Reported impact
Business outcome
Morefun turned observability into an operating system for reliability engineering. Core-link monitoring coverage rose from 65% to 99.5%, average fault-detection time dropped from 8 minutes to under 1 minute, and root-cause location fell from 15 minutes to 8 minutes. The company also reduced alert noise, automated much of routine incident handling, cut off-hours compute cost and cold-data storage cost, and improved service stability for high-concurrency ticketing peaks.
- Cost Reduction
- Customer Experience
- Productivity Improvement
- Risk Reduction
Solution
Morefun rebuilt its observability stack on Alibaba Cloud as an AIOps platform instead of separate monitoring tools. LoongCollector centralizes application logs into Log Service SLS, ARMS agents trace requests from page and API entry points through backend microservices, databases, caches, and message queues, and Managed Service for Prometheus collects infrastructure and container metrics across the fleet. Cloud Dial Test adds synthetic availability and latency checks from the user perspective, while Grafana surfaces shared dashboards. ARMS alert management normalizes multi-source alarms, routes them by business context, and preserves lifecycle records for SLO follow-up. On top of that observability data base, Morefun added an Observable MCP Server workflow that uses large-model reasoning to cluster related alarms, infer likely root causes, and generate diagnostic reports with remediation suggestions, turning incident response into an AI-assisted closed loop.
Data flow
Application logs are ingested by LoongCollector into SLS, traces are captured by ARMS agents, infrastructure and middleware metrics are written to Managed Service for Prometheus, and Cloud Dial Test adds external availability probes. Those streams are correlated in dashboards and alert pipelines, then Observable MCP Server applies large-model reasoning to cluster alarms, infer root causes, and emit diagnostic reports that guide follow-up actions.
