Fix cloud shotover metrics (#1449)

shotover-proxy/benches/windsock/cassandra/bench.rs

@@ -559,6 +559,9 @@ impl Bench for CassandraBench {
  let shotover_ip = shotover_instance
  .as_ref()
  .map(|x| x.instance.private_ip().to_string());
+ let shotover_connect_ip = shotover_instance
+ .as_ref()
+ .map(|x| x.instance.connect_ip().to_string());
 
  let mut profiler_instances: HashMap<String, &Ec2Instance> =
  [("bencher".to_owned(), &bench_instance.instance)].into();
@@ -581,9 +584,13 @@ impl Bench for CassandraBench {
  profiler_instances.insert("cassandra".to_owned(), &cassandra_instances[0].instance);
  }
  }
- let mut profiler =
- CloudProfilerRunner::new(self.name(), profiling, profiler_instances, &shotover_ip)
- .await;
+ let mut profiler = CloudProfilerRunner::new(
+ self.name(),
+ profiling,
+ profiler_instances,
+ &shotover_connect_ip,
+ )
+ .await;
 
  let cassandra_nodes = vec![
  AwsNodeInfo {

shotover-proxy/benches/windsock/cloud/aws.rs

@@ -22,6 +22,8 @@ impl AwsInstances {
  pub async fn new() -> Self {
  AwsInstances {
  aws: Aws::builder(CleanupResources::WithAppTag(AWS_THROWAWAY_TAG.to_owned()))
+ // shotover metrics port
+ .expose_ports_to_internet(vec![9001])
  .build()
  .await,
  }

shotover-proxy/benches/windsock/profilers.rs

@@ -104,7 +104,7 @@ impl CloudProfilerRunner {
  bench_name: String,
  profiling: Profiling,
  instances: HashMap<String, &Ec2Instance>,
- shotover_ip: &Option<String>,
+ shotover_connect_ip: &Option<String>,
  ) -> Self {
  let run_sys_monitor = profiling
  .profilers_to_use
@@ -124,7 +124,7 @@ impl CloudProfilerRunner {
  let shotover_metrics = if run_shotover_metrics {
  Some(ShotoverMetrics::new(
  bench_name.clone(),
- shotover_ip.as_ref().unwrap(),
+ shotover_connect_ip.as_ref().unwrap(),
  ))
  } else {
  None

shotover-proxy/benches/windsock/profilers/shotover_metrics.rs

@@ -148,14 +148,15 @@ impl ShotoverMetrics {
  tokio::select! {
  _ = shutdown_rx.recv() => break,
  _ = interval.tick() => {
- match tokio::time::timeout(Duration::from_secs(3), reqwest::get(url)).await.unwrap() {
- Ok(response) => {
+ match tokio::time::timeout(Duration::from_secs(3), reqwest::get(url)).await {
+ Ok(Ok(response)) => {
  results.push(RawPrometheusExposition {
  timestamp: OffsetDateTime::now_utc(),
  content: response.text().await.unwrap(),
  });
  }
- Err(err) => tracing::debug!("Failed to request from metrics endpoint, probably not up yet, error was {err:?}")
+ Ok(Err(err)) => tracing::debug!("Failed to request from metrics endpoint {url}, probably not up yet, error was {err:?}"),
+ Err(_) => panic!("Timed out request from metrics endpoint {url}")
  }
  }
  }