Platform For AI:FeatureDB overview

What is FeatureDB

FeatureDB is a distributed database storing features in KV and KKV formats with native Array and Map support. Structured storage outperforms serialized strings for reads, writes, and inference. FeatureDB handles offline features, real-time features, and behavior sequences.

Activate FeatureDB

Activate FeatureDB when creating a FeatureDB data store by following interface prompts.

Core capabilities

FeatureDB provides these capabilities:

• Read and write KV and KKV features.

• Read and write MaxCompute complex types (Array, Map).

• Pull all feature data in a FeatureView.

• Millisecond polling for real-time feature updates.

• Second-level TTL with automatic expired data cleanup.

• Pay-as-you-go billing based on actual operations.

FeatureDB shards data within a FeatureView and adjusts shard count based on performance needs. It supports replicas for data stability. Shard allocation depends on Estimated Order of Magnitude:

• Less than 10 million (default): 5 shards.

• Between 10 million and 100 million: 10 shards.

• More than 100 million: 20 shards.

Advantages

• Lower costs

  Reduces expenses for smaller feature storage needs.

• High-frequency updates

  Real-time statistical features update every few seconds to multiple EasyRec Processor instances.

• Complex type features

  Search and promotion businesses commonly use Array and Map features, behavior sequences, and SideInfo. String serialization hurts performance.

  FeatureDB stores complex types natively and synchronizes MaxCompute 2.0 complex data for high-performance reads.

• Elastic scaling

  Increase shards per feature view to scale read and write performance for larger deployments.

• Comprehensive monitoring

  Monitor read/write QPS, RT, data update latency, and storage at view-level granularity, eliminating blind spots when integrating third-party sources.

Usage

       // Initialize configuration with region, credentials, and project
       Configuration configuration = new Configuration("cn-beijing",
                Constants.accessId, Constants.accessKey,"fs_demo_featuredb" );

        // Set FeatureDB credentials
        configuration.setUsername(Constants.username);
        configuration.setPassword(Constants.password);

        // For public network access, set domain (omit for VPC)
        //configuration.setDomain(Constants.host);

        ApiClient client = new ApiClient(configuration);

        // For public network, set usePublicAddress = true (omit for VPC)
        // FeatureStoreClient featureStoreClient = new FeatureStoreClient(client, Constants.usePublicAddress);
        FeatureStoreClient featureStoreClient = new FeatureStoreClient(client );

        // Get project and feature view
        Project project = featureStoreClient.getProject("fs_demo_featuredb");
        if (null == project) {
            throw  new RuntimeException("project not found");
        }

        FeatureView featureView = project.getFeatureView("user_test_2");
        if (null == featureView) {
            throw  new RuntimeException("featureview not found");
        }

        // Prepare sample data
        List<Map<String, Object>> writeData = new ArrayList<>();
        for (int i = 0; i < 10; i++) {
            Map<String, Object> data = new HashMap<>();
            data.put("user_id", i);
            data.put("string_field", String.format("test_%d", i));
            data.put("int32_field", i);
            data.put("int64_field", Long.valueOf(i));
            data.put("float_field", Float.valueOf(i));
            data.put("double_field", Double.valueOf(i));
            data.put("boolean_field", i % 2 == 0);
            writeData.add(data);
        }

        // Write data in batches
        for (int i = 0; i < 100;i++) {
            featureView.writeFeatures(writeData);
        }

        // Flush all writes (call once after all writes complete)
        featureView.writeFlush();

Real-time feature data flow

FeatureStore storage service consists of three components: Feature Service (access layer), MSMQ (DataHub), and FeatureDB.

For real-time features, write data to FeatureDB via FeatureStore Java SDK or Flink Connector. Data written through feature service also synchronizes to your MaxCompute table for real-time feature export and model training.

Read feature data from FeatureDB via FeatureStore Java/Go SDK, or pull all features through EasyRec Processor and store in local cache for higher performance. Real-time features provide millisecond updates.

Performance benchmarks

These performance test results use FeatureStore Go SDK to read FeatureDB data. The feature table contains user-side data from a recommendation scenario with 17,689,586 rows. Test machine: 4 cores, 8 GiB memory. Results are for reference only.

• Configure VPC direct connection configured, online service in recommended zone:

  Number of feature fields (columns)	Number of keys read (rows)	Average latency	TP95	TP99
  260	1	0.89 milliseconds	1.20 milliseconds	1.45 milliseconds
  260	10	1.17 milliseconds	1.52 milliseconds	1.87 milliseconds
  260	50	1.91 milliseconds	2.56 milliseconds	2.92 milliseconds
  260	100	2.87 milliseconds	3.58 milliseconds	3.93 milliseconds
  260	200	4.43 milliseconds	5.25 milliseconds	5.80 milliseconds

• VPC direct connection configured, online service in non-recommended zone:

  Number of feature fields (columns)	Number of keys read (rows)	Average latency	TP95	TP99
  260	1	2.54 milliseconds	2.86 milliseconds	3.15 milliseconds
  260	10	2.75 milliseconds	3.12 milliseconds	3.56 milliseconds
  260	50	3.95 milliseconds	4.75 milliseconds	5.19 milliseconds
  260	100	4.82 milliseconds	5.66 milliseconds	6.21 milliseconds
  260	200	6.84 milliseconds	7.75 milliseconds	8.25 milliseconds

• VPC direct connection not configured:

  Number of feature fields (columns)	Number of keys read (rows)	Average latency	TP95	TP99
  260	1	3.62 milliseconds	3.83 milliseconds	4.27 milliseconds
  260	10	3.82 milliseconds	4.11 milliseconds	4.61 milliseconds
  260	50	4.54 milliseconds	5.19 milliseconds	5.60 milliseconds
  260	100	5.40 milliseconds	6.13 milliseconds	6.56 milliseconds
  260	200	7.15 milliseconds	7.93 milliseconds	8.47 milliseconds

Billing

For billing information, see Billing of FeatureStore.