feat: Show ignoredUnrecoerableEvents if timeout occurs

Fix #1041

Signed-off-by: Andrew Obuchowicz <aobuchow@redhat.com>

Author: AObuchow

controllers/workspace/devworkspace_controller.go

@@ -36,6 +36,7 @@ import (
 	"github.com/devfile/devworkspace-operator/pkg/library/flatten"
 	kubesync "github.com/devfile/devworkspace-operator/pkg/library/kubernetes"
 	"github.com/devfile/devworkspace-operator/pkg/library/projects"
+	"github.com/devfile/devworkspace-operator/pkg/library/status"
 	"github.com/devfile/devworkspace-operator/pkg/provision/automount"
 	"github.com/devfile/devworkspace-operator/pkg/provision/metadata"
 	"github.com/devfile/devworkspace-operator/pkg/provision/storage"
@@ -221,7 +222,14 @@ func (r *DevWorkspaceReconciler) Reconcile(ctx context.Context, req ctrl.Request
 		// encountered in main reconcile loop.
 		if err == nil {
 			if timeoutErr := checkForStartTimeout(clusterWorkspace); timeoutErr != nil {
-				reconcileResult, err = r.failWorkspace(workspace, timeoutErr.Error(), metrics.ReasonInfrastructureFailure, reqLogger, &reconcileStatus)
+				// Check if an ignoredUnrecoverableEvent occured and report it alongside the timeout notice
+				errMsg := status.CheckForIgnoredWorkspacePodEvents(workspace, clusterAPI)
+				if errMsg != "" {
+					failureMsg := fmt.Sprintf("%s. Reason: %s", timeoutErr.Error(), errMsg)
+					reconcileResult, err = r.failWorkspace(workspace, failureMsg, metrics.DetermineProvisioningFailureReason(errMsg), reqLogger, &reconcileStatus)
+				} else {
+					reconcileResult, err = r.failWorkspace(workspace, timeoutErr.Error(), metrics.ReasonInfrastructureFailure, reqLogger, &reconcileStatus)
+				}
 			}
 		}
 		if reconcileStatus.phase == dw.DevWorkspaceStatusRunning {
@@ -480,7 +488,7 @@ func (r *DevWorkspaceReconciler) Reconcile(ctx context.Context, req ctrl.Request
 	deploymentStatus := wsprovision.SyncDeploymentToCluster(workspace, allPodAdditions, serviceAcctName, clusterAPI)
 	if !deploymentStatus.Continue {
 		if deploymentStatus.FailStartup {
-			failureReason := metrics.DetermineProvisioningFailureReason(deploymentStatus)
+			failureReason := metrics.DetermineProvisioningFailureReason(deploymentStatus.Info())
 			return r.failWorkspace(workspace, deploymentStatus.Info(), failureReason, reqLogger, &reconcileStatus)
 		}
 		reqLogger.Info("Waiting on deployment to be ready")

controllers/workspace/metrics/deployment_provisioning.go

@@ -17,8 +17,6 @@ package metrics
 
 import (
 	"strings"
-
-	"github.com/devfile/devworkspace-operator/pkg/provision/workspace"
 )
 
 var badRequestFailures = []string{
@@ -37,15 +35,14 @@ var infrastructureFailures = []string{
 // DetermineProvisioningFailureReason scans a deployment provisioning status info message
 // and returns the corresponding failure reason.
 // If a failure reason cannot be found, an Unknown reason is returned.
-func DetermineProvisioningFailureReason(status workspace.DeploymentProvisioningStatus) FailureReason {
-	failStartupMsg := status.Info()
+func DetermineProvisioningFailureReason(statusMessage string) FailureReason {
 	for _, failure := range badRequestFailures {
-		if strings.Contains(failStartupMsg, failure) {
+		if strings.Contains(statusMessage, failure) {
 			return ReasonBadRequest
 		}
 	}
 	for _, failure := range infrastructureFailures {
-		if strings.Contains(failStartupMsg, failure) {
+		if strings.Contains(statusMessage, failure) {
 			return ReasonInfrastructureFailure
 		}
 	}

pkg/library/status/check.go

@@ -21,6 +21,7 @@ import (
 	"strings"
 
 	"github.com/devfile/devworkspace-operator/pkg/common"
+	"github.com/devfile/devworkspace-operator/pkg/constants"
 	"github.com/devfile/devworkspace-operator/pkg/infrastructure"
 	"github.com/devfile/devworkspace-operator/pkg/provision/sync"
 	appsv1 "k8s.io/api/apps/v1"
@@ -156,6 +157,22 @@ func CheckContainerStatusForFailure(containerStatus *corev1.ContainerStatus, ign
 	return true, ""
 }
 
+// Returns an error message if the workspace related pods are in an unrecoverable state, which may
+// have been caused  by an ignoredUnrecoverableEvent.
+// Otherwise, en empty string is returned.
+func CheckForIgnoredWorkspacePodEvents(workspace *common.DevWorkspaceWithConfig, clusterAPI sync.ClusterAPI) (errMsg string) {
+	workspaceIDLabel := k8sclient.MatchingLabels{constants.DevWorkspaceIDLabel: workspace.Status.DevWorkspaceId}
+	// CheckPodsState returns either a message or error, not both.
+	errMsg, checkErr := CheckPodsState(workspace.Status.DevWorkspaceId, workspace.Namespace, workspaceIDLabel, []string{}, clusterAPI)
+	if checkErr != nil {
+		return checkErr.Error()
+	}
+	if errMsg != "" {
+		return errMsg
+	}
+	return ""
+}
+
 func checkIfUnrecoverableEventIgnored(reason string, ignoredEvents []string) (ignored bool) {
 	for _, ignoredReason := range ignoredEvents {
 		if ignoredReason == reason {