...
- In your existing test plan, make sure that any variations in testing make use of variables. For example, if running a HTTP sampler, use HTTP Request Defaults to specify a host as "${+P(targetHost)}". Other useful places for variables might include number of threads, ramp-up period or scheduler duration in a thread group, using a format of ${+P(threadgroup.threads,500)} (The +P function is shorthand for +parameter. See the userguide for more info on using this parameter).
- Save your test plan and properties file to a directory.
- Create a properties file containing all your variables. E.g. mytest.properties could contain threadgroup.threads=100, targetHost=my-target-host.com
- The test plan does not need Listeners, as this will be configured via parameters. This will improve performance on the testing.
- Run the test mode in stand-alone mode (i.e. no remote servers):
jmeter -n -t load_test.jmx -l load_test_report.jtl -q mytest.properties -j mytest.log
- 1 small piece of functionality missing is the ability to set global variables from a property file. the "-q" parameter defines an additional property file, which is the equivalent of setting properties for the contents of the file using the -J parameter. This parameters are not global however, as would be set by the -G parameter. the following bash script (if your using Linux) would help with converting a properties file to global properties:
No Format |
---|
for var in `grep -vE "^(#|$)" mytest.properties`; do GLOBAL_VARS="-G${var} ${GLOBAL_VARS}" done |
...
- On all client machines, start up JMeter server, ensuring that firewall is not blocking connections (clients must all be on the same subnet).
- On JMeter controller (the host initializing the test), run the test with the -R parameter (can be run using -r and specfiying hosts in jmeter properties file):
jmeter -n -t load_test.jmx -l load_test_report.jtl -q mytest.properties -j mytest.log -R remotehost1,remotehost2 ${GLOBAL_VARS
}
Note that load injection to targets will be a increased as a ratio of number of threads per thread group multiplied by number of client machines.
A bash script similar to the following could be used to start up jmeter (called jmeter_test_wrapper.sh for example):
No Format |
---|
#!/bin/sh loadtest="myloadtest" GLOBAL_VARS="" REPORT_DIR=/tmp JMETER_PROPERTIES=jmeter.properties JMETER_CUSTOM_PROPERTIES=mytest.properties # If set as an environment var, then use it, otherwise leave it unset. HOST_LIST=${HOST_LIST:+"-R ${HOST_LIST}"} for var in `grep -vE "^#|^$" ${JMETER_CUSTOM_PROPERTIES}`; do GLOBAL_VARS="-G${var} ${GLOBAL_VARS}" done if [ -f ${loadtest}.jmx ]; then echo "Running ${loadtest} with:" echo -e " ----------------------- $JMETER -n -t ${loadtest}.jmx -l ${REPORT_DIR}/${loadtest}.jtl -p ${JMETER_PROPERTIES} ${GLOBAL_VARS} ${HOST_LIST} -j ${REPORT_DIR}/_${loadtest}.log -----------------------" $JMETER -n \ -t ${loadtest}.jmx \ -l ${REPORT_DIR}/${loadtest}.jtl \ -p ${JMETER_PROPERTIES} \ ${GLOBAL_VARS} \ ${HOST_LIST} \ -j ${REPORT_DIR}/_${loadtest}.log else echo "Could not find test plan for ${loadtest}" fi |
For a local test, you could run: ./jmeter_test_wrapper.sh
For a remote test, it can be modified as HOST_LIST="192.168.1.10,192.168.1.12" ./jmeter_test_wrapper.sh
Now all you need to do is make sure the servers are running and add the above command into scheduler.