v0.2.0

CHANGELOG.md

@@ -1,12 +1,26 @@
 # RustBox changelog
 
+### 0.2.0
+- Fixed nested database transactions during message move by reusing the caller-owned transaction for cache removal, header-sync invalidation, folder invalidation, and search outbox writes.
+- Kept `search_outbox` in sync between tenant and legacy-import table sets so schema/import parity tests pass.
+- Decoupled SQL, blob storage and search: PostgreSQL/MySQL may still use local `data_dir` backends in single-node mode, while cluster mode no longer requires persistent `data_dir` when cluster-safe backends are configured.
+- Added `BlobStorage` with filesystem/shared-POSIX and S3-compatible backends, streaming attachment I/O, SHA-256 metadata, S3 retry/timeouts/checksums/SSE options and storage readiness probes.
+- Added idempotent attachment migration from legacy filesystem paths to logical object keys (`rustbox storage migrate`).
+- Added OpenSearch as the external full-text backend with an async HTTP client, per-tenant filtering/routing, deterministic document IDs, health checks and rebuild support.
+- Added a durable SQL search outbox with caller-transaction participation, claim/lease concurrency, retry/backoff, dead-letter state and lag/dead-letter diagnostics.
+- Hardened clustered indexing against re-enqueue/lease races, sized worker leases from the OpenSearch timeout/failover budget, and made IMAP IDLE deletions enqueue index removals in the same SQL transaction.
+- Added runtime schema migration V047 for attachment checksums and the search outbox across SQLite, MySQL and PostgreSQL.
+- Added Docker variants for standalone SQLite, PostgreSQL + SILO (S3-compatible) + OpenSearch, and MySQL + NFS + OpenSearch.
+- Added tests for cluster capability validation, filesystem blob streaming, OpenSearch tenant scoping and search-outbox transaction rollback/claim/idempotency behavior.
+- Bumped the application/source release to 0.2.0.
+
 ### 0.1.2
 - Added `log_level` to TOML configuration (`trace`, `debug`, `info`, `warn`, `error`, `off`); `RUST_LOG` remains an explicit runtime override.
 - Audited the MySQL and PostgreSQL query catalogs, schema baselines and backend-specific server SQL without adding application tests.
 - Fixed a MySQL DDL incompatibility by changing indexed `pgp_keys.email` from `TEXT` to `VARCHAR(320)`; MySQL requires a prefix length when indexing `TEXT`/`BLOB` columns.
 - Fixed destructive MySQL cache upserts: folder refresh and message-header refresh now use `INSERT ... ON DUPLICATE KEY UPDATE` instead of `REPLACE INTO`, preserving child rows, message tags, cached bodies and other columns not part of the refresh.
-- Documented and enforced cluster storage behavior: mailbox metadata/body cache remains in the shared SQL backend, while draft attachments require shared `data_dir` mounted at the same path on all nodes.
-- Added `search_index_backend = "tantivy" | "disabled"`; cluster mode requires `disabled` because the built-in Tantivy index has no multi-node writer coordination.
+- Documented and enforced cluster storage behavior: mailbox metadata/body cache remains in shared SQL, while draft attachments use either S3-compatible storage or an explicitly shared POSIX filesystem.
+- Replaced the pre-0.2 storage/search compatibility switches with the final `[storage]` and `[search]` configuration model; cluster mode validates backend capabilities directly and rejects local Tantivy.
 - Added `pgp_enabled`; cluster mode requires OpenPGP disabled because private-key material is filesystem-backed.
 - Avoided provisioning per-user Tantivy directories when the search index backend is disabled and made diagnostics/deep indexing honor that setting.
 - Added a release audit document with SQL compatibility and distributed-storage findings.