Zum Hauptinhalt springen

Performance Optimization

At scale (many users or many rules), two common bottlenecks are throughput and policy size. Below are practical mitigations.

High request volume​

If one enforcer cannot keep up (e.g. 10k+ req/s):

  1. Multi-threading — Use SyncedEnforcer and run multiple worker threads/instances to use all cores.
  2. Cluster — Run multiple machines with Casbin and use Watchers to sync policy. You can combine both: e.g. 10 nodes × 5 threads each.

Large policy (many rules)​

With millions of rules, enforcement and load time can suffer:

  1. Model and policy design — Prefer RBAC and shared defaults so most tenants use a small rule set (e.g. < 100 rules); override only where needed. See GitHub #681 for discussion.
  2. Policy sharding — Load only the rules each enforcer needs (e.g. per tenant or per domain) via Policy subset loading.
  3. RBAC over direct user permissions — Assign permissions to roles; Casbin caches the role tree for O(1) role lookups and fewer policy rules.

Combine these where possible for best results.