TDengine Deployment for Sentinel Flow Control Data at DeWu

DeWu, an e‑commerce platform, needed persistent storage for massive flow‑control metrics generated by Alibaba Sentinel. The data volume can reach billions of records per day, making traditional relational databases unsuitable.

After evaluating several options (InfluxDB, OpenTSDB, Cassandra, TiDB), the team chose the open‑source time‑series database TDengine because of its high write throughput, efficient compression and native support for time‑series data.

TDengine’s data model recommends creating a separate table for each data collection point and grouping them under a super‑table. In this case each application‑resource pair becomes a table, and all tables share a common super‑table, allowing fast aggregation.

The architecture consists of Sentinel‑enabled services sending heartbeat and metric data to a control console, which aggregates the data and writes it in batches to a TDengine cluster via a JDBC connector. The cluster runs three physical nodes (16 CPU/64 GB RAM/1 TB storage each) with a replication factor of 1.

Database connection pooling uses Druid together with MyBatis for query operations; writes are performed with raw SQL for simplicity. The following SQL creates the database with appropriate retention and cache settings:

CREATE DATABASE sentinel KEEP 365 DAYS 1 blocks 16 cache 64;

Performance tests show write latency around 10 ms for batches of 400 rows and query latencies ranging from sub‑millisecond for simple look‑ups to a few seconds for large aggregations over billions of rows, demonstrating TDengine’s suitability for high‑frequency monitoring data.

Overall, TDengine provides low operational overhead, fast read/write performance, and good compression (≈10 % of original size), making it a strong candidate for time‑series storage in large‑scale monitoring scenarios.