Rate limit

EMQ X Broker specifies the limit on access speed and message speed. When the client’s connection request speed exceeds the specified limit, the establishment of a new connection is suspended; when the message reception speed exceeds the specified limit, the reception of messages is suspended.

Rate limit is a backpressure scheme that avoids system overload from the entrance and guarantees system stability and predictable throughput. The rate limit can be configured in etc/emqx.conf :

Configuration itemTypeDefault valueDescription
listener.tcp.external.max_conn_rateNumber1000The maximum allowable connection rate on this node (conn/s)
zone.external.publish_limitNumber,DurationNo limitMaximum allowable publish rate on a single connection (msg/s)
listener.tcp.external.rate_limitSize,DurationNo limitMaximum allowable packet rate on a single connection (bytes/s)
  • max_conn_rate is the rate limit for connection establishment on a single emqx node. 1000 means that 1000 clients can access at most.
  • publish_limit is the rate limit for receiving PUBLISH packets on a single connection. 100,10s means that the maximum PUBLISH message rate allowed on each connection is 100 every 10 seconds.
  • rate_limit is the rate limit for receiving TCP packets on a single connection. 100KB,10s means that the maximum TCP packet rate allowed on each connection is 100KB every 10 seconds.

publish_limit and rate_limit both provide limits for a single connection. EMQ X Broker currently does not provide a global message rate limit.

Rate limit explanation

EMQ X Broker uses the Token BucketRate limit - 图1 (opens new window) algorithm to control all Rate Limits. The logic of the token bucket algorithm is as follows:

image-20190604103907875

  • There is a bucket that can hold the maximum burst of the token. The maximum burst is abbreviated as b.
  • There is a rate for adding tokens to the bucket per second, abbreviated as r. When the bucket is full, no tokens are added to the bucket.
  • Whenever 1 (or N) request arrives, take 1 (or N) token from the bucket. If the token is not enough, it will be blocked and wait for the token to be generated.

It can be seen from this algorithm:

  • In the long run, the average value of the limited request rate is equal to the value of rate.

  • When the actual request reaching speed is M, and M> r, then the maximum (peak) rate that can be achieved in actual operation is M = b + r.

It is easy to think that the maximum rate M is the speed that can consume the full state token bucket in 1 unit of time. The consumption rate of token bucket is M-r, so we can see that: b / (M-r) = 1, and we get M = b + r

Application of Token Bucket Algorithm in EMQ X Broker

When the following configuration is used for packet rate limiting:

  1. listener.tcp.external.rate_limit = 100KB,10s

EMQ X Broker will initialize the rate-limit processor of each connection with two values:

  • rate = 100 KB / 10s = 10240 B/s
  • burst = 100 KB = 102400 B

According to the algorithm in Message Rate Limitation Explanation, it is known:

  • In the long run, the allowable average rate is limited to 10240 B/s
  • The allowable peak rate is 102400 + 10240 = 112640 B/s

To improve system throughput, the access module of EMQ X Broker does not read packets from the socket one by one, but reads N packets from the socket each time. The timing of the rate-limit check is after receiving these N messages and before preparing to continue to receive the next N messages. Therefore, the actual rate limit will not be as accurate as the algorithm. EMQ X Broker only provides a rough rate limit. The value of N can be configured in etc/emqx.conf :

Configuration itemTypeDefault valueDescription
listener.tcp.external.active_nNumber100how many messages are read from the TCP stack by emqx at a time