本页尚未翻译

本页内容仍为英文原文，源自官方文档 enterprise/tiered-index-cache。中文翻译正在分批进行中。

Tiered Index Cache

How LanceDB Enterprise caches deserialized index structures across memory and local disk.

Applies to: Embedded ❌, Enterprise ✅

The tiered index cache holds deserialized vector, full-text and scalar index structures in memory, backed by a local-disk tier beneath it. When memory fills, the cache serializes evicted entries and spills them to disk instead of dropping them; a later hit on a spilled entry promotes it back into memory. This lets a plan executor serve an index larger than its available memory without re-reading and re-deserializing it from object storage on every eviction. See Architecture.

%%{init: {"theme": "base", "themeVariables": {
  "background": "#ffffff", "primaryColor": "#ffffff", "primaryTextColor": "#221d19",
  "primaryBorderColor": "#ad9e95", "lineColor": "#6b625c", "textColor": "#221d19",
  "edgeLabelBackground": "#ffffff", "fontFamily": "Inter, sans-serif", "fontSize": "18px"
}}}%%
flowchart LR
    R["Index lookup"] --> M{"In memory?"}
    M -->|hit| Return["Served from memory"]
    M -->|miss| D{"Spilled to disk?"}
    D -->|hit| Promote["Deserialized,<br/>promoted into memory,<br/>served"]
    D -->|miss| O[("Object storage")]
    O --> Fill["Deserialized,<br/>loads into memory,<br/>served"]

    Evict["Memory eviction"] -.->|serialize, don't drop| D
    classDef default fill:#ffffff,stroke:#ad9e95,stroke-width:1px,color:#221d19

Sharing the disk budget

Where enabled, the index disk tier and the data page cache's disk tier draw from one combined, configurable disk allocation per executor, split between the two tiers so neither one needs its own dedicated volume.

Why it matters

Vector, full-text, and scalar indexes can easily outgrow available memory on large tables. Without this cache, an eviction means re-reading and re-deserializing that index from object storage on the next request -- slow, and repeated on every eviction. Spilling to local disk instead keeps index lookups fast even when the working set is larger than memory, without giving every executor enough memory to hold every index at once.

Contact us to learn more about how caching is sized and tuned for your workload.

Index cache interfaces

• Cache Prewarm for warming this cache ahead of the first request.
• Data Page Cache for the related cache that holds raw table data pages instead of deserialized index structures.
• Vector indexes, full-text indexes, and scalar indexes for the index types this cache serves.