Examples with PullQueryQueue io.confluent.ksql.query.PullQueryQueue used on opensource projects

Example 1 with PullQueryQueue

use of io.confluent.ksql.query.PullQueryQueue in project ksql by confluentinc.

the class HARouting method executeRounds.

private void executeRounds(final ServiceContext serviceContext, final PullPhysicalPlan pullPhysicalPlan, final ConfiguredStatement<Query> statement, final RoutingOptions routingOptions, final LogicalSchema outputSchema, final QueryId queryId, final List<KsqlPartitionLocation> locations, final PullQueryQueue pullQueryQueue, final CompletableFuture<Void> shouldCancelRequests, final Optional<ConsistencyOffsetVector> consistencyOffsetVector) throws InterruptedException {
    final ExecutorCompletionService<PartitionFetchResult> completionService = new ExecutorCompletionService<>(routerExecutorService);
    final int totalPartitions = locations.size();
    int processedPartitions = 0;
    final Map<Integer, List<Exception>> exceptionsPerPartition = new HashMap<>();
    for (final KsqlPartitionLocation partition : locations) {
        final KsqlNode node = getNodeForRound(partition, routingOptions);
        pullQueryMetrics.ifPresent(queryExecutorMetrics -> queryExecutorMetrics.recordPartitionFetchRequest(1));
        completionService.submit(() -> routeQuery.routeQuery(node, partition, statement, serviceContext, routingOptions, pullQueryMetrics, pullPhysicalPlan, outputSchema, queryId, pullQueryQueue, shouldCancelRequests, consistencyOffsetVector));
    }
    while (processedPartitions < totalPartitions) {
        final Future<PartitionFetchResult> future = completionService.take();
        try {
            final PartitionFetchResult fetchResult = future.get();
            if (fetchResult.isError()) {
                exceptionsPerPartition.computeIfAbsent(fetchResult.location.getPartition(), v -> new ArrayList<>()).add(fetchResult.exception.get());
                final KsqlPartitionLocation nextRoundPartition = nextNode(fetchResult.getLocation());
                final KsqlNode node = getNodeForRound(nextRoundPartition, routingOptions);
                pullQueryMetrics.ifPresent(queryExecutorMetrics -> queryExecutorMetrics.recordResubmissionRequest(1));
                completionService.submit(() -> routeQuery.routeQuery(node, nextRoundPartition, statement, serviceContext, routingOptions, pullQueryMetrics, pullPhysicalPlan, outputSchema, queryId, pullQueryQueue, shouldCancelRequests, consistencyOffsetVector));
            } else {
                Preconditions.checkState(fetchResult.getResult() == RoutingResult.SUCCESS);
                processedPartitions++;
            }
        } catch (final Exception e) {
            final MaterializationException exception = new MaterializationException("Unable to execute pull query: " + e.getMessage());
            for (Entry<Integer, List<Exception>> entry : exceptionsPerPartition.entrySet()) {
                for (Exception excp : entry.getValue()) {
                    exception.addSuppressed(excp);
                }
            }
            throw exception;
        }
    }
    pullQueryQueue.close();
}

Example 2 with PullQueryQueue

use of io.confluent.ksql.query.PullQueryQueue in project ksql by confluentinc.

the class HARouting method executeOrRouteQuery.

@SuppressWarnings("ParameterNumber")
@VisibleForTesting
static PartitionFetchResult executeOrRouteQuery(final KsqlNode node, final KsqlPartitionLocation location, final ConfiguredStatement<Query> statement, final ServiceContext serviceContext, final RoutingOptions routingOptions, final Optional<PullQueryExecutorMetrics> pullQueryMetrics, final PullPhysicalPlan pullPhysicalPlan, final LogicalSchema outputSchema, final QueryId queryId, final PullQueryQueue pullQueryQueue, final CompletableFuture<Void> shouldCancelRequests, final Optional<ConsistencyOffsetVector> consistencyOffsetVector) {
    final BiFunction<List<?>, LogicalSchema, PullQueryRow> rowFactory = (rawRow, schema) -> new PullQueryRow(rawRow, schema, Optional.ofNullable(routingOptions.getIsDebugRequest() ? node : null), Optional.empty());
    if (node.isLocal()) {
        try {
            LOG.debug("Query {} executed locally at host {} at timestamp {}.", statement.getStatementText(), node.location(), System.currentTimeMillis());
            pullQueryMetrics.ifPresent(queryExecutorMetrics -> queryExecutorMetrics.recordLocalRequests(1));
            synchronized (pullPhysicalPlan) {
                pullPhysicalPlan.execute(ImmutableList.of(location), pullQueryQueue, rowFactory);
                return new PartitionFetchResult(RoutingResult.SUCCESS, location, Optional.empty());
            }
        } catch (StandbyFallbackException | NotUpToBoundException e) {
            LOG.warn("Error executing query locally at node {}. Falling back to standby state which " + "may return stale results. Cause {}", node, e.getMessage());
            return new PartitionFetchResult(RoutingResult.STANDBY_FALLBACK, location, Optional.of(e));
        } catch (Exception e) {
            throw new KsqlException(String.format("Error executing query locally at node %s: %s", node.location(), e.getMessage()), e);
        }
    } else {
        try {
            LOG.debug("Query {} routed to host {} at timestamp {}.", statement.getStatementText(), node.location(), System.currentTimeMillis());
            pullQueryMetrics.ifPresent(queryExecutorMetrics -> queryExecutorMetrics.recordRemoteRequests(1));
            forwardTo(node, ImmutableList.of(location), statement, serviceContext, pullQueryQueue, rowFactory, outputSchema, shouldCancelRequests, consistencyOffsetVector);
            return new PartitionFetchResult(RoutingResult.SUCCESS, location, Optional.empty());
        } catch (StandbyFallbackException e) {
            LOG.warn("Error forwarding query to node {}. Falling back to standby state which may " + "return stale results", node.location(), e.getCause());
            return new PartitionFetchResult(RoutingResult.STANDBY_FALLBACK, location, Optional.of(e));
        } catch (Exception e) {
            throw new KsqlException(String.format("Error forwarding query to node %s: %s", node.location(), e.getMessage()), e);
        }
    }
}

Example 3 with PullQueryQueue

use of io.confluent.ksql.query.PullQueryQueue in project ksql by confluentinc.

the class HARouting method handlePullQuery.

public CompletableFuture<Void> handlePullQuery(final ServiceContext serviceContext, final PullPhysicalPlan pullPhysicalPlan, final ConfiguredStatement<Query> statement, final RoutingOptions routingOptions, final LogicalSchema outputSchema, final QueryId queryId, final PullQueryQueue pullQueryQueue, final CompletableFuture<Void> shouldCancelRequests, final Optional<ConsistencyOffsetVector> consistencyOffsetVector) {
    final List<KsqlPartitionLocation> allLocations = pullPhysicalPlan.getMaterialization().locator().locate(pullPhysicalPlan.getKeys(), routingOptions, routingFilterFactory, pullPhysicalPlan.getPlanType() == PullPhysicalPlanType.RANGE_SCAN);
    final Map<Integer, List<Host>> emptyPartitions = allLocations.stream().filter(loc -> loc.getNodes().stream().noneMatch(node -> node.getHost().isSelected())).collect(Collectors.toMap(KsqlPartitionLocation::getPartition, loc -> loc.getNodes().stream().map(KsqlNode::getHost).collect(Collectors.toList())));
    if (!emptyPartitions.isEmpty()) {
        final MaterializationException materializationException = new MaterializationException("Unable to execute pull query. " + emptyPartitions.entrySet().stream().map(kv -> String.format("Partition %s failed to find valid host. Hosts scanned: %s", kv.getKey(), kv.getValue())).collect(Collectors.joining(", ", "[", "]")));
        LOG.debug(materializationException.getMessage());
        throw materializationException;
    }
    // at this point we should filter out the hosts that we should not route to
    final List<KsqlPartitionLocation> locations = allLocations.stream().map(KsqlPartitionLocation::removeFilteredHosts).collect(Collectors.toList());
    final CompletableFuture<Void> completableFuture = new CompletableFuture<>();
    coordinatorExecutorService.submit(() -> {
        try {
            executeRounds(serviceContext, pullPhysicalPlan, statement, routingOptions, outputSchema, queryId, locations, pullQueryQueue, shouldCancelRequests, consistencyOffsetVector);
            completableFuture.complete(null);
        } catch (Throwable t) {
            completableFuture.completeExceptionally(t);
        }
    });
    return completableFuture;
}

Example 4 with PullQueryQueue

use of io.confluent.ksql.query.PullQueryQueue in project ksql by confluentinc.

the class HARouting method streamedRowsHandler.

private static Consumer<List<StreamedRow>> streamedRowsHandler(final KsqlNode owner, final PullQueryQueue pullQueryQueue, final BiFunction<List<?>, LogicalSchema, PullQueryRow> rowFactory, final LogicalSchema outputSchema, final Optional<ConsistencyOffsetVector> consistencyOffsetVector) {
    final AtomicInteger processedRows = new AtomicInteger(0);
    final AtomicReference<Header> header = new AtomicReference<>();
    return streamedRows -> {
        try {
            if (streamedRows == null || streamedRows.isEmpty()) {
                return;
            }
            final List<PullQueryRow> rows = new ArrayList<>();
            // If this is the first row overall, skip the header
            final int previousProcessedRows = processedRows.getAndAdd(streamedRows.size());
            for (int i = 0; i < streamedRows.size(); i++) {
                final StreamedRow row = streamedRows.get(i);
                if (i == 0 && previousProcessedRows == 0) {
                    final Optional<Header> optionalHeader = row.getHeader();
                    optionalHeader.ifPresent(h -> validateSchema(outputSchema, h.getSchema(), owner));
                    optionalHeader.ifPresent(header::set);
                    continue;
                }
                if (row.getErrorMessage().isPresent()) {
                    // If we receive an error that's not a network error, we let that bubble up.
                    throw new KsqlException(row.getErrorMessage().get().getMessage());
                }
                if (!row.getRow().isPresent()) {
                    parseNonDataRows(row, i, consistencyOffsetVector);
                    continue;
                }
                final List<?> r = row.getRow().get().getColumns();
                Preconditions.checkNotNull(header.get());
                rows.add(rowFactory.apply(r, header.get().getSchema()));
            }
            if (!pullQueryQueue.acceptRows(rows)) {
                LOG.error("Failed to queue all rows");
            }
        } catch (Exception e) {
            throw new KsqlException(e.getMessage(), e);
        }
    };
}

Example 5 with PullQueryQueue

use of io.confluent.ksql.query.PullQueryQueue in project ksql by confluentinc.

the class HARoutingTest method shouldCallRouteQuery_twoRound_networkError.

@Test
public void shouldCallRouteQuery_twoRound_networkError() throws InterruptedException, ExecutionException {
    // Given:
    locate(location2);
    when(ksqlClient.makeQueryRequest(eq(node2.location()), any(), any(), any(), any(), any())).thenAnswer(i -> {
        throw new RuntimeException("Network error!");
    });
    doAnswer(i -> {
        final PullQueryQueue queue = i.getArgument(1);
        queue.acceptRow(PQ_ROW1);
        return null;
    }).when(pullPhysicalPlan).execute(eq(ImmutableList.of(location2.removeHeadHost())), any(), any());
    // When:
    CompletableFuture<Void> future = haRouting.handlePullQuery(serviceContext, pullPhysicalPlan, statement, routingOptions, logicalSchema, queryId, pullQueryQueue, disconnect, Optional.empty());
    future.get();
    // Then:
    verify(ksqlClient, times(1)).makeQueryRequest(eq(node2.location()), any(), any(), any(), any(), any());
    verify(pullPhysicalPlan).execute(eq(ImmutableList.of(location2.removeHeadHost())), any(), any());
    assertThat(pullQueryQueue.size(), is(1));
    assertThat(pullQueryQueue.pollRow(1, TimeUnit.SECONDS).getRow(), is(ROW1));
    final double fetch_count = getMetricValue("-partition-fetch-count");
    final double resubmission_count = getMetricValue("-partition-fetch-resubmission-count");
    assertThat(fetch_count, is(2.0));
    assertThat(resubmission_count, is(1.0));
}