AgentSkillsCN

storage-sync

在 Turso(持久化存储层)与 redb(缓存存储层)之间同步记忆数据。当缓存数据出现陈旧、遭遇故障,或在定期维护期间,此方法尤为适用。

SKILL.md
--- frontmatter
name: storage-sync
description: Synchronize memories between Turso (durable) and redb (cache) storage layers. Use when cache appears stale, after failures, or during periodic maintenance.

Storage Sync

Synchronize memories between Turso (durable) and redb (cache) storage layers.

When to Sync

  1. On startup - After system initialization
  2. Periodic - Scheduled background sync
  3. Cache staleness - When redb appears outdated
  4. Recovery - After storage failures
  5. Manual trigger - When explicitly requested

Sync Process

  1. Check connection health:

    rust
    turso_client.ping().await?;
    redb_env.check_integrity()?;
    
  2. Fetch latest from Turso:

    rust
    let episodes = turso_client
        .query("SELECT * FROM episodes ORDER BY timestamp DESC LIMIT ?")
        .bind(max_episodes_cache)
        .await?;
    
  3. Update redb cache:

    rust
    tokio::task::spawn_blocking(move || {
        let write_txn = redb.begin_write()?;
        let mut table = write_txn.open_table(EPISODES_TABLE)?;
        for episode in episodes {
            table.insert(episode.id.as_bytes(), episode.to_bytes())?;
        }
        write_txn.commit()?;
    });
    
  4. Sync patterns and embeddings if enabled

Configuration

rust
pub struct SyncConfig {
    pub max_episodes_cache: usize,  // Default: 1000
    pub batch_size: usize,          // Default: 100
    pub sync_patterns: bool,        // Default: true
    pub sync_embeddings: bool,      // Default: true
}

Error Handling

ErrorHandling
Turso unavailableSkip sync, log warning, retry later
redb corruptionAttempt repair, rebuild from Turso
Partial syncTrack progress, resume from last point

Performance Tips

  • Batch small operations
  • Incremental sync (only changes)
  • Parallel fetch with Tokio
  • Write-ahead preparation

Validation

After sync, verify:

  • Episode count matches
  • Latest episodes present
  • Pattern counts consistent
  • No orphaned embeddings

Troubleshooting

IssueSolution
Slow syncsReduce cache size, increase batch
redb lock errorsUse dedicated write task
Memory pressureStream large results, smaller batches