Automated Resource Balancing and Migration for Redis Clusters

The DeWu Redis management platform oversees hundreds of clusters and thousands of Redis‑server nodes, keeping host memory usage at a high yet safe level; it automatically inspects hosts daily and initiates migration for any host exceeding a predefined threshold.

Manual migration proved labor‑intensive and error‑prone, prompting the development of an unattended, automated resource‑balancing scheduler that runs at low‑traffic hours (typically 5 am) to minimize business impact.

Node selection follows a prioritized algorithm: first choose instances with many nodes in the same cluster, then non‑P0 instances, preferring slave nodes, selecting medium‑sized instances (1‑4 GB, then 1‑5 GB, then others), while avoiding selecting nodes from the same cluster group in a single round; the process iterates through three rounds before falling back to any remaining nodes.

Reliability is ensured through a strict multi‑step workflow. The first step adds a slave node; after addition, the system checks replication status using info replication and verifies that state=online , offset != 0 , master_link_status:up and master_repl_offset != 0 . If the original node is a master, a promotion is performed with slaveof no one followed by updating other replicas via slaveof . The new master and all slaves are re‑checked for the same replication criteria. Once verified, the original node is safely taken offline after a second round of checks, ensuring no active client connections or proxy dependencies remain.

After a successful migration, a notification is sent to the responsible DBA. Migration tasks are displayed in a management UI where users can cancel, reschedule, or execute immediately, and view detailed node lists and historical records.

Overall, the automated system maintains a high average memory utilization across hosts while preserving sufficient headroom, reduces DBA workload, supports rapid vertical scaling, and lays the groundwork for further automation such as automatic cluster deployment, vertical expansion, and host‑failure recovery.