Update and rebuild an active index

Vertex AI Matching Engine has been renamed to Vector Search. We are in the process of updating content to reflect the new branding.

Overview

With large search queries, updating your indexes is important to always having the most accurate information. Today you can update your Vector Search indexes using a batch update, which lets you to insert and delete data points through a batch schedule, or with streaming update, which lets you update and query your index within a few seconds.

Additionally, you can use UpdateIndex to update your important metadata fields, like display_name, description and labels. You can also add optional tags to your index to help with diversifying results or filtering pre index query.

Note: You can invoke UpdateIndex to update either the index content or the metadata fields, but not both in a single invocation. For example, if you want to add data points as well as update the description, you need to with two separate invocations.

Update index content with Batch Updates

To update the content of an existing Index, use the IndexService.UpdateIndex method.

To replace the existing content of an existing Index:

• Set Index.metadata.contentsDeltaUri to the Cloud Storage URI that includes the vectors you want to update.
• Set isCompleteOverwrite to true.

If you set the contentsDeltaUri field when calling IndexService.UpdateIndex, then no other index fields (such as displayName, description, or userLabels) can be also updated as part of the same call.

gcloud

1. Update your index metadata file.
2. Use the gcloud ai indexes update command:

gcloud ai indexes update INDEX_ID \
  --metadata-file=LOCAL_PATH_TO_METADATA_FILE \
  --project=PROJECT_ID \
  --region=LOCATION

Replace the following:

• INDEX_ID: The ID of the index.
• LOCAL_PATH_TO_METADATA_FILE: The local path to the metadata file.
• PROJECT_ID: The ID of the project.
• LOCATION: The region where you are using Vertex AI.

REST

Before using any of the request data, make the following replacements:

• LOCATION: Your region.
• PROJECT_ID: Your project ID.
• INPUT_DIR: The Cloud Storage directory path of the index content.
• PROJECT_NUMBER: Project number for your project

HTTP method and URL:

PATCH https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/indexes/INDEX_ID

Request JSON body:

{
  "metadata": {
    "contentsDeltaUri": "INPUT_DIR",
    "isCompleteOverwrite": true
  }
}

To send your request, expand one of these options:

curl (Linux, macOS, or Cloud Shell)

Note: The following command assumes that you have logged in to the gcloud CLI with your user account by running gcloud init or gcloud auth login, or by using Cloud Shell, which automatically logs you into the gcloud CLI. You can check the currently active account by running gcloud auth list.

Save the request body in a file named request.json, and execute the following command:

curl -X PATCH \
    -H "Authorization: Bearer $(gcloud auth print-access-token)" \
    -H "Content-Type: application/json; charset=utf-8" \
    -d @request.json \
    "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/indexes/INDEX_ID"

PowerShell (Windows)

Note: The following command assumes that you have logged in to the gcloud CLI with your user account by running gcloud init or gcloud auth login, or by using Cloud Shell, which automatically logs you into the gcloud CLI. You can check the currently active account by running gcloud auth list.

Save the request body in a file named request.json, and execute the following command:

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method PATCH `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/indexes/INDEX_ID" | Select-Object -Expand Content

You should receive a JSON response similar to the following:

{
  "name": "projects/PROJECT_NUMBER/locations/LOCATION/indexes/INDEX_ID/operations/OPERATION_ID",
  "metadata": {
    "@type": "type.googleapis.com/google.cloud.aiplatform.v1.UpdateIndexOperationMetadata",
    "genericMetadata": {
      "createTime": "2022-01-12T23:56:14.480948Z",
      "updateTime": "2022-01-12T23:56:14.480948Z"
    }
  }
}

You can poll for the status of the operation until the response includes "done": true.

If the Index has any associated deployments (see the Index.deployed_indexes field), then when certain changes to the original Index are done, the DeployedIndex is automatically updated asynchronously in the background to reflect these changes.

To check whether the change has been propagated, compare the update index operation finish time and the DeployedIndex.index_sync_time.

Update an index using Streaming Updates

With Streaming Updates, you can update and query your index within a few seconds. At this time, you can't use Streaming Updates on an existing index, you must create a new index. See Create an index for Streaming Update to learn more.

You are charged $0.45 per GB used for Streaming Updates. To learn more about pricing, see the Vertex AI pricing page. Streaming Updates are directly applied to the deployed indexes in memory, which are then reflected in query results after a short delay.

Note: After you've created an index for Streaming Updates, you can't perform a Batch Update on this index. You can perform a complete overwrite to use the index or you can create a new index prepared for Batch Updates. See Create an index to learn more.

Upsert Data points

The throughput quota limit relates to the amount of data that is included in an upsert. If the data point ID exists in the index, the embedding is updated, otherwise, a new embedding is added.

DATAPOINT_ID_1=
DATAPOINT_ID_2=
curl -H "Content-Type: application/json" -H "Authorization: Bearer `gcloud auth print-access-token`" https://${LOCATION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/${LOCATION}/indexes/${INDEX_ID}:upsertDatapoints \
-d '{datapoints: [{datapoint_id: "'${DATAPOINT_ID_1}'", feature_vector: [...]},
{datapoint_id: "'${DATAPOINT_ID_2}'", feature_vector: [...]}]}'

Remove datapoints

curl -H "Content-Type: application/json" -H "Authorization: Bearer `gcloud auth print-access-token`" https://${LOCATION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/${REGION}/indexes/${INDEX_ID}:removeDatapoints -d '{datapoint_ids: ["'${DATAPOINT_ID_1}'", "'${DATAPOINT_ID_2}'"]}'

Compaction

Periodically, your index is rebuilt to account for all new updates since your last rebuild. This rebuild, or "compaction", improves query performance and reliability. Compactions occur for both Streaming Updates and Batch Updates.

• Streaming Update: Occurs when the uncompacted data size is > 1 GB or the oldest uncompacted data is at least three days old. You are billed for the cost of rebuilding the index at the same rate of a batch update, in addition to the Streaming Update costs.

• Batch Update: Occurs when the incremental dataset size is > 20% of the base dataset size.

Note: These compactions occur automatically, and when they occur, you receive an email to your account inbox. If you created the index with a service account, you won't receive a compaction notice email.

Rebuild and query your index

You can send Match/BatchMatch requests as usual with the grpc cli, the client library or the python SDK. When you rebuild the query you can expect to see your updates within a few seconds. To learn how to query an index, see Query indexes to get nearest neighbors.

Optional fields

When you create an index, there are some optional fields you can use to fine-tune your queries.

Upsert with restricts

Upserting your index and adding a restrict is a way of tagging your data points so they are already identified for filtering at query time. You might want to add restrict tags to limit the results that presents on your data before a query is sent. For example, a customer wants to run a query on an index, but wants to make sure the results only display items that match "red" in a search for footwear. In the example below, the index is being upserted and is filtering in all red shoes, but denying blue ones. This ensures the search filters in the best specific options from a large and varied index before running.

To learn more about filtering, see Vector Search for Indexing.

curl -H "Content-Type: application/json" -H "Authorization: Bearer `gcloud auth print-access-token`" https://${ENDPOINT}/v1/projects/${PROJECT_ID}/locations/us-central1/indexes/${INDEX_ID}:upsertDatapoints \
-d '{
datapoints: [
  {
    datapoint_id: "'${DATAPOINT_ID_1}'",
    feature_vector: [...],
    restricts: { namespace: "color", allow_list: ["red"], deny_list: ["blue"]}
  }
]}'

Upsert with crowding

The crowding tag limits similar results by improving result diversity. Crowding is a constraint on a neighbor list produced by a nearest neighbor search requiring that no more than some value, of a group of results, return the same value of crowding_attribute. As an example, let's say you were back online shopping for shoes. You want to see a wide variety of colors in the results, but maybe want them in a single style, like soccer cleats. You can ask that no more than 3 pairs of shoes with the same color is returned by setting per_crowding_attribute_num_neighbors = 3 in your query, assuming you set crowding_attribute to the color of the shoes when inserting the data point.

This field represents the allowed maximum number of matches with the same crowding tag.

curl -H "Content-Type: application/json" -H "Authorization: Bearer `gcloud auth print-access-token`" https://${ENDPOINT}/v1/projects/${PROJECT_ID}/locations/us-central1/indexes/${INDEX_ID}:upsertDatapoints \
-d '{
datapoints: [
  {
    datapoint_id: "'${DATAPOINT_ID_1}'",
    feature_vector: [...],
    restricts: { namespace: "type", allow_list: ["cleats"]}
    crowding_tag: { crowding_attribute: "red-shoe"},
  }
]}'