Improve caching when expanding nodes in hierarchical data (#8902)

Fixes #8790

client/src/main/java/com/vaadin/client/data/AbstractRemoteDataSource.java

@@ -49,6 +49,9 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
      * Callback used by
      * {@link AbstractRemoteDataSource#requestRows(int, int, RequestRowsCallback)}
      * to pass data to the underlying implementation when data has been fetched.
+     *
+     * @param <T>
+     *            the row type
      */
     public static class RequestRowsCallback<T> {
         private final Range requestedRange;
@@ -56,7 +59,7 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
         private final AbstractRemoteDataSource<T> source;
 
         /**
-         * Creates a new callback
+         * Creates a new callback.
          *
          * @param source
          *            the data source for which the request is made
@@ -181,6 +184,24 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
     private final HashMap<Integer, T> indexToRowMap = new HashMap<>();
     private final HashMap<Object, Integer> keyToIndexMap = new HashMap<>();
 
+    /**
+     * Map used to temporarily store rows invalidated by
+     * {@link #insertRowData(int, int)}. All invalidated rows are stored with
+     * their indices matching the state after row insertion. If the backend has
+     * pre-emptively pushed the row data for the added rows, these rows will be
+     * used again to fill the cache.
+     * <p>
+     * To avoid this cache invalidation issue from getting too big, this class
+     * does not attempt to track these rows indefinitely. Multiple row
+     * manipulations without filling the gaps in between will remove all
+     * invalidated rows and prevent any attempts to restore them. This is
+     * indicated by having the map empty, not {@code null}.
+     * <p>
+     * The map is set to {@code null} upon requesting the rows from the backend
+     * to indicate that future row additions can attempt to restore the cache.
+     */
+    private Map<Integer, T> invalidatedRows;
+
     private Set<DataChangeHandler> dataChangeHandlers = new LinkedHashSet<>();
 
     private CacheStrategy cacheStrategy = new CacheStrategy.DefaultCacheStrategy();
@@ -277,6 +298,9 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
 
         Profiler.enter("AbstractRemoteDataSource.checkCacheCoverage");
 
+        // Clean up invalidated data
+        invalidatedRows = null;
+
         Range minCacheRange = getMinCacheRange();
 
         if (!minCacheRange.intersects(cached) || cached.isEmpty()) {
@@ -496,6 +520,8 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
                  */
                 if (!cached.isEmpty()) {
                     cached = cached.combineWith(newUsefulData);
+                    // Attempt to restore invalidated items
+                    fillCacheFromInvalidatedRows(maxCacheRange);
                 } else {
                     cached = newUsefulData;
                 }
@@ -505,6 +531,7 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
                     cached.length()));
 
             updatePinnedRows(rowData);
+
         }
 
         if (!partition[0].isEmpty() || !partition[2].isEmpty()) {
@@ -533,6 +560,58 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
 
     }
 
+    /**
+     * Go through items invalidated by {@link #insertRowData(int, int)}. If the
+     * server has pre-emptively sent added row data immediately after informing
+     * of row addition, the invalid cache can be restored to proper index range.
+     *
+     * @param maxCacheRange
+     *            the maximum amount of rows that can cached
+     */
+    private void fillCacheFromInvalidatedRows(Range maxCacheRange) {
+        if (invalidatedRows == null || invalidatedRows.isEmpty()) {
+            // No old invalid cache available
+            return;
+        }
+
+        Range potentialCache = maxCacheRange.partitionWith(cached)[2];
+        int start = potentialCache.getStart();
+        int last = start;
+        try {
+            if (potentialCache.isEmpty()
+                    || invalidatedRows.containsKey(start - 1)) {
+                // Cache is already full or invalidated rows contains unexpected
+                // indices.
+                return;
+            }
+
+            for (int i = start; i < potentialCache.getEnd(); ++i) {
+                if (!invalidatedRows.containsKey(i)) {
+                    return;
+                }
+                T row = invalidatedRows.get(i);
+                indexToRowMap.put(i, row);
+                keyToIndexMap.put(getRowKey(row), i);
+                last = i;
+            }
+
+            // Cache filled from invalidated rows. Can continue as if it was
+            // never invalidated.
+            invalidatedRows = null;
+        } finally {
+            // Update cache range and clean up
+            if (invalidatedRows != null) {
+                invalidatedRows.clear();
+            }
+
+            Range updated = Range.between(start, last + 1);
+            cached = cached.combineWith(updated);
+
+            dataChangeHandlers.forEach(dch -> dch
+                    .dataUpdated(updated.getStart(), updated.length()));
+        }
+    }
+
     private Stream<DataChangeHandler> getHandlers() {
         Set<DataChangeHandler> copy = new LinkedHashSet<>(dataChangeHandlers);
         return copy.stream();
@@ -560,6 +639,12 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
     protected void removeRowData(int firstRowIndex, int count) {
         Profiler.enter("AbstractRemoteDataSource.removeRowData");
 
+        // Cache was not filled since previous insertRowData. The old rows are
+        // no longer useful.
+        if (invalidatedRows != null) {
+            invalidatedRows.clear();
+        }
+
         size -= count;
 
         Range removedRange = Range.withLength(firstRowIndex, count);
@@ -604,6 +689,12 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
     protected void insertRowData(int firstRowIndex, int count) {
         Profiler.enter("AbstractRemoteDataSource.insertRowData");
 
+        // Cache was not filled since previous insertRowData. The old rows are
+        // no longer useful.
+        if (invalidatedRows != null) {
+            invalidatedRows.clear();
+        }
+
         size += count;
 
         if (firstRowIndex <= cached.getStart()) {
@@ -625,7 +716,24 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
              * If holes were supported, we could shift the higher part of
              * "cached" and leave a hole the size of "count" in the middle.
              */
-            cached = cached.splitAt(firstRowIndex)[0];
+            Range[] splitAt = cached.splitAt(firstRowIndex);
+            cached = splitAt[0];
+            Range invalid = splitAt[1];
+
+            /*
+             * If we already have a map in invalidatedRows, we're in a state
+             * where multiple row manipulations without data received have
+             * happened and the cache restoration is prevented completely.
+             */
+
+            if (!invalid.isEmpty() && invalidatedRows == null) {
+                invalidatedRows = new HashMap<>();
+                // Store all invalidated items to a map. Indices are updated to
+                // match what they should be after the insertion.
+                for (int i = invalid.getStart(); i < invalid.getEnd(); ++i) {
+                    invalidatedRows.put(i + count, indexToRowMap.get(i));
+                }
+            }
 
             for (int i = firstRowIndex; i < oldCacheEnd; i++) {
                 T row = indexToRowMap.remove(Integer.valueOf(i));
@@ -657,7 +765,7 @@ public abstract class AbstractRemoteDataSource<T> implements DataSource<T> {
     }
 
     /**
-     * Gets the current range of cached rows
+     * Gets the current range of cached rows.
      *
      * @return the range of currently cached rows
      */

server/src/main/java/com/vaadin/data/provider/HierarchicalDataCommunicator.java

@@ -145,38 +145,41 @@ public class HierarchicalDataCommunicator<T> extends DataCommunicator<T> {
     private void loadRequestedRows() {
         final Range requestedRows = getPushRows();
         if (!requestedRows.isEmpty()) {
-            Stream<TreeLevelQuery> levelQueries = mapper
-                    .splitRangeToLevelQueries(requestedRows.getStart(),
-                            requestedRows.getEnd() - 1);
-
-            JsonObject[] dataObjects = new JsonObject[requestedRows.length()];
-            BiConsumer<JsonObject, Integer> rowDataMapper = (object,
-                    index) -> dataObjects[index
-                            - requestedRows.getStart()] = object;
-            List<T> fetchedItems = new ArrayList<>(dataObjects.length);
-
-            levelQueries.forEach(query -> {
-                List<T> results = doFetchQuery(query.startIndex, query.size,
-                        getKeyMapper().get(query.node.getParentKey()))
-                                .collect(Collectors.toList());
-                // TODO if the size differers from expected, all goes to hell
-                fetchedItems.addAll(results);
-                List<JsonObject> rowData = results.stream()
-                        .map(item -> createDataObject(item, query.depth))
-                        .collect(Collectors.toList());
-                mapper.reorderLevelQueryResultsToFlatOrdering(rowDataMapper,
-                        query, rowData);
-            });
-            verifyNoNullItems(dataObjects, requestedRows);
-
-            sendData(requestedRows.getStart(), Arrays.asList(dataObjects));
-            getActiveDataHandler().addActiveData(fetchedItems.stream());
-            getActiveDataHandler().cleanUp(fetchedItems.stream());
+            doPushRows(requestedRows);
         }
 
         setPushRows(Range.withLength(0, 0));
     }
 
+    private void doPushRows(final Range requestedRows) {
+        Stream<TreeLevelQuery> levelQueries = mapper.splitRangeToLevelQueries(
+                requestedRows.getStart(), requestedRows.getEnd() - 1);
+
+        JsonObject[] dataObjects = new JsonObject[requestedRows.length()];
+        BiConsumer<JsonObject, Integer> rowDataMapper = (object,
+                index) -> dataObjects[index
+                        - requestedRows.getStart()] = object;
+        List<T> fetchedItems = new ArrayList<>(dataObjects.length);
+
+        levelQueries.forEach(query -> {
+            List<T> results = doFetchQuery(query.startIndex, query.size,
+                    getKeyMapper().get(query.node.getParentKey()))
+                            .collect(Collectors.toList());
+            // TODO if the size differers from expected, all goes to hell
+            fetchedItems.addAll(results);
+            List<JsonObject> rowData = results.stream()
+                    .map(item -> createDataObject(item, query.depth))
+                    .collect(Collectors.toList());
+            mapper.reorderLevelQueryResultsToFlatOrdering(rowDataMapper, query,
+                    rowData);
+        });
+        verifyNoNullItems(dataObjects, requestedRows);
+
+        sendData(requestedRows.getStart(), Arrays.asList(dataObjects));
+        getActiveDataHandler().addActiveData(fetchedItems.stream());
+        getActiveDataHandler().cleanUp(fetchedItems.stream());
+    }
+
     /*
      * Verify that there are no null objects in the array, to fail eagerly and
      * not just on the client side.
@@ -387,8 +390,10 @@ public class HierarchicalDataCommunicator<T> extends DataCommunicator<T> {
 
         mapper.expand(expandedRowKey, expandedRowIndex, expandedNodeSize);
 
-        // TODO optimize by sending "enough" of the expanded items directly
         getClientRpc().insertRows(expandedRowIndex + 1, expandedNodeSize);
+        // TODO optimize by sending "just enough" of the expanded items directly
+        doPushRows(Range.withLength(expandedRowIndex + 1, expandedNodeSize));
+
         // expanded node needs to be updated to be marked as expanded
         // FIXME seems like a slight overkill to do this just for refreshing
         // expanded status

uitest/src/main/java/com/vaadin/tests/components/treegrid/TreeGridBasicFeatures.java

@@ -3,13 +3,18 @@ package com.vaadin.tests.components.treegrid;
 import java.util.Arrays;
 import java.util.LinkedHashMap;
 import java.util.List;
+import java.util.Optional;
+import java.util.stream.Stream;
 
 import com.vaadin.annotations.Theme;
 import com.vaadin.annotations.Widgetset;
 import com.vaadin.data.HierarchyData;
 import com.vaadin.data.provider.DataProvider;
+import com.vaadin.data.provider.HierarchicalDataProvider;
+import com.vaadin.data.provider.HierarchicalQuery;
 import com.vaadin.data.provider.InMemoryHierarchicalDataProvider;
 import com.vaadin.server.SerializablePredicate;
+import com.vaadin.shared.Range;
 import com.vaadin.tests.components.AbstractComponentTest;
 import com.vaadin.ui.TreeGrid;
 
@@ -20,6 +25,7 @@ public class TreeGridBasicFeatures extends AbstractComponentTest<TreeGrid> {
     private TreeGrid<HierarchicalTestBean> grid;
     private InMemoryHierarchicalDataProvider<HierarchicalTestBean> inMemoryDataProvider;
     private LazyHierarchicalDataProvider lazyDataProvider;
+    private HierarchicalDataProvider<HierarchicalTestBean, ?> loggingDataProvider;
 
     @Override
     public TreeGrid getComponent() {
@@ -82,6 +88,26 @@ public class TreeGridBasicFeatures extends AbstractComponentTest<TreeGrid> {
 
         inMemoryDataProvider = new InMemoryHierarchicalDataProvider<>(data);
         lazyDataProvider = new LazyHierarchicalDataProvider(3, 2);
+        loggingDataProvider = new InMemoryHierarchicalDataProvider<HierarchicalTestBean>(
+                data) {
+
+            @Override
+            public Stream<HierarchicalTestBean> fetchChildren(
+                    HierarchicalQuery<HierarchicalTestBean, SerializablePredicate<HierarchicalTestBean>> query) {
+                Optional<HierarchicalTestBean> parentOptional = query
+                        .getParentOptional();
+                if (parentOptional.isPresent()) {
+                    log("Children request: " + parentOptional.get() + " ; "
+                            + Range.withLength(query.getOffset(),
+                                    query.getLimit()));
+                } else {
+                    log("Root node request: " + Range
+                            .withLength(query.getOffset(), query.getLimit()));
+                }
+                return super.fetchChildren(query);
+            }
+        };
+
     }
 
     @SuppressWarnings("unchecked")
@@ -90,6 +116,7 @@ public class TreeGridBasicFeatures extends AbstractComponentTest<TreeGrid> {
         LinkedHashMap<String, DataProvider> options = new LinkedHashMap<>();
         options.put("LazyHierarchicalDataProvider", lazyDataProvider);
         options.put("InMemoryHierarchicalDataProvider", inMemoryDataProvider);
+        options.put("LoggingDataProvider", loggingDataProvider);
 
         createSelectAction("Set data provider", CATEGORY_FEATURES, options,
                 "LazyHierarchicalDataProvider",

uitest/src/test/java/com/vaadin/tests/components/treegrid/TreeGridExpandDataRequestTest.java

@@ -0,0 +1,52 @@
+package com.vaadin.tests.components.treegrid;
+
+import org.junit.Assert;
+import org.junit.Before;
+import org.junit.Test;
+
+import com.vaadin.testbench.elements.TreeGridElement;
+import com.vaadin.tests.tb3.SingleBrowserTest;
+
+public class TreeGridExpandDataRequestTest extends SingleBrowserTest {
+
+    @Override
+    protected Class<?> getUIClass() {
+        return TreeGridBasicFeatures.class;
+    }
+
+    TreeGridElement grid;
+
+    @Before
+    public void before() {
+        openTestURL();
+        grid = $(TreeGridElement.class).first();
+        selectMenuPath("Component", "Features", "Set data provider",
+                "LoggingDataProvider");
+        clearLog();
+    }
+
+    private void clearLog() {
+        selectMenuPath("Settings", "Clear log");
+    }
+
+    @Test
+    public void expand_node0_does_not_request_root_nodes() {
+        grid.expandWithClick(0);
+        Assert.assertFalse("Log should not contain request for root nodes.",
+                logContainsText("Root node request: "));
+    }
+
+    @Test
+    public void expand_node0_after_node1_does_not_request_children_of_node1() {
+        grid.expandWithClick(1);
+        Assert.assertFalse("Log should not contain request for root nodes.",
+                logContainsText("Root node request: "));
+        clearLog();
+        grid.expandWithClick(0);
+        Assert.assertFalse(
+                "Log should not contain request for children of '0 | 1'.",
+                logContainsText("Children request: 0 | 1"));
+        Assert.assertFalse("Log should not contain request for root nodes.",
+                logContainsText("Root node request: "));
+    }
+}