Skip to content
Software Architecture Guide

Synchronous vs Asynchronous Connector

Explanation

Section titled “Explanation”

A connector is the link that carries communication between two components. Its timing affects availability.

Synchronous connectors. The sender calls the receiver and waits. Both must be up and ready at the same moment. Remote calls can fail or hang until a timeout. While callers wait on an unresponsive service, they hold threads, memory and connections. Many waiting callers can use up these resources. Then unrelated parts of the system that share them can fail too. Failures cascade across systems this way.

Chains add risk. The combined availability of chained components is the product of their individual values. Each new dependency therefore lowers the total. See the Availability article for the term.

Asynchronous connectors. A queue sits between sender and receiver. The sender posts a message and moves on. The queue stores the message until the receiver picks it up. Neither side needs to be up at the same time as the other. The sender is not delayed when the receiver is down. It can keep posting messages. So the sender and the queue must be available, and the queue and the receiver must be available, but not both ends together.

Protecting synchronous calls. If you keep a synchronous connector, you can add safeguards:

  • A timeout stops a caller from waiting forever. Alone, it still lets blocked requests hold resources.
  • A retry repeats an operation. It suits temporary faults.
  • A circuit breaker trips once failures pass a threshold. After that, calls return an error at once without making the remote call. It prevents repeating operations likely to fail, and it can be combined with retry.

Message-driven designs often need a circuit breaker less. They have built-in failure isolation, and they can route failed messages to a dead letter queue.

Example

Section titled “Example”

An online shop takes an order and must send a confirmation email. In a synchronous design, the checkout calls the email service and waits. If the email service hangs, checkout requests pile up and the shop slows down. In an asynchronous design, checkout drops a message on a queue and finishes. The email service reads the message when it is back up.

Trade-offs

Section titled “Trade-offs”

Queues are one-way. If the sender needs an answer, you must build an extra reply mechanism. Queues are unsuitable when the caller needs a fast, synchronous response.

Delivery has costs too. At-least-once delivery means a message can arrive more than once, so receivers must tolerate duplicates. Order is not guaranteed when several receivers work in parallel. When messages arrive faster than receivers process them, the queue grows and latency rises.

A synchronous call needs both parties up together. A queue lets each party talk only to the queue.

Diagram source 
flowchart LR
  subgraph Synchronous
    A1[Sender] -->|call and wait| B1[Receiver]
  end
  subgraph Asynchronous
    A2[Sender] -->|post| Q[(Queue)]
    Q -->|read later| B2[Receiver]
  end

Checks

Section titled “Checks”
  • Each synchronous call chain is identified, and its composite availability is calculated as the product of the components’ availabilities.
  • Synchronous remote calls have timeouts and, where useful, retries or a circuit breaker.
  • Where a sender must keep working while the receiver is down, a queue or similar asynchronous connector is used.
  • For queue-based connectors, the document addresses duplicate delivery, ordering and queue growth.
  • Where a reply is needed over a queue, the reply mechanism is described.
Section titled “Related”

Sources

Section titled “Sources”