Advanced time-series database features matter most when requirements go beyond basic storage and queries. Teams may need data subscription, real-time computation, edge-cloud synchronization, zero-code ingestion, publishing, or AI analysis. The practical question is which capabilities to combine for a specific architecture, and which ones to leave out.
The seven advanced capabilities
1. Data subscription
For some internal data distribution workloads, data subscription can reduce the need for an external message queue, lowering component count and operational complexity. It supports both Native and MQTT subscription methods, with three topic types for different scenarios.
The Native method is suited for data synchronization between internal systems, offering low latency and high throughput. The MQTT method is suited for integration with external systems or edge devices, conforming to standard IoT protocols. In industrial data management platforms, data subscription is the foundational capability for data distribution and system decoupling.
2. Read cache
The built-in cache mechanism is a key tool for improving query performance. TDengine supports two cache modes: ‘last_row' and last_value. By caching the latest data points, it avoids frequent disk I/O operations.
In practical tests, the read cache delivers approximately 8x query performance improvement. For monitoring dashboards and real-time displays that need to frequently query the latest device status, the read cache significantly reduces query latency and improves user experience.
3. Stream processing
Stream processing uses SQL to define real-time data transformation pipelines, supporting six trigger modes and millisecond-level computation latency. Compared to external stream processing frameworks such as Flink or Spark Streaming, built-in stream processing reduces data transfer overhead between components and supports high-availability deployment.
The core advantage of built-in stream processing is its deep integration with the database. Computation results are written directly into database tables without additional data synchronization steps. For real-time databases, stream processing is the core engine behind real-time monitoring and alerting.
4. Edge-cloud collaboration
Edge-cloud collaboration supports bidirectional data synchronization between edge nodes and the cloud, handling millions of data points per second. It includes a resume-from-breakpoint mechanism that automatically continues incomplete transfers after network recovery, ensuring data integrity.
Zero-code configuration is a major advantage. Teams can define synchronization behavior without writing custom scripts for every edge-to-cloud path. In distributed industrial scenarios, edge-cloud collaboration is the key link for building three-tier “cloud-edge-endpoint” architectures.
5. Zero-code data ingestion
Zero-code data ingestion supports 17 common data sources, covering mainstream industrial IoT devices and protocols. With built-in ETL capabilities, it can perform format conversion, unit unification, and data filtering during the write process.
This feature significantly reduces the development cost of data ingestion, enabling non-developers to configure and manage data sources. For projects that need to bring multiple data sources online quickly, zero-code ingestion is the most efficient choice.
6. Data publishing
Data publishing pushes data from the database to external systems, supporting three destinations: MQTT, Kafka, and Flink. This feature complements data subscription: the former handles data inflow, the latter handles data outflow.
In scenarios that require integrating time-series data with external big data platforms or AI training platforms, data publishing provides a standardized data push channel, avoiding the complexity of point-to-point integrations.
7. TDgpt AI analysis
TDgpt provides four AI analysis capabilities: anomaly detection, predictive analysis, data imputation, and data classification. Through SQL function interfaces, users can invoke advanced analysis models without writing machine learning code.
AI analysis capabilities upgrade the time-series database from a pure data storage platform to an intelligent analysis platform, providing technical support for predictive maintenance, energy consumption optimization, and quality alerting.
Scenario-based selection recommendations
Real-time monitoring
Recommended combination: Read cache + Stream processing
Real-time monitoring requires low-latency data queries and real-time metric computation. The read cache ensures fast retrieval of the latest status data, while stream processing handles real-time aggregation and metric calculation.
Typical applications: production line monitoring dashboards, environmental monitoring displays, and real-time traffic statistics. In this architecture, stream processing aggregates raw data into minute-level or second-level metrics, while the read cache ensures dashboard refresh speed.
Data synchronization
Recommended combination: Data subscription + Edge-cloud collaboration
When time-series data needs to sync across multiple systems, data subscription and edge-cloud collaboration provide a complete solution. Data subscription handles same-tier system distribution; edge-cloud collaboration handles cross-tier data aggregation.
Typical applications: data sharing across multiple factories, automatic aggregation from edge gateways to the cloud, and disaster recovery data synchronization. Zero-code configuration reduces deployment and maintenance complexity.
Anomaly detection
Recommended combination: Stream processing + TDgpt
Anomaly detection requires combining real-time data streams with intelligent analysis. Stream processing transforms raw data into the feature data needed for analysis; TDgpt executes the anomaly detection algorithms.
Typical applications: equipment fault warning, quality anomaly detection, and energy consumption anomaly alerting. Stream processing can set event windows to identify persistent anomaly patterns, while TDgpt provides multiple detection algorithms for precise judgment.
Data aggregation
Recommended combination: Zero-code ingestion + Data publishing
The core of data aggregation is multi-source ingestion and controlled distribution. Zero-code ingestion quickly connects to diverse data sources; data publishing pushes the aggregated data to downstream systems.
Typical applications: data aggregation layers for industrial IoT platforms, data hubs for smart campuses, and data buses for multi-system integration. With 17 data source types and built-in ETL, the data aggregation build cycle shrinks from weeks to days.
Predictive analysis
Recommended combination: TDgpt + Stream processing
Predictive analysis needs both historical data accumulation and real-time data input. Stream processing continuously updates the datasets needed for analysis; TDgpt performs trend prediction based on historical and real-time data.
Typical applications: equipment lifespan prediction, energy load forecasting, and production yield prediction. Stream processing can periodically trigger model retraining, while TDgpt provides the full prediction capability from classical statistical models to deep learning models.
Architecture design principles
Match feature combinations to business needs. Each feature has its appropriate scenario. Do not adopt features simply to make the architecture look complete. Start from business requirements, identify the core data flows and processing logic, then select the best-matching feature combination. If only simple data aggregation is needed, zero-code ingestion alone is sufficient. There is no need to introduce the complexity of stream processing and AI analysis.
Avoid over-design. Keep the architecture simple. Use the simplest architecture that meets business requirements. Problems that can be solved with built-in features should not require external components. Logic that can be implemented with a single feature should not combine multiple features. Simple architectures are not only easier to develop and maintain but also less likely to encounter performance bottlenecks.
Use built-in features instead of external components. Built-in features such as data subscription (replacing Kafka) and stream processing (replacing Flink) significantly reduce system complexity. Fewer external components means fewer integration points, lower operational costs, and higher system reliability. When making technology choices, evaluate whether built-in features meet requirements first.
Ensure end-to-end data consistency. In data pipelines involving multiple feature components, pay special attention to data consistency. Use stream processing’s recalculation mechanism to handle out-of-order data, edge-cloud collaboration’s resume-from-breakpoint mechanism to ensure data completeness, and the WATERMARK option to control computation accuracy.
Summary
TDengine TSDB’s advanced features cover ingestion, synchronization, real-time computation, publishing, caching, and AI-assisted analysis. The best architecture is not the one that uses every feature. It is the one that matches the required data flow, keeps operations manageable, and leaves room to scale when business requirements become more demanding.
