fix: remove duplicate Docker Hub images, skip trailing cooldown, reliable exit code

- Remove library/{alpine,nginx,redis} from corpus -- already available
  via public ECR mirrors, saves Docker Hub rate limit quota
- Document Docker Hub image dependency and rate limit in corpus header
- Skip 30s cooldown after the last tool in each scenario
- Write bench exit code to a file instead of relying on wait (race-safe)
- Suppress ocync JSON stdout from operator stream (capture only)
- Add [tool_name] prefix to stderr stream for tool identification

Author: bryantbiggs

bench/corpus.yaml

@@ -1,8 +1,14 @@
 # Benchmark image corpus.
 #
-# Sources: Docker Hub (Jupyter + nvidia/cuda + tensorflow), Chainguard free
-# images, AWS Public ECR, SOCI workshop. Each image syncs to ECR under the
-# bench/ prefix.
+# Sources: Docker Hub, Chainguard (cgr.dev), AWS Public ECR, SOCI workshop.
+# Each image syncs to ECR under the bench/ prefix.
+#
+# Docker Hub images (docker.io): Jupyter, nvidia/cuda, tensorflow.
+# These require authenticated pulls (100/hr free tier). Credentials
+# are injected via DOCKERHUB_USERNAME and DOCKERHUB_ACCESS_TOKEN env vars.
+# The remaining Docker Hub official images (alpine, nginx, redis, postgres,
+# golang) are pulled from their public ECR mirrors instead to avoid
+# unnecessary rate limit consumption.
 #
 # Ordering: mount-exercising images first (Jupyter family shares GB-sized base
 # layers across distinct source names and therefore distinct target repos --
@@ -105,14 +111,6 @@ images:
   - source: "public.ecr.aws/docker/library/golang"
     tags: ["1.22", "1.21"]
 
-  # ── Docker Hub library images ──────────────────────────────────────────────
-  - source: "docker.io/library/alpine"
-    tags: ["3.20", "latest"]
-  - source: "docker.io/library/nginx"
-    tags: ["latest"]
-  - source: "docker.io/library/redis"
-    tags: ["latest"]
-
   # ── SOCI workshop images ──────────────────────────────────────────────────
   # Large multi-layer images from public.ecr.aws -- useful for chunked upload
   # and large-blob throughput; limited cross-repo mount potential (each image

xtask/src/bench/mod.rs

@@ -555,7 +555,7 @@ async fn run_cold(
     ));
     let mut runs = Vec::new();
 
-    for &tool in tools {
+    for (tool_idx, &tool) in tools.iter().enumerate() {
         let mut iteration_runs = Vec::new();
 
         for i in 0..args.iterations {
@@ -593,7 +593,9 @@ async fn run_cold(
         let median = iteration_runs.swap_remove(median_idx);
         runs.push(median);
 
-        cooldown().await;
+        if tool_idx + 1 < tools.len() {
+            cooldown().await;
+        }
     }
 
     Ok(ScenarioResult {
@@ -613,7 +615,7 @@ async fn run_warm(
     progress("bench: running warm scenario");
     let mut runs = Vec::new();
 
-    for &tool in tools {
+    for (tool_idx, &tool) in tools.iter().enumerate() {
         progress(&format!("  warm: {} (priming)", tool));
 
         ecr::create_repos(ecr_client, corpus).await?;
@@ -634,7 +636,9 @@ async fn run_warm(
 
         ecr::delete_repos(ecr_client, corpus).await?;
 
-        cooldown().await;
+        if tool_idx + 1 < tools.len() {
+            cooldown().await;
+        }
     }
 
     Ok(ScenarioResult {
@@ -655,7 +659,7 @@ async fn run_partial(
     progress("bench: running partial scenario");
     let mut runs = Vec::new();
 
-    for &tool in tools {
+    for (tool_idx, &tool) in tools.iter().enumerate() {
         progress(&format!("  partial: {} (priming)", tool));
 
         ecr::create_repos(ecr_client, base_corpus).await?;
@@ -678,7 +682,9 @@ async fn run_partial(
 
         ecr::delete_repos(ecr_client, base_corpus).await?;
 
-        cooldown().await;
+        if tool_idx + 1 < tools.len() {
+            cooldown().await;
+        }
     }
 
     Ok(ScenarioResult {

xtask/src/bench/remote.rs

@@ -188,10 +188,13 @@ mkdir -p "$TMPDIR"
 {registry_env}
 
 # Launch detached with output to log file.
+EXIT_FILE=bench/.bench-run.exit
 > "$LOG_FILE"
+rm -f "$EXIT_FILE"
 (
   cargo xtask bench {bench_args} 2>&1
   BENCH_EXIT=$?
+  echo "$BENCH_EXIT" > "$EXIT_FILE"
   echo "[bench-remote] exit code: $BENCH_EXIT"
   rm -f "$PID_FILE"
   exit $BENCH_EXIT
@@ -203,9 +206,13 @@ echo "bench-remote: started (PID $BENCH_PID), streaming log..."
 # Stream log until bench exits. --pid= makes tail exit when process dies.
 tail -n+1 -f "$LOG_FILE" --pid="$BENCH_PID"
 
-# Propagate exit code.
-wait "$BENCH_PID" 2>/dev/null
-EXIT=$?
+# Read exit code from file (reliable, not subject to wait race).
+if [ -f "$EXIT_FILE" ]; then
+  EXIT=$(cat "$EXIT_FILE")
+  rm -f "$EXIT_FILE"
+else
+  EXIT=1
+fi
 echo '[4/4] Benchmarks complete.'
 exit $EXIT
 "#