s4core/s4core v1.3.0-anti-entropy

on GitHub

Release v1.3.0-anti-entropy: Anti-Entropy, Shared Cluster State and Online Pool Growth

Release Overview
Replicas now recover missed changes through anti-entropy, while bucket settings and IAM state are shared across the cluster. Enterprise pools can grow with verified, resumable rebalancing.

🚀 New Features (All Editions)

• Anti-entropy and safe deletion: Replicas recover missed writes and deletes after hints expire. Deletion history is retained until repair confirms it is safe to collect.
• Shared bucket configuration: CORS, lifecycle, policy and encryption settings are stored and read through cluster quorums. Lifecycle rules run on every node holding object copies.
• Cluster IAM: Users, roles, S3 credentials and revocations are shared across nodes. S3 secrets are encrypted at rest and during replication; admin sessions follow the account's current permissions.
• Single-to-cluster migration: Existing buckets, objects and multipart ETags are automatically copied into a replicated cluster.

🏢 Enterprise Edition

• Online pool growth: Append nodes and let verified, resumable rebalancing move existing objects and active multipart uploads. Progress and pause/resume controls are available in the cluster admin API.
• Resilient bucket metadata: In an EC 4+2 pool, bucket existence records and settings survive any two node failures once their automatic migration completes.

🐛 Bug Fixes

• Cluster recovery and listings: Crashes, returning replicas and concurrent placement changes no longer lose newer writes or cause intermittent listing failures when quorum is available.
• EC deletion and promotion: DELETE refuses a missing namespace quorum instead of confirming an uncommitted deletion. Repairs made obsolete by a confirmed promotion no longer appear as permanent failures.

⚠️ Upgrade Notes

• Rolling upgrade: Upgrade one node at a time, preserving quorum, and finish the whole pool before expecting background repair and metadata migration. Cluster protocol is now 9; do not downgrade after metadata migration or a pool move begins.
• IAM: Log in again after upgrading. Review and adopt existing per-node accounts through /api/admin/iam/adoption before IAM changes resume. Keep existing shared secrets unchanged and identical across nodes; root is configured through the environment.
• Load balancers: Route clients using /minio/health/ready; keep /health or /minio/health/live for liveness. For single-to-cluster migration, wait for migration_pending_records to reach zero on every node before switching clients; versioned sources and EC destinations are refused.
• Cluster feature support: Configuration writes for versioning and Object Lock now return 501 instead of accepting protection that cluster mode does not enforce.
• Tombstone retention: Custom S4_GC_GRACE_DAYS must cover the configured rejoin and repair window; unsafe values refuse startup.
• EC availability: Hot objects and their namespace still use RF=3. Losing two of a key's namespace owners can make PUT, DELETE and LIST return 503, even when EC shards remain recoverable.

Artifacts

• s4-server-linux-x86_64.tar.gz
• s4-server-linux-x86_64.tar.gz.sha256

Full Changelog

v1.2.0-extended-metrics...v1.3.0-anti-entropy

Contributors

• Dmitry Rakov