Hydra heads, smaller state files could reduce the storage footprint of persisted head events. The reported replay result also points to a potentially smoother startup path for the tested database. This is operationally relevant because the migration changes stored data rather than only an in-memory representation.
The main trade-off is migration risk. The re-encode is one way, with no downgrade path stated for schema v2. Operators considering an upgrade should treat a backup of hydra.db and a test migration as practical safeguards, especially where a rollback may be needed.
What remains uncertain.
This is a merged change on master, not a tagged Hydra release or a mainnet ledger change. The 2.6x result comes from one real 3-node benchmark database, so it does not establish the same reduction or replay time for every Hydra deployment. The supplied record does not provide broader deployment results.
Watch for a tagged Hydra release and for operator reports from the one-way schema v1-to-v2 migration.
Upstream references and independent checks
Digest dated 2026-08-18 · upstream model claude-sonnet-4-6. Direct links are available for 1 of 2 upstream source IDs; unresolved IDs remain visible for audit.
- 1CBOR-1: Switch to cbor format in DB (#2767)Direct upstream source ·
f8626b6d185ac02bef488a92864a5fd70d54e056 - 2
0066e6b3f00c1566e32bca736b9a5ff0cf67419fUpstream reference; direct URL unavailable.
This Research brief was generated by Terra from a dated upstream research digest. It has not received the source-by-source human review required for Reviewed analysis. Material limit: The size and replay figures come from a single benchmark database, and the change is merged to master rather than a tagged release.
