sqs-tests
Overview
Two queue types, per the SQS developer guide:
| Type | Delivery semantics |
|---|
| Standard | "at-least-once message delivery" |
| FIFO | "exactly-once message processing" + "high-throughput" mode |
The semantic difference cascades through every test pattern: a
Standard-queue test must assume duplicates can occur; a FIFO test
must assume strict ordering.
When to use
- The repo has SQS-publishing or SQS-consuming code (AWS SDK).
- The user writes unit tests against mocked SQS clients.
- The user writes integration tests against LocalStack / real SQS
(in a sandbox account).
- A test verifies visibility-timeout, DLQ-routing, or
long-polling behavior.
Step 1 - Test approach: mock vs LocalStack vs real
Three approaches, ordered by isolation:
| Approach | Pros | Cons |
|---|
aws-sdk-client-mock (TS) / moto (Python) | Pure unit, no network | Doesn't catch AWS-side behavior (visibility timeouts, DLQ routing) |
| LocalStack (Docker SQS emulator) | Full SQS semantics locally | Slower; not 100% behavior parity with AWS SQS |
| Real SQS in sandbox AWS account | Highest fidelity | Costs money; per-PR queue cleanup needed |
For pure logic tests (does the code call SendMessage with the right
body?), use mocks. For semantic tests (does retry-on-failure work
end-to-end?), use LocalStack. For pre-prod smoke, use real SQS.
Step 2 - Mock-based unit test (TypeScript)
import { mockClient } from 'aws-sdk-client-mock';
import { SQSClient, SendMessageCommand } from '@aws-sdk/client-sqs';
const sqsMock = mockClient(SQSClient);
beforeEach(() => sqsMock.reset());
it('sends order-placed message to SQS', async () => {
sqsMock.on(SendMessageCommand).resolves({ MessageId: 'msg-123' });
await placeOrder({ customerId: 1 });
expect(sqsMock.commandCalls(SendMessageCommand)).toHaveLength(1);
expect(sqsMock.commandCalls(SendMessageCommand)[0].args[0].input).toMatchObject({
QueueUrl: expect.stringContaining('orders'),
MessageBody: expect.stringContaining('"customerId":1'),
});
});
Step 3 - Mock-based unit test (Python)
import boto3
from moto import mock_aws
@mock_aws
def test_send_order_message():
sqs = boto3.client('sqs', region_name='us-east-1')
queue_url = sqs.create_queue(QueueName='orders')['QueueUrl']
place_order(customer_id=1)
response = sqs.receive_message(QueueUrl=queue_url, MaxNumberOfMessages=10)
assert len(response['Messages']) == 1
body = json.loads(response['Messages'][0]['Body'])
assert body['customerId'] == 1
moto's @mock_aws decorator intercepts boto3 SQS calls; tests run
without network.
Step 4 - LocalStack integration test
# docker-compose.yml
services:
localstack:
image: localstack/localstack:latest
ports: [4566:4566]
environment:
SERVICES: sqs
sqs = boto3.client(
'sqs',
endpoint_url='http://localhost:4566',
region_name='us-east-1',
aws_access_key_id='test', aws_secret_access_key='test',
)
queue_url = sqs.create_queue(QueueName='orders')['QueueUrl']
# ... full SQS API works, including visibility timeouts, DLQ, FIFO
Step 5 - Test visibility-timeout behavior
Visibility timeout keeps an in-flight message invisible to other
receivers for a configurable window, preventing duplicate processing
(SQS developer guide).
Test pattern (LocalStack or real SQS, NOT mock):
sqs.send_message(QueueUrl=queue_url, MessageBody='test')
msg1 = sqs.receive_message(QueueUrl=queue_url, VisibilityTimeout=30)['Messages'][0]
# Within 30s, the message should be invisible to other receivers:
msg2 = sqs.receive_message(QueueUrl=queue_url)
assert msg2.get('Messages') is None
# After visibility timeout (or explicit ChangeMessageVisibility), it returns:
sqs.change_message_visibility(
QueueUrl=queue_url,
ReceiptHandle=msg1['ReceiptHandle'],
VisibilityTimeout=0,
)
msg3 = sqs.receive_message(QueueUrl=queue_url)
assert msg3['Messages'][0]['MessageId'] == msg1['MessageId']
Step 6 - Test DLQ routing
SQS routes a message to the dead-letter queue after maxReceiveCount
failed deliveries (poison-message isolation, per the SQS developer
guide). Full LocalStack recipe:
references/localstack-recipes.md.
Step 7 - Test FIFO ordering + dedup
FIFO queues (FifoQueue: 'true') preserve per-MessageGroupId order
and drop identical bodies within a 5-minute dedup window
(ContentBasedDeduplication). Full LocalStack recipe:
references/localstack-recipes.md.
Step 8 - Message retention
SQS auto-deletes messages older than the retention period: default 4
days, configurable from 60 seconds to 1,209,600 seconds (14 days)
(SQS developer guide). Tests rarely verify retention
directly; document the expected retention in queue setup (Terraform /
CloudFormation) and review per-team.
Step 9 - CI integration
services:
localstack:
image: localstack/localstack:latest
ports: [4566:4566]
env: { SERVICES: sqs }
steps:
- run: pytest tests/integration/sqs/ -v
Anti-patterns
| Anti-pattern | Why it fails | Fix |
|---|
| Test visibility-timeout via mock | Mock doesn't track invisibility window; tests pass-by-accident | Use LocalStack (Step 4 - 5) |
| Skip DLQ-routing test | Poison-message handling unverified; production incidents | Always cover DLQ for production queues (Step 6) |
| Use Standard-queue body assertions sensitive to delivery order | At-least-once = duplicates + reordering | Assert per-message processing idempotency, not order |
| Hard-code queue URLs in tests | Tests break when account / region changes | Pull from env vars / fixtures |
Limitations
- LocalStack ≠ real SQS - some behaviors (extreme latency, AWS
throttling) are not emulated.
- FIFO throughput limits (300 messages/sec without
high-throughput mode) need real SQS to verify.
- Cost-effective sandbox testing requires per-PR queue cleanup; use
random queue-name suffixes + cleanup hooks.
References
- sqs-dg - official SQS developer guide; standard vs FIFO,
visibility timeout, DLQ, retention
- docs.aws.amazon.com/AWSSimpleQueueService - full SQS docs
- localstack.cloud - LocalStack SQS emulator
- pypi.org/project/moto - Python AWS mock
- npmjs.com/package/aws-sdk-client-mock - TypeScript AWS mock
sidekiq-tests,
celery-tests,
bullmq-tests,
rabbitmq-tests - sister tools
idempotency-test-author -
critical companion for at-least-once SQS Standard queues