The shinyloadtest
package
and the accompanying
shinycannon
command line
tool make it possible to load test deployed Shiny apps. Load testing
helps developers and administrators estimate how many users their
application can support. If an application requires tuning, load testing
and load test result analysis can be used to identify performance
bottlenecks and to guide changes to infrastructure, configuration, or
code.
Scientific load testing helps put to rest the common misconception that “Shiny doesn’t scale”. As rstudio::conf(2018) Sean Lopp presented on Scaling Shiny which shows how to to horizontally scale an app to handle tens of thousands of users.
To perform a load test you’ll need two pieces of software:
shinyloadtest
is an R package used to generate recordings and analyze results. You should install it on your development machine withinstall.packages("shinyloadtest")
.shinycannon
is a command-line used to replay recordings in parallel. You can install it on your development machine for testing, but for best results we recommend installing it on a server, and preferably not the one the application under test is also on. See installation instructions for operating specific install instructions..
The process for load testing a Shiny application consists of three steps:
- Record a typical user session for the app.
- Replay the session in parallel, simulating many simultaneous users accessing the app.
- Analyze the results of the load test and determine if the app performed well enough.
Rinse and repeat as necessary. Each step is described below.
Record a session using shinyloadtest::record_session()
, which takes
the URL of the deployed application as an argument:
shinyloadtest::record_session("https://shinyapp.example.com/")
Running the function will open a browser displaying the app. Once open,
interact with the application as a typical user might then close the
browser. After closing the app, a file (recording.log
by default) will
be created that contains a recording of the session. This recording will
serve as the basis for the load test.
If your application requires authentication, consult the authentication article. Also be aware that certain Shiny features are not compatible with shinyloadtest.
With the recording in hand, we’re ready to run the load test. The actual
test is conducted outside of R using the shinycannon
command-line
tool. You can run it using your system’s terminal or console program, or
you can run it from the RStudio IDE’s terminal tab. A typical run looks
like this:
shinycannon recording.log https://shinyapp.example.com/ --workers 5 --loaded-duration-minutes 2 --output-dir run1
(On Windows, you will need to replace “shinycannon
” with
java -jar shinycannon-VERSION.jar
.)
See the shinycannon article for details.
Now we can analyse our results by reading the data into
shinyloadtest::load_runs()
and create a report with
shinyloadtest_report()
:
df <- shinyloadtest::load_runs("run1")
shinyloadtest::shinyloadtest_report(df, "run1.html")
This self contained html report will be opened in your browser for inspection. For further analysis explanation, please visit Analysing load test logs.
Analysis Example