RecoveryStorage

Persists WAL consumer state to the catalog (etcd) and object storage. Core invariant: from any WAL position + the corresponding persisted state, RecoveryStorage can replay the WAL forward and recover a fully consistent in-memory state.

Persisted State

• WALCheckpoint (etcd): MessageID (= LastConfirmedMessageID of last consumed message), TimeTick, ReplicateCheckpoint (for secondary clusters), AlterWalState (for WAL backend migration).
• VChannel metadata (etcd): Per-VChannel collection info, partition list, schema history, state (NORMAL / DROPPED).
• Segment assignments (etcd): Per-segment growing/flushed status with row count and binary size stats.
• Segment data (object storage): Sealed segment binlog, indexes, and stats files.

Recovery Flow

1. Persist recovery (recoverRecoveryInfoFromMeta): Load checkpoint, VChannel metadata, and segment assignments from catalog in parallel.
2. Stream recovery (recoverFromStream): Build a RecoveryStream from the checkpoint's MessageID to the current WAL position. Replay all messages to reconstruct in-memory state. Extract uncommitted TxnBuffer.

Key Packages

• internal/streamingnode/server/wal/recovery/ — RecoveryStorage, RecoverySnapshot, WALCheckpoint, background persist task
• internal/streamingnode/server/flusher/flusherimpl/ — WALFlusherImpl, segment data flush to object storage