Offline Sync Strategies

LivestockAI's "Action Era" requires robust handling of data conflicts beyond simple "last-write-wins".

The Problem: Dependent Mutations

A user offline might:

1. Create Batch A (Temp ID: tmp_123)
2. Log Feed for Batch A (Refers to tmp_123)
3. Record Mortality for Batch A (Refers to tmp_123)

If the sync order is wrong, or if Batch A creation fails on the server, the subsequent records will be orphaned or invalid.

Strategy 1: Optimistic ID Generation

Always generate UUIDs on the client using crypto.randomUUID(). Do not rely on server-side ID generation for primary entities.

// ✅ Correct
const batchId = crypto.randomUUID()
const batch = { id: batchId, ...data }
await db.batches.add(batch) // IndexedDB
await syncQueue.add({ type: 'CREATE_BATCH', payload: batch })

// ❌ Wrong (Server ID dependency)
const response = await api.createBatch(data)
const batchId = response.id // Fails if offline

Strategy 2: Operation Replay

The Sync Queue must process operations serially per entity.

interface SyncOperation {
  id: string
  entityId: string // Partition key for ordering
  type: 'CREATE' | 'UPDATE' | 'DELETE'
  payload: any
  occurredAt: number
}

Rule: If CREATE fails, all subsequent UPDATE operations for that entityId must be effectively paused or moved to a "Dead Letter Queue" for manual intervention.

Strategy 3: Conflict Resolution Modes

Field-Level Merge (JSON Patch)

Used for: Settings, Profile updates. User A changes Name. User B changes Email. Both succeed.

Last-Write-Wins (LWW)

Used for: Sensor readings, Status updates. The latest timestamp prevails.

Append-Only (Commutative)

Used for: Feed Logs, Mortality Records, Sales. These are "facts" that occurred. They don't update state; they append events. Even if two users log feed at the same time, both logs are valid.

Strategy 4: The "Soft Delete" Handling

Never hard delete entities on the client. Mark as deletedAt: timestamp. Server syncs the deletion. Other clients receive "tombstone" updates.

UI Patterns for Conflict

When automated resolution is impossible (e.g. User A sets Price=500, User B sets Price=600 for the same sale):

1. Quarantine: The item stays local.
2. Notification: "Conflict detected in Batch A."
3. Resolution UI: Show Side-by-Side comparison. "Keep Yours" vs "Keep Server's".

Related Skills

• offline-first - The mechanism (IndexedDB/TanStack Query)
• batch-centric-design - The UI context for these operations