Redis-Backed Engine

With Preset set, Redis stays the source of truth while every instance serves reads from its own automaton. This page covers how that stays consistent. Which preset to pick is Preset-Optimized Engine.

Redis or Valkey. ACOR connects over RESP via go-redis v9, so Redis 3.0+ or Valkey 7.2+ works in Standalone, Sentinel, Cluster, and Ring topologies. Cross-instance invalidation uses server Pub/Sub, which behaves identically on both. CI covers both.

Architecture

                    Write Path
Instance A ──Add()──▶ Lua Script (optimistic lock) ──▶ Redis
                                                  │
                       Pub/Sub invalidate ◀────────┘
                            │
Instance B ◀────────────────┘
      │
      └─ ensureValid() ──▶ reload from Redis ──▶ rebuild local engine

                    Read Path
Instance A ──Find()──▶ local engine (0 RTT)

Invalidation safety

Pub/Sub is best effort — a disconnected subscriber misses invalidations. In multi-instance deployments set InvalidationPollInterval:

args := &acor.AhoCorasickArgs{
    Addr:                     "localhost:6379",
    Name:                     "my-collection",
    Preset:                   acor.PresetBalanced,
    InvalidationPollInterval: 30 * time.Second,
}
_ = args

Zero disables polling, and polling applies to Preset mode only. Each poll reads just the version hash field; a change marks the engine stale and the next search loads the full snapshot.

The interval is not a freshness bound. Redis failures, query latency, and rebuild time all delay recovery. This is eventual refresh, not strong consistency.

Concurrent stale reads share one reload job while each request observes its own context cancellation — cancelling one waiter leaves the others running, and all waiters leaving (or Close) cancels the job. Redis reads and engine builds run outside the state lock, and a generation check rejects snapshots overtaken by local writes.

Watch CacheStats().PresetReloadFailures (once per failed shared job) and PresetPollFailures (per failed version poll) alongside your search errors. Cancellation is excluded from both; both stay zero outside Preset mode. A retained previous engine does not turn a failed reload into a successful response. See Monitoring.

Topologies and connection tuning

All four topologies are configured through the connection fields on AhoCorasickArgs — see Redis topologies.

DialTimeout, ReadTimeout, WriteTimeout, MaxRetries, and PoolSize pass straight to go-redis for every topology. Zero keeps the go-redis default; -1 disables the read/write timeouts or command retries where supported.

Preset mode versus plain V2

Preset is unset by default (PresetNone), which runs the original mode. Choose it when you need the fastest reads across several instances and can accept V2-only, no-Suggest.