Amazon MQ for RabbitMQ sizing guidelines

You can choose the broker instance type that best supports your application. When choosing an instance type, it is important to consider factors that will affect broker performance:

• the number of clients and queues

• the volume of messages sent

• messages kept in memory

• redundant messages

Smaller broker instance types (t3.micro) are recommended only for testing application performance. We recommend larger broker instance types (m5.large and above) for production levels of clients and queues, high throughput, messages in memory, and redundant messages.

It is important to test your brokers to determine the appropriate instance type and size for your workload messaging requirements. Use the following sizing guidelines to determine the best appropriate instance type for your application.

Sizing guidelines for single instance deployment

The following table shows the maximum limit values for each instance type for single instance brokers.

Instance Type	Connections	Channels	Queues	Consumers per channel	Shovels
t3.micro	500	1,500	2,500	1,000	150
m5.large	5,000	15,000	30,000	1,000	250
m5.xlarge	10,000	30,000	60,000	1,000	500
m5.2xlarge	20,000	60,000	120,000	1,000	1,000
m5.4xlarge	40,000	120,000	240,000	1,000	2,000

Sizing guidelines for cluster deployment

The following table shows the maximum limit values for each instance type for cluster brokers.

Instance Type	Connections	Channels	Queues	Consumers per channel	Shovels
m5.large	15,000	45,000	10,000	1,000	150
m5.xlarge	30,000	90,000	15,000	1,000	300
m5.2xlarge	60,000	180,000	20,000	1,000	600
m5.4xlarge	120,000	360,000	30,000	1,000	1200

The connection, channel, and shovel limits are applied per node. The exact limit values for a cluster broker may be lower than the indicated value depending on the number of available nodes and how RabbitMQ distributes resources among the available nodes.

Error messages

The following error messages are returned when limits are exceeded. All values are based on the m5.large single instance limits.

Note

The error codes for the following messages may change based on the client library you are using.

Connection

ConnectionClosedByBroker 500 "NOT_ALLOWED - connection refused: node connection limit (500) is reached"

Channel

ConnectionClosedByBroker 1500 "NOT_ALLOWED - number of channels opened on node '[e-mail]' has reached the maximum allowed limit of (15,000)"

Consumer

ConnectionClosedByBroker: (530, 'NOT_ALLOWED - reached maximum (1,000) of consumers per channel')

Note

The following error messages use the HTTP Management API format.

Queue

{"error":"bad_request","reason":"cannot declare queue 'my_queue': queue limit in cluster (30,000) is reached"}]

Shovel

{"error":"bad_request","reason":"Validation failed\n\ncomponent shovel is limited to 250 per node\n"}

Vhost

{"error":"bad_request","reason":"cannot create vhost 'my_vhost': vhost limit of 4,000 is reached"}