Crate swanling[−][src]

Expand description

Swanling

Have you ever been attacked by a swanling?

Swanling is a load testing framework inspired by Locust. User behavior is defined with standard Rust code.

Swanling load tests, called Swanling Attacks, are built by creating an application with Cargo, and declaring a dependency on the Swanling library.

Swanling uses reqwest to provide a convenient HTTP client.

Documentation

Creating and running a Swanling load test

Creating a simple Swanling load test

First create a new empty cargo application, for example:

$ cargo new loadtest
     Created binary (application) `loadtest` package
$ cd loadtest/

Add Swanling as a dependency in Cargo.toml:

[dependencies]
swanling = "0.12"

Add the following boilerplate use declaration at the top of your src/main.rs:

use swanling::prelude::*;

Using the above prelude will automatically add the following use statements necessary for your load test, so you don’t need to manually add them:

use swanling::swanling::{
    SwanlingTask, SwanlingTaskError, SwanlingTaskFunction, SwanlingTaskResult, SwanlingTaskSet, SwanlingUser,
};
use swanling::metrics::SwanlingMetrics;
use swanling::{
    task, taskset, SwanlingAttack, SwanlingDefault, SwanlingDefaultType, SwanlingError, SwanlingScheduler,
};

Below your main function (which currently is the default Hello, world!), add one or more load test functions. The names of these functions are arbitrary, but it is recommended you use self-documenting names. Load test functions must be async. Each load test function must accept a reference to a SwanlingUser object and return a SwanlingTaskResult. For example:

use swanling::prelude::*;

async fn loadtest_foo(user: &SwanlingUser) -> SwanlingTaskResult {
  let _swanling = user.get("/path/to/foo").await?;

  Ok(())
}

In the above example, we’re using the SwanlingUser helper get to load a path on the website we are load testing. This helper creates a reqwest::RequestBuilder object and uses it to build and execute a request for the above path. If you want access to the RequestBuilder object, you can instead use the swanling_get helper, for example to set a timeout on this specific request:

use std::time;

use swanling::prelude::*;

async fn loadtest_bar(user: &SwanlingUser) -> SwanlingTaskResult {
    let request_builder = user.swanling_get("/path/to/bar").await?;
    let _swanling = user.swanling_send(request_builder.timeout(time::Duration::from_secs(3)), None).await?;

    Ok(())
}

We pass the RequestBuilder object to swanling_send which builds and executes it, also collecting useful metrics. The .await at the end is necessary as swanling_send is an async function.

Once all our tasks are created, we edit the main function to initialize swanling and register the tasks. In this very simple example we only have two tasks to register, while in a real load test you can have any number of task sets with any number of individual tasks.

use swanling::prelude::*;

fn main() -> Result<(), SwanlingError> {
    let _swanling_metrics = SwanlingAttack::initialize()?
        .register_taskset(taskset!("LoadtestTasks")
            // Register the foo task, assigning it a weight of 10.
            .register_task(task!(loadtest_foo).set_weight(10)?)
            // Register the bar task, assigning it a weight of 2 (so it
            // runs 1/5 as often as bar). Apply a task name which shows up
            // in metrics.
            .register_task(task!(loadtest_bar).set_name("bar").set_weight(2)?)
        )
        // You could also set a default host here, for example:
        .set_default(SwanlingDefault::Host, "http://dev.local/")?
        // We set a default run time so this test runs to completion.
        .set_default(SwanlingDefault::RunTime, 1)?
        .execute()?;

    Ok(())
}

// A task function that loads `/path/to/foo`.
async fn loadtest_foo(user: &SwanlingUser) -> SwanlingTaskResult {
    let _swanling = user.get("/path/to/foo").await?;

    Ok(())
}

// A task function that loads `/path/to/bar`.
async fn loadtest_bar(user: &SwanlingUser) -> SwanlingTaskResult {
    let _swanling = user.get("/path/to/bar").await?;

    Ok(())
}

Swanling now spins up a configurable number of users, each simulating a user on your website. Thanks to reqwest, each user maintains its own web client state, handling cookies and more so your “users” can log in, fill out forms, and more, as real users on your sites would do.

Running the Swanling load test

Attempts to run our example will result in an error, as we have not yet defined the host against which this load test should be run. We intentionally do not hard code the host in the individual tasks, as this allows us to run the test against different environments, such as local development, staging, and production.

$ cargo run --release
   Compiling loadtest v0.1.0 (~/loadtest)
    Finished release [optimized] target(s) in 1.52s
     Running `target/release/loadtest`
Error: InvalidOption { option: "--host", value: "", detail: "A host must be defined via the --host option, the SwanlingAttack.set_default() function, or the SwanlingTaskSet.set_host() function (no host defined for WebsiteUser)." }

Pass in the -h flag to see all available run-time options. For now, we’ll use a few options to customize our load test.

$ cargo run --release -- --host http://dev.local -t 30s -v

The first option we specified is --host, and in this case tells Swanling to run the load test against a VM on my local network. The -t 30s option tells Swanling to end the load test after 30 seconds (for real load tests you’ll certainly want to run it longer, you can use h, m, and s to specify hours, minutes and seconds respectively. For example, -t1h30m would run the load test for 1 hour 30 minutes). Finally, the -v flag tells swanling to display INFO and higher level logs to stdout, giving more insight into what is happening. (Additional -v flags will result in considerably more debug output, and are not recommended for running actual load tests; they’re only useful if you’re trying to debug Swanling itself.)

Running the test results in the following output (broken up to explain it as it goes):

   Finished release [optimized] target(s) in 0.05s
    Running `target/release/loadtest --host 'http://dev.local' -t 30s -v`
15:42:23 [ INFO] Output verbosity level: INFO
15:42:23 [ INFO] Logfile verbosity level: WARN

If we set the --log-file flag, Swanling will write a log file with WARN and higher level logs as you run the test from (add a -g flag to log all INFO and higher level logs).

15:42:23 [ INFO] concurrent users defaulted to 8 (number of CPUs)
15:42:23 [ INFO] run_time = 30
15:42:23 [ INFO] hatch_rate = 1

Swanling will default to launching 1 user per available CPU core, and will launch them all in one second. You can change how many users are launched with the -u option, and you can change how many users are launched per second with the -r option. For example, -u30 -r2 would launch 30 users over 15 seconds (two users per second).

15:42:23 [ INFO] global host configured: http://dev.local/
15:42:23 [ INFO] initializing user states...
15:42:23 [ INFO] launching user 1 from LoadtestTasks...
15:42:24 [ INFO] launching user 2 from LoadtestTasks...
15:42:25 [ INFO] launching user 3 from LoadtestTasks...
15:42:26 [ INFO] launching user 4 from LoadtestTasks...
15:42:27 [ INFO] launching user 5 from LoadtestTasks...
15:42:28 [ INFO] launching user 6 from LoadtestTasks...
15:42:29 [ INFO] launching user 7 from LoadtestTasks...
15:42:30 [ INFO] launching user 8 from LoadtestTasks...
15:42:31 [ INFO] launched 8 users...
15:42:31 [ INFO] printing running metrics after 8 seconds...

Each user is launched in its own thread with its own user state. Swanling is able to make very efficient use of server resources. By default Swanling resets the metrics after all users are launched, but first it outputs the metrics collected while ramping up:

15:42:31 [ INFO] printing running metrics after 8 seconds...

 === PER TASK METRICS ===
 ------------------------------------------------------------------------------
 Name                     |   # times run |        # fails |   task/s |  fail/s
 ------------------------------------------------------------------------------
 1: LoadtestTasks         |
   1:                     |         2,033 |         0 (0%) |   254.12 |    0.00
   2: bar                 |           407 |         0 (0%) |    50.88 |    0.00
 -------------------------+---------------+----------------+----------+--------
 Aggregated               |         2,440 |         0 (0%) |   305.00 |    0.00
 ------------------------------------------------------------------------------
 Name                     |    Avg (ms) |        Min |         Max |     Median
 ------------------------------------------------------------------------------
 1: LoadtestTasks         |
   1:                     |       14.23 |          6 |          32 |         14
   2: bar                 |       14.13 |          6 |          30 |         14
 -------------------------+-------------+------------+-------------+-----------
 Aggregated               |       14.21 |          6 |          32 |         14

 === PER REQUEST METRICS ===
 ------------------------------------------------------------------------------
 Name                     |        # reqs |        # fails |    req/s |  fail/s
 ------------------------------------------------------------------------------
 GET /                    |         2,033 |         0 (0%) |   254.12 |    0.00
 GET bar                  |           407 |         0 (0%) |    50.88 |    0.00
 -------------------------+---------------+----------------+----------+--------
 Aggregated               |         2,440 |         0 (0%) |   305.00 |    0.00
 ------------------------------------------------------------------------------
 Name                     |    Avg (ms) |        Min |        Max |      Median
 ------------------------------------------------------------------------------
 GET /                    |       14.18 |          6 |          32 |         14
 GET bar                  |       14.08 |          6 |          30 |         14
 -------------------------+-------------+------------+-------------+-----------
 Aggregated               |       14.16 |          6 |          32 |         14

All 8 users hatched, resetting metrics (disable with --no-reset-metrics).

Swanling can optionally display running metrics if started with --running-metrics INT where INT is an integer value in seconds. For example, if Swanling is started with --running-metrics 15 it will display running values approximately every 15 seconds. Running metrics are broken into several tables. First are the per-task metrics which are further split into two sections. The first section shows how many requests have been made, how many of them failed (non-2xx response), and the corresponding per-second rates.

This table shows details for all Task Sets and all Tasks defined by your load test, regardless of if they actually run. This can be useful to ensure that you have set up weighting as intended, and that you are simulating enough users. As our first task wasn’t named, it just showed up as “1:”. Our second task was named, so it shows up as the name we gave it, “bar”.

15:42:46 [ INFO] printing running metrics after 15 seconds...

 === PER TASK METRICS ===
 ------------------------------------------------------------------------------
 Name                     |   # times run |        # fails |   task/s |  fail/s
 ------------------------------------------------------------------------------
 1: LoadtestTasks         |
   1:                     |         4,618 |         0 (0%) |   307.87 |    0.00
   2: bar                 |           924 |         0 (0%) |    61.60 |    0.00
 -------------------------+---------------+----------------+----------+--------
 Aggregated               |         5,542 |         0 (0%) |   369.47 |    0.00
 ------------------------------------------------------------------------------
 Name                     |    Avg (ms) |        Min |         Max |     Median
 ------------------------------------------------------------------------------
 1: LoadtestTasks         |
   1:                     |       21.17 |          8 |         151 |         19
   2: bar                 |       21.62 |          9 |         156 |         19
 -------------------------+-------------+------------+-------------+-----------
 Aggregated               |       21.24 |          8 |         156 |         19

The second table breaks down the same metrics by request instead of by Task. For our simple load test, each Task only makes a single request, so the metrics are the same. There are two main differences. First, metrics are listed by request type and path or name. The first request shows up as GET /path/to/foo as the request was not named. The second request shows up as GET bar as the request was named. The times to complete each are slightly smaller as this is only the time to make the request, not the time for Swanling to execute the entire task.

 === PER REQUEST METRICS ===
 ------------------------------------------------------------------------------
 Name                     |        # reqs |        # fails |    req/s |  fail/s
 ------------------------------------------------------------------------------
 GET /path/to/foo         |         4,618 |         0 (0%) |   307.87 |    0.00
 GET bar                  |           924 |         0 (0%) |    61.60 |    0.00
 -------------------------+---------------+----------------+----------+--------
 Aggregated               |         5,542 |         0 (0%) |   369.47 |    0.00
 ------------------------------------------------------------------------------
 Name                     |    Avg (ms) |        Min |        Max |      Median
 ------------------------------------------------------------------------------
 GET /path/to/foo         |       21.13 |          8 |         151 |         19
 GET bar                  |       21.58 |          9 |         156 |         19
 -------------------------+-------------+------------+-------------+-----------
 Aggregated               |       21.20 |          8 |         156 |         19

Note that Swanling respected the per-task weights we set, and foo (with a weight of 10) is being loaded five times as often as bar (with a weight of 2). On average each page is returning within 21.2 milliseconds. The quickest page response was for foo in 8 milliseconds. The slowest page response was for bar in 156 milliseconds.

15:43:02 [ INFO] stopping after 30 seconds...
15:43:02 [ INFO] waiting for users to exit
15:43:02 [ INFO] exiting user 3 from LoadtestTasks...
15:43:02 [ INFO] exiting user 4 from LoadtestTasks...
15:43:02 [ INFO] exiting user 5 from LoadtestTasks...
15:43:02 [ INFO] exiting user 8 from LoadtestTasks...
15:43:02 [ INFO] exiting user 2 from LoadtestTasks...
15:43:02 [ INFO] exiting user 7 from LoadtestTasks...
15:43:02 [ INFO] exiting user 6 from LoadtestTasks...
15:43:02 [ INFO] exiting user 1 from LoadtestTasks...
15:43:02 [ INFO] printing metrics after 30 seconds...

Our example only runs for 30 seconds, so we only see running metrics once. When the test completes, we get more detail in the final summary. The first two tables are the same as what we saw earlier, however now they include all metrics for the entire length of the load test:

 === PER TASK METRICS ===
 ------------------------------------------------------------------------------
 Name                     |   # times run |        # fails |   task/s |  fail/s
 ------------------------------------------------------------------------------
 1: LoadtestTasks         |
   1:                     |         9,974 |         0 (0%) |   332.47 |    0.00
   2: bar                 |         1,995 |         0 (0%) |    66.50 |    0.00
 -------------------------+---------------+----------------+----------+--------
 Aggregated               |        11,969 |         0 (0%) |   398.97 |    0.00
 ------------------------------------------------------------------------------
 Name                     |    Avg (ms) |        Min |         Max |     Median
 ------------------------------------------------------------------------------
 1: LoadtestTasks         |
   1:                     |       19.65 |          8 |         151 |         18
   2: bar                 |       19.92 |          9 |         156 |         18
 -------------------------+-------------+------------+-------------+-----------
 Aggregated               |       19.69 |          8 |         156 |         18

 === PER REQUEST METRICS ===
 ------------------------------------------------------------------------------
 Name                     |        # reqs |        # fails |    req/s |  fail/s
 ------------------------------------------------------------------------------
 GET /                    |         9,974 |         0 (0%) |   332.47 |    0.00
 GET bar                  |         1,995 |         0 (0%) |    66.50 |    0.00
 -------------------------+---------------+----------------+----------+--------
 Aggregated               |        11,969 |         0 (0%) |   398.97 |    0.00
 ------------------------------------------------------------------------------
 Name                     |    Avg (ms) |        Min |        Max |      Median
 ------------------------------------------------------------------------------
 GET /                    |       19.61 |          8 |         151 |         18
 GET bar                  |       19.88 |          9 |         156 |         18
 -------------------------+-------------+------------+-------------+-----------
 Aggregated               |       19.66 |          8 |         156 |         18
 ------------------------------------------------------------------------------

The ratio between foo and bar remained 5:2 as expected.

 ------------------------------------------------------------------------------
 Slowest page load within specified percentile of requests (in ms):
 ------------------------------------------------------------------------------
 Name                     |    50% |    75% |    98% |    99% |  99.9% | 99.99%
 ------------------------------------------------------------------------------
 GET /                    |     18 |     21 |     29 |     79 |    140 |    140
 GET bar                  |     18 |     21 |     29 |    120 |    150 |    150
 -------------------------+--------+--------+--------+--------+--------+-------
 Aggregated               |     18 |     21 |     29 |     84 |    140 |    156

A new table shows additional information, breaking down response-time by percentile. This shows that the slowest page loads only happened in the slowest 1% of page loads, so were an edge case. 98% of the time page loads happened in 29 milliseconds or less.

License

Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.