#Vectorstores

[govindhtech]

Spanner Instance Use New Spanner Dual-Region Configurations

Spanner instance in Google console use Spanner dual-region configurations.

Presenting Spanner dual-region solutions that ensure both high availability and data residency.

Spanner is a database managed by Google that is spread globally, has a high throughput, and can scale essentially without limits. At its peak, Spanner is capable of processing more than 4 billion queries per second. With its sophisticated features automated sharding, zero downtime, and great consistency spanner effectively manages challenging, global projects. Along with being heavily used inside Google, it fits a broad spectrum of businesses including financial services, gambling, retail, and others. These tasks can have relational as well as non-relational character.

Spanner separates compute resources from data storage, enabling seamless scaling of processing resources. Each additional compute capability has the ability to handle both read and write operations, allowing for easy horizontal scalability. Spanner enhances performance by autonomously managing the distribution of data over multiple servers, ensuring data redundancy, and efficiently processing transactions.

Handling failovers and controlling the internal mechanisms

The Spanner dual-region configurations are specifically designed to ensure that Google can provide the same level of availability and zero recovery-point objective (RPO) assurances as the Spanner multi-region deployments, which usually cover three distinct regions.

Spanner instance

Let consider a case where a Spanner instance is running in a newly launched dual-region setup in India. These occurrences duplicate data to Mumbai and Delhi. Each region consists of three replicas, with two being read/write replicas and one being a witness replica. The Spanner instance, when considering both regions together, comprises a total of six replicas. This setup is referred to as dual-region mode.

If one of the three replicas in a region has a zone outage, Spanner continues to operate in dual-region mode. In this mode, Spanner maintains a quorum of at least two replicas in each region, ensuring that your database remains accessible even during a zonal outage. If an entire area has an outage, databases in dual-region mode will lose their availability. In order to restore availability, databases need to transition to a single-region configuration, which entails having three copies located within a single region.

Initiating Spanner dual-region configurations

When instantiating a new Spanner object, you have the option to select the recently introduced dual-region setups in the Google Cloud console.

Aside from generating new Spanner instances, you can also transfer your current regional or multi-regional setups utilising the newly introduced self-serve instance move function of Spanner. Relocating your instance does not result in any period of time when the service is unavailable, and Spanner maintains its standard transaction assurances, such as strong consistency, during the relocation process.

Spanner

Enhance cost-efficiency and enhance dependability for databases of any magnitude.

Synchronous replication and maintenance are inherent and automated. Performing schema modifications and maintenance tasks entirely online, without any interruption to the ongoing traffic, ensuring continuous operation.

Integrate the scalability and stability features of Spanner with the familiar and portable PostgreSQL interface. Utilise the existing talents and tools of your teams, ensuring the long-term viability of your investment and providing reassurance.

Eliminate the need for manual resharding of your database. The use of sharding in the system automatically distributes data in order to maximise performance and ensure availability. Expand and contract seamlessly.

Utilise Spanner’s exact nearest neighbour (KNN) vector search (in preview) to search vector embeddings at a large scale. This feature is particularly useful for highly partitionable workloads, where each search is limited to data linked to a specific user. Spanner’s integrated KNN search capabilities are well-suited for these workloads, since they enable Spanner to effectively narrow down the search area and deliver precise, real-time results with minimal delays.

Spanner Data Boost allows users to execute analytical queries, batch processing tasks, or data export procedures more quickly without impacting the current transactional burden. Data Boost, which is fully managed by Google Cloud, eliminates the need for capacity planning or management. The system is consistently warm and prepared to directly handle user queries on data stored in Spanner’s distributed storage system, Colossus. This flexible and autonomous computing resource allows users to efficiently manage diverse workloads and securely share data without any concerns.

Integrate LangChain to effortlessly develop advanced AI applications that exhibit enhanced accuracy, transparency, and reliability. Spanner incorporates three LangChain integrations: Document loader, which facilitates the loading and storage of information from documents; Vector stores, which enable semantic search capabilities; and Chat Messages Memory, which allows chains to remember and retrieve past discussions.

In the past, if you wanted to ensure that your data was stored in countries that only had two Google Cloud regions, you were only able to use regional Spanner configurations that had a 99.99% availability. This is because Spanner multi-region arrangements necessitate the use of three regions, with one of them being situated outside of the country. The new Spanner dual-region configurations allow you to benefit from Spanner’s exceptional 99.999% availability and meet data residency requirements. The newly introduced dual-region setups, now accessible in Australia, Germany, India, and Japan, guarantee that your data is stored exclusively within the chosen nation.

Read more on govindhtech.com

[kaobei-engineer]

#純靠北工程師83g

----------

施主，我懂 document versioning 是啥，但這個 document，跟 LangChain 的 Document，跟 ChromaDB 的 Document，甚至候選的某某 SQL 的 Document，好像不是同一個概念？一邊搞 LangChain 一邊拿那個 VectorStore 做 CRUD 是不是... 有點跳脫？ （尤其是 LangChain 的文檔跟 LLM 一樣，存在大量幻覺）

----------

💖 純靠北工程師 官方 Discord 歡迎在這找到你的同溫層！

👉 https://discord.gg/tPhnrs2

----------

💖 全平台留言、文章詳細內容

👉 https://init.engineer/cards/show/10492

[netmarkjp]

#ばばさん通信ダイジェスト : ChatGPT x LangChain で独自ドキュメントのベクターストア検索をチューニングする

賛否関わらず話題になった/なりそうなものを共有しています。

ChatGPT x LangChain で独自ドキュメントのベクターストア検索をチューニングする

https://recruit.gmo.jp/engineer/jisedai/blog/chatgpt-vectorstore-tuning/

[faridilhamoglu]

Pumpkin patterns collection etsy: FImarket #farid_ilhamoglu_etsy #etsylove #etsyseller #etsyfinds #etsypatterns #etsyautumn #autumn #pumpkin #halloween #leaves #etsydigital #patterndesign #pattern #digitallyart #arts #этсипродвижение #wherebuy #гдекупить #фриланс #freelancerdesigner #autumnlover #green #orange #buyer #seller #vectorillustration #shutterstockcontributor #stocker #vectorstore https://www.instagram.com/p/B2P59euAHAr/?igshid=1s80g4dwi0iph

[govindhtech]

Spring AI 1.0 and Google Cloud to Build Intelligent Apps

Spring AI 1.0

After extensive development, Spring AI 1.0 provides a robust and dependable AI engineering solution for your Java ecosystem. This is calculated to position Java and Spring at the forefront of the AI revolution, not just another library.

Spring Boot is used by so many enterprises that integrating AI into business logic and data has never been easier. Spring AI 1.0 lets developers effortlessly integrate cutting-edge AI models into their apps, bringing up new possibilities. Prepare to implement smart JVM app features!

Spring AI 1.0 is a powerful and comprehensive Java AI engineering solution. Its goal is to lead the AI revolution with Java and Spring. Spring AI 1.0 integrates AI into business logic and data without the integration issues many Spring Boot-using enterprises confront. It lets developers use cutting-edge AI models in their apps, expanding possibilities.

Spring AI supports multiple AI models:

Images produced by text-command image models.

Audio-to-text transcription models.

Vectors are formed by embedding models that transform random data into them for semantic similarity search.

Chat models can edit documents and write poetry, but they are tolerant and easily sidetracked.

The following elements in Spring AI 1.0 enable conversation models overcome their limits and improve:

Use system prompts to set and manage model behaviour.

Memory is added to the model to capture conversational context and memory.

Making tool calling feasible for AI models to access external features.

Including confidential information in the request with rapid filling.

Retrieval Augmented Generation (RAG) uses vector stores to retrieve and use business data to inform the model's solution.

Evaluation to ensure output accuracy employs a different model.

Linking AI apps to other services using the Model Context Protocol (MCP), which works with all programming languages, to develop agentic workflows for complex tasks.

Spring AI integrates seamlessly with Spring Boot and follows Spring developers' convention-over-configuration setup by providing well-known abstractions and startup dependencies via Spring Initialisation. This lets Spring Boot app developers quickly integrate AI models utilising their logic and data.

When using Gemini models in Vertex AI, Google Cloud connectivity is required. A Google Cloud environment must be created by establishing or selecting a project, enabling the Vertex AI API in the console, billing, and the gcloud CLI.

Use gcloud init, config set project, and auth application-default login to configure local development authentication.

The Spring Initialiser must generate GraalVM Native Support, Spring Web, Spring Boot Actuator, Spring Data JDBC, Vertex AI Gemini, Vertex AI Embeddings, PGvector Vector Database, MCP Client, and Docker Compose Support to build a Spring AI and Google Cloud application. The site recommends using the latest Java version, especially GraalVM, which compiles code into native images instead of JRE-based apps to save RAM and speed up startup. Set application properties during configuration.characteristics for application name, database connection options, Vertex AI project ID and location for chat and embedding models (gemini-2.5-pro-preview-05-06), actuator endpoints, Docker Compose administration, and PGvector schema initialisation.

PostgreSQL database with a vector type plugin that stores data with Spring AI's VectorStore abstraction? Database schema and data can be initialised on startup using schema.sql and data.sql files, and Spring Boot's Docker Compose can start the database container automatically. Spring Data JDBC creates database interaction and data access entities.

The ChatClient manages chat model interactions and is a one-stop shop. ChatClients need autoconfigured ChatModels like Google's Gemini. Developers can create several ChatClients with different parameters and conditions using ChatClient.Builder. ChatClients can be used with PromptChatMemoryAdvisor or QuestionAnswerAdvisor to handle chat memory or VectorStore data for RAG.

Spring AI simplifies tool calls by annotating methods and arguments with @Tool and @ToolParam. The model evaluates tool relevance using annotation descriptions and structure. New default tools can be added to ChatClient.

Spring AI also integrates with the Model Context Protocol (MCP), which separates tools into services and makes them available to LLMs in any language.

For Google Cloud production deployments, Spring AI apps work with pgVector-supporting databases like Google Cloud SQL or AlloyDB. AlloyDB is built for AI applications with its high performance, availability (99.99% SLA including maintenance), and scalability.

FAQ

How is spring AI?

The Spring AI application framework simplifies Spring ecosystem AI application development. It allows Java developers to easily integrate AI models and APIs without retraining, inspired by LangChain and LlamaIndex.

Essentials and Ideas:

AI integration:

Spring AI integrates AI models with enterprise data and APIs.

Abstraction and Portability:

Its portable APIs work across vector database and AI model manufacturers.

Spring Boot compatibility:

It integrates with Spring Boot and provides observability tools, starters, and autoconfiguration.

Support the Model:

It supports text-to-image, embedding, chat completion, and other AI models.

Quick Templates:

Template engines let Spring AI manage and produce AI model prompts.

Vector databases:

It uses popular vector database providers to store and retrieve embeddings.

Tools and Function Calling:

Models can call real-time data access functions and tools.

Observability:

It tracks AI activity with observability solutions.

Assessing and Preventing Hallucinations:

Spring AI helps evaluate content and reduce hallucinations.