mirror of
https://github.com/luxfi/geth.git
synced 2026-07-27 01:59:25 +00:00
Hanzo AI
670f2b00cf
Implement ethdb.ResettableAncientStore over a single KV backend (the
canonical luxfi/zapdb). Replaces the upstream geth freezer
(.cdat/.ridx files per table) with one ZAP database per chain. Snappy
compression per-block (~10x cold data); single-byte kind markers keep
prefix scans efficient.
Key layout (single ZAP database per chain):
['a'][kind byte][big-endian uint64 number] -> snappy value
['m']['h'][kind byte] -> head pointer
['m']['t'][kind byte] -> tail pointer
ZAP is canonical Lux storage (luxfi/zap + luxfi/zapdb). PebbleDB stays
only as legacy in-place upgrade. New deployments use ZAP everywhere.
Surface:
- core/rawdb/zap_ancient.go (ZapAncientStore implementation)
- core/rawdb/zap_ancient_migration.go (exported wrappers for migrate-ancient)
- core/rawdb/zap_ancient_rlp.go (rlp.EncodeToBytes split out)
- core/rawdb/zap_ancient_test.go (7 tests: roundtrip, out-of-order,
range maxBytes, truncate head/tail,
snappy, persist-across-open)
- core/rawdb/zap_ancient_bench_test.go (write + read throughput)
- cmd/migrate-ancient/ (one-shot upstream-freezer → ZAP migrator)
- cmd/fleet-bench/ (CSV harness for write/read latency)
Benchmarks (in-memory backend, M1 Pro):
WriteThroughput: 6860 ns/op, 4 puts per block (header+hash+body+receipt)
ReadHot: 920 ns/op
Migration command for an existing chain:
migrate-ancient --src /data/db/<net>/chainData/<chainID>/ancient \
--dst /data/ancient/<chainID>
The luxfi/operator NodeFleet CRD ships archive pods with
--ancient-store-backend=zap as the default.
Tests: 7 zap-ancient tests + full core/rawdb suite race-clean.