add support for aborting workspaces

Author: sagor999

components/ws-manager-api/core.proto

@@ -121,6 +121,7 @@ message StopWorkspaceRequest {
 enum StopWorkspacePolicy {
     NORMALLY = 0;
     IMMEDIATELY = 1;
+    ABORT = 2;
 }
 
 // StopWorkspaceResponse is the answer to a stop workspace request
@@ -423,6 +424,9 @@ message WorkspaceConditions {
 
     // volume_snapshot contains info about volume snapshot that was used to save persistent volume
     VolumeSnapshotInfo volume_snapshot = 12;
+
+    // aborted is true if StopWorkspace was called with StopWorkspacePolicy set to ABORT
+    WorkspaceConditionBool aborted = 13;
 }
 
 // WorkspaceConditionBool is a trinary bool: true/false/empty

components/ws-manager-api/go/core.pb.go

@@ -840,6 +840,8 @@ export class WorkspaceConditions extends jspb.Message {
     clearVolumeSnapshot(): void;
     getVolumeSnapshot(): VolumeSnapshotInfo | undefined;
     setVolumeSnapshot(value?: VolumeSnapshotInfo): WorkspaceConditions;
+    getAborted(): WorkspaceConditionBool;
+    setAborted(value: WorkspaceConditionBool): WorkspaceConditions;
 
     serializeBinary(): Uint8Array;
     toObject(includeInstance?: boolean): WorkspaceConditions.AsObject;
@@ -864,6 +866,7 @@ export namespace WorkspaceConditions {
         headlessTaskFailed: string,
         stoppedByRequest: WorkspaceConditionBool,
         volumeSnapshot?: VolumeSnapshotInfo.AsObject,
+        aborted: WorkspaceConditionBool,
     }
 }
 
@@ -1227,6 +1230,7 @@ export namespace WorkspaceClass {
 export enum StopWorkspacePolicy {
     NORMALLY = 0,
     IMMEDIATELY = 1,
+    ABORT = 2,
 }
 
 export enum AdmissionLevel {

components/ws-manager-api/typescript/src/core_pb.d.ts

@@ -6666,7 +6666,8 @@ proto.wsman.WorkspaceConditions.toObject = function(includeInstance, msg) {
     firstUserActivity: (f = msg.getFirstUserActivity()) && google_protobuf_timestamp_pb.Timestamp.toObject(includeInstance, f),
     headlessTaskFailed: jspb.Message.getFieldWithDefault(msg, 10, ""),
     stoppedByRequest: jspb.Message.getFieldWithDefault(msg, 11, 0),
-    volumeSnapshot: (f = msg.getVolumeSnapshot()) && proto.wsman.VolumeSnapshotInfo.toObject(includeInstance, f)
+    volumeSnapshot: (f = msg.getVolumeSnapshot()) && proto.wsman.VolumeSnapshotInfo.toObject(includeInstance, f),
+    aborted: jspb.Message.getFieldWithDefault(msg, 13, 0)
   };
 
   if (includeInstance) {
@@ -6749,6 +6750,10 @@ proto.wsman.WorkspaceConditions.deserializeBinaryFromReader = function(msg, read
       reader.readMessage(value,proto.wsman.VolumeSnapshotInfo.deserializeBinaryFromReader);
       msg.setVolumeSnapshot(value);
       break;
+    case 13:
+      var value = /** @type {!proto.wsman.WorkspaceConditionBool} */ (reader.readEnum());
+      msg.setAborted(value);
+      break;
     default:
       reader.skipField();
       break;
@@ -6857,6 +6862,13 @@ proto.wsman.WorkspaceConditions.serializeBinaryToWriter = function(message, writ
       proto.wsman.VolumeSnapshotInfo.serializeBinaryToWriter
     );
   }
+  f = message.getAborted();
+  if (f !== 0.0) {
+    writer.writeEnum(
+      13,
+      f
+    );
+  }
 };
 
 
@@ -7096,6 +7108,24 @@ proto.wsman.WorkspaceConditions.prototype.hasVolumeSnapshot = function() {
 };
 
 
+/**
+ * optional WorkspaceConditionBool aborted = 13;
+ * @return {!proto.wsman.WorkspaceConditionBool}
+ */
+proto.wsman.WorkspaceConditions.prototype.getAborted = function() {
+  return /** @type {!proto.wsman.WorkspaceConditionBool} */ (jspb.Message.getFieldWithDefault(this, 13, 0));
+};
+
+
+/**
+ * @param {!proto.wsman.WorkspaceConditionBool} value
+ * @return {!proto.wsman.WorkspaceConditions} returns this
+ */
+proto.wsman.WorkspaceConditions.prototype.setAborted = function(value) {
+  return jspb.Message.setProto3EnumField(this, 13, value);
+};
+
+
 
 
 
@@ -9756,7 +9786,8 @@ proto.wsman.WorkspaceClass.prototype.setDisplayname = function(value) {
  */
 proto.wsman.StopWorkspacePolicy = {
   NORMALLY: 0,
-  IMMEDIATELY: 1
+  IMMEDIATELY: 1,
+  ABORT: 2
 };
 
 /**

components/ws-manager-api/typescript/src/core_pb.js

@@ -80,6 +80,9 @@ const (
 	// stoppedByRequestAnnotation is set on a pod when it was requested to stop using a StopWorkspace call
 	stoppedByRequestAnnotation = "gitpod.io/stoppedByRequest"
 
+	// abortRequestAnnotation is set if StopWorkspace was called with ABORT StopWorkspacePolicy
+	abortRequestAnnotation = "gitpod.io/abortRequest"
+
 	// attemptingToCreatePodAnnotation is set when ws-manager is trying to create pod and is removed when pod is successfully scheduled on the node
 	attemptingToCreatePodAnnotation = "gitpod.io/attemptingToCreate"
 )

components/ws-manager/pkg/manager/annotations.go

@@ -43,6 +43,7 @@ import (
 	wsk8s "github.com/gitpod-io/gitpod/common-go/kubernetes"
 	"github.com/gitpod-io/gitpod/common-go/log"
 	"github.com/gitpod-io/gitpod/common-go/tracing"
+	"github.com/gitpod-io/gitpod/common-go/util"
 	csapi "github.com/gitpod-io/gitpod/content-service/api"
 	"github.com/gitpod-io/gitpod/content-service/pkg/layer"
 	regapi "github.com/gitpod-io/gitpod/registry-facade/api"
@@ -733,6 +734,12 @@ func (m *Manager) StopWorkspace(ctx context.Context, req *api.StopWorkspaceReque
 	gracePeriod := stopWorkspaceNormallyGracePeriod
 	if req.Policy == api.StopWorkspacePolicy_IMMEDIATELY {
 		gracePeriod = stopWorkspaceImmediatelyGracePeriod
+	} else if req.Policy == api.StopWorkspacePolicy_ABORT {
+		gracePeriod = stopWorkspaceImmediatelyGracePeriod
+		err = m.markWorkspace(ctx, req.Id, addMark(abortRequestAnnotation, util.BooleanTrueString))
+		if err != nil {
+			clog.WithError(err).Error("failed to mark workspace for abort")
+		}
 	}
 
 	err = m.markWorkspace(ctx, req.Id, addMark(stoppedByRequestAnnotation, gracePeriod.String()))

components/ws-manager/pkg/manager/manager.go

@@ -265,6 +265,8 @@ func (m *metrics) OnChange(status *api.WorkspaceStatus) {
 			reason = "tab-closed"
 		} else if strings.Contains(status.Message, "workspace timed out") {
 			reason = "timeout"
+		} else if status.Conditions.Aborted == api.WorkspaceConditionBool_TRUE {
+			reason = "aborted"
 		} else if status.Conditions.Failed != "" {
 			reason = "failed"
 		} else {

components/ws-manager/pkg/manager/metrics.go

@@ -493,6 +493,7 @@ func actOnPodEvent(ctx context.Context, m actingManager, manager *Manager, statu
 		span.LogKV("disposalStatusAnnotation", ds)
 		if terminated && !ds.Status.IsDisposed() {
 			if wso.Pod.Annotations[workspaceFailedBeforeStoppingAnnotation] == util.BooleanTrueString && wso.Pod.Annotations[workspaceNeverReadyAnnotation] == util.BooleanTrueString {
+				span.LogKV("event", "failed before stopping and never ready")
 				// The workspace is never ready, so there is no need for a finalizer.
 				if _, ok := pod.Annotations[workspaceExplicitFailAnnotation]; !ok {
 					failMessage := status.Conditions.Failed
@@ -511,6 +512,14 @@ func actOnPodEvent(ctx context.Context, m actingManager, manager *Manager, statu
 					return err
 				}
 				return m.modifyFinalizer(ctx, workspaceID, gitpodFinalizerName, false)
+			} else if wso.Pod.Annotations[abortRequestAnnotation] == util.BooleanTrueString {
+				span.LogKV("event", "workspace was aborted")
+				// The workspace is aborted, so there is no need to finalize content
+				if err = manager.deleteWorkspacePVC(ctx, pod.Name); err != nil {
+					log.Error(err)
+					return err
+				}
+				return m.modifyFinalizer(ctx, workspaceID, gitpodFinalizerName, false)
 			} else {
 				// We start finalizing the workspace content only after the container is gone. This way we ensure there's
 				// no process modifying the workspace content as we create the backup.

components/ws-manager/pkg/manager/monitor.go

@@ -330,6 +330,9 @@ func (m *Manager) extractStatusFromPod(result *api.WorkspaceStatus, wso workspac
 	if _, sbr := pod.Annotations[stoppedByRequestAnnotation]; sbr {
 		result.Conditions.StoppedByRequest = api.WorkspaceConditionBool_TRUE
 	}
+	if _, abr := pod.Annotations[abortRequestAnnotation]; abr {
+		result.Conditions.Aborted = api.WorkspaceConditionBool_TRUE
+	}
 	if wso.IsWorkspaceHeadless() {
 		for _, cs := range pod.Status.ContainerStatuses {
 			if cs.State.Terminated != nil && cs.State.Terminated.Message != "" {