HTAP OLTP and OLAP Platform
An HTAP platform on a distributed SQL database serves OLTP from a row store and OLAP from a synchronized columnar replica, eliminating ETL lag. Raft replication gives consistency, scale-out, and resilience across hybrid deployments.
HTAP OLTP and OLAP Platform
Hybrid Transactional and Analytical Processing (HTAP) runs operational transactions (OLTP) and analytical queries (OLAP) on the same data without an ETL delay. This platform uses a distributed SQL database that keeps a row store for transactions and a synchronized columnar store for analytics. Deploy it in a hybrid environment to serve applications and fresh dashboards from one system. Use HTAP when analytics must reflect the latest writes and you want to avoid maintaining separate transactional and analytical copies.
Components
- TiDB: the stateless SQL layer that parses queries and routes them to the right storage engine.
- TiKV: the distributed, transactional row store backing OLTP workloads with strong consistency.
- TiFlash: the columnar replica that serves analytical queries, kept in sync via the Raft log.
- Placement Driver (PD): the cluster brain that manages metadata, data placement, and load balancing.
- Kafka: streams change events out to downstream lakes and warehouses when needed.
- Load balancer: distributes client connections across SQL nodes.
- Grafana: dashboards over cluster and query metrics.
Data Flow
Applications issue SQL through the load balancer to TiDB nodes. Writes commit to TiKV with multi-version concurrency control and Raft replication. TiFlash subscribes to the same Raft log, so columnar replicas stay continuously consistent with the row store. The optimizer routes point lookups and updates to TiKV and large scans and aggregations to TiFlash, sometimes within one query. Optional Kafka change feeds export data to a broader analytics estate.
Scaling and Resilience
Storage and compute scale horizontally by adding TiKV, TiFlash, or TiDB nodes; PD rebalances regions automatically. Raft replication across nodes and zones tolerates failures with no data loss and automatic leader election. Because TiFlash is a replica, analytical load does not slow transactions. Online schema changes and rolling upgrades avoid downtime, and cross-cluster replication supports disaster recovery.
Security
The platform enforces TLS for all internal and client traffic, role-based access control, and encryption at rest. Authentication integrates with directory services, and audit logging records access. In a hybrid deployment, private networking connects on-prem and cloud nodes, and least-privilege service accounts limit component permissions.
Trade-offs and Alternatives
HTAP eliminates ETL lag and operational complexity of separate systems, but maintaining row and columnar replicas costs extra storage and the engines are less specialized than a dedicated OLTP database or a tuned warehouse. At very large analytical scale, a purpose-built warehouse still wins. Alternatives include SingleStore, Google AlloyDB, or pairing PostgreSQL with a CDC pipeline to a warehouse. Choose HTAP when fresh, unified transactional-plus-analytical access on one system is the dominant requirement.