Removed GC blobs from channel context (DDS). The GC blobs in data sto…

…re is used to generate initial GC data for DDS (#5007)

packages/runtime/container-runtime/src/dataStoreContext.ts

@@ -621,7 +621,7 @@ export abstract class FluidDataStoreContext extends TypedEventEmitter<IFluidData
 
     protected abstract getInitialSnapshotDetails(): Promise<ISnapshotDetails>;
 
-    protected abstract getInitialGCSummaryDetails(): Promise<IGarbageCollectionSummaryDetails>;
+    public abstract getInitialGCSummaryDetails(): Promise<IGarbageCollectionSummaryDetails>;
 
     public reSubmit(contents: any, localOpMetadata: unknown) {
         assert(!!this.channel, "Channel must exist when resubmitting ops");
@@ -768,7 +768,7 @@ export class RemotedFluidDataStoreContext extends FluidDataStoreContext {
         return this.initialSnapshotDetailsP;
     }
 
-    protected async getInitialGCSummaryDetails(): Promise<IGarbageCollectionSummaryDetails> {
+    public async getInitialGCSummaryDetails(): Promise<IGarbageCollectionSummaryDetails> {
         return this.gcDetailsInInitialSummaryP;
     }
 
@@ -861,7 +861,7 @@ export class LocalFluidDataStoreContextBase extends FluidDataStoreContext {
         };
     }
 
-    protected async getInitialGCSummaryDetails(): Promise<IGarbageCollectionSummaryDetails> {
+    public async getInitialGCSummaryDetails(): Promise<IGarbageCollectionSummaryDetails> {
         // Local data store does not have initial summary.
         return {};
     }

packages/runtime/datastore/src/channelContext.ts

@@ -7,11 +7,9 @@ import { IChannel } from "@fluidframework/datastore-definitions";
 import { IDocumentStorageService } from "@fluidframework/driver-definitions";
 import { ISequencedDocumentMessage, ISnapshotTree } from "@fluidframework/protocol-definitions";
 import {
-    gcBlobKey,
     IChannelSummarizeResult,
     IContextSummarizeResult,
     IGarbageCollectionData,
-    IGarbageCollectionSummaryDetails,
 } from "@fluidframework/runtime-definitions";
 import { addBlobToSummary } from "@fluidframework/runtime-utils";
 import { ChannelDeltaConnection } from "./channelDeltaConnection";
@@ -76,13 +74,5 @@ export function summarizeChannel(
 
     // Add the channel attributes to the returned result.
     addBlobToSummary(summarizeResult, attributesBlobKey, JSON.stringify(channel.attributes));
-
-    // Add GC details to the summary.
-    const gcDetails: IGarbageCollectionSummaryDetails = {
-        usedRoutes: [""],
-        gcData: summarizeResult.gcData,
-    };
-    addBlobToSummary(summarizeResult, gcBlobKey, JSON.stringify(gcDetails));
-
     return summarizeResult;
 }

packages/runtime/datastore/src/dataStoreRuntime.ts

@@ -22,6 +22,7 @@ import {
 import {
     assert,
     Deferred,
+    LazyPromise,
     TypedEventEmitter,
     unreachableCase,
 } from "@fluidframework/common-utils";
@@ -48,6 +49,7 @@ import {
     IFluidDataStoreContext,
     IFluidDataStoreChannel,
     IGarbageCollectionData,
+    IGarbageCollectionSummaryDetails,
     IInboundSignalMessage,
     ISummaryTreeWithStats,
 } from "@fluidframework/runtime-definitions";
@@ -66,7 +68,13 @@ import {
     IChannelFactory,
     IChannelAttributes,
 } from "@fluidframework/datastore-definitions";
-import {  GCDataBuilder, getChildNodesUsedRoutes } from "@fluidframework/garbage-collector";
+import {
+    cloneGCData,
+    GCDataBuilder,
+    getChildNodesGCData,
+    getChildNodesUsedRoutes,
+    removeRouteFromAllNodes,
+} from "@fluidframework/garbage-collector";
 import { v4 as uuid } from "uuid";
 import { IChannelContext, summarizeChannel } from "./channelContext";
 import { LocalChannelContext } from "./localChannelContext";
@@ -183,6 +191,14 @@ IFluidDataStoreChannel, IFluidDataStoreRuntime, IFluidHandleContext {
     private readonly audience: IAudience;
     public readonly logger: ITelemetryLogger;
 
+    // A map of child channel context ids to the context's used routes in the initial summary of this data store. This
+    // is used to initialize the context with data from the previous summary.
+    private readonly initialChannelUsedRoutesP: LazyPromise<Map<string, string[]>>;
+
+    // A map of child channel context ids to the channel context's GC data in the initial summary of this data store.
+    // This is used to initialize the context with data from the previous summary.
+    private readonly initialChannelGCDataP: LazyPromise<Map<string, IGarbageCollectionData>>;
+
     public constructor(
         private readonly dataStoreContext: IFluidDataStoreContext,
         private readonly sharedObjectRegistry: ISharedObjectRegistry,
@@ -200,6 +216,35 @@ IFluidDataStoreChannel, IFluidDataStoreRuntime, IFluidHandleContext {
 
         const tree = dataStoreContext.baseSnapshot;
 
+        this.initialChannelUsedRoutesP = new LazyPromise(async () => {
+            // back-compat: 0.35.0. getInitialGCSummaryDetails is added to IFluidDataStoreContext in 0.35.0. Remove
+            // undefined check when N > 0.36.0.
+            const gcDetailsInInitialSummary = await this.dataStoreContext.getInitialGCSummaryDetails?.();
+            if (gcDetailsInInitialSummary?.usedRoutes !== undefined) {
+                // Remove the route to this data store, if it exists.
+                const usedRoutes = gcDetailsInInitialSummary.usedRoutes.filter(
+                    (id: string) => { return id !== "/" && id !== ""; },
+                );
+                return getChildNodesUsedRoutes(usedRoutes);
+            }
+            return new Map();
+        });
+
+        this.initialChannelGCDataP = new LazyPromise(async () => {
+            // back-compat: 0.35.0. getInitialGCSummaryDetails is added to IFluidDataStoreContext in 0.35.0. Remove
+            // undefined check when N > 0.36.0.
+            const gcDetailsInInitialSummary = await this.dataStoreContext.getInitialGCSummaryDetails?.();
+            if (gcDetailsInInitialSummary?.gcData !== undefined) {
+                const gcData = cloneGCData(gcDetailsInInitialSummary.gcData);
+                // Remove GC node for this data store, if any.
+                delete gcData.gcNodes["/"];
+                // Remove the back route to this data store that was added when generating each child's GC nodes.
+                removeRouteFromAllNodes(gcData.gcNodes, this.absolutePath);
+                return getChildNodesGCData(gcData);
+            }
+            return new Map();
+        });
+
         // Must always receive the data store type inside of the attributes
         if (tree?.trees !== undefined) {
             Object.keys(tree.trees).forEach((path) => {
@@ -246,7 +291,8 @@ IFluidDataStoreChannel, IFluidDataStoreRuntime, IFluidHandleContext {
                         this.dataStoreContext.getCreateChildSummarizerNodeFn(
                             path,
                             { type: CreateSummarizerNodeSource.FromSummary },
-                        ));
+                        ),
+                        async () => this.getChannelInitialGCDetails(path));
                 }
                 const deferred = new Deferred<IChannelContext>();
                 deferred.resolve(channelContext);
@@ -512,6 +558,7 @@ IFluidDataStoreChannel, IFluidDataStoreRuntime, IFluidHandleContext {
                                 snapshot: attachMessage.snapshot,
                             },
                         ),
+                        async () => this.getChannelInitialGCDetails(id),
                         attachMessage.type);
 
                     this.contexts.set(id, remoteChannelContext);
@@ -638,6 +685,30 @@ IFluidDataStoreChannel, IFluidDataStoreRuntime, IFluidHandleContext {
         }
     }
 
+    /**
+     * Returns the GC details in initial summary for the channel with the given id. The initial summary of the data
+     * store contains the GC details of all the child channel contexts that were created before the summary was taken.
+     * We find the GC details belonging to the given channel context and return it.
+     * @param channelId - The id of the channel context that is asked for the initial GC details.
+     * @returns the requested channel's GC details in the initial summary.
+     */
+    private async getChannelInitialGCDetails(channelId: string): Promise<IGarbageCollectionSummaryDetails> {
+        const channelInitialUsedRoutes = await this.initialChannelUsedRoutesP;
+        const channelInitialGCData = await this.initialChannelGCDataP;
+
+        let channelUsedRoutes = channelInitialUsedRoutes.get(channelId);
+        // Currently, channel context's are always considered used. So, it there is no used route for it, we still
+        // need to mark it as used. Add self-route (empty string) to the channel context's used routes.
+        if (channelUsedRoutes === undefined || channelUsedRoutes.length === 0) {
+            channelUsedRoutes = [""];
+        }
+
+        return {
+            usedRoutes: channelUsedRoutes,
+            gcData: channelInitialGCData.get(channelId),
+        };
+    }
+
     /**
      * Returns a summary at the current sequence number.
      * @param fullTree - true to bypass optimizations and force a full summary tree

packages/runtime/datastore/src/remoteChannelContext.ts

@@ -3,7 +3,7 @@
  * Licensed under the MIT License.
  */
 
-import { assert, LazyPromise } from "@fluidframework/common-utils";
+import { assert } from "@fluidframework/common-utils";
 import { CreateContainerError, DataCorruptionError } from "@fluidframework/container-utils";
 import {
     IChannel,
@@ -19,7 +19,6 @@ import {
 } from "@fluidframework/protocol-definitions";
 import {
     CreateChildSummarizerNodeFn,
-    gcBlobKey,
     IContextSummarizeResult,
     IFluidDataStoreContext,
     IGarbageCollectionData,
@@ -49,17 +48,6 @@ export class RemoteChannelContext implements IChannelContext {
     };
     private readonly summarizerNode: ISummarizerNodeWithGC;
 
-    /**
-     * This loads the GC details from the base snapshot of this context.
-     */
-    private readonly gcDetailsInInitialSummaryP = new LazyPromise<IGarbageCollectionSummaryDetails>(async () => {
-        if (await this.services.objectStorage.contains(gcBlobKey)) {
-            return readAndParse<IGarbageCollectionSummaryDetails>(this.services.objectStorage, gcBlobKey);
-        } else {
-            return {};
-        }
-    });
-
     constructor(
         private readonly runtime: IFluidDataStoreRuntime,
         private readonly dataStoreContext: IFluidDataStoreContext,
@@ -71,6 +59,7 @@ export class RemoteChannelContext implements IChannelContext {
         private readonly registry: ISharedObjectRegistry,
         extraBlobs: Map<string, string> | undefined,
         createSummarizerNode: CreateChildSummarizerNodeFn,
+        gcDetailsInInitialSummary: () => Promise<IGarbageCollectionSummaryDetails>,
         private readonly attachMessageType?: string,
     ) {
         this.services = createServiceEndpoints(
@@ -88,7 +77,7 @@ export class RemoteChannelContext implements IChannelContext {
         this.summarizerNode = createSummarizerNode(
             thisSummarizeInternal,
             async () => this.getGCDataInternal(),
-            async () => this.gcDetailsInInitialSummaryP,
+            async () => gcDetailsInInitialSummary(),
         );
     }
 

packages/runtime/garbage-collector/src/utils.ts

@@ -22,25 +22,69 @@ export function cloneGCData(gcData: IGarbageCollectionData): IGarbageCollectionD
 }
 
 /**
- * Helper function that generates the used routes of the children from a given node's used routes.
+ * Helper function that generates the used routes of children from a given node's used routes.
  * @param usedRoutes - The used routes of a node.
  * @returns A map of used routes of each children of the the given node.
  */
 export function getChildNodesUsedRoutes(usedRoutes: string[]) {
-    const usedNodesRoutes: Map<string, string[]> = new Map();
+    const childUsedRoutesMap: Map<string, string[]> = new Map();
     for (const route of usedRoutes) {
         assert(route.startsWith("/"), "Used route should always be an absolute route");
         const childId = route.split("/")[1];
         const childUsedRoute = route.slice(childId.length + 1);
 
-        const childUsedRoutes = usedNodesRoutes.get(childId);
+        const childUsedRoutes = childUsedRoutesMap.get(childId);
         if (childUsedRoutes !== undefined) {
             childUsedRoutes.push(childUsedRoute);
         } else {
-            usedNodesRoutes.set(childId, [ childUsedRoute ]);
+            childUsedRoutesMap.set(childId, [ childUsedRoute ]);
+        }
+    }
+    return childUsedRoutesMap;
+}
+
+/**
+ * Helper function that generates the GC data of children from a given node's GC data.
+ * @param gcData - The GC data of a node.
+ * @returns A map of GC data of each children of the the given node.
+ */
+export function getChildNodesGCData(gcData: IGarbageCollectionData) {
+    const childGCDataMap: Map<string, IGarbageCollectionData> = new Map();
+    for (const [id, outboundRoutes] of Object.entries(gcData.gcNodes)) {
+        assert(id.startsWith("/"), "id should always be an absolute route");
+        const childId = id.split("/")[1];
+        let childGCNodeId = id.slice(childId.length + 1);
+        // GC node id always begins with "/". Handle the special case where the id in parent's GC nodes is of the
+        // for `/root`. This would make `childId=root` and `childGCNodeId=""`.
+        if (childGCNodeId === "") {
+            childGCNodeId = "/";
+        }
+
+        // Create a copy of the outbound routes array in the parents GC data.
+        const childOutboundRoutes = Array.from(outboundRoutes);
+
+        let childGCData = childGCDataMap.get(childId);
+        if (childGCData === undefined) {
+            childGCData = { gcNodes: {} };
+        }
+        childGCData.gcNodes[childGCNodeId] = childOutboundRoutes;
+        childGCDataMap.set(childId, childGCData);
+    }
+    return childGCDataMap;
+}
+
+/**
+ * Removes the given route from the outbound routes of all the given GC nodes.
+ * @param gcNodes - The nodes from which the route is to be removed.
+ * @param outboundRoute - The route to be removed.
+ */
+export function removeRouteFromAllNodes(gcNodes: { [ id: string ]: string[] }, outboundRoute: string) {
+    for (const outboundRoutes of Object.values(gcNodes)) {
+        const index = outboundRoutes.indexOf(outboundRoute);
+        if (index > -1) {
+            outboundRoutes.splice(index, 1);
         }
     }
-    return usedNodesRoutes;
 }
 
 export class GCDataBuilder implements IGarbageCollectionData {

packages/runtime/runtime-definitions/src/dataStoreContext.ts

@@ -347,6 +347,12 @@ IEventProvider<IFluidDataStoreContextEvents>, Partial<IProvideFluidDataStoreRegi
     ): CreateChildSummarizerNodeFn;
 
     uploadBlob(blob: ArrayBufferLike): Promise<IFluidHandle<ArrayBufferLike>>;
+
+    /**
+     * Returns the GC details in the initial summary of this data store. This is used to initialize the data store
+     * and its children with the GC details from the previous summary.
+     */
+    getInitialGCSummaryDetails(): Promise<IGarbageCollectionSummaryDetails>;
 }
 
 export interface IFluidDataStoreContextDetached extends IFluidDataStoreContext {