Connection pooling in HLD - Scalability & System Analysis

Scalability Analysis - Connection pooling

Growth Table: Connection Pooling at Different Scales

Users	Connections Opened	DB Load	Latency Impact	Resource Usage
100 users	~10-50 pooled connections	Low	Minimal	Low CPU & Memory
10,000 users	~200-500 pooled connections	Moderate	Noticeable if no pooling	Moderate CPU & Memory
1,000,000 users	~1000-2000 pooled connections	High	High without pooling	High CPU & Memory, DB saturation risk
100,000,000 users	~5000+ pooled connections (sharded)	Very High	Severe without pooling and sharding	Very High, requires distributed pooling

First Bottleneck

The database server is the first bottleneck. Opening many direct connections for each user request overloads the DB CPU and memory. Without connection pooling, the DB spends too much time managing connections instead of queries.

Scaling Solutions

• Connection Pooling: Reuse a limited number of DB connections for many requests to reduce overhead.
• Horizontal Scaling: Add more app servers each with its own pool to distribute load.
• Read Replicas: Use read-only DB replicas to spread read queries.
• Sharding: Split data across multiple DB instances to reduce single DB load.
• Caching: Cache frequent queries to reduce DB hits.
• Timeouts and Limits: Set max pool size and connection timeouts to avoid resource exhaustion.

Back-of-Envelope Cost Analysis

• 1 server can handle ~1000-5000 concurrent DB connections.
• DB can handle ~5000-10000 queries per second (QPS) depending on hardware.
• Without pooling, each user request may open a new connection causing overhead and latency.
• Pooling reduces connection overhead by reusing connections, improving throughput and reducing CPU usage.
• Bandwidth impact is minimal for connections but important for query result size.

Interview Tip

Start by explaining what connection pooling is and why it matters. Then describe how the DB becomes a bottleneck as users grow. Discuss how pooling reduces connection overhead. Finally, mention complementary scaling techniques like read replicas and sharding.

Self Check

Your database handles 1000 QPS. Traffic grows 10x to 10,000 QPS. What do you do first?

Answer: Implement or increase connection pooling to reuse DB connections efficiently and reduce overhead. Then consider adding read replicas or sharding to distribute load.

Key Result

Connection pooling prevents database overload by reusing a limited number of connections, enabling the system to scale from hundreds to millions of users efficiently.

Practice

1. What is the main purpose of connection pooling in system design?

easy

A. To reuse existing connections and reduce the overhead of creating new ones

B. To create a new connection for every request to ensure freshness

C. To store user data temporarily for faster access

D. To encrypt all data sent between client and server

Solution

1. Step 1: Understand connection creation cost

   Creating new connections for every request is slow and resource-heavy.

2. Step 2: Role of connection pooling

   Connection pooling keeps connections open and reuses them to save time and resources.

3. Final Answer:

   To reuse existing connections and reduce the overhead of creating new ones -> Option A

4. Quick Check:

   Connection pooling = reuse connections [OK]

Hint: Pooling means reuse connections to save time [OK]

Common Mistakes:

• Thinking pooling creates new connections each time
• Confusing pooling with caching user data
• Assuming pooling encrypts data

2. Which of the following is the correct way to configure a connection pool size in a system design?

easy

A. Set pool size to the maximum integer value to allow unlimited connections

B. Set pool size to zero to disable pooling

C. Set pool size to a fixed number based on expected load, e.g., 50

D. Set pool size to negative number to auto-scale connections

Solution

1. Step 1: Understand pool size meaning

   Pool size controls how many connections can be reused simultaneously.

2. Step 2: Choose appropriate pool size

   Setting a fixed number based on expected load balances resource use and performance.

3. Final Answer:

   Set pool size to a fixed number based on expected load, e.g., 50 -> Option C

4. Quick Check:

   Pool size = fixed number for load [OK]

Hint: Pool size should match expected load, not zero or negative [OK]

Common Mistakes:

• Setting pool size to zero disables pooling
• Using negative or unlimited values causes errors
• Ignoring load when setting pool size

3. Consider this pseudocode for connection pooling:

pool = createPool(size=3)
conn1 = pool.getConnection()
conn2 = pool.getConnection()
conn3 = pool.getConnection()
conn4 = pool.getConnection()

What happens when conn4 is requested?

medium

A. conn4 closes one existing connection and reuses it

B. conn4 creates a new connection ignoring the pool size

C. conn4 returns null immediately because pool is full

D. conn4 waits until one connection is released back to the pool

Solution

1. Step 1: Analyze pool size and connections taken

   The pool size is 3, and 3 connections are already taken (conn1, conn2, conn3).

2. Step 2: Behavior on extra connection request

   Requesting conn4 must wait until a connection is released because pool is full.

3. Final Answer:

   conn4 waits until one connection is released back to the pool -> Option D

4. Quick Check:

   Pool full = wait for release [OK]

Hint: Extra requests wait if pool is full [OK]

Common Mistakes:

• Assuming new connections are created beyond pool size
• Expecting null return immediately
• Thinking pool forcibly closes connections

4. A system uses connection pooling but experiences slow responses under high load. Which of the following is the most likely cause?

medium

A. Connections are not reused and closed immediately

B. Pool size is too small, causing requests to wait for connections

C. Pool size is too large, causing too many open connections

D. Timeout settings are too high, causing connections to stay open too long

Solution

1. Step 1: Identify impact of small pool size

   A small pool size means limited connections, so requests wait longer.

2. Step 2: Understand effect on response time

   Waiting for connections causes slow responses under high load.

3. Final Answer:

   Pool size is too small, causing requests to wait for connections -> Option B

4. Quick Check:

   Small pool = slow responses [OK]

Hint: Small pool size causes waiting delays [OK]

Common Mistakes:

• Blaming large pool size for slowness
• Ignoring connection reuse behavior
• Confusing timeout with pool size

5. You design a connection pool for a web service expected to handle 1000 concurrent users. Which combination of settings best balances performance and resource use?

hard

A. Pool size = 100, Timeout = 30 seconds, Max wait time = 10 seconds

B. Pool size = 1000, Timeout = 5 minutes, No max wait time

C. Pool size = 10, Timeout = 1 hour, Max wait time = 1 second

D. Pool size = 5000, Timeout = 10 seconds, No max wait time

Solution

1. Step 1: Evaluate pool size for concurrency

   Pool size of 100 allows many connections but avoids resource overload compared to 1000 or 5000.

2. Step 2: Assess timeout and wait time

   Timeout of 30 seconds frees unused connections quickly; max wait time of 10 seconds prevents long waits.

3. Step 3: Compare other options

   Too large pool sizes waste resources; too small pool or short wait times cause failures.

4. Final Answer:

   Pool size = 100, Timeout = 30 seconds, Max wait time = 10 seconds -> Option A

5. Quick Check:

   Balanced pool size and timeouts = Pool size = 100, Timeout = 30 seconds, Max wait time = 10 seconds [OK]

Hint: Balance pool size and timeouts for best performance [OK]

Common Mistakes:

• Choosing pool size equal to max users wastes resources
• Ignoring max wait time causes request failures
• Setting too long timeout holds connections unnecessarily