eMQTT-Bench is a lightweight MQTT v5.0 benchmark tool written in Erlang.
You can download pre-built binary packeges from https://github.com/emqx/emqtt-bench/releases.
NOTE: Requires Erlang/OTP 22.3+ to build.
emqtt-bench requires libatomic
# centos 7
sudo yum install libatomic# ubuntu 20.04
sudo apt install libatomic1 git clone https://github.com/emqx/emqtt-bench.git
cd emqtt-bench
makeOptional, you could disable QUIC support if you have problem with compiling
BUILD_WITHOUT_QUIC=1 make$ ./emqtt_bench conn --help
Usage: emqtt_bench conn [--help <help>] [-d <dist>] [-h [<host>]]
[-p [<port>]] [-V [<version>]] [-c [<count>]]
[-n [<startnumber>]] [--load-qst <nst_dets_file>]
[-Q [<qoe>]] [-i [<interval>]] [-u <username>]
[-P <password>] [-k [<keepalive>]] [-C [<clean>]]
[-x [<expiry>]] [-S [<ssl>]]
[--certfile <certfile>] [--keyfile <keyfile>]
[--quic [<quic>]] [--ifaddr <ifaddr>]
[--prefix <prefix>] [-s [<shortids>]]
[-l <lowmem>]
[--num-retry-connect [<num_retry_connect>]]
[-R [<conn_rate>]]
[--force-major-gc-interval [<force_major_gc_interval>]]
[--log_to [<log_to>]]
--help help information
-d, --dist enable distribution port
-h, --host mqtt server hostname or comma-separated
hostnames [default: localhost]
-p, --port mqtt server port number [default: 1883]
-V, --version mqtt protocol version: 3 | 4 | 5 [default: 5]
-c, --count max count of clients [default: 200]
-n, --startnumber start number [default: 0]
-Q, --qoe Enable QoE tracking [default: false]
-i, --interval interval of connecting to the broker
[default: 10]
-u, --username username for connecting to server
-P, --password password for connecting to server
-k, --keepalive keep alive in seconds [default: 300]
-C, --clean clean session [default: true]
-x, --session-expiry Set 'Session-Expiry' for persistent sessions
(seconds) [default: 0]
-S, --ssl ssl socoket for connecting to server
[default: false]
--certfile client certificate for authentication, if
required by server
--keyfile client private key for authentication, if
required by server
--quic QUIC transport [default: false]
--load-qst load quic session tickets from dets file
--ifaddr local ipaddress or interface address
--prefix client id prefix
-s, --shortids use short ids for client ids [default: false]
-l, --lowmem enable low mem mode, but use more CPU
--num-retry-connect number of times to retry estabilishing a
connection before giving up [default: 0]
-R, --connrate connection rate(/s), default: 0, fallback to
use --interval [default: 0]
--force-major-gc-interval interval in milliseconds in which a major GC
will be forced on the bench processes. a
value of 0 means disabled (default). this
only takes effect when used together with
--lowmem. [default: 0]
--log_to Control where the log output goes. console:
directly to the console null: quietly,
don't output any logs. [default: console]
For example, create 50K concurrent connections at the arrival rate of 100/sec:
./emqtt_bench conn -c 50000 -i 10$ ./emqtt_bench sub --help
Usage: emqtt_bench sub [--help <help>] [-d <dist>] [-h [<host>]] [-p [<port>]] [-V [<version>]] [-c [<count>]]
[-n [<startnumber>]] [-i [<interval>]] [-t <topic>] [--payload-hdrs [<payload_hdrs>]] [-q [<qos>]]
[-Q [<qoe>]] [-u <username>] [-P <password>] [-k [<keepalive>]] [-C [<clean>]] [-x [<expiry>]]
[-S [<ssl>]] [--certfile <certfile>] [--keyfile <keyfile>] [--ws [<ws>]] [--quic [<quic>]]
[--load-qst <nst_dets_file>] [--ifaddr <ifaddr>] [--prefix <prefix>] [-s [<shortids>]] [-l <lowmem>]
[--num-retry-connect [<num_retry_connect>]] [-R [<conn_rate>]]
[--force-major-gc-interval [<force_major_gc_interval>]] [--log_to [<log_to>]]
--help help information
-d, --dist enable distribution port
-h, --host mqtt server hostname or comma-separated hostnames [default: localhost]
-p, --port mqtt server port number [default: 1883]
-V, --version mqtt protocol version: 3 | 4 | 5 [default: 5]
-c, --count max count of clients [default: 200]
-n, --startnumber The start point when assigning sequence numbers to clients. This is useful when running
multiple emqtt-bench instances to test the same broker (cluster), so the start number can be
planned to avoid client ID collision [default: 0]
-i, --interval interval of connecting to the broker [default: 10]
-t, --topic topic subscribe, support %u, %c, %i variables
--payload-hdrs Handle the payload header from received message. Publish side must have the same option enabled
in the same order. cnt64: Check the counter is strictly increasing. ts: publish latency counting.
[default: ]
-q, --qos subscribe qos [default: 0]
-Q, --qoe Enable QoE tracking [default: false]
-u, --username username for connecting to server
-P, --password password for connecting to server
-k, --keepalive keep alive in seconds [default: 300]
-C, --clean clean start [default: true]
-x, --session-expiry Set 'Session-Expiry' for persistent sessions (seconds) [default: 0]
-S, --ssl ssl socoket for connecting to server [default: false]
--certfile client certificate for authentication, if required by server
--keyfile client private key for authentication, if required by server
--ws websocket transport [default: false]
--quic QUIC transport [default: false]
--load-qst load quic session tickets from dets file
--ifaddr local ipaddress or interface address
--prefix Client ID prefix. If not provided '$HOST_bench_(pub|sub)_$RANDOM_$N' is used, where $HOST is
either the host name or the IP address provided in the --ifaddr option, $RANDOM is a random
number and $N is the sequence number assigned for each client. If provided, the $RANDOM suffix
will not be added.
-s, --shortids Use short client ID. If --prefix is provided, the prefix is added otherwise client ID is the
assigned sequence number. [default: false]
-l, --lowmem enable low mem mode, but use more CPU
--num-retry-connect number of times to retry estabilishing a connection before giving up [default: 0]
-R, --connrate connection rate(/s), default: 0, fallback to use --interval [default: 0]
--force-major-gc-interval interval in milliseconds in which a major GC will be forced on the bench processes. a value of
0 means disabled (default). this only takes effect when used together with --lowmem. [default:
0]
--log_to Control where the log output goes. console: directly to the console null: quietly, don't
output any logs. [default: console]For example, create 50K concurrent connections at the arrival rate of 100/sec:
./emqtt_bench sub -c 50000 -i 10 -t bench/%i -q 2Usage: emqtt_bench pub [--help <help>] [-d <dist>] [-h [<host>]] [-p [<port>]] [-V [<version>]] [-c [<count>]]
[-n [<startnumber>]] [-i [<interval>]] [-I [<interval_of_msg>]] [-u <username>] [-P <password>]
[-t <topic>] [--payload-hdrs [<payload_hdrs>]] [-s [<size>]] [-m <message>] [-q [<qos>]]
[-Q [<qoe>]] [-r [<retain>]] [-k [<keepalive>]] [-C [<clean>]] [-x [<expiry>]] [-L [<limit>]]
[-S [<ssl>]] [--certfile <certfile>] [--keyfile <keyfile>] [--ws [<ws>]] [--quic [<quic>]]
[--load-qst <nst_dets_file>] [--ifaddr <ifaddr>] [--prefix <prefix>] [-s [<shortids>]] [-l <lowmem>]
[-F [<inflight>]] [-w [<wait_before_publishing>]] [--max-random-wait [<max_random_wait>]]
[--min-random-wait [<min_random_wait>]] [--num-retry-connect [<num_retry_connect>]]
[-R [<conn_rate>]] [--force-major-gc-interval [<force_major_gc_interval>]] [--log_to [<log_to>]]
--help help information
-d, --dist enable distribution port
-h, --host mqtt server hostname or comma-separated hostnames [default: localhost]
-p, --port mqtt server port number [default: 1883]
-V, --version mqtt protocol version: 3 | 4 | 5 [default: 5]
-c, --count max count of clients [default: 200]
-n, --startnumber The start point when assigning sequence numbers to clients. This is useful when running
multiple emqtt-bench instances to test the same broker (cluster), so the start number can be
planned to avoid client ID collision [default: 0]
-i, --interval interval of connecting to the broker [default: 10]
-I, --interval_of_msg interval of publishing message(ms) [default: 1000]
-u, --username username for connecting to server
-P, --password password for connecting to server
-t, --topic topic subscribe, support %u, %c, %i, %s variables
--payload-hdrs If set, add optional payload headers. cnt64: strictly increasing counter(64bit) per publisher
ts: Timestamp when emit example: --payload-hdrs cnt64,ts [default: ]
-s, --size payload size [default: 256]
-m, --message set the message content for publish
-q, --qos subscribe qos [default: 0]
-Q, --qoe Enable QoE tracking [default: false]
-r, --retain retain message [default: false]
-k, --keepalive keep alive in seconds [default: 300]
-C, --clean clean start [default: true]
-x, --session-expiry Set 'Session-Expiry' for persistent sessions (seconds) [default: 0]
-L, --limit The max message count to publish, 0 means unlimited [default: 0]
-S, --ssl ssl socoket for connecting to server [default: false]
--certfile client certificate for authentication, if required by server
--keyfile client private key for authentication, if required by server
--ws websocket transport [default: false]
--quic QUIC transport [default: false]
--load-qst load quic session tickets from dets file
--ifaddr One or multiple (comma-separated) source IP addresses
--prefix Client ID prefix. If not provided '$HOST_bench_(pub|sub)_$RANDOM_$N' is used, where $HOST is
either the host name or the IP address provided in the --ifaddr option, $RANDOM is a random
number and $N is the sequence number assigned for each client. If provided, the $RANDOM
suffix will not be added.
-s, --shortids Use short client ID. If --prefix is provided, the prefix is added otherwise client ID is the
assigned sequence number. [default: false]
-l, --lowmem enable low mem mode, but use more CPU
-F, --inflight maximum inflight messages for QoS 1 an 2, value 0 for 'infinity' [default: 1]
-w, --wait-before-publishing wait for all publishers to have (at least tried to) connected before starting publishing
[default: false]
--max-random-wait maximum randomized period in ms that each publisher will wait before starting to publish
(uniform distribution) [default: 0]
--min-random-wait minimum randomized period in ms that each publisher will wait before starting to publish
(uniform distribution) [default: 0]
--num-retry-connect number of times to retry estabilishing a connection before giving up [default: 0]
-R, --connrate connection rate(/s), default: 0, fallback to use --interval [default: 0]
--force-major-gc-interval interval in milliseconds in which a major GC will be forced on the bench processes. a value
of 0 means disabled (default). this only takes effect when used together with --lowmem.
[default: 0]
--log_to Control where the log output goes. console: directly to the console null: quietly,
don't output any logs. [default: console]For example, create 100 connections and each publishes messages at the rate of 100 msg/sec.
./emqtt_bench pub -c 100 -I 10 -t bench/%i -s 256./emqtt_bench pub --ifaddr 192.168.1.10
./emqtt_bench sub --ifaddr 192.168.2.10./emqtt_bench sub -c 100 -i 10 -t bench/%i -p 8883 --ssl
./emqtt_bench pub -c 100 -I 10 -t bench/%i -p 8883 -s 256 --ssl./emqtt_bench sub -c 100 -i 10 -t bench/%i -p 8883 --ssl --certfile path/to/client-cert.pem --keyfile path/to/client-key.pem
./emqtt_bench pub -c 100 -i 10 -t bench/%i -s 256 -p 8883 --ssl --certfile path/to/client-cert.pem --keyfile path/to/client-key.pemYou should not set '-c' option more than 64K for TCP ports limit on one source addresses, however you can send messages from multiple source IP Addresses with '--ifaddr ' such like
./emqtt_bench sub -c 200000 -t "perf/test" --ifaddr 192.168.200.18,192.168.200.19,192.168.200.20,192.168.200.21
Make sure to increase resource usage limits and expand the port range like following on Linux.
ulimit -n 200000
sudo sysctl -w net.ipv4.ip_local_port_range="1025 65534"EMQX Team.