HYPERGRYPH was founded in 2017, based in Shanghai, China, born with a passion for creating games that are challenging and possess aesthetic value. We are passionate creators of a unique world where logic enlightens art, talents delight creativity.

Arknights, the debut work of HYPERGRYPH, is a tower defense mobile game. Its unique worldview and character design have earned widespread acclaim from players and industry professionals, and top-ranked best-selling game lists in many countries and regions.
We continue to explore spaces for creativity and hope to publish more anime mobile games across the globe.

As a live-service game, Arknights runs frequent in-game events with diverse gameplay mechanics, creating high-frequency, tidal data workloads. The resulting data demands are not limited to traditional BI reporting; data is deeply embedded into game mechanics and live operations, requiring a highly stable and performant compute engine.

Hypergryph’s legacy data platform faced several limitations:

● Lack of external catalog support and integration with popular schedulers, such as DolphinScheduler, creates workflow silos.
● Low community compatibility leads to engine stability issues.
● Disk pressure during large shuffle operations due to the absence of a Remote Shuffle Service.
● Inadequate technical support from the previous provider, with slow issue response time and no clear product iteration roadmap.

Hypergryph chose Alibaba Cloud for its EMR Serverless Spark, which offered the exact combination of open-source compatibility, elastic scalability, and enterprise-grade reliability that it required.

● As a fully managed, Spark-compatible Lakehouse product, EMR Serverless Spark provided integration with both Airflow and DolphinScheduler, native Paimon Catalog and Hive MetaStore support, and a built-in Celeborn Remote Shuffle Service to eliminate disk bottlenecks.
● EMR Serverless Spark maintained 100% Apache Spark community compatibility with a high-performance Fusion SQL engine, ensuring both stability and execution speed.
● Alibaba Cloud’s responsive technical support and transparent product roadmap strengthened Hypergryph’s confidence in a long-term partnership.

Hypergryph built a modern, layered data architecture with EMR Serverless Spark as the core offline compute engine. Data is ingested through a custom instrumentation SDK for log collection and Flink CDC for real-time database synchronization, ensuring freshness and accuracy as the basis for downstream analytics.

For offline scheduling, Hypergryph adopted a dual orchestration model: Airflow for code-savvy engineers and DolphinScheduler for analysts, both seamlessly connected to EMR Serverless Spark as the unified batch compute engine. Job artifacts are developed in local Git repositories, deployed to OSS via CI pipelines, and submitted as SQL files or JAR jobs through DolphinScheduler’s native ALIYUN_SERVERLESS_SPARK job type.

The built-in Celeborn shuffle service resolved disk limitations in large shuffle tasks, and the status remained consistent with the scheduling system, therefore requiring no secondary confirmation.

For online serving, EMR Serverless StarRocks powers real-time analytics, with metrics visualized through smart BI dashboards and an integrated business analytics platform. Spark Thrift Server enables JDBC-based ad-hoc queries with dynamic resource allocation, supporting both operations teams and data warehouse pipelines.

Post-migration, Hypergryph achieved measurable results:

● 50% reduction in key metric computation time. For example, user-wide table processing dropped from 30 minutes to 15 minutes.
● Core SLA pipeline delivery was accelerated by 1.5 hours, significantly strengthening data availability guarantees.
● Enhanced development efficiency through streamlined Spark SQL session and DolphinScheduler production scheduling workflow, successfully ensuring data delivery for multiple critical in-game events.
● Significant reduction in operational overhead through fully managed, elastic clusters with multi-version management and rapid upgrade capabilities.

As the partnership grows, Alibaba Cloud will continue supporting Hypergryph’s rapid expansion. Hypergryph has already begun building a similar data architecture on Alibaba Cloud for its overseas game operations, extending the same efficiency and reliability to a global audience.

Looking ahead, Hypergryph expects EMR Serverless Spark to continue evolving its Lakehouse ecosystem with an open-first approach, including unified catalog management and broader coverage of exploratory analytics scenarios. Together with Alibaba Cloud, Hypergryph will continue to forge the digital future of global game operations.