The configuration guide of EMQX Rate Limit

Before reading this guide, we assume that you have known simple MQTT and MQTT broker.

EMQX Broker starts supporting the function rate limit from version V3, including the limit on the PUBLISH packet receiving rate and the TCP data package receiving rate. This article will introduce the use and configuration of this function in detail.

Configuration items

MQTT PUBLISH packet receiving rate

This configuration locate on emqx.conf:

zone.external.publish_limit = 10,1m

The format of configuration is: <Number>,<Duration>, which means the maximum number <Number> of PUBLISH packet that can be allowed to receive during the time <Duration>.

TCP data package receiving rate

This configuration locate on emqx.conf:

listener.tcp.external.rate_limit = 1024,4096

The format of configuration is: <Rate>,<Burst>, which means the allowed average data package receiving rate is <Rate>. However, the allowed maximum number is depending on the value of <Burst>. See the next section for details: rate limit algorithm token bucket-algorithm

active_ n

This configuration locate on emqx.conf:

listener.tcp.external.active_n = 100

active_n actually means that the number of datagrams allowed to be read in the underlying asynchronous I/O. Every time, when the asynchronous read operation reaches this limit, it will be temporarily switched to the Sync mode. Every time, when it switches to the Sync mode, it will check the rate limit one time. Therefore, if the value is larger, the throughput performance of this system is better; if the value is smaller, rate checking is more accurate and the inflow is more stable and the security of the system is higher.

Rate limit algorithm - token bucket

Introduction

The publish_limit and rate_limit we mentioned are implemented by the token bucket algorithm. The algorithm logic is as follows:

1. There is a bucket of a token that can hold up the number of burst at most.
2. We will add tokens into this bucket at a rate of N per second. If the bucket is full, no more tokens are added to this bucket. We call this rate as rate.
3. When there are requests, we will take the corresponding number of tokens from the bucket. If the bucket is empty, block it until enough tokens are put in.

Function

Through token bucket algorithm, we can:

• In the case of running for a long time, the average value of the limited request rate is equal to the rate of adding tokens, the rate.
• Allow a certain level of peak flow. If the request rate is M and is greater than rate, the rate of token decreasing is M - rate. Therefore, the time for us to take all the tokens from a full bucket is burst / (M - rate), and the number of the requests be accepted is burst / (M - rate) * M during this period.

All in all, it can be simply understood as rate is the average request rate and burst is is the instantaneous maximum request rate.

The implementation of EMQX rate limit

Based on the token bucket algorithm, EMQX's implementation logic for rate limit is as follows:

Its meaning is:

1. The socket performs a rate checking every time it receives n TCP data packets, and will record the total length of the received n TCP data packets as s.
2. If the number of left tokens in the bucket is greater and equal to s, will update the number of tokens on the corresponding bucket and active_n will be performed through continue activating socket.
3. If the number of left tokens in the bucket r is less than s, will active socket after waiting for (s - r) / rate seconds.

The example of rate limit configuration

The configuration of rate limit

listener.tcp.external.active_n = 100
listener.tcp.external.rate_limit = 1024,1024000

The above configuration means:

• Check the rate every 100 TCP packets are received.
• The average rate limit is 1024 byte/s.
• Bucket size is 1000KB. If the total length of this 100 TCP packets is greater than 1000KB, the rate limit will be triggered.

Therefore, users need to set <Burst> according to the size of the real packet. EMQX highly recommends that configure it as (max_packet_size * active_n) / 2 to prevent from block.

listener.tcp.external.active_n = 100
zone.external.publish_limit = 10,1m

The above configuration means:

• Check the rate every 100 TCP packets are received.
• Only 10 PUBLISH packet will be allowed to receive within one minute.

You can transfer <Number>,<Duration> into the form of <Rate>,<Burst>, namely <Number> / <Duration>, <Numebr>.

Other configurations

In addition to the above Rate Limit, it is also supported that perform the following limits for the TCP, WebSocket and other connections:

listener.tcp.external.max_connections = 1024000

The maximum connection number that exists at the same time allowed, namely the maximum number of simultaneous online clients.

listener.tcp.external.max_conn_rate = 1000

The maximum number of concurrent connections allowed per second.