Code Summary: Creating a Vector Search Index Using MongoDB Shell

Creating a Vector Search Index Using MongoDB Shell:

Use the db.collection.createSearchIndex command to build an index on a collection. You must specify the type of index as vectorSearch, and you will need to define the fields you wish to index, including the type, number of dimensions, and similarity.

db.movies.createSearchIndex(
  "movies_vector_index", 
  "vectorSearch", 
  {
    "fields": [
      {
        "type": "vector",
        "numDimensions": 1536,
        "path": "plot_embedding",
        "similarity": "cosine",
      }
    ],
  }
);

After running the command, you will receive a confirmation message with the name of the new index:

movies_vector_index

Viewing an Existing Index:

Use the db.collection.getSearchIndexes command and leave it blank in order to see all indexes on the specified collection:

db.movies.getSearchIndexes();

If you wish to specify a particular index, pass in the name of the index as an argument:

db.movies.getSearchIndexes("movies_vector_index");

This will return a lot of information about the index, including the status, which will let you know if the index is ready or still being prepared. It will also show the current definition of the index.

[
  {
    id: "6671e934b362ed3c6ad84512",
    name: "movies_vector_index",
    type: "vectorSearch",
    status: "READY",
    queryable: true,
    latestDefinitionVersion: { version: 0, createdAt: ISODate("2024-06-18T20:08:20.678Z") },
    latestDefinition: {
      fields: [
        {
          type: "vector",
          numDimensions: 1536,
          path: "plot_embedding",
          similarity: "cosine"
        },
      ],
    },
    statusDetail: [
      . . .
    ],
  },
];

Note the statusDetail array in the above example. This array contains an index status array for each node in the cluster.

[
  {
    . . .
    statusDetail: [
      {
        hostname: "atlas-11yyiw-shard-00-02",
        status: "READY",
        queryable: true,
        mainIndex: {
          status: "READY",
          queryable: true,
          definitionVersion: {
            version: 0,
            createdAt: ISODate("2024-06-18T20:08:20.000Z"),
          },
          definition: { fields: [[Object]] },
        },
      },
      . . .
    ]
  },
];

Editing an Existing Vector Search Index:

Update the definition of an existing vector search index with the db.collection.updateSearchIndex command. In this example, we’ll add a filter on the year field.

db.movies.updateSearchIndex(
  "movies_vector_index", 
  {
    "fields": [
      {
        "type": "vector",
        "numDimensions": 1536,
        "path": "embedding",
        "similarity": "cosine",
      },
      {
        "type": "filter",
        "path": "year",
      }
    ],
  }
);

Using db.movies.getSearchIndexes("movies_vector_index"), we can confirm that the update was successful:

[
  {
    id: "6671e934b362ed3c6ad84512",
    name: "movies_vector_index",
    type: "vectorSearch",
    status: "READY",
    queryable: true,
    latestDefinitionVersion: { version: 1, createdAt: ISODate("2024-06-18T20:08:25.348Z") },
    latestDefinition: {
      fields: [
        {
          type: "vector",
          numDimensions: 1536,
          path: "plot_embedding",
          similarity: "cosine"
        },
     { type: 'filter', path: 'year' }
      ],
    },
    statusDetail: [
      . . .
    ],
  },
];

Deleting a Vector Search Index

Delete a vector search index with the db.collection.dropSearchIndex command and pass the name of the vector along as an argument:

db.movies.dropSearchIndex("movies_vector_index");

Running this command will not produce any feedback from the console, so to confirm that the deletion was successful, you can use the db.collection.getSearchIndexes command:

db.movies.getSearchIndexes("movies_vector_index");

This will return the information we’re accustomed to seeing when we run this command. In this case, we can see that the status is listed as “DELETING”, so we know the deletion was successful. We can also run getSearchIndexes again later without specifying an index name, and see that it is no longer present on the collection.

[
  {
    id: "6671e934b362ed3c6ad84512",
    name: "movies_vector_index",
    type: "vectorSearch",
    status: "DELETING",
    queryable: true,
    latestDefinitionVersion: { version: 1, createdAt: ISODate("2024-06-18T20:08:25.348Z") },
    latestDefinition: {
      fields: [
        {
          type: "vector",
          numDimensions: 1536,
          path: "plot_embedding",
          similarity: "cosine"
        },
     { type: 'filter', path: 'year' }
      ],
    },
    statusDetail: [
      . . .
    ],
  },
];