# OpenSearch Vector Search Edition > OpenSearch Vector Search Edition is Alibaba's self-developed distributed vector search engine powering Taobao, Tmall, Cainiao and Youku, offering vector table creation, multi-shard index building, real-time incremental writes (second-level visibility) and C++ high-concurrency retrieval; supports MaxCompute, Saro and API data sources with cluster scaling and ops via the control system; sibling editions include Retrieval Engine, Industry Algorithm, High-Performance Search and LLM Intelligent Q&A. This documentation applies to the Alibaba Cloud China site. For international site documentation, please refer to https://www.alibabacloud.com/help/llms.txt ## Product overview - [Product overview](https://help.aliyun.com/en/open-search/vector-search-edition/vector-search-product-overview.md): OpenSearch Vector Search Edition is a large-scale distributed search engine built by Alibaba Group, offering high availability, real-time data freshness, and cost-effective vector search for demanding production workloads. - [Release notes](https://help.aliyun.com/en/open-search/vector-search-edition/release-notes-8.md): Chronological log of new features (CAGRA GPU search, hybrid search, DiskANN, QGraph, DLF data sources) and optimizations by release date - [Engine version features](https://help.aliyun.com/en/open-search/vector-search-edition/engine-version-properties.md): Compare vector_service_1.x.x (current) vs ha3_3.10.0 (legacy) engines: data sources, query types, vectorization models, SDK references, and version history - [Billing](https://help.aliyun.com/en/open-search/vector-search-edition/billing.md) - [Billing overview](https://help.aliyun.com/en/open-search/vector-search-edition/billing-overview-of-vector-search-edition.md): Pay-as-you-go (hourly) and subscription billing. Six cost components: instance rental, QRS workers, data nodes, storage, index storage, data update resources. Pricing by region and spec family - [Specification calculator](https://help.aliyun.com/en/open-search/vector-search-edition/specification-calculator.md): Estimate QRS and Searcher worker specs based on vector doc count, dimension, algorithm (HNSW/QC/Linear), target QPS, and response time requirements - [Purchase an instance](https://help.aliyun.com/en/open-search/vector-search-edition/buy-opensearch-vector-retrieve-version-of-the-instance.md): Console purchase flow: select region, configure QRS/Searcher worker specs and count, set VPC/vSwitch, and create credentials. Requires AccessKey pair and RAM permissions (AliyunSearchEngineFullAccess) - [Glossary](https://help.aliyun.com/en/open-search/vector-search-edition/terms-2.md): Definitions for instance terms (QRS workers, Searcher workers, shards), table terms (data sources, fields, vector indexes), and O&M terms (reindexing, FSM state machine) - [Limits](https://help.aliyun.com/en/open-search/vector-search-edition/vector-restrictions-on-use.md): Service limits: max 10-30 tables per instance (by core count), 2 MB single batch push, 10 MB/s push rate, 30 KB GET query length. HTTP GET and POST supported ## Getting started - [Getting started](https://help.aliyun.com/en/open-search/vector-search-edition/getting-started.md): Learn how to deploy and use OpenSearch Vector Search Edition, and choose the right vector index algorithm for your use case. - [Vector indexes](https://help.aliyun.com/en/open-search/vector-search-edition/vector-index.md): OpenSearch Vector Search Edition supports multiple vector index algorithms and distance metrics for various retrieval scenarios. - [Introduction to vectors](https://help.aliyun.com/en/open-search/vector-search-edition/vector-introduction.md): How vector retrieval works: encode unstructured data as vectors, compute distance for similarity. Five algorithms (Linear, QC, HNSW, QGraph, CAGRA) and two distance metrics (cosine, inner product) - [Vector retrieval algorithm selection](https://help.aliyun.com/en/open-search/vector-search-edition/selection-of-vector-retrieval-algorithms.md): Compare 7 algorithms: Linear (exact), HNSW (high recall), QC (low memory), QGraph (billions), DiskANN (disk-based), CagraHnsw/CAGRA (GPU). Selection guide by data scale - [Common vector index configurations](https://help.aliyun.com/en/open-search/vector-search-edition/vector-index-general-configuration.md): Configure vector dimension, distance type (cosine/inner product/Euclidean), algorithm selection (FLAT/HNSW/DiskANN/IVF_SQ8), real-time indexing toggle, and invalid data handling - [Quantized clustering (QC) configuration](https://help.aliyun.com/en/open-search/vector-search-edition/quantitative-clustering-configuration.md): QcBuilder params control centroid count and quantization (FP16/INT8). QcSearcher params control scan ratio and candidate size at query time. Supports GPU-based real-time retrieval - [HNSW configuration](https://help.aliyun.com/en/open-search/vector-search-edition/hnsw-hierarchical-navigable-small-world-configuration.md): Build params: max_neighbor_count (default 100), efconstruction. Search params: efsearch controls recall vs latency trade-off. Prefix: proxima.hnsw.builder/searcher - [QGraph configuration](https://help.aliyun.com/en/open-search/vector-search-edition/qgraph-quantized-graph-configuration.md): HNSW graph with automatic quantization reducing index to 1/8 of raw size. Build params control graph construction; search params balance recall vs speed for billion-scale datasets - [CAGRA configuration](https://help.aliyun.com/en/open-search/vector-search-edition/gpu-algorithm-cagra-configuration.md): GPU-accelerated graph-based ANN algorithm. Build params control graph degree and construction threads; search params (itopk_size, search_width) tune recall vs latency on GPU nodes - [CagraHnsw configuration](https://help.aliyun.com/en/open-search/vector-search-edition/cagrahnsw-cs.md): Hybrid GPU+CPU index: GPU builds the graph (10x faster than CPU), HNSW serves online queries. Build params fixed at creation; search params adjustable per query without rebuild - [DiskANN configuration](https://help.aliyun.com/en/open-search/vector-search-edition/diskann.md): Disk-based ANN for datasets exceeding memory. Vamana graph on disk, PQ compresses vectors in RAM. Key params: pq_dimension_scale, max_degree, list_size, io_limit - [Single-node quick start](https://help.aliyun.com/en/open-search/vector-search-edition/single-node-quick-start.md): Purchase and configure a single-node instance: set QRS workers to 0, configure data node specs, add table with shard count fixed at 1. Includes field configuration and index schema setup - [High-availability Edition quick start](https://help.aliyun.com/en/open-search/vector-search-edition/high-availability-version-quick-start.md): Cross-zone deployment with mutual backup. Minimum 2 Searcher + 2 QRS workers across two AZs. Create new HA instance or upgrade from Standard Edition via console - [General-purpose Edition quick start](https://help.aliyun.com/en/open-search/vector-search-edition/general-edition-quick-start.md): End-to-end setup: purchase instance, create table, configure fields and index schema, push data, build index, and run first vector query - [Quick start for image search](https://help.aliyun.com/en/open-search/vector-search-edition/figure-search-scene-quick-start-1.md): Set up image-to-image and text-to-image search: purchase instance, configure Vector Image Search template, upload image data, build vector index, and query by image or text - [Quick start: Text vector search](https://help.aliyun.com/en/open-search/vector-search-edition/text-vector-scene-quickstart-1.md): Set up text semantic search: purchase instance, use Vector Text Semantics template, configure text-to-vector model, push documents, and query by natural language - [Quick start: Video search](https://help.aliyun.com/en/open-search/vector-search-edition/video-search-quick-start.md): Search videos in OSS by text, image, or video clip. Extracts key frames, generates embeddings, builds searchable index. Supports cross-modal queries without manual tagging - [Quick Start for Video Semantic Retrieval](https://help.aliyun.com/en/open-search/vector-search-edition/quick-start-guide-to-video-semantic-retrieval.md): OpenSearch Vector Search Edition uses AI Search Open Platform offline services to segment, summarize, and vectorize OSS videos (qwen3-vl-embedding, 2560-dim). Supports text, image, and mixed-mode retrieval of video clips. Available only in China East 1, China East 2, and China North 2 regions; fees apply for video segmentation, summarization, and embedding services. ## User guide - [User guide](https://help.aliyun.com/en/open-search/vector-search-edition/llm-operation-guide.md) - [Account management](https://help.aliyun.com/en/open-search/vector-search-edition/account-management.md): RAM user creation and authorizationCreate a RAM userA RAM user is an identity in Resource Access Management (RAM) that represents an operator or an ap... - [Vector Search Edition - Instance Management (Create/Scale/Whitelist/API Endpoint)](https://help.aliyun.com/en/open-search/vector-search-edition/instance-management-easy-to-use-edition.md): Manage Vector Search Edition instances: create, scale query/data nodes, public access whitelist (IP/CIDR), API endpoint domain, password reset (13+ chars). - [Upgrade or downgrade instance configurations](https://help.aliyun.com/en/open-search/vector-search-edition/change-the-configurations-of-an-instance.md): Scale QRS/Searcher worker count and specs, switch between Standard and High-availability Edition. Pay-as-you-go supports upgrades only; subscription supports both - [Configure public access whitelist](https://help.aliyun.com/en/open-search/vector-search-edition/public-whitelist-configuration.md): Allow public network or on-premises access to VPC-deployed instances by adding IP addresses to the public access whitelist. Required for cross-network API calls - [Metric monitoring](https://help.aliyun.com/en/open-search/vector-search-edition/monitoring-indicator-description.md): Monitor key metrics such as document count, request success and failure rates, latency, and vector recall rate through charts. - [Table metrics](https://help.aliyun.com/en/open-search/vector-search-edition/table-indicators.md): Per-table real-time monitoring: core health status, QRS worker metrics, Searcher worker metrics, and data update latency/throughput - [Instance metrics](https://help.aliyun.com/en/open-search/vector-search-edition/indicator-market-description.md): Instance-level monitoring: query throughput (QPS), latency percentiles, CPU/memory usage, error rates, and real-time indexing activity. Use to detect bottlenecks - [Alert management](https://help.aliyun.com/en/open-search/vector-search-edition/alert-management.md): OpenSearch Vector Search Edition provides alert management capabilities to help you monitor system status and respond to anomalies promptly. - [Alert rule groups](https://help.aliyun.com/en/open-search/vector-search-edition/alert-rule-groups.md): Configure metric thresholds and receive alerts via phone, SMS, or DingTalk when instance health degrades. Integrates with CloudMonitor - [OpenSearch Vector Search Edition: Alert Contact Management (Create/Activate/Disable/Groups)](https://help.aliyun.com/en/open-search/vector-search-edition/contact-management.md): Create alert contacts (phone/DingTalk robot webhook), activate via SMS code, edit, disable, and manage contact groups; disabled contacts stop receiving alert notifications. - [Configure DingTalk group alerts](https://help.aliyun.com/en/open-search/vector-search-edition/configure-dingtalk-group-alarm.md): Add a DingTalk chatbot webhook URL to an alert contact for group notifications. No alert rule changes needed after setup - [Log records](https://help.aliyun.com/en/open-search/vector-search-edition/logging.md): OpenSearch Vector Search Edition allows you to query and write logs and track change history for instance maintenance operations. - [Table aliases](https://help.aliyun.com/en/open-search/vector-search-edition/table-alias-management.md): Create alternative names for tables usable in queries and writes. Supports multiple aliases per table (up to 20 per instance) and switching an alias to a different table without client changes - [Table management](https://help.aliyun.com/en/open-search/vector-search-edition/table-management.md): You can view table details, field configurations, index schemas, and advanced configurations for vector fields and indexes in the OpenSearch console. - [Add a table](https://help.aliyun.com/en/open-search/vector-search-edition/add-table.md): Select a data sync method based on your data source:For MaxCompute data sources, use a MaxCompute + API data source to sync data to your OpenSearch in... - [Add table: MaxCompute + API data source](https://help.aliyun.com/en/open-search/vector-search-edition/maxcompute-api-data-source.md): Use MaxCompute partitions for full data import and API for incremental updates. Configure project, table, partition column, and field mapping during table creation - [Vector Search Edition: Add Table with API Data Source](https://help.aliyun.com/en/open-search/vector-search-edition/api-dataseouce.md): Add table via API push data source: primary key and multi-value float vector field with ha3 separator ^], index algorithms HNSW/linear/Quantized Clustering. Full rebuild clears pushed data. - [Add table: OSS + API data source](https://help.aliyun.com/en/open-search/vector-search-edition/oss-api-data-source.md): OSS bucket (HA3 or JSON format) for full index build, API for incremental updates after index goes live. Configure bucket path, file format, and partition settings - [Add table: DLF (Data Lake Formation) data source](https://help.aliyun.com/en/open-search/vector-search-edition/data-lake-construction-dlf.md): Pull multimodal data from DLF, auto-vectorize with built-in models. Supports Paimon, Lance, and Object Table formats. Image/text/video search without separate ingestion pipeline - [Modify a table](https://help.aliyun.com/en/open-search/vector-search-edition/modify-table.md): Change shard count, data sources, fields, or index schema. Shard/field/schema changes require full reindex; Empty Data method clears all previous data. Check impact matrix before modifying - [Configure Model Studio embeddings](https://help.aliyun.com/en/open-search/vector-search-edition/configure-model-studio-embedding.md): Integrate Model Studio models for auto-vectorization: qwen3-vl-embedding (multimodal, image+text) and text-embedding-v4 (text). Converts raw data to vectors during preprocessing - [Configure fused vectors](https://help.aliyun.com/en/open-search/vector-search-edition/fusion-vector.md): Combine multiple fields (text+image) into one vector via multimodal models (qwen3-vl-embedding). Enables cross-modal retrieval. Requires engine v1.5.8+ - [Delete a table](https://help.aliyun.com/en/open-search/vector-search-edition/delete-table.md): Remove a table from the instance. Must stop the table before deletion. Irreversible operation that removes all table data and index - [Increase data update resources](https://help.aliyun.com/en/open-search/vector-search-edition/data-update-resources.md): Scale up from the default 2 free resources (4-core 8 GB each) per table when incremental sync latency is high. Additional resources are billed separately - [Reindex a table](https://help.aliyun.com/en/open-search/vector-search-edition/index-rebuild.md): Rebuild full index from complete dataset. Method varies by data source: MaxCompute/OSS reimport with partition+timestamp; API clears all data and starts fresh (last 3 days only) - [Configure table loading policy](https://help.aliyun.com/en/open-search/vector-search-edition/index-table-load-policy.md): Control how index files load into memory via regex-matched policies. Strategies: mmap (lock hot indexes in RAM) or cache (direct_io for SSD, remote for distributed storage). First-match evaluation - [Enable or disable a table](https://help.aliyun.com/en/open-search/vector-search-edition/stop-using-resume-using.md): Stop a table to free resources or resume it for queries and writes. Ensure no active traffic before disabling. Disabled tables cannot serve queries - [Configure vector cache](https://help.aliyun.com/en/open-search/vector-search-edition/configure-vector-cache.md): In OpenSearch Vector Search Edition, the vector cache feature reduces resource consumption and cost using a cache table to avoid repeated vectorization. When data is written, the system automatically checks whether field content has changed and only recomputes vectors for changed data, significantly lowering model invocation volume. - [Nested query](https://help.aliyun.com/en/open-search/vector-search-edition/nested-query.md): Query OBJECT-type nested fields with is_nested: true to match conditions within the same sub-object. Prevents cross-object false matches for user profiles, orders, or product attributes - [Recover data from indexes](https://help.aliyun.com/en/open-search/vector-search-edition/recover-data-from-indexes.md): After a full indexing operation, API-pushed data can only be traced back up to 3 days. OpenSearch Vector Search Edition provides the index recovery feature that recovers data from the current full index version, ensuring that previously pushed data is retained in the next full indexing. - [Model management](https://help.aliyun.com/en/open-search/vector-search-edition/model-management.md) - [Model list](https://help.aliyun.com/en/open-search/vector-search-edition/model-list.md): Vector Search Edition supports built-in models, AI Search Open Platform models, and user-defined models. Use the Model list page to view and manage them. - [Custom models](https://help.aliyun.com/en/open-search/vector-search-edition/custom-model.md): Register external HTTP embedding services (text_embedding, image_embedding, sparse, image_analyze). Provide endpoint URL; platform auto-calls at index and query time - [OpenSearch Vector Search Edition Data Management (Add/Delete/Update/Query)](https://help.aliyun.com/en/open-search/vector-search-edition/table-operation-and-maintenance.md): OpenSearch Vector Search Edition data management: add, delete, update and query vector data, covering import, preprocessing, storage, retrieval and optimization. - [Vector management](https://help.aliyun.com/en/open-search/vector-search-edition/vector-management.md): OpenSearch Vector Search Edition provides vector management for storing, indexing, and retrieving high-dimensional vectors to enable similarity search. - [Query tests](https://help.aliyun.com/en/open-search/vector-search-edition/query-test.md): OpenSearch Vector Search Edition supports vector queries, primary key queries, and vector-text hybrid queries to handle different search scenarios. - [Query test: Vector query](https://help.aliyun.com/en/open-search/vector-search-edition/vector-query-test.md): Console tool to test vector queries in Form or Developer mode. Set table, index, TopK, filter conditions, namespace, and query type (vector/image/text) - [Query test: Primary key query](https://help.aliyun.com/en/open-search/vector-search-edition/primary-key-query-test.md): Retrieve records by primary key IDs in the console. Use to verify specific documents were indexed correctly after reindexing completes - [Query test: Hybrid vector-text query](https://help.aliyun.com/en/open-search/vector-search-edition/vector-text-hybrid-query.md): Combine dense vector similarity with text-based filtering in one query. Configure TopK, filter conditions, namespace, sort order, and dense vector input via Form or Developer mode - [Insert data via console](https://help.aliyun.com/en/open-search/vector-search-edition/vector-management-add-data.md): Manually insert records into a table using the console Insert Data tab. Supports form mode and developer mode for vector, text-embedding, and image-embedding data types. - [Delete data via console](https://help.aliyun.com/en/open-search/vector-search-edition/vector-management-delete-data.md): Delete records by primary key using the console Delete Data tab. Supports form mode and developer mode. - [Update data via API](https://help.aliyun.com/en/open-search/vector-search-edition/update-data-api.md): Push data to a table using the Python SDK (alibabacloud-ha3engine-vector). Covers table creation with OSS/MaxCompute/DLF data sources, whitelist config, and push code examples. - [Query data](https://help.aliyun.com/en/open-search/vector-search-edition/query-test-1.md): Run vector similarity, primary key, and filter queries via SDK or console. Includes Python SDK setup with alibabacloud-ha3engine-vector and query code examples. ## Use cases - [Use cases](https://help.aliyun.com/en/open-search/vector-search-edition/vector-best-practice.md) - [End-to-end image search solution](https://help.aliyun.com/en/open-search/vector-search-edition/end-to-end-graph-search-solution-1.md): Build an image search engine without pre-computed vectors. Supports OSS+MaxCompute, Base64 via API, and DLF pipelines for image vectorization and retrieval. - [Multi-object detection for image search](https://help.aliyun.com/en/open-search/vector-search-edition/multi-agent-identification-best-practices.md): Detect and match multiple objects in a single query image. Control via crop and range parameters. Returns results for any detected object rather than only the dominant one. - [Hybrid search with sparse and dense vectors](https://help.aliyun.com/en/open-search/vector-search-edition/hybrid-search-best-practices.md): Combine dense vectors (semantic search) and sparse vectors (keyword matching) in a unified index. Covers external model integration, index configuration, and weighted scoring. - [Hybrid query with text inverted index and dense vectors](https://help.aliyun.com/en/open-search/vector-search-edition/text-dense-vector-hybrid-query-best-practices.md): Combine token-based inverted index queries with vector similarity search using adjustable weights. Supports built-in embedding models or custom vectors for RAG and e-commerce scenarios. - [RAG with Vector Search and LLM Conversational Search](https://help.aliyun.com/en/open-search/vector-search-edition/the-vector-retrieval-sersion-supports-the-rag-scheme.md): Build a RAG pipeline pairing Vector Search Edition (retrieval) with LLM Conversational Search Edition (inference). Covers ingestion, vector search, and knowledge-llm API. - [RAG with Vector Search and DeepSeek](https://help.aliyun.com/en/open-search/vector-search-edition/vector-retrieval-version-deepseek-quickly-build-rag-scheme.md): Combine AI Search Open Platform (document parsing, embedding, retrieval) with DeepSeek reasoning model to build RAG. Covers service enablement, workflow creation, and end-to-end testing. - [Filter optimization for vector retrieval](https://help.aliyun.com/en/open-search/vector-search-edition/vector-recall-filtering-optimization.md): Enable via vector_service.search.enable_filter_optimize in searchParams. Pre-queries inverted indexes when filter matches are sparse (prefetch_size=500, prefetch_coverage=0.8). - [OpenSearch Vector Index Load Test (gist-960-euclidean + wrk)](https://help.aliyun.com/en/open-search/vector-search-edition/vector-index-pressure-test-steps.md): Benchmark OpenSearch Vector Search Edition: push 1M vectors from gist-960-euclidean via prepare_data.py/push_data.py, then load-test QPS and latency with wrk. - [Call AI Search Open Platform model services](https://help.aliyun.com/en/open-search/vector-search-edition/call-the-search-development-workbench-model-service.md): Use AI Search Open Platform to vectorize images/text for search-by-image, text-to-image, and semantic search. Supports Vector: Image Search template with OSS+API source. - [Offline data sync via DataWorks](https://help.aliyun.com/en/open-search/vector-search-edition/vector-retrieval-edition-single-table-offline-data-synchronization.md): Bulk-load data into Vector Search Edition using DataWorks Data Integration. Resource group must be in the same region as the instance. Supports 40+ source types. - [Access OpenSearch from PAI-DSW](https://help.aliyun.com/en/open-search/vector-search-edition/access-the-opensearch-engine-through-dsw.md): Configure DSW instance network settings to connect to OpenSearch over private VPC. Place DSW in the same VPC/vSwitch as OpenSearch and allowlist the instance IP. - [Access OpenSearch from Function Compute](https://help.aliyun.com/en/open-search/vector-search-edition/access-the-opensearch-engine-through-fc.md): Configure FC service VPC settings to reach OpenSearch private endpoint. Enable Access to VPC and select the OpenSearch VPC in FC service network configuration. - [Cross-VPC access via CEN with CIDR conflict](https://help.aliyun.com/en/open-search/vector-search-edition/multi-vpc-access-through-cen-multi-vpc-network-segment-conflict.md): Route traffic between VPCs with overlapping CIDR blocks using CEN Transit Router + VPC NAT Gateway (SNAT). Add a secondary non-overlapping CIDR to the source VPC for address translation. - [Cross-VPC access via CEN without CIDR conflict](https://help.aliyun.com/en/open-search/vector-search-edition/multi-vpc-access-through-cen-multiple-vpc-network-segments-do-not-conflict.md): Connect non-overlapping VPCs to OpenSearch using CEN Transit Routers. Supports same-region and cross-region access without public internet exposure. - [Cross-VPC access via VPC peering with CIDR conflict](https://help.aliyun.com/en/open-search/vector-search-edition/multi-vpc-access-to-opensearch-engine-through-vpc-peer-to-peer-connection.md): Resolve overlapping CIDR blocks between peered VPCs using a secondary CIDR block, VPC NAT gateway (SNAT), and custom route entries to reach OpenSearch private endpoint. - [Category queries with namespaces and filters](https://help.aliyun.com/en/open-search/vector-search-edition/vector-search-edition-best-practices-for-category-query.md): Partition vector indexes by namespace or apply filter conditions for category-based retrieval. Namespace isolates data per tenant; filters narrow by attributes like product category. ## Developer guide - [Developer guide](https://help.aliyun.com/en/open-search/vector-search-edition/development-guide.md) - [SDK reference](https://help.aliyun.com/en/open-search/vector-search-edition/sdk-example.md) - [SDK statistical syntax](https://help.aliyun.com/en/open-search/vector-search-edition/vector-engine-statistical-syntax.md): Run aggregate queries via SDK (Java/Python/Go) with groupKeys, aggFuncs (max, min, avg), and orderBy parameters. Returns grouped statistics on numeric fields. - [SDK version overview](https://help.aliyun.com/en/open-search/vector-search-edition/version-description-test.md): SDK packages for async Java, Java, Python, and Go (version 1.1.18). Cover Traffic API only (data push and search). Use Management and Control SDKs for other features. - [SDK push data](https://help.aliyun.com/en/open-search/vector-search-edition/update-data.md): Push documents via Java, Python, or Go SDK. Each request can add, overwrite, or delete records in bulk. Includes client config and code examples per language. - [SDK query data](https://help.aliyun.com/en/open-search/vector-search-edition/query-vector-data.md): Query vectors using SDK (Java/Python/Go) with multiple strategies: vector similarity, primary key lookup, filter expressions, multi-vector, batch, and prediction queries. - [SDK data statistics](https://help.aliyun.com/en/open-search/vector-search-edition/vector-statistics.md): Retrieve table statistics via SDK stats API: vector counts and index state. Monitor data ingestion progress and verify index readiness before querying. - [SDK index table operations](https://help.aliyun.com/en/open-search/vector-search-edition/index-table-operations.md): Manage index tables programmatically via Java, Python, and Go SDK. Create, list, describe, and delete index tables through the Traffic API. - [Data API reference](https://help.aliyun.com/en/open-search/vector-search-edition/data-api.md) - [Data API update data](https://help.aliyun.com/en/open-search/vector-search-edition/vector-update-data-api.md): POST /vector-service/push endpoint for bulk document upload and deletion. Supports add, overwrite, and delete operations in a single HTTP request with JSON body. - [Query data](https://help.aliyun.com/en/open-search/vector-search-edition/query-data.md): OpenSearch Vector Search Edition supports multiple data query methods to meet diverse query requirements. - [Vector-based query API](https://help.aliyun.com/en/open-search/vector-search-edition/vector-query.md): POST /vector-service/query for k-nearest neighbor retrieval by vector similarity. Supports namespace filtering. Falls back to prediction query if no pre-computed vectors exist. - [Multi-vector query for single document](https://help.aliyun.com/en/open-search/vector-search-edition/single-doc-multi-vector-retrieval.md): Store multiple vectors per document in one vector field to avoid information loss from pooling. Any matching vector retrieves the full document without splitting into separate records. - [Multi-query API](https://help.aliyun.com/en/open-search/vector-search-edition/multi-query-query.md): Run multiple vector queries in one request and merge results into a unified ranked list. Search across namespaces, indexes, and dense+sparse vector combinations. - [Batch query API](https://help.aliyun.com/en/open-search/vector-search-edition/batch-query.md): POST /vector-service/batch-query to run multiple queries in a single HTTP request. Uses JSON body with HTTP Basic authentication (Base64-encoded username:password). - [Prediction query API](https://help.aliyun.com/en/open-search/vector-search-edition/prediction-query.md): POST /vector-service/inference-query converts text, images, or videos to vectors using built-in embedding models, then retrieves similar documents. No external vector model required. - [Sparse vector-based query API](https://help.aliyun.com/en/open-search/vector-search-edition/hybrid-retrieval.md): Submit hybrid search combining dense and sparse vector scoring in a single request. Use sparse vectors for keyword matching alongside dense vectors for semantic similarity. - [Primary key query API](https://help.aliyun.com/en/open-search/vector-search-edition/primary-key-query.md): POST /vector-service/fetch to retrieve specific documents by primary key. Returns full document fields for point lookups and data verification. - [Filter expression syntax](https://help.aliyun.com/en/open-search/vector-search-edition/filter-expression.md): Syntax: filter:"expression" with relational operators (>, <, =, >=, <=, !=). Combine conditions with AND/OR. Supports numeric, string, and multi-value field filtering. - [Hybrid query API (vector + text)](https://help.aliyun.com/en/open-search/vector-search-edition/inverted-query.md): Combine kNN vector search with text keyword search in one request. Ideal for queries mixing natural language with exact terms, product names, or IDs where neither approach alone suffices. - [Data API aggregation syntax](https://help.aliyun.com/en/open-search/vector-search-edition/api-vector-engine-statistical-syntax.md): POST /vector-service/aggregate for grouped statistics (max, min, avg, sum, count) on table fields. Uses HTTP Basic auth with JSON request body. - [Data API stats endpoint](https://help.aliyun.com/en/open-search/vector-search-edition/statistics.md): POST /vector-service/stats to retrieve table-level statistics including vector counts and index state. HTTP protocol with JSON format and Basic authentication. - [Management and Control API reference](https://help.aliyun.com/en/open-search/vector-search-edition/description-of-the-control-api.md): Service identifier: searchengine-2021-10-25. Manage instances, tables, and indexes programmatically. Links to operation list, endpoints, RAM auth, and error codes. - [OpenSearch Vector Search - API Push Data Error Codes (1000/3001/6015)](https://help.aliyun.com/en/open-search/vector-search-edition/error-code-description.md): Common error codes for OpenSearch Vector Search Edition API data push: 1000 timeout retry, 3001 empty document, 3008 too frequent requests, 3013/3014 push rate exceeded, 3015 single push count exceeded, 3016 total docs over quota, 4007 JSON decode failure, 6015 LCU quota exceeded (Dropped by over compute resource quota; scale up to resolve). ## AISearch Classic Edition (Discontinued) - [AISearch Classic Edition (Discontinued)](https://help.aliyun.com/en/open-search/vector-search-edition/vector-retrieval-general-edition.md) - [Getting started](https://help.aliyun.com/en/open-search/vector-search-edition/getting-started-2.md): This guide helps you get started with OpenSearch Vector Search Edition. Before you begin, you must create and authorize Resource Access Management (RA... - [Quick start for General-purpose Edition](https://help.aliyun.com/en/open-search/vector-search-edition/getting-started-1.md): End-to-end setup: purchase instance, configure table and index, push data via API, and run query test. Requires VPC environment and AliyunSearchEngineFullAccess permission. - [Image search quick start](https://help.aliyun.com/en/open-search/vector-search-edition/quick-start-figure-search-scene.md): Set up an image search instance with built-in vectorization. Covers instance purchase, image data import, and search-by-image query testing. - [Text search quick start](https://help.aliyun.com/en/open-search/vector-search-edition/text-vector-scene-quickstart.md): Set up a text semantic search instance with built-in text embedding. Covers instance configuration, data ingestion, and text-to-vector query testing. - [Create and authorize RAM users (1)](https://help.aliyun.com/en/open-search/vector-search-edition/create-and-authorize-ram-users-1.md): Grant RAM users AliyunSearchEngineFullAccess (management) or AliyunSearchEngineReadOnlyAccess (read-only) system policies. Also supports custom policies for monitoring-only access. - [Service-linked role AliyunServiceRoleForSearchEngine](https://help.aliyun.com/en/open-search/vector-search-edition/opensearch-vector-search-version-service-association-role.md): Auto-created RAM role granting OpenSearch access to OSS (ListObjects, GetObject) for data source integration. Required when configuring OSS-based data sources. - [Instance management](https://help.aliyun.com/en/open-search/vector-search-edition/instance-management-2.md): Instances pageAt the top of the page are the Create Instance button and a search bar to search by instance ID. The instance list includes the followin... - [O&M center](https://help.aliyun.com/en/open-search/vector-search-edition/operation-and-maintenance-center-1.md): The OpenSearch Operation Center integrates multiple operations management features to improve stability.Deployment management: You can view the cluste... - [Operations management](https://help.aliyun.com/en/open-search/vector-search-edition/operation-and-maintenance-management-1.md): In OpenSearch, you can use the following operations and features to manage and maintain the performance and stability of indexes and clusters.Modify a... - [Modify an index table](https://help.aliyun.com/en/open-search/vector-search-edition/modify-an-index-table-1.md): Edit index_config_edit schema, save a new version, update configuration, and rebuild index. Version number auto-increments (e.g., index_config_v4). - [Update configurations](https://help.aliyun.com/en/open-search/vector-search-edition/update-configurations.md): Apply configuration changes after modifying data sources, index tables, or advanced settings. Select data source, index table version, dictionary version, and destination cluster before rebuilding. - [Reindex an index](https://help.aliyun.com/en/open-search/vector-search-edition/perform-reindexing-1.md): Trigger full reindexing: select a partition for MaxCompute sources or use empty data/timestamp for API sources. Creates a new index generation. - [Restore data from an index version](https://help.aliyun.com/en/open-search/vector-search-edition/restore-data-from-an-index-version.md): Roll back to a previous index version when API-pushed data expires (retained only 3 days after full indexing). Restores data from a saved index generation. - [OpenSearch Vector Search Edition: Quick Shard Scaling (Data Node Expansion)](https://help.aliyun.com/en/open-search/vector-search-edition/scale-shards.md): Scale shards to expand data nodes, effective immediately. Index tables must share shard count; vector-index tables unsupported; run force switch afterward. - [Add a cluster](https://help.aliyun.com/en/open-search/vector-search-edition/add-a-cluster.md): Add a second cluster for capacity expansion or dual-cluster high availability. New cluster shares the same instance but processes queries independently. - [Version rollback](https://help.aliyun.com/en/open-search/vector-search-edition/roll-back-to-a-previous-version.md): Roll back to a previous index generation when the current version has issues. Each rebuild creates a new generation; rollback reverts to an earlier one. - [Deployment management](https://help.aliyun.com/en/open-search/vector-search-edition/deployment-management.md): View cluster topology graph and modify online configuration. Displays node distribution, resource allocation, and allows runtime configuration changes. - [Change History](https://help.aliyun.com/en/open-search/vector-search-edition/change-history.md): View operation records for index building, table creation, resource scaling, and configuration updates on a Vector Search Edition instance. - [Configuration center](https://help.aliyun.com/en/open-search/vector-search-edition/configuration-center-1.md) - [Index Schema](https://help.aliyun.com/en/open-search/vector-search-edition/index-schema-1.md): Define index table schema with fields, inverted indexes (term-to-DocID), forward indexes (attributes for sorting/filtering), and summary indexes for result display. Supports zstd compression. - [Advanced settings](https://help.aliyun.com/en/open-search/vector-search-edition/advanced-settings.md): In OpenSearch Vector Search Edition, advanced configuration includes the following areas:Dictionary configuration: You can define and manage custom di... - [Dictionary Configuration](https://help.aliyun.com/en/open-search/vector-search-edition/dictionary-configurations-1.md): Customize tokenization by adding intervention entries to built-in dictionaries (Chinese, e-commerce, game, education analyzers). Upload .dict/.txt files up to 5 MB. - [Query Configuration](https://help.aliyun.com/en/open-search/vector-search-edition/query-configurations.md): Tune QRS parameters: connection timeout (qrs_connection_timeout), return hits limit, compression type, default index, rank/rerank sizes. Publish versioned configs and apply to clusters in real time. - [Data Processing Configuration](https://help.aliyun.com/en/open-search/vector-search-edition/data-processing-configuration-1.md): Scale out data processing resources by modifying process_partition_count. Default free quota is two resources per data source. Versioned configs (api_v1, api_edit) with sequential naming. - [Data sources](https://help.aliyun.com/en/open-search/vector-search-edition/data-sources-1.md): OpenSearch uses data sources to retrieve and integrate external data.MaxCompute data source: MaxCompute is a big data computing platform. By connectin... - [MaxCompute Data Source](https://help.aliyun.com/en/open-search/vector-search-edition/maxcompute-data-source.md): Connect a MaxCompute (ODPS) table as data source. Requires describe/select/download permissions. Supported field types: STRING, BOOLEAN, DOUBLE, BIGINT, DATETIME. - [API Push Data Source](https://help.aliyun.com/en/open-search/vector-search-edition/api-data-source-z.md): Add or delete an API push data source. Cannot delete if an index table is configured. Full data import clears all previously pushed data and syncs from specified timestamp. - [OSS Data Source](https://help.aliyun.com/en/open-search/vector-search-edition/object-storage-oss.md): Connect an OSS bucket as data source. Folder name must contain "opensearch". Configure Data Source Name, OSS Path, and Bucket parameters. - [Best practices](https://help.aliyun.com/en/open-search/vector-search-edition/best-practices.md) - [End-to-End Image Search](https://help.aliyun.com/en/open-search/vector-search-edition/the-end-to-end-graph-search-solution.md): Build image search without pre-computed vectors. Three ingestion methods: OSS+MaxCompute, MaxCompute with Base64 images, or API push. Supports search-by-image and text-to-image retrieval. - [Conversational Search with LLM (RAG)](https://help.aliyun.com/en/open-search/vector-search-edition/integrate-opensearch-with-an-llm-for-intelligent-operation-and-maintenance.md): Build a RAG-based Q&A system combining vector search with an LLM. Offline: embed text and build vector index. Online: query vectorization, top-N retrieval, prompt assembly, and LLM answer generation. - [Documentation of Vector Search Edition V3.9.0](https://help.aliyun.com/en/open-search/vector-search-edition/documentation-of-vector-search-edition-v3-9.md) - [Instance configuration](https://help.aliyun.com/en/open-search/vector-search-edition/instance-configuration.md): In OpenSearch, you can configure Vector Search Edition instances online or offline:Online configuration: You can use the OpenSearch console or an Appl... - [Online configuration](https://help.aliyun.com/en/open-search/vector-search-edition/online-configuration.md) - [Index Table Online Services Configuration](https://help.aliyun.com/en/open-search/vector-search-edition/index-table-online-services-configuration.md): Configure real-time indexing with online_index_config: build_total_memory, max_realtime_memory_use, load_config (mmap strategy with slice/lock params), and async dump settings. - [Index Table Loading Policy](https://help.aliyun.com/en/open-search/vector-search-edition/the-index-table-loading-strategy.md): Define per-file loading strategies (mmap or cache) using file_patterns matching. Configure lock, partial_lock, slice size, warmup_strategy, and lifecycle (hot) parameters. - [Offline configuration](https://help.aliyun.com/en/open-search/vector-search-edition/offline-configuration.md): In OpenSearch Vector Search Edition, offline configuration consists of two main parts:Index table configuration: Defines the data format of original d... - [Index Table Configuration](https://help.aliyun.com/en/open-search/vector-search-edition/configure-an-index-table.md): Define index table schema: table_name, fields, inverted indexes (indexs), forward indexes (attributes), summary indexes (summarys), and TTL settings (enable_ttl, default_ttl in seconds). - [Offline Cluster Configuration](https://help.aliyun.com/en/open-search/vector-search-edition/offline-cluster-configuration.md): Configure offline builder and merger: customized_merge_config for merge strategies (optimize, align_version), period scheduling, merge_parallel_num, and keep_version_count settings. - [Data types](https://help.aliyun.com/en/open-search/vector-search-edition/data-types-1.md): Built-in field types for Vector Search EditionField type introductionField typeDescriptionSupports multiple valuesCan be used in a forward indexCan be... - [Document Formats](https://help.aliyun.com/en/open-search/vector-search-edition/document-formats.md): JSON format for real-time document push. Each document has cmd (add/delete) and fields parameters. Add replaces existing documents with same ID. - [Analyzers](https://help.aliyun.com/en/open-search/vector-search-edition/analyzer.md): Alibaba Cloud OpenSearch Vector Search Edition provides a variety of analyzers to meet different language and application requirements:General-purpose... - [Indexes](https://help.aliyun.com/en/open-search/vector-search-edition/indexes.md): Each document consists of multiple fields, and each field contains a series of words. Building an index speeds up data retrieval. Indexes are classifi... - [Introduction to inverted indexes](https://help.aliyun.com/en/open-search/vector-search-edition/inverted-indexes.md): What is an inverted index?An inverted index, also known as a reverse index, is an indexing method used for full-text search. It maps words to their lo... - [Inverted Index Types](https://help.aliyun.com/en/open-search/vector-search-edition/types-of-inverted-indexes.md): PACK index: multi-field TEXT index with section info, truncation, and tf-bitmaps. Supports df, ttf, position, docpayload, and termpayload options. - [Forward indexes](https://help.aliyun.com/en/open-search/vector-search-edition/forward-indexes.md): Introduction to indexesWhat is a forward index?A forward index, also called an attribute or profile index, stores the value of a specific field for a ... - [Summary Indexes](https://help.aliyun.com/en/open-search/vector-search-edition/summary-indexes.md): Store complete document info indexed by DocID for result display. Supports zlib compression. Retrieve summary only for displayed documents to minimize overhead. - [Vector indexes](https://help.aliyun.com/en/open-search/vector-search-edition/vector-indexes-2.md): The vector-based retrieval mechanism expresses commodity data and content data in the form of vectors and builds a vector index library. You can specify one or more user vectors or commodity vectors in a vector index library to retrieve a top-k list of commodities or content based on vector distance. - [Proxima Builder Parameters](https://help.aliyun.com/en/open-search/vector-search-edition/proxima-builder.md): Vector index build parameters for LinearBuilder, QcBuilder (centroid_count, cluster_class, quantizer), and HnswBuilder (max_neighbor_count, efconstruction, thread_count). - [Proxima Searcher Parameters](https://help.aliyun.com/en/open-search/vector-search-edition/proxima-searcher.md): Vector index search parameters for LinearSearcher (read_block_size), QcSearcher, and HnswSearcher. Controls memory read size, scan ratio, and search precision at query time. - [Proxima Cluster Parameters](https://help.aliyun.com/en/open-search/vector-search-edition/proxima-cluster-parameters.md): Clustering algorithm parameters: KmeansCluster/BatchKmeansCluster (centroid count, iterations), OptKmeansCluster, and HierarchicalCluster configurations for vector index building. - [SQL statements](https://help.aliyun.com/en/open-search/vector-search-edition/sql-statements.md): query clauseThe query clause is the most basic and indispensable part of a statement. The query clause defines the specific content to be queried base... - [Query syntax](https://help.aliyun.com/en/open-search/vector-search-edition/query-syntax-3.md): Query syntaxTuring SQL supports the `query` and `kvpair` clauses. The `query` clause lets you write SQL queries. The `kvpair` clause lets you specify ... - [SQL Data Types](https://help.aliyun.com/en/open-search/vector-search-edition/data-types.md): Three SQL field type categories: AtomicType (basic scalar types), ArrayType (multi-value fields), and MultiSetType. Reference for field declarations in TuringSQL queries. - [Query statements](https://help.aliyun.com/en/open-search/vector-search-edition/query-statements-1.md) - [SELECT Statement](https://help.aliyun.com/en/open-search/vector-search-edition/select.md): Core SQL query statement supporting DISTINCT, WHERE, GROUP BY, ORDER BY, and LIMIT clauses. Available in Vector Search Edition V3.7.0+. Includes CASE WHEN expressions. - [WHERE Clause (SQL)](https://help.aliyun.com/en/open-search/vector-search-edition/where-1.md): Conditionally filter rows in SELECT statements. Requires Ha3 version >= 3.7.0. Supports boolean expressions for row-level filtering. - [UNION Clause (SQL)](https://help.aliyun.com/en/open-search/vector-search-edition/union-1.md): Merge multiple result sets with identical schema (same field types and order). UNION ALL keeps duplicates; UNION removes them. - [Dynamic Parameters (SQL)](https://help.aliyun.com/en/open-search/vector-search-edition/dynamic-parameters-1.md): Use placeholder (?) syntax in SQL statements with Iquan auto-replacement. Pass parameter values at query time for flexible statement reuse. - [Hint Syntax (SQL)](https://help.aliyun.com/en/open-search/vector-search-edition/hint.md): TuringSQL Hint semantics for advanced SQL customization. Requires Ha3 version >= 3.7.3. Override query planner behavior per statement. - [Table-Valued Functions (TVF)](https://help.aliyun.com/en/open-search/vector-search-edition/table-valued-function.md): TuringSQL TVFs operate on entire tables, accepting scalar parameters and a table input, returning a new table. Use for custom table-level transformations in queries. - [Common Table Expressions (CTE)](https://help.aliyun.com/en/open-search/vector-search-edition/common-table-expressions-1.md): Temporary named result sets referenceable in subsequent queries. Requires Ha3 version >= 3.7.5. Simplify complex queries by breaking them into named subqueries. - [SQL Query Limits](https://help.aliyun.com/en/open-search/vector-search-edition/limits-3.md): Unsupported DDL/DML syntax restrictions. ORDER BY requires LIMIT clause. Documents known SQL limitations for the Vector Search Edition query engine. - [User-defined functions](https://help.aliyun.com/en/open-search/vector-search-edition/user-defined-functions-1.md): OpenSearch Vector Search Edition supports two types of user-defined functions:User-defined function (UDF): Allows you to create custom operation funct... - [UDF](https://help.aliyun.com/en/open-search/vector-search-edition/udf-1.md): Built-in UDFsComplex type functionsFunction nameDescriptionVersioncontainChecks if a field value is in a given set.ALLnotcontainChecks if a field valu... - [UDAF](https://help.aliyun.com/en/open-search/vector-search-edition/udaf.md): UsageList of built-in UDAFsHA3 provides several built-in user-defined aggregate functions (UDAFs):sum: Calculates the sum.avg: Calculates the average.... - [Query Result Formats](https://help.aliyun.com/en/open-search/vector-search-edition/interpret-query-results.md): Four return formats: string (human-readable debugging), json, full_json, and flatbuffers. Set via config file (global) or kvpair clause (per-query) with formatType parameter. - [Query DSL of OpenSearch Vector Search Edition](https://help.aliyun.com/en/open-search/vector-search-edition/query-dsl-of-opensearch-vector-search-edition.md): The query domain-specific language (DSL) for OpenSearch Vector Search Edition provides the following powerful and flexible query features:Query syntax... - [Query syntax](https://help.aliyun.com/en/open-search/vector-search-edition/query-syntax.md) - [Havenask query Clause](https://help.aliyun.com/en/open-search/vector-search-edition/query-clause.md): Core Havenask search clause specifying index and search terms. Syntax: index_name:search_query^boost. Supports AND, OR, ANDNOT, RANK operators for combining conditions. - [Havenask config Clause](https://help.aliyun.com/en/open-search/vector-search-edition/config-clause.md): Required Havenask clause for runtime query control: starting position, hit count, result format, and cluster selection. Provides real-time query parameter override. - [Havenask cluster Clause](https://help.aliyun.com/en/open-search/vector-search-edition/cluster-clause.md): Required Havenask clause to specify target cluster name or specific shards. Route queries to designated partitions within a multi-shard deployment. - [Havenask filter Clause](https://help.aliyun.com/en/open-search/vector-search-edition/filter-clause.md): Post-retrieval document filtering in Havenask queries using conditional expressions on attribute fields. Applied after query clause narrows initial result set. - [Havenask sort Clause](https://help.aliyun.com/en/open-search/vector-search-edition/sort-clause.md): Sort results with +field (ascending) or -field (descending) syntax. Supports multi-dimensional sorting (semicolon-separated), arithmetic expressions, and RANK relevance sorting. - [Aggregate clause](https://help.aliyun.com/en/open-search/vector-search-edition/aggregate-clause.md): Group results by field with count/sum/max/min functions, range-based distribution, sampling thresholds, and max_group limit. Syntax: aggregate=group_key:field,agg_fun:func - [Distinct clause](https://help.aliyun.com/en/open-search/vector-search-edition/distinct-clause.md): Deduplicate results by field to ensure diversity. Params: dist_key, dist_count, dist_times, dist_filter, grade. Syntax: distinct=dist_key:field,dist_count:N,dist_times:N - [Kvpairs clause](https://help.aliyun.com/en/open-search/vector-search-edition/kvpairs-clause.md): Pass key-value parameters to sort expressions and feature functions at query time. Syntax: kvpairs=key1:value1,key2:value2. Supports tag_match and duniqfield usage - [Attribute clause](https://help.aliyun.com/en/open-search/vector-search-edition/attribute-clause.md): Return specified property field values in query results. Syntax: attribute=field1,field2. Nonexistent attributes cause the entire query to fail - [Virtual attribute clause](https://help.aliyun.com/en/open-search/vector-search-edition/virtual-attribute-clause.md): Define reusable named expressions (arithmetic, logical, function) usable in distinct, filter, aggregate, attribute, and sort clauses. Syntax: virtual_attribute=va_name:va_expression - [Layer clause](https://help.aliyun.com/en/open-search/vector-search-edition/layer-clause.md): Control document recall by specifying scan ranges, retrieval order, and per-layer quotas. Supports multi-layer recall with different query clauses per range - [Analyzer clause](https://help.aliyun.com/en/open-search/vector-search-edition/analyzer-clause.md): Control query tokenization per index. Params: global_analyzer, specific_index_analyzer, no_tokenize_indexes. Priority: no_tokenize > specific > global > schema default - [searcher_cache clause](https://help.aliyun.com/en/open-search/vector-search-edition/searcher-cache-clause.md): Cache sorted query results to reduce re-ranking overhead. Params: use (yes/no), key, expire_time, cur_time, cache_filter, cache_doc_num_limit, refresh_attributes - [fetch_summary clause](https://help.aliyun.com/en/open-search/vector-search-edition/fetch-summary-clause.md): Perform independent phase-2 queries to fetch document summaries by docid, primary key hash, or primary key value. Set fetch_summary_type in config clause - [Advanced queries](https://help.aliyun.com/en/open-search/vector-search-edition/advanced-queries.md): Independent phase-1 queriesYou can include the no_summary parameter in a config clause to specify whether to perform only a phase-1 query. The default... - [Query Results and Parsing](https://help.aliyun.com/en/open-search/vector-search-edition/query-result-formats.md): Query result formatsHavenask supports query results in XML and JSON formats. The XML format is easy to read but less efficient and is often used for e... - [Parse JSON format results](https://help.aliyun.com/en/open-search/vector-search-edition/query-results-in-json-format.md): JSON response structure: searchtime, totalHits, numHits, coveredPercent, items (fields, property, attribute, variableValue). Set format:json in config clause - [Parse protobuf format results](https://help.aliyun.com/en/open-search/vector-search-edition/parse-protobuf-format-results.md): Deserialize protobuf responses into PBResult. Proto defines PBAttrKVPair, PBInt64Attribute, PBDoubleAttribute, PBBytesAttribute message types - [Parse XML format results](https://help.aliyun.com/en/open-search/vector-search-edition/query-results-in-xml-format.md): XML response structure for normal, standalone phase-one, and standalone phase-two query types. Covers result parsing for each query classification - [Custom calculation logic](https://help.aliyun.com/en/open-search/vector-search-edition/custom-calculation-logic-1.md): Sort expressionsExpression syntaxHavenask lets you customize scoring formulas using built-in operators. The following table lists the supported operat... - [Dynamic summary](https://help.aliyun.com/en/open-search/vector-search-edition/dynamic-summary.md): Configure highlighted search result snippets via freeschema_extract_fields in kvpairs. Specify field name, fragment count, fragment length, highlight tags, and separator - [Metrics](https://help.aliyun.com/en/open-search/vector-search-edition/metrics.md): In OpenSearch, monitoring metrics are used to track and evaluate system performance and health.QRS worker metrics: QRS workers process user queries an... - [QRS node metrics](https://help.aliyun.com/en/open-search/vector-search-edition/qrs-metrics.md): QRS worker monitoring: CPU, memory, query QPS (normal/phase-1/phase-2), latency (session/wait/process), SQL query metrics, and error QPS breakdown - [Searcher node metrics](https://help.aliyun.com/en/open-search/vector-search-edition/searcher-metrics.md): Searcher monitoring: CPU, memory, query latency (rank/rerank/extraRank), index stats (partitionDocCount, segmentCount), real-time build QPS, vector index recall ratio - [Engine performance optimization](https://help.aliyun.com/en/open-search/vector-search-edition/engine-performance-tuning.md): Lock policy for vector indexesConfigure the lock policy for vector indexesFor business scenarios that require high-performance vector retrieval, you c... - [Developer guide (ha3_3.10.0)](https://help.aliyun.com/en/open-search/vector-search-edition/developer-guide.md): ha3_3.10.0 is the engine version identifier for the traditional vector search engine in Alibaba Cloud OpenSearch Vector Search Edition. For more infor... - [Traffic API reference](https://help.aliyun.com/en/open-search/vector-search-edition/traffic-api-reference-3.md): The traffic API reference for OpenSearch Vector Search Edition covers the following topics:ha search processing: Build query requests using URL parame... - [HA search request API](https://help.aliyun.com/en/open-search/vector-search-edition/initiate-search-requests-2.md): HTTP GET search endpoint using HA query syntax. Params: query, config (start, hit, format), cluster, sort. Requires Base64-encoded authorization header - [RESTful search request API](https://help.aliyun.com/en/open-search/vector-search-edition/restful-api-search-processing.md): HTTP POST endpoint for RESTful search. URL: /{indexName}/search. JSON request body format with Base64-encoded authorization header - [Data processing API](https://help.aliyun.com/en/open-search/vector-search-edition/process-data-2.md): HTTP POST endpoint for bulk data operations (add/delete). URL: /update/$table_name/actions/bulk. Requires X-Opensearch-Swift-PK-Field header and Base64 authorization - [Control API guide](https://help.aliyun.com/en/open-search/vector-search-edition/management-and-control-api-reference.md): Index page for management APIs covering endpoints, authorization, API reference, common error codes, and versioning for ha3_3.10.0 engine - [SDK for Java](https://help.aliyun.com/en/open-search/vector-search-edition/sdk-for-java.md): This Java SDK encapsulates traffic API operations, such as data processing and search. For other features, use the management SDK. - [SDK clients](https://help.aliyun.com/en/open-search/vector-search-edition/sdk-clients.md): This topic describes the classes in the Java SDK client, including their methods and parameters. This information helps you understand how to use each... - [Config class](https://help.aliyun.com/en/open-search/vector-search-edition/config-class.md): SDK Config class for building Client objects. Constructor params: endpoint, instanceId, accessUserName, accessPassWord, httpProxy. Thread-safe - [Client class](https://help.aliyun.com/en/open-search/vector-search-edition/client-class.md): SDK Client class for data push and search. Methods: pushDocuments(dataSourceName, keyField, request), Search(). Not thread-safe. Requires Config object - [PushDocumentsRequestModel class](https://help.aliyun.com/en/open-search/vector-search-edition/pushdocumentsrequestmodel-class.md): Request model for Client.pushDocuments() method. Thread-safe. Set headers and document body via setHeaders() and setBody() methods - [PushDocumentsResponseModel class](https://help.aliyun.com/en/open-search/vector-search-edition/pushdocumentsresponsemodel-class.md): Response model returned by Client.pushDocuments(). Contains push result status, headers, and body for document operations - [SearchQuery class](https://help.aliyun.com/en/open-search/vector-search-edition/searchquery-class.md): Build search requests with HA3 query syntax or SQL. Methods: setQuery(String) for HA3, setSql(String) for SQL mode. Used to construct SearchRequestModel - [SearchResponseModel class](https://help.aliyun.com/en/open-search/vector-search-edition/searchresponsemodel-class.md): Response object returned by Client.Search(). Contains search result body, headers, and status information for query responses - [SearchRequestModel class](https://help.aliyun.com/en/open-search/vector-search-edition/searchrequestmodel-class.md): Request parameter for Client.Search() method. Wraps SearchQuery with optional headers and table routing configuration - [Java SDK data push demo](https://help.aliyun.com/en/open-search/vector-search-edition/data-push-demo-2.md): Java code example for real-time data sync (add/delete) using Config, Client, and PushDocumentsRequestModel classes. Includes endpoint and auth setup - [Java SDK document search demo](https://help.aliyun.com/en/open-search/vector-search-edition/document-search-demo-2.md): Java code for querying data via four methods: SQL QUERY, SQL string, HA QUERY, and HA query string. Uses SearchQuery and Client.Search() - [Java SDK RESTful search demo](https://help.aliyun.com/en/open-search/vector-search-edition/restful-api-document-search-demo-1.md): Java code example for searching via RESTful API using the SDK to call the Havenask engine. Alternative to HA query and SQL methods - [Java SDK parse retrieval results](https://help.aliyun.com/en/open-search/vector-search-edition/parsing-of-retrieval-results.md): Parse protobuf and flatbuffers binary responses in Java. Includes Maven dependencies (protobuf 3.21.5, grpc 1.6.1) and deserialization code - [Java SDK control operations demo](https://help.aliyun.com/en/open-search/vector-search-edition/management-and-control-sdk-demo.md): Java code examples for management API operations including instance and table control method calls via the SDK - [SDK for Python](https://help.aliyun.com/en/open-search/vector-search-edition/sdk-for-python.md): This Python SDK encapsulates traffic API operations, such as data processing and search. For other features, use the management SDK. - [Python SDK data push demo](https://help.aliyun.com/en/open-search/vector-search-edition/data-push-demo-1.md): Python code for real-time document sync (upload/delete) to Vector Search Edition. Uses SDK client with endpoint and auth configuration - [Python SDK document search demo](https://help.aliyun.com/en/open-search/vector-search-edition/document-search-demo-1.md): Python code for querying data from Vector Search Edition using the SDK. Demonstrates search request construction and result parsing - [Python SDK RESTful search demo](https://help.aliyun.com/en/open-search/vector-search-edition/pythonrestful.md): Python code for searching via RESTful API endpoint. Alternative to HA query and SQL methods for data retrieval from Vector Search Edition - [SDK for Go](https://help.aliyun.com/en/open-search/vector-search-edition/sdk-for-go.md): This Go software development kit (SDK) encapsulates API operations for traffic, data processing, and search. For other features, you can use the management SDK. - [Data structure](https://help.aliyun.com/en/open-search/vector-search-edition/data-structure.md): This topic describes the structure of the response body for Havenask management APIs. ## Legal Resources - [Legal Resources](https://help.aliyun.com/en/open-search/vector-search-edition/legal-resources.md) - [OpenSearch SLA](https://help.aliyun.com/en/open-search/vector-search-edition/opensearch-sla-2.md): Service Level Agreement for OpenSearch Retrieval Engine Edition and Vector Search Edition, effective December 1, 2022