Haier Cloud Chain Technology is a technology-driven platform serving consumer finance scenarios. To meet growing demand for timely post-loan data, the company upgraded its analytics architecture with StarRocks, simplifying data processing and improving access to financial insights for business teams and external partners.

Solution

Adopt StarRocks as a unified analytical engine to replace the Apache HBase and Apache Phoenix stack for real-time processing and part of the Apache Hive, Alluxio, and Presto stack for offline analytics. Support real-time ingestion, data updates, multi-table joins, and concurrent queries through a simpler architecture.

Results

Haier Cloud Chain Technology reported 3–7× faster query performance and overall server costs reduced to nearly half those of the previous architecture.

In the reported comparison using StarRocks v1.19.0 on CentOS 7.8, 90% of queries completed within 1 second and 99% within 3 seconds, compared with 5 seconds and 10 seconds previously. For Hive external table queries, the time within which 99% completed decreased from 35 seconds to 5 seconds. Direct debit queries improved from approximately 10 seconds to under 3 seconds, with most completing within 1 second.

These results reflect the company’s reported workloads, not general product benchmarks.

Challenge

Post-loan operations generated large volumes of frequently changing financial data. Nightly accounting and monthly posting created ingestion spikes, causing loading delays, high CPU utilization, and degraded query performance.

Different departments and external partners required flexible access to current and historical records. Supporting these requirements through extensive secondary indexes and denormalized tables increased ETL complexity, data redundancy, and governance costs. Maintaining multiple analytics components also drove up infrastructure expenditure and operational overhead.

Approach

After evaluating OLAP engines in mid-2021, the team selected StarRocks based on its data update capabilities and performance in workload-specific join and concurrency tests.

The upgraded platform ingested source data in real time without first combining it into denormalized tables. Multi-table joins supported calculations across loan principal, interest, penalty interest, and repayment records while preserving original business semantics.

The team used the Aggregate table model for aggregation workloads and the Duplicate Key table model for append-only direct debit records. Automatic date partitioning and the removal of redundant fields reduced individual table sizes and simplified data management.

Business impact

Business teams could retrieve current financial data and aggregate reports directly during meetings rather than preparing extracts in advance. Faster queries and more timely data access improved the responsiveness of post-loan reporting.

Reducing denormalization simplified ETL pipelines and data quality traceability, while workload-specific table models addressed ingestion and query bottlenecks. Consolidating services on StarRocks also lowered server costs and simplified deployment, scaling, and monitoring.