indexer-reorg-idempotent

name: indexer-reorg-idempotent description: Enforce reorg-aware, idempotent indexing with stable event identity keys, replay safety, and clear optimistic vs confirmed consistency tiers. Use when touching DB writes or replay logic.

Indexer Reorg + Idempotency

When to use

• Adding/changing event handlers, DB schema, or replay/re-sync behavior.
• Any modification to how you store logs, claims, or day/month aggregates.

Goal (Definition of Done)

• ✅ Every handler is safe to replay (idempotent).
• ✅ Reorgs do not leave “ghost” rows or double-counts.
• ✅ You can rebuild projections from raw events.

Canonical identity keys

• Use unique keys:
  • (txHash, logIndex) as primary event identity (preferred)
  • or (blockNumber, logIndex) only if txHash is unavailable (rare)
• Store raw logs in event_raw (or equivalent) to support rebuilds.

Reorg handling

• On reorg, revert projections derived from orphaned blocks.
• Prefer strategies:
  • “Ponder-managed reorg rollback” + idempotent upserts
  • or explicit “confirmed depth” tables if you separate tiers

Consistency tiers (recommended)

• Optimistic: fast UX, may change on reorg.
• Confirmed: only after N confirmations; used for final rankings/payout history.

Anti-patterns

• ❌ INSERT without unique constraints (duplicates on replay).
• ❌ Mutating aggregates without a deterministic source-of-truth event stream.
• ❌ Assuming “finality” at head block.

Testing checklist

• Replay test: ingest same events twice → identical DB state.
• Reorg sim: orphan a block range → projections roll back correctly.
• Constraint tests: unique violation never crashes the indexer (use upsert).