Two-phase commit (and why to avoid it) in Microservices - Architecture Diagram

System Overview - Two-phase commit (and why to avoid it)

This system demonstrates a two-phase commit protocol coordinating a transaction across multiple microservices to ensure data consistency. The key requirement is atomicity: either all services commit the transaction or none do, preventing partial updates.

However, two-phase commit can cause delays and blocking issues, making it less ideal for scalable microservices architectures.

Architecture Diagram

User
  |
  v
Load Balancer
  |
  v
Transaction Coordinator (2PC)
  /           \
 /             \
Service A     Service B
  |             |
Database A    Database B
  ^             ^
  |             |
 Cache A      Cache B

Components

User

user

Initiates transaction requests

Load Balancer

load_balancer

Distributes user requests to coordinator

Transaction Coordinator (2PC)

service

Manages two-phase commit protocol across services

Service A

service

Handles part of the transaction and manages Database A

Service B

service

Handles part of the transaction and manages Database B

Database A

database

Stores data for Service A

Database B

database

Stores data for Service B

Cache A

cache

Speeds up reads for Database A

Cache B

cache

Speeds up reads for Database B

Request Flow - 20 Hops

User → Load Balancer

Load Balancer → Transaction Coordinator (2PC)

Transaction Coordinator (2PC) → Service A

Service A → Database A

Database A → Service A

Service A → Transaction Coordinator (2PC)

Transaction Coordinator (2PC) → Service B

Service B → Database B

Database B → Service B

Service B → Transaction Coordinator (2PC)

Transaction Coordinator (2PC) → Service A

Service A → Database A

Database A → Service A

Service A → Transaction Coordinator (2PC)

Transaction Coordinator (2PC) → Service B

Service B → Database B

Database B → Service B

Service B → Transaction Coordinator (2PC)

Transaction Coordinator (2PC) → Load Balancer

Load Balancer → User

Failure Scenario

Component Fails:Transaction Coordinator (2PC)

Impact:If the coordinator crashes during commit phase, services may remain locked waiting for commit or rollback, causing blocking and potential deadlocks.

Mitigation:Use coordinator recovery logs to resume or abort transactions; alternatively, avoid 2PC in favor of eventual consistency or saga patterns to reduce blocking.

Architecture Quiz - 3 Questions

Which component is responsible for coordinating the two-phase commit?

A Load Balancer

B Transaction Coordinator (2PC)

C Service A

D Cache A

Design Principle

Two-phase commit ensures strong consistency by coordinating all services to commit or abort together, but it introduces blocking and latency that hurt scalability. Modern microservices often prefer eventual consistency or saga patterns to avoid these issues.

Practice

1. What is the main purpose of the two-phase commit protocol in microservices?

easy

A. To automatically retry failed requests

B. To speed up communication between services

C. To allow services to work independently without coordination

D. To ensure all services agree on a transaction before committing

Solution

1. Step 1: Understand the role of two-phase commit

   Two-phase commit is designed to make sure all parts of a distributed transaction agree to commit or abort together.

2. Step 2: Identify the main goal in microservices

   Its main goal is to keep data consistent across multiple services by coordinating their commit decisions.

3. Final Answer:

   To ensure all services agree on a transaction before committing -> Option D

4. Quick Check:

   Two-phase commit = agreement before commit [OK]

Hint: Two-phase commit means all must say yes before commit [OK]

Common Mistakes:

• Thinking it speeds up communication
• Believing services act independently
• Assuming it retries failed requests automatically

2. Which of the following correctly describes the two phases in the two-phase commit protocol?

easy

A. Abort phase where coordinator asks, Prepare phase where services finalize

B. Prepare phase where coordinator asks, Commit phase where services finalize

C. Commit phase where coordinator asks, Prepare phase where services finalize

D. Prepare phase where services finalize, Commit phase where coordinator asks

Solution

1. Step 1: Recall the two phases names and order

   The first phase is the prepare phase where the coordinator asks all services if they can commit.

2. Step 2: Understand the commit phase

   If all agree, the coordinator sends a commit command to finalize the transaction.

3. Final Answer:

   Prepare phase where coordinator asks, Commit phase where services finalize -> Option B

4. Quick Check:

   Prepare then commit = correct phase order [OK]

Hint: Prepare asks, commit finalizes transaction [OK]

Common Mistakes:

• Mixing up the order of prepare and commit phases
• Confusing abort with prepare phase
• Thinking services finalize before coordinator asks

3. Consider a microservices system using two-phase commit. If one service fails to respond during the prepare phase, what is the expected outcome?

medium

A. The coordinator ignores the failure and proceeds

B. The coordinator commits the transaction anyway

C. The coordinator aborts the transaction and tells all services to rollback

D. The coordinator retries the prepare phase indefinitely

Solution

1. Step 1: Analyze failure during prepare phase

   If any service fails to respond or votes no during prepare, the coordinator must abort to keep consistency.

2. Step 2: Understand coordinator's action

   The coordinator sends abort commands to all services to rollback any partial changes.

3. Final Answer:

   The coordinator aborts the transaction and tells all services to rollback -> Option C

4. Quick Check:

   Failure in prepare = abort transaction [OK]

Hint: Any no or failure in prepare means abort [OK]

Common Mistakes:

• Assuming commit happens despite failure
• Thinking coordinator retries forever
• Ignoring failure and proceeding anyway

4. A developer notices that their two-phase commit implementation causes long delays and system hangs when a service crashes. What is the most likely cause?

medium

A. The coordinator is waiting indefinitely for responses from crashed services

B. The services are committing too quickly without coordination

C. The coordinator is skipping the prepare phase

D. The services are not logging their transactions

Solution

1. Step 1: Identify cause of delays and hangs

   In two-phase commit, the coordinator waits for all services to respond during prepare phase.

2. Step 2: Understand impact of crashed services

   If a service crashes, the coordinator may wait indefinitely, causing delays and system hangs.

3. Final Answer:

   The coordinator is waiting indefinitely for responses from crashed services -> Option A

4. Quick Check:

   Waiting on crashed service = system hang [OK]

Hint: Coordinator waits forever if service crashes [OK]

Common Mistakes:

• Thinking services commit too fast causes hangs
• Believing skipping prepare phase causes delays
• Assuming missing logs cause system hangs

5. Why is two-phase commit often avoided in modern microservices architectures despite ensuring consistency?

hard

A. Because it causes blocking, reduces availability, and hurts scalability

B. Because it does not guarantee data consistency

C. Because it requires no coordination between services

D. Because it is too simple and lacks fault tolerance

Solution

1. Step 1: Understand drawbacks of two-phase commit

   Two-phase commit blocks resources while waiting, reducing system availability and scalability.

2. Step 2: Recognize why modern systems avoid it

   Modern microservices prefer eventual consistency and non-blocking patterns to improve performance and fault tolerance.

3. Final Answer:

   Because it causes blocking, reduces availability, and hurts scalability -> Option A

4. Quick Check:

   Blocking and low availability = avoid two-phase commit [OK]

Hint: Two-phase commit blocks and limits scalability [OK]

Common Mistakes:

• Thinking it does not guarantee consistency
• Believing it requires no coordination
• Assuming it is too simple