Dynamic Maglev Table Sizing Based on Application Instance Count

Proposed Change

As a developer of applications with varying numbers of instances using hash-based routing
I want Maglev lookup table sizes to be calculated dynamically based on the number of application instances per route pool
So that I can ensure better hash distribution for large pools.

The current implementation uses a hard-coded lookup table size of 3001 for the Maglev consistent hashing algorithm across all route pools. According to the Maglev paper recommendations, the optimal table size should be approximately 100× the number of application instances per pool. The size of 3001 may provide insufficient distribution for large pools.

Acceptance criteria

• Maglev lookup table size is dynamically calculated as the nearest prime number to 100 × number_of_instances per route pool
• When instances are added, table size is calculated with a buffer to prevent frequent resizing when instance counts fluctuate slightly; table only resizes when actual instance count exceeds the buffered capacity
• When instances are removed, the table size does not change to prevent frequent remapping of hashes
• Minimum table size is defined
• Maximum table size is defined to prevent excessive memory usage
• Table size changes are logged with appropriate level

Related links

• Implement Hash-Based Routing
• [RFC0042] Hash-Based Load Balancing Algorithm for CF Routing

Start with src/code.cloudfoundry.org/gorouter/route/maglev.go, especially the current hard-coded table size, then review the linked hash-based routing pull request and RFC0042. Define and implement the sizing, buffering, minimum, maximum, removal, and logging behavior so every acceptance criterion is met without unnecessary remapping.