This is general knowledge or an external-engine reference, not an Apollo architecture or implementation claim. See Apollo's architecture status for target terminology and verified implementation boundaries.
Q89: How to Design Rate Limiting and Anti-Abuse Mechanisms?
Core Conclusion
The goal of rate limiting and anti-abuse is not just to block "too many requests," but to distinguish:
- Normal high-frequency behavior
- Malicious or abnormal repetitive behavior
- Critical resource protection needs
A truly effective system usually doesn't just apply one token bucket algorithm, but works in layers:
- Access-layer rate limiting
- Session-layer rate limiting
- Business-layer anti-abuse
I. First Distinguish Rate Limiting from Anti-Abuse
The two are related, but not exactly the same.
1. Rate Limiting
Focuses more on:
- Resource protection
- System stability
2. Anti-Abuse
Focuses more on:
- Malicious profit-seeking
- Automated abuse
- Abnormal repeated operations
Some requests have low volume but abnormally high profit, which also falls under anti-abuse.
II. Rate Limiting Granularity Must Be Layered
Common granularities include:
- IP
- Account
- Character
- Session
- Interface or message type
Different layers solve different problems.
For example:
- IP is more about external attack and access protection
- Character and interface granularity is closer to business farming behavior
III. Business-Layer Anti-Abuse Matters More Than the Algorithm
Many high-value vulnerabilities are not triggered by large traffic, but by:
- Repeated clicking
- Automated scripts
- Boundary-state retries
So the business layer usually also needs:
- Idempotency
- Cooldowns
- State machine validation
- Profit caps
These are more critical than a simple "how many times per second."
IV. Different Interfaces Should Use Different Strategies
For example:
- The login interface suits strict access-layer rate limiting
- Chat suits dual frequency and content control
- Rewards and purchases suit idempotency plus frequency limits
- Movement messages cannot be handled with ordinary interface rate limiting thinking
If all requests use the same bucket parameters, neither experience nor security will be good.
V. Post-Rate-Limiting Handling Must Also Be Designed
Common handling approaches include:
- Direct rejection
- Delayed response
- Degraded service
- CAPTCHA or manual review
- Recording into risk-control scores
Not every over-limit case should be dealt with a one-size-fits-all ban.
VI. A More Robust Engineering Combination
Common practice is:
- Coarse rate limiting at the access layer
- Frequency control at the session layer
- Idempotency, state machines, and profit limits at the business layer
- Continuous anomaly statistics at the risk-control layer
Only this way can both the system and business profits be protected.
VII. Common Misconceptions
1. Rate limiting is the token bucket
The token bucket is very common, but it is only a basic tool, not a complete anti-abuse solution.
2. Low request volume doesn't count as abuse
Many abusers exploit exactly the low-frequency, high-profit vulnerabilities.
3. Banning on over-limit is the safest
In many cases, gradual handling and continuous observation are more reasonable.
References
- Rate limiting, idempotency, and risk-control scoring practice materials
