feat(stresstest): TTL after 1h and make cleanup configurable (#223)

Stresstests are frequently canceled or crash as we test objectstore's
limits. This leaves objects in the backends permanently. To fix this, we
now configure a default 1h TTL for all objects created by the
stresstest.

Since we can usually rely on that TTL, we no longer have to run the
cleanup. This saves significant time, as the deletions usually take
longer than the test itself.

To make configuration and usage a bit easier, the stresstest is now
converted to the command pattern.

Author: jan-auer

objectstore-server/tests/stresstest.rs

@@ -3,8 +3,8 @@ use std::time::Duration;
 
 use nix::sys::signal::{self, Signal};
 use nix::unistd::Pid;
-use stresstest::Workload;
 use stresstest::http::HttpRemote;
+use stresstest::{Stresstest, Workload};
 
 const OBJECTSTORE_EXE: &str = env!("CARGO_BIN_EXE_objectstore");
 
@@ -60,7 +60,11 @@ async fn test_basic() {
         .action_weights(8, 1, 1)
         .build();
 
-    stresstest::run(remote, vec![workload], Duration::from_secs(2))
+    Stresstest::new(remote)
+        .duration(Duration::from_secs(2))
+        .workload(workload)
+        .cleanup(true)
+        .run()
         .await
         .expect("Failed to run stress test");
 

stresstest/src/config.rs

@@ -12,6 +12,9 @@ pub struct Config {
     pub duration: Duration,
 
     pub workloads: Vec<Workload>,
+
+    #[serde(default)]
+    pub cleanup: bool,
 }
 
 #[derive(Debug, Deserialize)]

stresstest/src/lib.rs

@@ -12,12 +12,16 @@
 //!
 //! *Read* or *delete* actions are using a *zipfian* distribution, meaning that
 //! more recently written blobs are the ones that will be read/deleted.
+//!
+//! To run a stresstest, instantiate the [`Stresstest`] struct with a remote backend
+//! implementation, such as the [`HttpRemote`](http::HttpRemote) and configure it with
+//! the desired duration and workload.
 #![warn(missing_docs)]
 #![warn(missing_debug_implementations)]
 
 pub mod http;
 pub mod stresstest;
 pub mod workload;
 
-pub use crate::stresstest::run;
+pub use crate::stresstest::Stresstest;
 pub use crate::workload::Workload;

stresstest/src/main.rs

@@ -21,6 +21,7 @@ use anyhow::Context;
 use argh::FromArgs;
 use stresstest::Workload;
 use stresstest::http::HttpRemote;
+use stresstest::stresstest::Stresstest;
 
 use crate::config::Config;
 
@@ -46,22 +47,23 @@ async fn main() -> anyhow::Result<()> {
         serde_yaml::from_reader(config_file).context("failed to parse config YAML")?;
 
     let remote = HttpRemote::new(&config.remote);
+    let mut stresstest = Stresstest::new(remote)
+        .duration(config.duration)
+        .cleanup(config.cleanup);
 
-    let workloads = config
-        .workloads
-        .into_iter()
-        .map(|w| {
-            Workload::builder(w.name)
-                .concurrency(w.concurrency)
-                .organizations(w.organizations)
-                .mode(w.mode)
-                .size_distribution(w.file_sizes.p50.0, w.file_sizes.p99.0)
-                .action_weights(w.actions.writes, w.actions.reads, w.actions.deletes)
-                .build()
-        })
-        .collect();
+    for w in config.workloads {
+        let workload = Workload::builder(w.name)
+            .concurrency(w.concurrency)
+            .organizations(w.organizations)
+            .mode(w.mode)
+            .size_distribution(w.file_sizes.p50.0, w.file_sizes.p99.0)
+            .action_weights(w.actions.writes, w.actions.reads, w.actions.deletes)
+            .build();
 
-    stresstest::run(remote, workloads, config.duration).await?;
+        stresstest = stresstest.workload(workload);
+    }
+
+    stresstest.run().await?;
 
     Ok(())
 }