Examples with GroupRowsByShard io.crate.execution.engine.indexing.GroupRowsByShard used on opensource projects

Example 1 with GroupRowsByShard

use of io.crate.execution.engine.indexing.GroupRowsByShard in project crate by crate.

the class InsertFromValues method executeBulk.

@Override
public List<CompletableFuture<Long>> executeBulk(DependencyCarrier dependencies, PlannerContext plannerContext, List<Row> bulkParams, SubQueryResults subQueryResults) {
    DocTableInfo tableInfo = dependencies.schemas().getTableInfo(writerProjection.tableIdent(), Operation.INSERT);
    String[] updateColumnNames;
    Assignments assignments;
    if (writerProjection.onDuplicateKeyAssignments() == null) {
        assignments = null;
        updateColumnNames = null;
    } else {
        assignments = Assignments.convert(writerProjection.onDuplicateKeyAssignments(), dependencies.nodeContext());
        updateColumnNames = assignments.targetNames();
    }
    InputFactory inputFactory = new InputFactory(dependencies.nodeContext());
    InputFactory.Context<CollectExpression<Row, ?>> context = inputFactory.ctxForInputColumns(plannerContext.transactionContext());
    var allColumnSymbols = InputColumns.create(writerProjection.allTargetColumns(), new InputColumns.SourceSymbols(writerProjection.allTargetColumns()));
    ArrayList<Input<?>> insertInputs = new ArrayList<>(allColumnSymbols.size());
    for (Symbol symbol : allColumnSymbols) {
        insertInputs.add(context.add(symbol));
    }
    ArrayList<Input<?>> partitionedByInputs = new ArrayList<>(writerProjection.partitionedBySymbols().size());
    for (Symbol partitionedBySymbol : writerProjection.partitionedBySymbols()) {
        partitionedByInputs.add(context.add(partitionedBySymbol));
    }
    ArrayList<Input<?>> primaryKeyInputs = new ArrayList<>(writerProjection.ids().size());
    for (Symbol symbol : writerProjection.ids()) {
        primaryKeyInputs.add(context.add(symbol));
    }
    Input<?> clusterByInput;
    if (writerProjection.clusteredBy() != null) {
        clusterByInput = context.add(writerProjection.clusteredBy());
    } else {
        clusterByInput = null;
    }
    var indexNameResolver = IndexNameResolver.create(writerProjection.tableIdent(), writerProjection.partitionIdent(), partitionedByInputs);
    ShardUpsertRequest.Builder builder = new ShardUpsertRequest.Builder(plannerContext.transactionContext().sessionSettings(), BULK_REQUEST_TIMEOUT_SETTING.get(dependencies.settings()), writerProjection.isIgnoreDuplicateKeys() ? ShardUpsertRequest.DuplicateKeyAction.IGNORE : ShardUpsertRequest.DuplicateKeyAction.UPDATE_OR_FAIL, // continueOnErrors
    true, updateColumnNames, writerProjection.allTargetColumns().toArray(new Reference[0]), null, plannerContext.jobId(), true);
    var shardedRequests = new ShardedRequests<>(builder::newRequest, RamAccounting.NO_ACCOUNTING);
    HashMap<String, InsertSourceFromCells> validatorsCache = new HashMap<>();
    IntArrayList bulkIndices = new IntArrayList();
    List<CompletableFuture<Long>> results = createUnsetFutures(bulkParams.size());
    for (int bulkIdx = 0; bulkIdx < bulkParams.size(); bulkIdx++) {
        Row param = bulkParams.get(bulkIdx);
        final Symbol[] assignmentSources;
        if (assignments != null) {
            assignmentSources = assignments.bindSources(tableInfo, param, subQueryResults);
        } else {
            assignmentSources = null;
        }
        GroupRowsByShard<ShardUpsertRequest, ShardUpsertRequest.Item> grouper = createRowsByShardGrouper(assignmentSources, insertInputs, indexNameResolver, context, plannerContext, dependencies.clusterService());
        try {
            Iterator<Row> rows = evaluateValueTableFunction(tableFunctionRelation.functionImplementation(), tableFunctionRelation.function().arguments(), writerProjection.allTargetColumns(), tableInfo, param, plannerContext, subQueryResults);
            while (rows.hasNext()) {
                Row row = rows.next();
                grouper.accept(shardedRequests, row);
                checkPrimaryKeyValuesNotNull(primaryKeyInputs);
                checkClusterByValueNotNull(clusterByInput);
                checkConstraintsOnGeneratedSource(row.materialize(), indexNameResolver.get(), tableInfo, plannerContext, validatorsCache);
                bulkIndices.add(bulkIdx);
            }
        } catch (Throwable t) {
            for (CompletableFuture<Long> result : results) {
                result.completeExceptionally(t);
            }
            return results;
        }
    }
    validatorsCache.clear();
    var actionProvider = dependencies.transportActionProvider();
    createIndices(actionProvider.transportBulkCreateIndicesAction(), shardedRequests.itemsByMissingIndex().keySet(), dependencies.clusterService(), plannerContext.jobId()).thenCompose(acknowledgedResponse -> {
        var shardUpsertRequests = resolveAndGroupShardRequests(shardedRequests, dependencies.clusterService()).values();
        return execute(dependencies.nodeLimits(), dependencies.clusterService().state(), shardUpsertRequests, actionProvider.transportShardUpsertAction(), dependencies.scheduler());
    }).whenComplete((response, t) -> {
        if (t == null) {
            long[] resultRowCount = createBulkResponse(response, bulkParams.size(), bulkIndices);
            for (int i = 0; i < bulkParams.size(); i++) {
                results.get(i).complete(resultRowCount[i]);
            }
        } else {
            for (CompletableFuture<Long> result : results) {
                result.completeExceptionally(t);
            }
        }
    });
    return results;
}

Example 2 with GroupRowsByShard

use of io.crate.execution.engine.indexing.GroupRowsByShard in project crate by crate.

the class InsertFromValues method execute.

@Override
public void execute(DependencyCarrier dependencies, PlannerContext plannerContext, RowConsumer consumer, Row params, SubQueryResults subQueryResults) {
    DocTableInfo tableInfo = dependencies.schemas().getTableInfo(writerProjection.tableIdent(), Operation.INSERT);
    // For instance, the target table of the insert from values
    // statement is the table with the following schema:
    // 
    // CREATE TABLE users (
    // dep_id TEXT,
    // name TEXT,
    // id INT,
    // country_id INT,
    // PRIMARY KEY (dep_id, id, country_id))
    // CLUSTERED BY (dep_id)
    // PARTITIONED BY (country_id)
    // 
    // The insert from values statement below would have the column
    // index writer projection of its plan that contains the column
    // idents and symbols required to create corresponding inputs.
    // The diagram below shows the projection's column symbols used
    // in the plan and relation between symbols sub-/sets.
    // 
    // +------------------------+
    // |          +-------------+  PK symbols
    // cluster by +------+ |          |      +------+
    // symbol            | |          |      |
    // + +          +      +
    // INSERT INTO users (dep_id, name, id, country_id) VALUES (?, ?, ?, ?)
    // +      +    +     +   +
    // +-------+      |    |     |   |
    // all target  +--------------+    |     |   +---+  partitioned by
    // column      +-------------------+     |          symbols
    // symbols     +-------------------------+
    InputFactory inputFactory = new InputFactory(dependencies.nodeContext());
    InputFactory.Context<CollectExpression<Row, ?>> context = inputFactory.ctxForInputColumns(plannerContext.transactionContext());
    var allColumnSymbols = InputColumns.create(writerProjection.allTargetColumns(), new InputColumns.SourceSymbols(writerProjection.allTargetColumns()));
    ArrayList<Input<?>> insertInputs = new ArrayList<>(allColumnSymbols.size());
    for (Symbol symbol : allColumnSymbols) {
        insertInputs.add(context.add(symbol));
    }
    ArrayList<Input<?>> partitionedByInputs = new ArrayList<>(writerProjection.partitionedBySymbols().size());
    for (Symbol partitionedBySymbol : writerProjection.partitionedBySymbols()) {
        partitionedByInputs.add(context.add(partitionedBySymbol));
    }
    ArrayList<Input<?>> primaryKeyInputs = new ArrayList<>(writerProjection.ids().size());
    for (Symbol symbol : writerProjection.ids()) {
        primaryKeyInputs.add(context.add(symbol));
    }
    Input<?> clusterByInput;
    if (writerProjection.clusteredBy() != null) {
        clusterByInput = context.add(writerProjection.clusteredBy());
    } else {
        clusterByInput = null;
    }
    String[] updateColumnNames;
    Symbol[] assignmentSources;
    if (writerProjection.onDuplicateKeyAssignments() == null) {
        updateColumnNames = null;
        assignmentSources = null;
    } else {
        Assignments assignments = Assignments.convert(writerProjection.onDuplicateKeyAssignments(), dependencies.nodeContext());
        assignmentSources = assignments.bindSources(tableInfo, params, subQueryResults);
        updateColumnNames = assignments.targetNames();
    }
    var indexNameResolver = IndexNameResolver.create(writerProjection.tableIdent(), writerProjection.partitionIdent(), partitionedByInputs);
    GroupRowsByShard<ShardUpsertRequest, ShardUpsertRequest.Item> grouper = createRowsByShardGrouper(assignmentSources, insertInputs, indexNameResolver, context, plannerContext, dependencies.clusterService());
    ArrayList<Row> rows = new ArrayList<>();
    evaluateValueTableFunction(tableFunctionRelation.functionImplementation(), tableFunctionRelation.function().arguments(), writerProjection.allTargetColumns(), tableInfo, params, plannerContext, subQueryResults).forEachRemaining(rows::add);
    List<Symbol> returnValues = this.writerProjection.returnValues();
    ShardUpsertRequest.Builder builder = new ShardUpsertRequest.Builder(plannerContext.transactionContext().sessionSettings(), BULK_REQUEST_TIMEOUT_SETTING.get(dependencies.settings()), writerProjection.isIgnoreDuplicateKeys() ? ShardUpsertRequest.DuplicateKeyAction.IGNORE : ShardUpsertRequest.DuplicateKeyAction.UPDATE_OR_FAIL, // continueOnErrors
    rows.size() > 1, updateColumnNames, writerProjection.allTargetColumns().toArray(new Reference[0]), returnValues.isEmpty() ? null : returnValues.toArray(new Symbol[0]), plannerContext.jobId(), false);
    var shardedRequests = new ShardedRequests<>(builder::newRequest, RamAccounting.NO_ACCOUNTING);
    HashMap<String, InsertSourceFromCells> validatorsCache = new HashMap<>();
    for (Row row : rows) {
        grouper.accept(shardedRequests, row);
        try {
            checkPrimaryKeyValuesNotNull(primaryKeyInputs);
            checkClusterByValueNotNull(clusterByInput);
            checkConstraintsOnGeneratedSource(row.materialize(), indexNameResolver.get(), tableInfo, plannerContext, validatorsCache);
        } catch (Throwable t) {
            consumer.accept(null, t);
            return;
        }
    }
    validatorsCache.clear();
    var actionProvider = dependencies.transportActionProvider();
    createIndices(actionProvider.transportBulkCreateIndicesAction(), shardedRequests.itemsByMissingIndex().keySet(), dependencies.clusterService(), plannerContext.jobId()).thenCompose(acknowledgedResponse -> {
        var shardUpsertRequests = resolveAndGroupShardRequests(shardedRequests, dependencies.clusterService()).values();
        return execute(dependencies.nodeLimits(), dependencies.clusterService().state(), shardUpsertRequests, actionProvider.transportShardUpsertAction(), dependencies.scheduler());
    }).whenComplete((response, t) -> {
        if (t == null) {
            if (returnValues.isEmpty()) {
                consumer.accept(InMemoryBatchIterator.of(new Row1((long) response.numSuccessfulWrites()), SENTINEL), null);
            } else {
                consumer.accept(InMemoryBatchIterator.of(new CollectionBucket(response.resultRows()), SENTINEL, false), null);
            }
        } else {
            consumer.accept(null, t);
        }
    });
}

Example 3 with GroupRowsByShard

use of io.crate.execution.engine.indexing.GroupRowsByShard in project crate by crate.

the class InsertFromValues method createRowsByShardGrouper.

private GroupRowsByShard<ShardUpsertRequest, ShardUpsertRequest.Item> createRowsByShardGrouper(Symbol[] assignmentSources, ArrayList<Input<?>> insertInputs, Supplier<String> indexNameResolver, InputFactory.Context<CollectExpression<Row, ?>> collectContext, PlannerContext plannerContext, ClusterService clusterService) {
    InputRow insertValues = new InputRow(insertInputs);
    Function<String, ShardUpsertRequest.Item> itemFactory = id -> new ShardUpsertRequest.Item(id, assignmentSources, insertValues.materialize(), null, null, null);
    var rowShardResolver = new RowShardResolver(plannerContext.transactionContext(), plannerContext.nodeContext(), writerProjection.primaryKeys(), writerProjection.ids(), writerProjection.clusteredByIdent(), writerProjection.clusteredBy());
    return new GroupRowsByShard<>(clusterService, rowShardResolver, new TypeGuessEstimateRowSize(), indexNameResolver, collectContext.expressions(), itemFactory, true);
}