Bigtable

Bigtable is Google Cloud's managed wide-column NoSQL database with HBase and Cassandra APIs, billed per node, for apps needing fast key lookups on huge tables.

Bigtable holds very large tables of structured, semi-structured and unstructured data for applications that read and write by row key. Google runs it as a managed service on Google Cloud, aimed at machine learning, operational analytics and user-facing applications. Its data model is key-value and wide-column, with very wide tables grouped into column families. It is most similar to Apache HBase and Cassandra, the open source projects it inspired, and it offers APIs for both. Each table is sorted by its row key and split into tablets, which Bigtable splits, merges and rebalances on its own as data grows. Replication keeps copies in clusters across several regions, and multi-cluster routing lets applications write through any of them. When two clusters change the same cell, a last-write-wins rule based on server time settles it. A SQL query API built on GoogleSQL adds extensions for the wide-column model that resemble Cassandra Query Language, and it runs K-nearest neighbor searches over vector embeddings. Change streams capture data changes as they happen, garbage collection policies drop old cells by age or version count, and tiered storage moves cold data off SSD. Bigtable is built for latency-sensitive workloads such as personalization, ad tech, time series and IoT data, and the feature stores behind machine learning predictions. It suits engineering teams holding terabytes or petabytes who need very high query rates without operating the cluster themselves. Teams leaving self-managed HBase or Cassandra get the HBase replication library and Cassandra Proxy for live migrations without downtime. Bigtable Data Bridge and the Aerospike Migration Tool cover moves from Amazon DynamoDB and Aerospike. Capacity is sized in nodes per cluster, and autoscaling changes the node count as demand moves. Bigtable serves as the storage layer behind user-facing applications, holding data that other systems produce and stream in. Dataflow pipelines enrich streaming events with low-latency lookups against its tables, while BigQuery queries the same data for ad-hoc analysis. Pub/Sub, Kafka Connect, Apache Beam, Flink and Spark link it to event streams and batch jobs. A LangChain integration brings its vector search into retrieval for generative AI applications. It fits product teams whose app must fetch one user's profile or one device's history quickly while the data keeps growing.