chore(pbs): add status checks (#213)

Author: ltitanb

crates/cli/src/docker_init.rs

@@ -15,6 +15,7 @@ use cb_common::{
         SIGNER_DIR_SECRETS_ENV, SIGNER_KEYS_ENV, SIGNER_MODULE_NAME, SIGNER_PORT_ENV,
         SIGNER_URL_ENV,
     },
+    pbs::{BUILDER_API_PATH, GET_STATUS_PATH},
     signer::{ProxyStore, SignerLoader},
     types::ModuleId,
     utils::random_jwt,
@@ -295,6 +296,17 @@ pub async fn handle_docker_init(config_path: String, output_dir: String) -> Resu
         networks: pbs_networs,
         volumes: pbs_volumes,
         environment: Environment::KvPair(pbs_envs),
+        healthcheck: Some(Healthcheck {
+            test: Some(HealthcheckTest::Single(format!(
+                "curl -f http://localhost:{}{}{}",
+                cb_config.pbs.pbs_config.port, BUILDER_API_PATH, GET_STATUS_PATH
+            ))),
+            interval: Some("30s".into()),
+            timeout: Some("5s".into()),
+            retries: 3,
+            start_period: Some("5s".into()),
+            disable: false,
+        }),
         ..Service::default()
     };
 
@@ -413,10 +425,10 @@ pub async fn handle_docker_init(config_path: String, output_dir: String) -> Resu
                     test: Some(HealthcheckTest::Single(format!(
                         "curl -f http://localhost:{signer_port}/status"
                     ))),
-                    interval: Some("5s".into()),
+                    interval: Some("30s".into()),
                     timeout: Some("5s".into()),
-                    retries: 5,
-                    start_period: Some("0s".into()),
+                    retries: 3,
+                    start_period: Some("5s".into()),
                     disable: false,
                 }),
                 ..Service::default()

crates/pbs/src/service.rs

@@ -1,9 +1,15 @@
-use cb_common::constants::COMMIT_BOOST_VERSION;
+use std::time::Duration;
+
+use cb_common::{
+    constants::COMMIT_BOOST_VERSION,
+    pbs::{BUILDER_API_PATH, GET_STATUS_PATH},
+};
 use cb_metrics::provider::MetricsProvider;
-use eyre::{Context, Result};
+use eyre::{bail, Context, Result};
 use prometheus::core::Collector;
 use tokio::net::TcpListener;
 use tracing::info;
+use url::Url;
 
 use crate::{
     api::BuilderApi,
@@ -16,15 +22,30 @@ pub struct PbsService;
 
 impl PbsService {
     pub async fn run<S: BuilderApiState, A: BuilderApi<S>>(state: PbsState<S>) -> Result<()> {
-        let address = state.config.endpoint;
+        let addr = state.config.endpoint;
         let events_subs =
             state.config.event_publisher.as_ref().map(|e| e.n_subscribers()).unwrap_or_default();
-        info!(version = COMMIT_BOOST_VERSION, ?address, events_subs, chain =? state.config.chain, "starting PBS service");
+        info!(version = COMMIT_BOOST_VERSION, ?addr, events_subs, chain =? state.config.chain, "starting PBS service");
 
         let app = create_app_router::<S, A>(state);
-        let listener = TcpListener::bind(address).await?;
+        let listener = TcpListener::bind(addr).await?;
+
+        let task =
+            tokio::spawn(
+                async move { axum::serve(listener, app).await.wrap_err("PBS server exited") },
+            );
+
+        // wait for the server to start
+        tokio::time::sleep(Duration::from_millis(250)).await;
+        let local_url =
+            Url::parse(&format!("http://{}{}{}", addr, BUILDER_API_PATH, GET_STATUS_PATH))?;
+
+        let status = reqwest::get(local_url).await?;
+        if !status.status().is_success() {
+            bail!("PBS server failed to start. Are the relays properly configured?");
+        }
 
-        axum::serve(listener, app).await.wrap_err("PBS server exited")
+        task.await?
     }
 
     pub fn register_metric(c: Box<dyn Collector>) {

docker/pbs.Dockerfile

@@ -22,6 +22,7 @@ RUN apt-get update && apt-get install -y \
   ca-certificates \
   libssl3 \
   libssl-dev \
+  curl \
   && apt-get clean autoclean \
   && rm -rf /var/lib/apt/lists/*