[bridge] Introduce AppClusterWorkspaceInstanceController

Author: geropl

components/ws-manager-bridge/src/app-cluster-instance-controller.ts

@@ -0,0 +1,86 @@
+/**
+ * Copyright (c) 2022 Gitpod GmbH. All rights reserved.
+ * Licensed under the GNU Affero General Public License (AGPL).
+ * See License-AGPL.txt in the project root for license information.
+ */
+
+import { WorkspaceDB } from "@gitpod/gitpod-db/lib/workspace-db";
+import { Disposable, DisposableCollection } from "@gitpod/gitpod-protocol";
+import { log } from "@gitpod/gitpod-protocol/lib/util/logging";
+import { repeat } from "@gitpod/gitpod-protocol/lib/util/repeat";
+import { TraceContext } from "@gitpod/gitpod-protocol/lib/util/tracing";
+import { inject, injectable } from "inversify";
+import { Configuration } from "./config";
+import { WorkspaceInstanceController } from "./workspace-instance-controller";
+
+/**
+ * The WorkspaceInstance lifecycle is split between application clusters and workspace clusters on the transition from
+ * pending/building -> starting (cmp. WorkspacePhases here:
+ * https://github.com/gitpod-io/gitpod/blob/008ea3fadc89d4817cf3effc8a5b30eaf469fb1c/components/gitpod-protocol/src/workspace-instance.ts#L111).
+ *
+ * Before the transition, WorkspaceInstances belong to the respective app cluster, denoted by "instance.region === 'eu02'", for exmaple.
+ * After a WorkspaceInstance has been moved over to a workspace cluster, that moved "ownership" is reflected in said field.
+ * We maintain a constant connection (called "bridge") to all workspace clusters to be able to keep reality (workspace
+ * side) in sync with what we have in our DB/forward to clients.
+ *
+ * This class is meant to take the same responsibility for all WorkspaceInstances that have not (yet) been passed over
+ * to a workspace cluster for whatever reason. Here's a list of examples, prefixed by phase:
+ *  - "preparing": failed cleanup after failed call to wsManager.StartWorkspace
+ *  - "building": failed cleanup after failed image-build (which is still controlled by the application cluster,
+ *     although that might change in the future)
+ */
+@injectable()
+export class AppClusterWorkspaceInstancesController implements Disposable {
+    @inject(Configuration) protected readonly config: Configuration;
+
+    @inject(WorkspaceDB) protected readonly workspaceDb: WorkspaceDB;
+
+    @inject(WorkspaceInstanceController) protected readonly workspaceInstanceController: WorkspaceInstanceController;
+
+    protected readonly dispoables = new DisposableCollection();
+
+    public async start() {
+        const disposable = repeat(
+            async () => this.controlAppClusterManagedWorkspaceInstances(),
+            this.config.controllerIntervalSeconds * 1000,
+        );
+        this.dispoables.push(disposable);
+    }
+
+    protected async controlAppClusterManagedWorkspaceInstances() {
+        const installation = this.config.installation;
+
+        const span = TraceContext.startSpan("controlAppClusterManagedWorkspaceInstances");
+        const ctx = { span };
+        try {
+            log.info("Controlling app cluster instances", { installation });
+
+            const notStoppedInstances = await this.workspaceDb.findRunningInstancesWithWorkspaces(
+                installation,
+                undefined,
+                false,
+            );
+            await this.workspaceInstanceController.controlNotStoppedAppClusterManagedInstanceTimeouts(
+                ctx,
+                notStoppedInstances,
+                installation,
+            );
+
+            log.info("Done controlling app cluster instances", {
+                installation,
+                instancesCount: notStoppedInstances.length,
+            });
+        } catch (err) {
+            log.error("Error controlling app cluster instances", err, {
+                installation,
+            });
+            TraceContext.setError(ctx, err);
+        } finally {
+            span.finish();
+        }
+    }
+
+    public dispose() {
+        this.dispoables.dispose();
+    }
+}

components/ws-manager-bridge/src/container-module.ts

@@ -39,6 +39,7 @@ import { Client } from "@gitpod/gitpod-protocol/lib/experiments/types";
 import { getExperimentsClientForBackend } from "@gitpod/gitpod-protocol/lib/experiments/configcat-server";
 import { ClusterSyncService } from "./cluster-sync-service";
 import { WorkspaceInstanceController, WorkspaceInstanceControllerImpl } from "./workspace-instance-controller";
+import { AppClusterWorkspaceInstancesController } from "./app-cluster-instance-controller";
 
 export const containerModule = new ContainerModule((bind) => {
     bind(MessagebusConfiguration).toSelf().inSingletonScope();
@@ -94,4 +95,6 @@ export const containerModule = new ContainerModule((bind) => {
     bind(Client).toDynamicValue(getExperimentsClientForBackend).inSingletonScope();
 
     bind(WorkspaceInstanceController).to(WorkspaceInstanceControllerImpl).inTransientScope();
+
+    bind(AppClusterWorkspaceInstancesController).toSelf().inSingletonScope();
 });

components/ws-manager-bridge/src/main.ts

@@ -15,6 +15,7 @@ import { TracingManager } from "@gitpod/gitpod-protocol/lib/util/tracing";
 import { ClusterServiceServer } from "./cluster-service-server";
 import { BridgeController } from "./bridge-controller";
 import { ClusterSyncService } from "./cluster-sync-service";
+import { AppClusterWorkspaceInstancesController } from "./app-cluster-instance-controller";
 
 log.enableJSONLogging("ws-manager-bridge", undefined, LogrusLogLevel.getFromEnv());
 
@@ -52,6 +53,11 @@ export const start = async (container: Container) => {
         const clusterSyncService = container.get<ClusterSyncService>(ClusterSyncService);
         clusterSyncService.start();
 
+        const appClusterInstanceController = container.get<AppClusterWorkspaceInstancesController>(
+            AppClusterWorkspaceInstancesController,
+        );
+        appClusterInstanceController.start();
+
         process.on("SIGTERM", async () => {
             log.info("SIGTERM received, stopping");
             bridgeController.dispose();
@@ -64,6 +70,7 @@ export const start = async (container: Container) => {
                 });
             }
             clusterServiceServer.stop().then(() => log.info("gRPC shutdown completed"));
+            appClusterInstanceController.dispose();
         });
         log.info("ws-manager-bridge is up and running");
         await new Promise((rs, rj) => {});