Collections Service

Overview

The Collections service provides gRPC methods for managing collection lifecycle, configuration, and cluster operations.

Service Definition

From collections_service.proto:8-41:

service Collections {
  rpc Get(GetCollectionInfoRequest) returns (GetCollectionInfoResponse);
  rpc List(ListCollectionsRequest) returns (ListCollectionsResponse);
  rpc Create(CreateCollection) returns (CollectionOperationResponse);
  rpc Update(UpdateCollection) returns (CollectionOperationResponse);
  rpc Delete(DeleteCollection) returns (CollectionOperationResponse);
  rpc UpdateAliases(ChangeAliases) returns (CollectionOperationResponse);
  rpc ListCollectionAliases(ListCollectionAliasesRequest) returns (ListAliasesResponse);
  rpc ListAliases(ListAliasesRequest) returns (ListAliasesResponse);
  rpc CollectionClusterInfo(CollectionClusterInfoRequest) returns (CollectionClusterInfoResponse);
  rpc CollectionExists(CollectionExistsRequest) returns (CollectionExistsResponse);
  rpc UpdateCollectionClusterSetup(UpdateCollectionClusterSetupRequest) returns (UpdateCollectionClusterSetupResponse);
  rpc CreateShardKey(CreateShardKeyRequest) returns (CreateShardKeyResponse);
  rpc DeleteShardKey(DeleteShardKeyRequest) returns (DeleteShardKeyResponse);
}

Create Collection

Create a new collection with vector configuration:

Python Example

from qdrant_client import QdrantClient, models

client = QdrantClient("localhost", grpc_port=6334, prefer_grpc=True)

client.create_collection(
    collection_name="my_collection",
    vectors_config=models.VectorParams(
        size=384,
        distance=models.Distance.COSINE
    ),
    hnsw_config=models.HnswConfigDiff(
        m=16,
        ef_construct=100
    ),
    optimizers_config=models.OptimizersConfigDiff(
        default_segment_number=2
    ),
    replication_factor=2,
    write_consistency_factor=1,
    on_disk_payload=False
)

Go Example

import (
    "context"
    pb "github.com/qdrant/go-client/qdrant"
)

_, err := client.Create(context.Background(), &pb.CreateCollection{
    CollectionName: "my_collection",
    VectorsConfig: &pb.VectorsConfig{
        Config: &pb.VectorsConfig_Params{
            Params: &pb.VectorParams{
                Size:     384,
                Distance: pb.Distance_Cosine,
            },
        },
    },
    HnswConfig: &pb.HnswConfigDiff{
        M:          ptrUint64(16),
        EfConstruct: ptrUint64(100),
    },
    ReplicationFactor: ptrUint32(2),
})

Rust Example

use qdrant_client::{
    client::QdrantClient,
    qdrant::{
        vectors_config::Config, CreateCollection, Distance, HnswConfigDiff,
        VectorParams, VectorsConfig,
    },
};

let client = QdrantClient::from_url("http://localhost:6334")
    .build()
    .await?;

client
    .create_collection(&CreateCollection {
        collection_name: "my_collection".into(),
        vectors_config: Some(VectorsConfig {
            config: Some(Config::Params(VectorParams {
                size: 384,
                distance: Distance::Cosine.into(),
                hnsw_config: Some(HnswConfigDiff {
                    m: Some(16),
                    ef_construct: Some(100),
                    ..Default::default()
                }),
                ..Default::default()
            })),
        }),
        replication_factor: Some(2),
        ..Default::default()
    })
    .await?;

Named Vectors

Create collections with multiple named vectors:

from qdrant_client import models

client.create_collection(
    collection_name="multi_vector",
    vectors_config={
        "text": models.VectorParams(
            size=768,
            distance=models.Distance.COSINE
        ),
        "image": models.VectorParams(
            size=512,
            distance=models.Distance.EUCLID
        )
    }
)

Vector Parameters

From collections.proto:20-38:

message VectorParams {
  uint64 size = 1;  // Dimension of vectors
  Distance distance = 2;  // Distance function
  optional HnswConfigDiff hnsw_config = 3;
  optional QuantizationConfig quantization_config = 4;
  optional bool on_disk = 5;  // Store vectors on disk
  optional Datatype datatype = 6;  // float32, uint8, float16
  optional MultiVectorConfig multivector_config = 7;
}

Distance Functions

From collections.proto:137-143:

enum Distance {
  Cosine = 1;    // Cosine similarity
  Euclid = 2;    // Euclidean distance
  Dot = 3;       // Dot product
  Manhattan = 4; // Manhattan distance
}

Get Collection Info

info = client.get_collection(collection_name="my_collection")

print(f"Status: {info.status}")
print(f"Vectors count: {info.vectors_count}")
print(f"Points count: {info.points_count}")
print(f"Segments: {info.segments_count}")
print(f"Disk size: {info.disk_data_size}")

List Collections

collections = client.get_collections()

for collection in collections.collections:
    print(f"Collection: {collection.name}")

Update Collection

Modify collection configuration:

from qdrant_client import models

client.update_collection(
    collection_name="my_collection",
    optimizers_config=models.OptimizersConfigDiff(
        indexing_threshold=50000,
        memmap_threshold=100000
    ),
    hnsw_config=models.HnswConfigDiff(
        ef_construct=200
    )
)

Delete Collection

client.delete_collection(collection_name="my_collection")

Collection Aliases

Manage collection aliases for zero-downtime updates:

from qdrant_client import models

# Create alias
client.update_collection_aliases(
    change_aliases_operations=[
        models.CreateAliasOperation(
            create_alias=models.CreateAlias(
                collection_name="my_collection",
                alias_name="production"
            )
        )
    ]
)

# List aliases
aliases = client.get_collection_aliases(collection_name="my_collection")

# Rename alias
client.update_collection_aliases(
    change_aliases_operations=[
        models.RenameAliasOperation(
            rename_alias=models.RenameAlias(
                old_alias_name="production",
                new_alias_name="prod_v2"
            )
        )
    ]
)

# Delete alias
client.update_collection_aliases(
    change_aliases_operations=[
        models.DeleteAliasOperation(
            delete_alias=models.DeleteAlias(
                alias_name="production"
            )
        )
    ]
)

HNSW Configuration

From collections.proto:202-231:

message HnswConfigDiff {
  optional uint64 m = 1;  // Edges per node (default: 16)
  optional uint64 ef_construct = 2;  // Neighbors during build (default: 100)
  optional uint64 full_scan_threshold = 3;  // Threshold for full scan in KB
  optional uint64 max_indexing_threads = 4;  // Parallel indexing threads
  optional bool on_disk = 5;  // Store index on disk
  optional uint64 payload_m = 6;  // Payload-aware links
}

HNSW Tuning Example

from qdrant_client import models

client.create_collection(
    collection_name="optimized",
    vectors_config=models.VectorParams(
        size=768,
        distance=models.Distance.COSINE,
        hnsw_config=models.HnswConfigDiff(
            m=32,  # Higher M = better recall, more memory
            ef_construct=200,  # Higher ef_construct = better index quality
            full_scan_threshold=10000,  # Use HNSW for segments >10K vectors
            on_disk=False  # Keep index in memory for speed
        )
    )
)

Quantization

Enable vector quantization to reduce memory usage:

Scalar Quantization

from qdrant_client import models

client.create_collection(
    collection_name="quantized",
    vectors_config=models.VectorParams(
        size=768,
        distance=models.Distance.COSINE,
        quantization_config=models.ScalarQuantization(
            type=models.ScalarType.INT8,
            quantile=0.99,
            always_ram=True
        )
    )
)

Product Quantization

client.create_collection(
    collection_name="pq_collection",
    vectors_config=models.VectorParams(
        size=768,
        distance=models.Distance.COSINE,
        quantization_config=models.ProductQuantization(
            compression=models.CompressionRatio.X16,
            always_ram=True
        )
    )
)

Sharding

Configure sharding for distributed collections:

from qdrant_client import models

client.create_collection(
    collection_name="sharded",
    vectors_config=models.VectorParams(size=384, distance=models.Distance.COSINE),
    shard_number=4,  # Number of shards
    replication_factor=2,  # Replicas per shard
    write_consistency_factor=1,  # Acks needed for write
    sharding_method=models.ShardingMethod.AUTO
)

Custom Shard Keys

from qdrant_client import models

# Create collection with custom sharding
client.create_collection(
    collection_name="tenant_sharded",
    vectors_config=models.VectorParams(size=384, distance=models.Distance.COSINE),
    sharding_method=models.ShardingMethod.CUSTOM
)

# Create shard key
client.create_shard_key(
    collection_name="tenant_sharded",
    request=models.CreateShardKey(
        shard_key=models.ShardKey(keyword="tenant_a"),
        shards_number=2,
        replication_factor=2
    )
)

Collection Status

From collections.proto:145-155:

enum CollectionStatus {
  Green = 1;   // All segments ready
  Yellow = 2;  // Optimization in progress
  Red = 3;     // Error occurred
  Grey = 4;    // Optimization pending
}

Cluster Operations

Get cluster information for a collection:

cluster_info = client.get_collection_cluster_info(
    collection_name="my_collection"
)

print(f"Peer ID: {cluster_info.peer_id}")
print(f"Shard count: {cluster_info.shard_count}")
print(f"Local shards: {len(cluster_info.local_shards)}")
print(f"Remote shards: {len(cluster_info.remote_shards)}")

Collection Exists

exists = client.collection_exists(collection_name="my_collection")
if exists:
    print("Collection exists")

Best Practices

Performance Tips:

Use on_disk_payload=True for large payloads to save memory
Enable quantization for >100M vectors to reduce memory 4-16x
Set shard_number to number of nodes for distributed setup
Use replication_factor=2 or higher for production

Configuration Immutability:

Vector size and distance function cannot be changed after creation
To change these, create a new collection and migrate data
HNSW and optimizer configs can be updated on existing collections

Next Steps

Points Service

Learn about point operations

Python Client

Python client documentation

Overview

REST API

gRPC API

Client Libraries

Collections Service

Overview

Service Definition

Create Collection

Python Example

Go Example

Rust Example

Named Vectors

Vector Parameters

Distance Functions

Get Collection Info

List Collections

Update Collection

Delete Collection

Collection Aliases

HNSW Configuration

HNSW Tuning Example

Quantization

Scalar Quantization

Product Quantization

Sharding

Custom Shard Keys

Collection Status

Cluster Operations

Collection Exists

Best Practices

Next Steps

Points Service

Python Client

Overview

REST API

gRPC API

Client Libraries

​Overview

​Service Definition

​Create Collection

​Python Example

​Go Example

​Rust Example

​Named Vectors

​Vector Parameters

​Distance Functions

​Get Collection Info

​List Collections

​Update Collection

​Delete Collection

​Collection Aliases

​HNSW Configuration

​HNSW Tuning Example

​Quantization

​Scalar Quantization

​Product Quantization

​Sharding

​Custom Shard Keys

​Collection Status

​Cluster Operations

​Collection Exists

​Best Practices

​Next Steps

Points Service

Python Client

Overview

Service Definition

Create Collection

Python Example

Go Example

Rust Example

Named Vectors

Vector Parameters

Distance Functions

Get Collection Info

List Collections

Update Collection

Delete Collection

Collection Aliases

HNSW Configuration

HNSW Tuning Example

Quantization

Scalar Quantization

Product Quantization

Sharding

Custom Shard Keys

Collection Status

Cluster Operations

Collection Exists

Best Practices

Next Steps