NewAISAQIndex()

This function creates an AISAQ index configuration, a DiskANN variant that keeps product-quantized codes inline with the graph to cut the number of random reads per hop.

func NewAISAQIndex(metricType MetricType) *aisaqIndex

PARAMETERS:

• metricType (MetricType)

  The distance metric type for similarity search (e.g., entity.COSINE, entity.L2, entity.IP).

RETURNS:

Index

An index configuration instance. Pass this to CreateIndex() via the index option. Every build parameter has a server-side default, so the rest are opt-in via the With* methods.

BUILDER METHODS:

• WithInlinePQ(inlinePQ int) *aisaqIndex

  Sets how many compressed vectors are stored inline in a node. Capped by the graph’s max degree; range [0, 2048].

• WithPQCacheSize(bytes int) *aisaqIndex

  Sets the compressed-vector cache size in bytes.

• WithRearrange(rearrange bool) *aisaqIndex

  Enables the compressed-vector reordering search optimization.

• WithPQReadIOEngine(engine string) *aisaqIndex

  Selects the IO engine used to read PQ vectors, either “aio” (server default) or "uring".

• WithNumEntryPoints(numEntryPoints int) *aisaqIndex

  Sets how many entry points are generated and stored when the graph is built. This is a build param, not a per-search knob.

• WithPQReadPageCacheSize(bytes int) *aisaqIndex

  Sets the per-thread read-page cache size in bytes. Honored at both build and search time.

Example

import (
    "github.com/milvus-io/milvus/client/v3/index"
    "github.com/milvus-io/milvus/client/v3/entity"
    "github.com/milvus-io/milvus/client/v3/milvusclient"
)

// Create index configuration
idx := index.NewAISAQIndex(entity.COSINE)

// Use with CreateIndex
createIdxOption := milvusclient.NewCreateIndexOption("collection_name", "vector_field", idx)
task, err := client.CreateIndex(ctx, createIdxOption)