Pika Snapshot Backup Solution

Principle

Unlike Redis, Pika's data is primarily stored on disk, giving it a natural advantage for data backup — backups can be accomplished directly through file copying.

Implementation

Process

• Take a snapshot: Block writes (prevent clients from writing to the DB), and obtain the snapshot contents during this process.
• Asynchronous thread copies files: Copies snapshot files using a modified RocksDB BackupEngine, which prevents file deletion during this process.

Snapshot Contents

• All file names of the current DB
• Size of the manifest file
• sequence_number
• Sync point
  • binlog filenum
  • offset