v4.2.0
Overview
Migration difficulty: TODO
Breaking changes
New features
Blob retrieval improvements
Blob retrieval now uses a unified retry loop that alternates between consensus clients and file stores, replacing the previous multi-phase approach. This reduces retrieval latency from ~12s to ~1.5-3s when blobs aren't immediately available in file stores.
Non-supernode consensus hosts are automatically detected at startup and skipped during blob fetching, avoiding wasted requests.
Configuration:
BLOB_PREFER_FILESTORES=false # Try file stores before consensus (default: false)
BLOB_FILE_STORE_TIMEOUT_MS=10000 # HTTP timeout for blob file store requests in ms (default: 10000)
Set BLOB_PREFER_FILESTORES=true if your file stores are faster or more reliable than your consensus clients.