Merge pull request #2010 from buildkite/pdp-663-exit-with-non-zero-ex…

…it-status-from-a

agent/job_runner.go

@@ -404,6 +404,18 @@ func (r *JobRunner) Run(ctx context.Context) error {
 			exitStatus = "-1"
 			signalReason = "process_run_error"
 		} else {
+			// Intended to capture situations where the job-exec (aka bootstrap) container did not
+			// start. Normally such errors are hidden in the Kubernetes events. Let's feed them up
+			// to the user as they may be the caused by errors in the pipeline definition.
+			if r.cancelled && !r.stopped {
+				k8sProcess, ok := r.process.(*kubernetes.Runner)
+				if ok && k8sProcess.ClientStateUnknown() {
+					r.logStreamer.Process([]byte(
+						"Some containers had unknown exit statuses. Perhaps they were in ImagePullBackOff.",
+					))
+				}
+			}
+
 			// Add the final output to the streamer
 			r.logStreamer.Process(r.output.ReadAndTruncate())
 

kubernetes/kubernetes.go

@@ -159,17 +159,29 @@ func (w waitStatus) Signaled() bool {
 }
 
 func (r *Runner) WaitStatus() process.WaitStatus {
-	var ws process.WaitStatus
+	ws := waitStatus{}
 	for _, client := range r.clients {
 		if client.ExitStatus != 0 {
 			return waitStatus{Code: client.ExitStatus}
 		}
-		// just return any ExitStatus if we don't find any "interesting" ones
-		ws = waitStatus{Code: client.ExitStatus}
+
+		// use an unusual status code to distinguish this unusual state
+		if client.State == stateUnknown {
+			ws.Code -= 10
+		}
 	}
 	return ws
 }
 
+func (r *Runner) ClientStateUnknown() bool {
+	for _, client := range r.clients {
+		if client.State == stateUnknown {
+			return true
+		}
+	}
+	return false
+}
+
 // ==== sidecar api ====
 
 type Empty struct{}