Table of Contents

1. Introduction
   • What is a Rate Limiter?
   • Advantages of an API Rate Limiter
   • Understanding the Problem and Requirements
     • Key Considerations
     • Step 1: Requirements
     • Functional and Non-Functional Requirements
     • Step 2: High Level Design (HLD)
     • Placement of the Rate Limiter
     • Example Scenario
     • API Gateway
     • Implementation Considerations
     • High Level Architecture
     • Storage
     • Assumptions and Requirements
     • Storage Requirements
     • Overall Storage required when we handle large traffic
     • Request Flow
     • Step 3: Design Deep Dive
     • Rate Limiting Rules
     • Handling Exceeded Limits
     • Rate Limiter Headers
     • Detailed Design
2. Distributed Environment Challenges
   • Heavy/Big Rate Limit Service Server
     • Race Condition
     • Synchronization
     • Synchronizing User Request Counts Across Hosts
     • Push vs Pull Synchronization
     • Traffic Considerations for Hosts
   • Synchronization Algorithms
     • All to All Communication
     • Gossip Protocol
     • External Storage or Coordination Service
     • Random Leader Selection
3. Performance Optimization
   • Multi-Data Center Setup
   • Eventual Consistency
4. Rate Limiting Algorithms
   • Token Bucket
   • Leaking Bucket
   • Fixed Window Counter
   • Sliding Window Log
   • Sliding Window Counter
5. Rate Limiter Using Sorted Set in Redis
   • How Sorted Sets in Redis Work for Rate Limiting
   • Key Concepts
   • Steps to Implement Rate Limiting Using Sorted Sets
   • Example
     • Step by Step Implementation
   • Handling Concurrency
   • Benefits of Using Sorted Sets in Redis
6. Conclusion
7. FAQ'S

Feel free to modify or expand any section as needed!

Introduction

What is a Rate Limiter?

In a network system, a rate limiter controls the rate of traffic sent by a client or service. In the context of HTTP, a rate limiter restricts the number of client requests that can be sent over a specified period. If the number of API requests exceeds the threshold defined by the rate limiter, the excess calls are blocked. Here are some examples:

• A user can write no more than 2 posts per second.
• A maximum of 10 accounts can be created per day from the same IP address.
• Rewards can be claimed no more than 5 times per week from the same device.

Advantages of an API Rate Limiter

Prevent Resource Starvation

Rate limiters help prevent resource starvation caused by Denial of Service (DoS) attacks. Many large tech companies enforce some form of rate limiting. For example, Twitter limits the number of tweets to 300 per 3 hours, and Google Docs APIs have a default limit of 300 requests per user per 60 seconds for read requests. By blocking excess calls, a rate limiter can prevent DoS attacks, whether intentional or unintentional.

Reduce Costs

Limiting excess requests reduces the number of servers needed and allocates more resources to high-priority APIs. This is crucial for companies using paid third-party APIs, where each call incurs a cost. For instance, checking credit, making payments, or retrieving health records often involves per-call charges, making it essential to limit the number of calls to reduce costs.

Prevent Server Overload

To reduce server load, a rate limiter filters out excess requests caused by bots or user misbehavior, ensuring that the servers are not overwhelmed.

Understanding the Problem and Requirements

Key Considerations

• Type of Rate Limiter: We will focus on a server-side API rate limiter.
• Throttling Criteria: The rate limiter should be flexible enough to support different sets of throttle rules, such as IP address or user ID.
• System Scale: The system must handle a large number of requests.
• Distributed Environment: The rate limiter should work in a distributed environment.
• Implementation: It can be a separate service or integrated into the application code.
• User Notifications: Users should be informed when their requests are throttled.

Step 1: Requirements

Functional and Non-Functional Requirements

Functional requirements to Design a Rate Limiter API:

• The API should allow the definition of multiple rate-limiting rules.
• The API should provide the ability to customize the response to clients when rate limits are exceeded.
• The API should allow for the storage and retrieval of rate-limit data.
• The API should be implemented with proper error handling as in when the threshold limit of requests are crossed for a single server or across different combinations, the client should get a proper error message.

Non-functional requirements to Design a Rate Limiter API:

• The API should be highly available and scalable. Availability is the main pillar in the case of request fetching APIs.
• The API should be secure and protected against malicious attacks.
• The API should be easy to integrate with existing systems.
• There should be low latency provided by the rate limiter to the system, as performance is one of the key factors in the case of any system.

Step 2: High Level Design (HLD)

Placement of the Rate Limiter

Instead of implementing the rate limiter on the client or server side, we can create a rate limiter middleware that throttles requests to APIs.

Example Scenario

Assume our API allows 2 requests per second. If a client sends 3 requests within a second, the first two requests are routed to API servers, but the third request is throttled by the middleware and returns an HTTP status code 429, indicating too many requests.

HTTP 429 "Too Many Requests" — The user has sent too many requests in a certain amount of time. This response status code tells the user their request has been rate limited.

API Gateway

In a microservices architecture, rate limiting is often implemented within a component called an API gateway. This is a fully managed service that supports rate limiting, SSL termination, authentication, IP whitelisting, and servicing static content.

Implementation Considerations

• Technology Stack: Ensure the current technology stack, including the programming language and cache service, supports efficient rate limiting.
• Rate Limiting Algorithm: Choose an algorithm that fits business needs. Server-side implementation provides full control, but third-party gateways might have limitations.
• Microservice Architecture: If using an API gateway for authentication and other services, adding a rate limiter there can be beneficial.
• Engineering Resources: Building a rate-limiting service requires time and resources. If limited, consider using a commercial API gateway.

High Level Architecture

Rate limiting algorithms are simple at a high level. A counter keeps track of the number of requests from the same user or IP address. If the counter exceeds the limit, the request is disallowed.

Storage

Using a database is not ideal due to the slowness of disk access. In-memory cache (e.g., Redis) is preferred for its speed and support for time-based expiration. Redis offers commands like INCR (increment the counter) and EXPIRE (set a timeout for the counter).

To estimate the capacity requirements for our rate limiter service, consider the following:

Assumptions and Requirements:

• Scale: The service must handle billions of daily requests.
• User Base: There are one billion users, each potentially generating up to 100 requests at any given moment.
• Data to Store:
  • Each user ID is 64 bits (8 bytes).
  • Each timestamp in the queue is also 64 bits (8 bytes).
  • The rate limiter needs to track these timestamps for up to 100 services (100 timestamps per user per service).
• Retention Period: The rate limiter only needs to store data for a short period, typically 10 seconds, as decisions are based on recent activity.

Storage Requirements:

• Data Per User:

  • For each user, we store 100 timestamps.
  • Storage per user per service: (100 \times 8) bytes = 800 bytes.
  • For 100 services: (100 \times 800) bytes = 80,000 bytes (80 KB).

• Active Users:

  • Assume that at any given moment, we have a maximum of 10 million active users (a conservative estimate).
  • Storage for 10 million users: (10^7 \times 80) KB = 800,000,000 KB = 800 GB.

• Total Storage Requirement:

  • Primary Storage: 800 GB to handle 10 million users within a 10-second window.
  • Overhead: Additional storage may be required due to the delay in removing stale data. The exact amount depends on how efficiently the system purges old data, but the worst-case scenario is slightly above 800 GB.

Overall Storage required when we handle large traffic:

To scale the rate limiter for billions of daily requests, with a retention period of 10 seconds and an assumption of 10 million concurrent active users, the estimated storage requirement is approximately 808 GB. This storage size assumes efficient data management, like using a database such as Redis, where each user entry (with 100 timestamps) consumes 800 bytes. The actual storage might vary based on the system's ability to promptly remove outdated data.

Request Flow

1. The client sends a request to the rate-limiting middleware.
2. The middleware fetches the counter from Redis and checks if the limit is reached.
3. If the limit is reached, the request is rejected.
4. If the limit is not reached, the request is sent to API servers, the counter is incremented, and the updated value is saved back to Redis.

Step 3: Design Deep Dive

Rate Limiting Rules

Rate limiting rules define the conditions under which requests are throttled. These rules are generally written in configuration files and stored on disk. For example, allowing a maximum of 5 marketing messages per day or restricting logins to 5 times per minute.

Putting these requirements together, we need the following:

A database with fast reads and writes for counts. The schema will be simple; it is unlikely to be much more complex than (user ID, service ID). We can use an in-memory database like Redis.

A service where rules can be defined and retrieved, which we call the Rules service.

A service that makes requests to the Rules service and the Redis database, which we can call the Backend service.

The two services are separate because requests to the Rules service for adding or modifying rules should not interfere with requests to the rate limiter that determine if a request should be rate limited.

Handling Exceeded Limits

When a request exceeds the rate limit, the API returns an HTTP response code 429. Depending on the use case, rate-limited requests may be enqueued for later processing.

Rate Limiter Headers

Clients can be informed about rate limits via HTTP response headers:

• X-Ratelimit-Remaining: The remaining number of allowed requests.
• X-Ratelimit-Limit: The maximum number of requests allowed per time window.
• X-Ratelimit-Retry-After: The number of seconds to wait before making another request.

Detailed Design

1. Rules are stored on the disk and frequently pulled by workers into the cache.
2. A client request goes to the rate limiter middleware.
3. The middleware loads rules from the cache and fetches counters from Redis.
4. Depending on the counter value:
   • If the request is not rate limited, it is forwarded to API servers.
   • If the request is rate limited, an HTTP 429 error is returned, and the request may be dropped or queued.

Distributed Environment Challenges

Heavy/Big Rate Limit Service Server

As discussed above in terms of storage requirements to handle large traffic we have to scale up our server & we ended up having a Rate Limit server with a huge storage requirement.
How can we reduce the storage requirement of Rate Limit service server? We can reduce our 808 GB storage requirement to 8.08 GB ≈ 8 GB by creating a new instance of our rate-limiting service for each of the ~100 services that uses it and use the frontend to route requests by service to the appropriate service. 8 GB can fit into a host’s memory. Due to our high request rate, we cannot use a single host for rate limiting. If we use 128 hosts, each host will store only 64 MB. The final number we decide upon will likely be between 1 and 128. We will split this rate limit server & based on that we will use Level 4 load balancer to route request to these rate limit server. we can achieve a scalable, efficient, and memory-friendly solution. We will discuss above sync between these server below.

Race Condition

Race conditions occur when multiple threads read and write shared data concurrently. For example, two requests reading the counter value before either writes it back can lead to an incorrect count. Solutions include using locks or Redis features like Lua scripts and sorted sets.

Synchronization

In a distributed environment, synchronization ensures that multiple rate limiter servers have consistent data. Sticky sessions or centralized data stores like Redis can help maintain synchronization.

Synchronizing User Request Counts Across Hosts

When designing a rate limiter in a distributed system, it’s crucial to ensure that all hosts have synchronized user request counts. Below, we explore several algorithms that can be used for this synchronization. It’s important to note that not all algorithms are practical for this purpose. For instance, the all-to-all communication approach is generally not suitable for our rate limiter due to its scalability issues.

Push vs. Pull Synchronization

One of the first decisions to make is whether the synchronization mechanism should be push-based or pull-based. A push-based approach is often preferable as it provides a balance between performance, resource efficiency, and system simplicity. In a push system, hosts proactively send their data to others, trading off some consistency and accuracy for speed and reduced overhead.

If a host fails, the system can simply ignore its contribution to the request counts, allowing users to make additional requests before being rate-limited. Given these trade-offs, we can conclude that the hosts should asynchronously share their timestamps using a lightweight protocol like UDP rather than TCP.

Traffic Considerations for Hosts

When implementing synchronization, we must account for two primary types of requests that hosts need to handle:

1. Rate Limiting Decisions: These are requests from users that need a response on whether they are allowed to proceed based on their request rate. The load balancer manages these requests, and we can provision additional hosts as necessary to handle increased traffic.

2. Synchronization of Timestamps: These are requests for updating and synchronizing timestamps across hosts. The synchronization mechanism should be designed to prevent any single host from being overwhelmed by a high volume of these requests, especially as the cluster size increases.

Synchronization Algorithms

All to All Communication

In an all-to-all communication model, every host in the system sends messages to every other host. This model is more general than broadcasting, where a message is sent to all recipients simultaneously. In a full mesh topology, as illustrated in figure 8.8, every node is connected to every other node.

While this approach ensures that all hosts are fully synchronized, it scales poorly. The communication overhead grows quadratically with the number of hosts. For example, in a system with 128 hosts, the all-to-all communication would require an enormous amount of data transfer, making it impractical for large-scale systems.

Gossip Protocol

The gossip protocol offers a more scalable alternative. In this model, each host periodically selects another host at random and exchanges information. Over time, this random exchange ensures that all hosts have a consistent view of the system’s state.

Yahoo’s distributed rate limiter, for instance, employs a gossip protocol to synchronize request counts across its hosts. This method also trades off strict consistency for better performance and lower resource consumption. However, it introduces additional complexity in managing the protocol.

Both the all-to-all and gossip protocol require that every host knows the IP addresses of all other hosts. As nodes are added or removed from the cluster, each host must query a configuration service (like ZooKeeper) to update its knowledge of the cluster's current state.

External Storage or Coordination Service

Another approach is to use an external component, such as a storage or coordination service, to facilitate synchronization. As depicted in figure 8.10, this setup allows hosts to communicate indirectly through a centralized service.

A common implementation involves selecting a leader host, designated by a configuration service such as ZooKeeper. All other hosts only need to communicate with this leader, which simplifies the system. The leader host periodically updates its list of active hosts and manages the synchronization process.

Random Leader Selection

A simpler alternative for leader election is random leader selection. This approach, shown in figure 8.11, can lead to the election of multiple leaders, which increases messaging overhead. Despite this inefficiency, the system remains functional, as all hosts eventually receive the necessary updates. This method is a trade-off between resource consumption and simplicity, avoiding the complexity of more sophisticated leader election algorithms.

Overall, choosing the right synchronization mechanism depends on the specific requirements and constraints of the rate-limiting system, such as the desired trade-offs between consistency, performance, and resource usage.

Performance Optimization

Multi-Data Center Setup

To reduce latency for users far from the data center, use edge servers geographically distributed around the world. Cloud service providers like Cloudflare have many edge servers to route traffic to the nearest one.

Eventual Consistency

Synchronize data with an eventual consistency model to handle large-scale systems efficiently.

Rate limiting is a technique used to control the rate at which requests are allowed in a system. Let's dive into the details of each algorithm with examples:

Token Bucket

Description:

• A token bucket algorithm has a bucket that holds a certain number of tokens.
• Tokens are added to the bucket at a fixed rate.
• Each incoming request consumes a token.
• If there are no tokens available, the request is denied.
• This allows bursts of requests up to the bucket size while maintaining a steady request rate over time.

Example:

• Imagine a bucket that can hold 10 tokens, and 1 token is added every second.
• If the system receives 10 requests at the beginning, all will be served.
• After the bucket is empty, new tokens will be added at a rate of 1 per second.
• If a request comes in when there are no tokens, it will be denied until a new token is added.

Pros:

• Allows bursts of traffic.
• Simple to implement.

Cons:

• Potential for bursts to overload the system if the bucket size is too large.

Leaking Bucket

Description:

• This algorithm uses a fixed-size bucket where requests are added.
• Requests leak out of the bucket at a fixed rate.
• If the bucket is full, incoming requests are denied.
• Ensures a steady flow of requests, preventing sudden bursts.

Example:

• A bucket with a capacity of 10 requests leaks requests at a rate of 1 per second.
• If 10 requests arrive immediately, they fill the bucket.
• The system processes these requests at a rate of 1 per second.
• Additional incoming requests are denied until space is available.

Pros:

• Smoothens traffic flow.
• Prevents system overload from bursts.

Cons:

• May deny legitimate burst traffic.

Fixed Window Counter

Description:

• Divides time into fixed windows (e.g., 1 minute).
• Counts the number of requests within each window.
• If the count exceeds a predefined limit, requests are denied until the next window.

Example:

• A rate limit of 100 requests per minute.
• If 100 requests arrive in the first minute, further requests are denied until the start of the next minute.
• At the beginning of each minute, the counter resets.

Pros:

• Simple and easy to implement.
• Effective for evenly distributed traffic.

Cons:

• Boundary problem: spikes at window boundaries can exceed the rate limit.

Sliding Window Log

Description:

• Tracks request timestamps in a log.
• Counts requests within a sliding time window (e.g., 1 minute).
• Provides a more accurate count compared to fixed windows.

Example:

• A limit of 100 requests per minute.
• Requests and their timestamps are logged.
• For each incoming request, the system checks the number of requests in the last 60 seconds.
• If the count is within the limit, the request is allowed; otherwise, it is denied.

Pros:

• Smooth rate limiting without boundary issues.
• More accurate than fixed windows.

Cons:

• Requires more memory to store request logs.
• Potentially higher computational cost.

Sliding Window Counter

Description:

• Combines fixed window counter and sliding window log.
• Divides time into small sub-windows and maintains a count of requests in each sub-window.
• Computes the total count in the current window by summing sub-window counts.

Example:

• A limit of 100 requests per minute with sub-windows of 10 seconds.
• Requests are counted in each 10-second sub-window.
• For each incoming request, the system sums the counts of the last 6 sub-windows.
• If the total is within the limit, the request is allowed; otherwise, it is denied.

Pros:

• Balances accuracy and simplicity.
• Avoids boundary issues of fixed windows.

Cons:

• Requires careful configuration of sub-window size.
• Slightly more complex than fixed window counters.

Each algorithm has its own use cases and trade-offs, and the choice depends on the specific requirements of the system being designed.

Rate Limiter Using Sorted Set in Redis

Certainly! Implementing a rate limiter using a Sorted Set in Redis is a robust method, especially in a distributed network. Here’s an in-depth explanation of how it works, along with an example:

How Sorted Sets in Redis Work for Rate Limiting

A Sorted Set in Redis is a collection of unique elements, each associated with a score. The elements are sorted by their scores, which can be any floating-point value. This sorting allows for efficient range queries, which is crucial for implementing a rate limiter.

Key Concepts

1. Elements: Represent the client’s requests.
2. Scores: Represent the timestamps of the requests.

Steps to Implement Rate Limiting Using Sorted Sets

1. Initialize the Sorted Set: For each client (identified by a unique key like IP address or user ID), maintain a Sorted Set in Redis where the elements are request identifiers (could be a unique request ID or just a count) and the scores are the timestamps of the requests.

2. Adding a Request: When a client makes a request, add an entry to the Sorted Set with the current timestamp as the score.

3. Trimming Old Requests: Periodically or on each request, remove entries from the Sorted Set that fall outside the allowed time window.

4. Counting Requests: Count the number of elements in the Sorted Set to check if it exceeds the allowed limit.

5. Enforcing Limits: If the count exceeds the limit, deny the request. If not, process the request and add the new entry to the set.

Example

Let's say we want to limit a user to 100 requests per hour.

Step by Step Implementation

const Redis = require('ioredis');
const redis = new Redis();

async function handleRequest(clientId, limit, window) {
  const currentTimestamp = Math.floor(Date.now() / 1000);
  const windowStart = currentTimestamp - window;

  // Define the key for the sorted set
  const key = `rate:${clientId}`;

  // Start a transaction (multi)
  const pipeline = redis.multi();

  // Add the current request
  pipeline.zadd(key, currentTimestamp, currentTimestamp);

  // Remove requests older than the window
  pipeline.zremrangebyscore(key, 0, windowStart);

  // Get the count of requests in the current window
  pipeline.zcount(key, windowStart, currentTimestamp);

  // Execute the pipeline
  const results = await pipeline.exec();

  // results[2][1] contains the count of requests in the current window
  const requestCount = results[2][1];

  if (requestCount > limit) {
    return false;  // Rate limit exceeded
  } else {
    return true;   // Request allowed
  }
}

// Example usage
const clientId = "user123";
const limit = 100;  // Maximum 100 requests
const window = 3600;  // Time window of 1 hour

handleRequest(clientId, limit, window).then(allowed => {
  if (allowed) {
    console.log("Request allowed");
  } else {
    console.log("Rate limit exceeded");
  }
}).catch(error => {
  console.error("Error handling request:", error);
});

Handling Concurrency

In a distributed environment, multiple servers might handle requests for the same client concurrently. Redis handles the atomicity of commands, ensuring that even with concurrent access, the operations (like adding a request or trimming old requests) are performed safely.

Benefits of Using Sorted Sets in Redis

• Efficiency: Sorted sets in Redis are optimized for quick insertion, deletion, and range queries, which are essential for rate limiting.
• Atomic Operations: Redis supports atomic operations, ensuring that concurrent modifications are handled safely.
• Scalability: Redis is designed to handle high-throughput scenarios, making it suitable for distributed environments.

By leveraging Redis's sorted sets and atomic operations, you can build a robust and scalable rate limiter that effectively controls the rate of client requests in a distributed network.

Conclusion

Designing an effective rate limiter involves understanding requirements, choosing the right algorithms, and considering performance and synchronization challenges in a distributed environment. Using tools like Redis can simplify implementation and ensure high performance.

FAQ'S

Q: How would you handle distributed rate limiting?
A: By using a common read/write cache (e.g., Redis) for consistency or a different read/write cache setup to improve read performance and handle scalability with coordination for writes.

Q: What is the difference between a stateful and stateless rate limiter?
A: A stateful rate limiter maintains state (counters, timestamps) across requests, ensuring accurate limits. A stateless rate limiter does not maintain state, often using algorithms or cryptographic techniques to enforce limits without storing data between requests.

Q: How will you make the rate limiter distributed in system design?
A: Implement the rate limiter using a distributed caching system like Redis or Memcached to store counters and timestamps. Ensure all nodes in the system interact with the same distributed cache to maintain consistency.

Q: What challenges do you foresee if a distributed rate limiter is implemented?
A: Challenges include ensuring data consistency across nodes, handling cache synchronization, managing race conditions, dealing with network partitions, and preventing single points of failure in the distributed cache.

Q: How would you handle failure when the rate limiter service goes down?
A: Implement fallback mechanisms such as default allowing or denying requests, using secondary caches, or employing circuit breakers to handle temporary failures and prevent cascading issues.

Q: How can Redis be used in rate limiting?
A: Redis can be used to store rate limit counters and timestamps with operations like INCR, EXPIRE, and LUA scripts to perform atomic updates and checks, ensuring accurate rate limiting across distributed systems.

Q: How can you inform a user or service about their rate limit status?
A: By including rate limit status in the HTTP response headers (e.g., X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset) or through custom API responses detailing the current rate limit status.

Q: What are the issues with using sticky sessions in rate limiting?
A: Sticky sessions can lead to uneven load distribution, session affinity issues, and complications in scaling out as the rate limiting state is tied to a specific server, reducing the benefits of a distributed system.

Q: Why is an IP address preferred over any other parameter to limit users?
A: IP addresses are readily available in network requests and provide a straightforward way to identify and limit clients. However, they can be spoofed or shared by multiple users, leading to potential inaccuracies.

Q: Why is it preferred to implement rate limiting on the server side rather than the client side?
A: Server-side implementation ensures control over enforcement, prevents clients from bypassing limits, and centralizes management and monitoring, providing a more secure and reliable rate limiting mechanism.

Q) Why don’t we scale out the service by monitoring the load and adding more hosts when needed, instead of doing rate limiting?

We can design our service to be horizontally scalable, so it will be straightforward to add more hosts to serve the additional load. We can consider auto-scaling.

The process of adding new hosts when a traffic spike is detected may be too slow. Adding a new host involves steps that take time, like provisioning the host hardware, downloading the necessary Docker containers, starting up the service on the new host, then updating our load balancer configuration to direct traffic to the new host. This process may be too slow, and our service may already have crashed by the time the new hosts are ready to serve traffic. Even auto-scaling solutions may be too slow.

Q) A load balancer can limit the number of requests sent to each host. Why don’t we use our load balancer to ensure that hosts are not overloaded and drop requests when our cluster has no more spare capacity?

We should not serve malicious requests, as we mentioned previously. Rate limiting guards against such requests by detecting their IP addresses and simply dropping them. As discussed later, our rate limiter can usually return 429 Too Many Requests, but if we are sure that certain requests are malicious, we can choose either of these options:

Drop the request and not return any response and allow the attacker to think that our service is experiencing an outage.

Shadow ban the user by returning 200 with an empty or misleading response.

Q) Why does rate limiting need to be a separate service? Can each host independently track the request rate from its requestors, and rate limit them?

The reason is that certain requests are more expensive than others. Certain users may make requests that return more data, require more expensive filtering and aggregation, or involve JOIN operations between larger datasets. A host may become slow from processing expensive requests from particular clients.

Note: A level 4 load balancer cannot process a request’s contents. We will need a level 7 load balancer for sticky sessions (to route requests from a user to the same host), which introduces cost and complexity.
If we do not have other use cases for a level 7 load balancer, it may not be worth it to use level 7 load balancer just for this purpose, and a dedicated and shared rate limiting service may be a better solution.

More Details:

Get all articles related to system design
Hastag: SystemDesignWithZeeshanAli

systemdesignwithzeeshanali

Git: https://github.com/ZeeshanAli-0704/SystemDesignWithZeeshanAli