Constraints and limits

更新时间: 2026-01-18 19:36:19

Data consistency

Because of the read/write splitting architecture, a Doc may not be immediately available for Retrieve Doc and Get Doc operations after an Insert Doc, Update Doc, or Insert or Update Doc operation. Similarly, a Doc may not be immediately hidden from Retrieve Doc and Get Doc operations after a Delete Doc operation.

DashVector ensures eventual data consistency. All Doc operations are strictly executed in the order they are received.

Free instance (cluster) limits

Note

To adjust the limits for free instances (clusters), such as the number of collections, maximum number of Docs, or Queries Per Second (QPS) limits, or if you have any requests, feedback, or need technical support, contact us through the following channels:

Limit

Condition

Number of free clusters per UID

1

Number of collections per free cluster

2

Free cluster usage duration

1 month. After the cluster expires or is manually released, you can request to create a new free cluster.

Get Doc

  • QPS limit: 100

Delete Doc

  • QPS limit: 100

Insert Doc

  • QPS limit: 100

Update Doc

  • QPS limit: 100

Insert or Update Doc

  • QPS limit: 100

Collections in a free cluster have a fixed amount of storage space. The maximum number of Docs and the QPS for Retrieve Doc adjust dynamically based on the vector dimensions. The current maximum QPS for free instances is 30. Use the following table to estimate the maximum number of Docs and the maximum QPS for Retrieve Doc:

Vector dimensions

Theoretical maximum number of Docs in a free collection

Theoretical maximum QPS for Retrieve Doc in a free collection

64

2,400,000

30

128

1,200,000

30

256

600,000

30

512

300,000

15

768

200,000

10

1024

150,000

7

1536

100,000

5

2048

75,000

3

4096

37,500

1

8192

18,750

1

16384

9,375

1

20000

7680

1

Paid instance (cluster) limits

Note
  • The capacity and QPS of paid clusters do not have predefined hard limits. Instead, their upper limits are determined by factors such as the selected cluster specifications, data volume, and vector dimensions. For more information, see Product specifications.

  • A paid cluster can have a maximum of 32 collections. The number of partitions you can create increases with the instance specifications and is independent of the number of collections.

For example, a P.2xlarge cluster can have up to 2,048 partitions. If there is only one collection, that collection can have 2,048 partitions. If there are 32 collections, each collection can have an average of 64 partitions.

Instance type

Specifications

Number of collections

Number of partitions

Compute-optimized

P.small

32

512

P.large

32

1024

P.2xlarge

32

2048

P.4xlarge

32

4096

P.8xlarge

32

8192

P.16xlarge

32

16384

Storage-optimized

S.small

32

512

S.large

32

1024

S.2xlarge

32

2048

S.4xlarge

32

4096

S.8xlarge

32

8192

S.16xlarge

32

16384

Serverless

-

32

128

General limits

Limit

Limitations

Maximum number of API keys per UID

3

Collection name

Must consist of uppercase and lowercase letters, digits, underscores (_), and hyphens (-). The length must be from 3 to 32 characters.

Collection vector dimension limit

(1, 20000]

Partition name

Must consist of uppercase letters, lowercase letters, digits, and underscores (_). The length must be from 3 to 32 characters.

Field name

Must consist of uppercase letters, lowercase letters, digits, and underscores (_). The length must be from 1 to 32 characters.

Doc ID

Must consist of uppercase letters, lowercase letters, digits, and the following characters: _, -, !, @, #, $, %, +, =, .. The length must be from 1 to 64 characters.

Get Doc

  • Maximum batch size: 1,024. A size of 128 or less is recommended.

Delete Doc

  • Maximum batch size: 1,024. A size of 128 or less is recommended.

Insert Doc

  • Maximum batch size: 1,024. A size of 128 or less is recommended.

Update Doc

  • Maximum batch size: 1,024. A size of 128 or less is recommended.

Insert or Update Doc

  • Maximum batch size: 1,024. A size of 128 or less is recommended.

Maximum TopK for Retrieve Doc

1024

Maximum group_count for grouped vector search

64

Maximum group_topk for grouped vector search

16

Maximum body size for all API requests

2 MB. A size of 50 KB or less is recommended.

Maximum number of fields

1,024. 64 or fewer is recommended.

Sparse feature

The number of non-zero values cannot exceed 4,095. The index must be between 0 and 4,294,967,295, inclusive.

上一篇: Data types 下一篇: DashVector public endpoints