Collection #

A collection in SemaDB is a group of points / documents that share a common index schema. You may create several collections in a single database, each with its own index schema to organise and group your data.

A collection defines a search boundary such that all indexed points are searchable together. This is important to remember when deciding how to organise your data. It’s quite common to have a single collection for each logical type of data you have such as products, documents etc and in general most applications have a single collection to search over. Recall that SemaDB isn’t a relational database but more of a search engine.

Please refer to indexing to learn more about how to define the schema for a collection.

Collection ID #

The collection ID is a unique identifier per user that is used to reference the collection in all operations. It is an alpha numeric string and must be unique within the user’s account. There is usually a limit on the length of the collection ID and it is recommended to keep it short and descriptive.

Shards #

A collection is divided into multiple shards to distribute the data across the cluster. Each shard is a self-contained index that can be searched independently and the collection level actions orchestrate the operations across all the required shards.

title: Collection Sharding
graph TD
    subgraph NodeA
    Collection --> Shard1[Shard 1]
    Collection --> Shard2[Shard 2]
    Collection --> Shard3[Shard 3]
    subgraph NodeB
    Collection -- RPC --> Shard4[Shard 4]
    Collection -- RPC --> Shard5[Shard 5]

The sharding happens automatically based on the configuration of what the maximum shard size should be. Multiple shards can exist on a single node or across multiple nodes in the cluster. This translates to either concurrent multi-threaded operations on a single node or distributed operations via remote procedure calls across multiple nodes.

There is no requirement to run SemaDB on multiple nodes. A single traditional server can run SemaDB with multiple shards and still benefit from the parallelism.