org.apache.beam.sdk.nexmark.NexmarkUtils.java Source code

Java tutorial

Introduction

Here is the source code for org.apache.beam.sdk.nexmark.NexmarkUtils.java

Source

/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
package org.apache.beam.sdk.nexmark;

import static org.apache.beam.vendor.guava.v26_0_jre.com.google.common.base.Preconditions.checkArgument;

import com.fasterxml.jackson.databind.ObjectMapper;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.charset.StandardCharsets;
import java.util.Iterator;
import java.util.List;
import java.util.stream.Collectors;
import org.apache.beam.sdk.Pipeline;
import org.apache.beam.sdk.coders.AvroCoder;
import org.apache.beam.sdk.coders.ByteArrayCoder;
import org.apache.beam.sdk.coders.Coder;
import org.apache.beam.sdk.coders.CoderException;
import org.apache.beam.sdk.coders.CoderRegistry;
import org.apache.beam.sdk.coders.CustomCoder;
import org.apache.beam.sdk.coders.SerializableCoder;
import org.apache.beam.sdk.io.FileSystems;
import org.apache.beam.sdk.io.GenerateSequence;
import org.apache.beam.sdk.io.Read;
import org.apache.beam.sdk.io.TextIO;
import org.apache.beam.sdk.metrics.Counter;
import org.apache.beam.sdk.metrics.Metrics;
import org.apache.beam.sdk.nexmark.model.Auction;
import org.apache.beam.sdk.nexmark.model.AuctionBid;
import org.apache.beam.sdk.nexmark.model.AuctionCount;
import org.apache.beam.sdk.nexmark.model.AuctionPrice;
import org.apache.beam.sdk.nexmark.model.Bid;
import org.apache.beam.sdk.nexmark.model.BidsPerSession;
import org.apache.beam.sdk.nexmark.model.CategoryPrice;
import org.apache.beam.sdk.nexmark.model.Done;
import org.apache.beam.sdk.nexmark.model.Event;
import org.apache.beam.sdk.nexmark.model.IdNameReserve;
import org.apache.beam.sdk.nexmark.model.KnownSize;
import org.apache.beam.sdk.nexmark.model.NameCityStateId;
import org.apache.beam.sdk.nexmark.model.Person;
import org.apache.beam.sdk.nexmark.model.SellerPrice;
import org.apache.beam.sdk.nexmark.sources.BoundedEventSource;
import org.apache.beam.sdk.nexmark.sources.UnboundedEventSource;
import org.apache.beam.sdk.nexmark.sources.generator.Generator;
import org.apache.beam.sdk.nexmark.sources.generator.GeneratorConfig;
import org.apache.beam.sdk.state.StateSpec;
import org.apache.beam.sdk.state.StateSpecs;
import org.apache.beam.sdk.state.ValueState;
import org.apache.beam.sdk.transforms.Combine;
import org.apache.beam.sdk.transforms.Create;
import org.apache.beam.sdk.transforms.DoFn;
import org.apache.beam.sdk.transforms.MapElements;
import org.apache.beam.sdk.transforms.PTransform;
import org.apache.beam.sdk.transforms.ParDo;
import org.apache.beam.sdk.transforms.SimpleFunction;
import org.apache.beam.sdk.transforms.windowing.AfterPane;
import org.apache.beam.sdk.transforms.windowing.BoundedWindow;
import org.apache.beam.sdk.transforms.windowing.GlobalWindows;
import org.apache.beam.sdk.transforms.windowing.Window;
import org.apache.beam.sdk.values.KV;
import org.apache.beam.sdk.values.PBegin;
import org.apache.beam.sdk.values.PCollection;
import org.apache.beam.sdk.values.TimestampedValue;
import org.apache.beam.vendor.guava.v26_0_jre.com.google.common.base.Splitter;
import org.apache.beam.vendor.guava.v26_0_jre.com.google.common.base.Strings;
import org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableList;
import org.apache.beam.vendor.guava.v26_0_jre.com.google.common.hash.Hashing;
import org.joda.time.Duration;
import org.joda.time.Instant;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

/** Odd's 'n Ends used throughout queries and driver. */
public class NexmarkUtils {
    private static final Logger LOG = LoggerFactory.getLogger(NexmarkUtils.class);

    /** Mapper for (de)serializing JSON. */
    public static final ObjectMapper MAPPER = new ObjectMapper();

    /** Possible sources for events. */
    public enum SourceType {
        /** Produce events directly. */
        DIRECT,
        /** Read events from an Avro file. */
        AVRO,
        /** Read from a PubSub topic. It will be fed the same synthetic events by this pipeline. */
        PUBSUB,
        /**
         * Read events from a Kafka topic. It will be fed the same synthetic events by this pipeline.
         */
        KAFKA
    }

    /** Possible sinks for query results. */
    public enum SinkType {
        /** Discard all results. */
        COUNT_ONLY,
        /** Discard all results after converting them to strings. */
        DEVNULL,
        /** Write to a PubSub topic. It will be drained by this pipeline. */
        PUBSUB,
        /** Write to a Kafka topic. It will be drained by this pipeline. */
        KAFKA,
        /** Write to a text file. Only works in batch mode. */
        TEXT,
        /** Write raw Events to Avro. Only works in batch mode. */
        AVRO,
        /** Write raw Events to BigQuery. */
        BIGQUERY
    }

    /** Pub/sub mode to run in. */
    public enum PubSubMode {
        /** Publish events to pub/sub, but don't run the query. */
        PUBLISH_ONLY,
        /** Consume events from pub/sub and run the query, but don't publish. */
        SUBSCRIBE_ONLY,
        /** Both publish and consume, but as separate jobs. */
        COMBINED
    }

    /** Possible side input sources. */
    public enum SideInputType {
        /** Produce the side input via {@link Create}. */
        DIRECT,
        /** Read side input from CSV files. */
        CSV
    }

    /** Coder strategies. */
    public enum CoderStrategy {
        /** Hand-written. */
        HAND,
        /** Avro. */
        AVRO,
        /** Java serialization. */
        JAVA
    }

    /** How to determine resource names. */
    public enum ResourceNameMode {
        /** Names are used as provided. */
        VERBATIM,
        /** Names are suffixed with the query being run. */
        QUERY,
        /** Names are suffixed with the query being run and a random number. */
        QUERY_AND_SALT,
        /** Names are suffixed with the runner being used and a the mode (streaming/batch). */
        QUERY_RUNNER_AND_MODE
    }

    /** Return a query name with query language (if applicable). */
    static String fullQueryName(String queryLanguage, String query) {
        return queryLanguage != null ? query + "_" + queryLanguage : query;
    }

    /** Return a BigQuery table spec. */
    static String tableSpec(NexmarkOptions options, String queryName, long now, String version) {
        return String.format("%s:%s.%s", options.getProject(), options.getBigQueryDataset(),
                tableName(options, queryName, now, version));
    }

    /** Return a BigQuery table name. */
    static String tableName(NexmarkOptions options, String queryName, long now, String version) {
        String baseTableName = options.getBigQueryTable();
        if (Strings.isNullOrEmpty(baseTableName)) {
            throw new RuntimeException("Missing --bigQueryTable");
        }

        switch (options.getResourceNameMode()) {
        case VERBATIM:
            return String.format("%s_%s", baseTableName, version);
        case QUERY:
            return String.format("%s_%s_%s", baseTableName, queryName, version);
        case QUERY_AND_SALT:
            return String.format("%s_%s_%s_%d", baseTableName, queryName, version, now);
        case QUERY_RUNNER_AND_MODE:
            String runnerName = options.getRunner().getSimpleName();
            boolean isStreaming = options.isStreaming();

            String tableName = String.format("%s_%s_%s_%s", baseTableName, queryName, runnerName,
                    processingMode(isStreaming));

            return version != null ? String.format("%s_%s", tableName, version) : tableName;
        }
        throw new RuntimeException("Unrecognized enum " + options.getResourceNameMode());
    }

    private static String processingMode(boolean isStreaming) {
        return isStreaming ? "streaming" : "batch";
    }

    /** Units for rates. */
    public enum RateUnit {
        PER_SECOND(1_000_000L), PER_MINUTE(60_000_000L);

        RateUnit(long usPerUnit) {
            this.usPerUnit = usPerUnit;
        }

        /** Number of microseconds per unit. */
        private final long usPerUnit;

        /** Number of microseconds between events at given rate. */
        public long rateToPeriodUs(long rate) {
            return (usPerUnit + rate / 2) / rate;
        }
    }

    /** Shape of event rate. */
    public enum RateShape {
        SQUARE, SINE;

        /** Number of steps used to approximate sine wave. */
        private static final int N = 10;

        /**
         * Return inter-event delay, in microseconds, for each generator to follow in order to achieve
         * {@code rate} at {@code unit} using {@code numGenerators}.
         */
        public long interEventDelayUs(int rate, RateUnit unit, int numGenerators) {
            return unit.rateToPeriodUs(rate) * numGenerators;
        }

        /**
         * Return array of successive inter-event delays, in microseconds, for each generator to follow
         * in order to achieve this shape with {@code firstRate/nextRate} at {@code unit} using {@code
         * numGenerators}.
         */
        public long[] interEventDelayUs(int firstRate, int nextRate, RateUnit unit, int numGenerators) {
            if (firstRate == nextRate) {
                long[] interEventDelayUs = new long[1];
                interEventDelayUs[0] = unit.rateToPeriodUs(firstRate) * numGenerators;
                return interEventDelayUs;
            }

            switch (this) {
            case SQUARE: {
                long[] interEventDelayUs = new long[2];
                interEventDelayUs[0] = unit.rateToPeriodUs(firstRate) * numGenerators;
                interEventDelayUs[1] = unit.rateToPeriodUs(nextRate) * numGenerators;
                return interEventDelayUs;
            }
            case SINE: {
                double mid = (firstRate + nextRate) / 2.0;
                double amp = (firstRate - nextRate) / 2.0; // may be -ve
                long[] interEventDelayUs = new long[N];
                for (int i = 0; i < N; i++) {
                    double r = (2.0 * Math.PI * i) / N;
                    double rate = mid + amp * Math.cos(r);
                    interEventDelayUs[i] = unit.rateToPeriodUs(Math.round(rate)) * numGenerators;
                }
                return interEventDelayUs;
            }
            }
            throw new RuntimeException(); // switch should be exhaustive
        }

        /**
         * Return delay between steps, in seconds, for result of {@link #interEventDelayUs}, so as to
         * cycle through the entire sequence every {@code ratePeriodSec}.
         */
        public int stepLengthSec(int ratePeriodSec) {
            int n = 0;
            switch (this) {
            case SQUARE:
                n = 2;
                break;
            case SINE:
                n = N;
                break;
            }
            return (ratePeriodSec + n - 1) / n;
        }
    }

    /** Set to true to capture all info messages. The logging level flags don't currently work. */
    private static final boolean LOG_INFO = false;

    /**
     * Set to true to log directly to stdout. If run using Google Dataflow, you can watch the results
     * in real-time with: tail -f /var/log/dataflow/streaming-harness/harness-stdout.log
     */
    private static final boolean LOG_TO_CONSOLE = false;

    /** Log info message. */
    public static void info(String format, Object... args) {
        if (LOG_INFO) {
            LOG.info(String.format(format, args));
            if (LOG_TO_CONSOLE) {
                System.out.println(String.format(format, args));
            }
        }
    }

    /** Log message to console. For client side only. */
    public static void console(String format, Object... args) {
        System.out.printf("%s %s%n", Instant.now(), String.format(format, args));
    }

    /** Label to use for timestamps on pub/sub messages. */
    public static final String PUBSUB_TIMESTAMP = "timestamp";

    /** Label to use for windmill ids on pub/sub messages. */
    public static final String PUBSUB_ID = "id";

    /** All events will be given a timestamp relative to this time (ms since epoch). */
    private static final long BASE_TIME = Instant.parse("2015-07-15T00:00:00.000Z").getMillis();

    /**
     * Instants guaranteed to be strictly before and after all event timestamps, and which won't be
     * subject to underflow/overflow.
     */
    public static final Instant BEGINNING_OF_TIME = new Instant(0).plus(Duration.standardDays(365));

    public static final Instant END_OF_TIME = BoundedWindow.TIMESTAMP_MAX_VALUE.minus(Duration.standardDays(365));

    /** Setup pipeline with codes and some other options. */
    public static void setupPipeline(CoderStrategy coderStrategy, Pipeline p) {
        CoderRegistry registry = p.getCoderRegistry();
        switch (coderStrategy) {
        case HAND:
            registry.registerCoderForClass(Auction.class, Auction.CODER);
            registry.registerCoderForClass(AuctionBid.class, AuctionBid.CODER);
            registry.registerCoderForClass(AuctionCount.class, AuctionCount.CODER);
            registry.registerCoderForClass(AuctionPrice.class, AuctionPrice.CODER);
            registry.registerCoderForClass(Bid.class, Bid.CODER);
            registry.registerCoderForClass(CategoryPrice.class, CategoryPrice.CODER);
            registry.registerCoderForClass(Event.class, Event.CODER);
            registry.registerCoderForClass(IdNameReserve.class, IdNameReserve.CODER);
            registry.registerCoderForClass(NameCityStateId.class, NameCityStateId.CODER);
            registry.registerCoderForClass(Person.class, Person.CODER);
            registry.registerCoderForClass(SellerPrice.class, SellerPrice.CODER);
            registry.registerCoderForClass(Done.class, Done.CODER);
            registry.registerCoderForClass(BidsPerSession.class, BidsPerSession.CODER);
            break;
        case AVRO:
            registry.registerCoderProvider(AvroCoder.getCoderProvider());
            break;
        case JAVA:
            registry.registerCoderProvider(SerializableCoder.getCoderProvider());
            break;
        }
    }

    /** Return a generator config to match the given {@code options}. */
    private static GeneratorConfig standardGeneratorConfig(NexmarkConfiguration configuration) {
        return new GeneratorConfig(configuration,
                configuration.useWallclockEventTime ? System.currentTimeMillis() : BASE_TIME, 0,
                configuration.numEvents, 0);
    }

    /** Return an iterator of events using the 'standard' generator config. */
    public static Iterator<TimestampedValue<Event>> standardEventIterator(NexmarkConfiguration configuration) {
        return new Generator(standardGeneratorConfig(configuration));
    }

    /** Return a transform which yields a finite number of synthesized events generated as a batch. */
    public static PTransform<PBegin, PCollection<Event>> batchEventsSource(NexmarkConfiguration configuration) {
        return Read.from(
                new BoundedEventSource(standardGeneratorConfig(configuration), configuration.numEventGenerators));
    }

    /**
     * Return a transform which yields a finite number of synthesized events generated on-the-fly in
     * real time.
     */
    public static PTransform<PBegin, PCollection<Event>> streamEventsSource(NexmarkConfiguration configuration) {
        return Read.from(new UnboundedEventSource(NexmarkUtils.standardGeneratorConfig(configuration),
                configuration.numEventGenerators, configuration.watermarkHoldbackSec, configuration.isRateLimited));
    }

    /** Return a transform to pass-through events, but count them as they go by. */
    public static ParDo.SingleOutput<Event, Event> snoop(final String name) {
        return ParDo.of(new DoFn<Event, Event>() {
            final Counter eventCounter = Metrics.counter(name, "events");
            final Counter newPersonCounter = Metrics.counter(name, "newPersons");
            final Counter newAuctionCounter = Metrics.counter(name, "newAuctions");
            final Counter bidCounter = Metrics.counter(name, "bids");
            final Counter endOfStreamCounter = Metrics.counter(name, "endOfStream");

            @ProcessElement
            public void processElement(ProcessContext c) {
                eventCounter.inc();
                if (c.element().newPerson != null) {
                    newPersonCounter.inc();
                } else if (c.element().newAuction != null) {
                    newAuctionCounter.inc();
                } else if (c.element().bid != null) {
                    bidCounter.inc();
                } else {
                    endOfStreamCounter.inc();
                }
                info("%s snooping element %s", name, c.element());
                c.output(c.element());
            }
        });
    }

    /** Return a transform to count and discard each element. */
    public static <T> ParDo.SingleOutput<T, Void> devNull(final String name) {
        return ParDo.of(new DoFn<T, Void>() {
            final Counter discardedCounterMetric = Metrics.counter(name, "discarded");

            @ProcessElement
            public void processElement(ProcessContext c) {
                discardedCounterMetric.inc();
            }
        });
    }

    /** Return a transform to log each element, passing it through unchanged. */
    public static <T> ParDo.SingleOutput<T, T> log(final String name) {
        return ParDo.of(new DoFn<T, T>() {
            @ProcessElement
            public void processElement(ProcessContext c) {
                LOG.info("%s: %s", name, c.element());
                c.output(c.element());
            }
        });
    }

    /** Return a transform to format each element as a string. */
    public static <T> ParDo.SingleOutput<T, String> format(final String name) {
        return ParDo.of(new DoFn<T, String>() {
            final Counter recordCounterMetric = Metrics.counter(name, "records");

            @ProcessElement
            public void processElement(ProcessContext c) {
                recordCounterMetric.inc();
                c.output(c.element().toString());
            }
        });
    }

    /** Return a transform to make explicit the timestamp of each element. */
    public static <T> ParDo.SingleOutput<T, TimestampedValue<T>> stamp(String name) {
        return ParDo.of(new DoFn<T, TimestampedValue<T>>() {
            @ProcessElement
            public void processElement(ProcessContext c) {
                c.output(TimestampedValue.of(c.element(), c.timestamp()));
            }
        });
    }

    /** Return a transform to reduce a stream to a single, order-invariant long hash. */
    public static <T> PTransform<PCollection<T>, PCollection<Long>> hash(final long numEvents, String name) {
        return new PTransform<PCollection<T>, PCollection<Long>>(name) {
            @Override
            public PCollection<Long> expand(PCollection<T> input) {
                return input
                        .apply(Window.<T>into(new GlobalWindows())
                                .triggering(AfterPane.elementCountAtLeast((int) numEvents))
                                .withAllowedLateness(Duration.standardDays(1)).discardingFiredPanes())
                        .apply(name + ".Hash", ParDo.of(new DoFn<T, Long>() {
                            @ProcessElement
                            public void processElement(ProcessContext c) {
                                long hash = Hashing.murmur3_128().newHasher().putLong(c.timestamp().getMillis())
                                        .putString(c.element().toString(), StandardCharsets.UTF_8).hash().asLong();
                                c.output(hash);
                            }
                        })).apply(Combine.globally(new Combine.BinaryCombineFn<Long>() {
                            @Override
                            public Long apply(Long left, Long right) {
                                return left ^ right;
                            }
                        }));
            }
        };
    }

    private static final long MASK = (1L << 16) - 1L;
    private static final long HASH = 0x243F6A8885A308D3L;
    private static final long INIT_PLAINTEXT = 50000L;

    /** Return a transform to keep the CPU busy for given milliseconds on every record. */
    public static <T> ParDo.SingleOutput<T, T> cpuDelay(String name, final long delayMs) {
        return ParDo.of(new DoFn<T, T>() {
            @ProcessElement
            public void processElement(ProcessContext c) {
                long now = System.currentTimeMillis();
                long end = now + delayMs;
                while (now < end) {
                    // Find plaintext which hashes to HASH in lowest MASK bits.
                    // Values chosen to roughly take 1ms on typical workstation.
                    long p = INIT_PLAINTEXT;
                    while (true) {
                        long t = Hashing.murmur3_128().hashLong(p).asLong();
                        if ((t & MASK) == (HASH & MASK)) {
                            break;
                        }
                        p++;
                    }
                    now = System.currentTimeMillis();
                }
                c.output(c.element());
            }
        });
    }

    private static final int MAX_BUFFER_SIZE = 1 << 24;

    private static class DiskBusyTransform<T> extends PTransform<PCollection<T>, PCollection<T>> {

        private long bytes;

        private DiskBusyTransform(long bytes) {
            this.bytes = bytes;
        }

        @Override
        public PCollection<T> expand(PCollection<T> input) {
            // Add dummy key to be able to use State API
            PCollection<KV<Integer, T>> kvCollection = input.apply("diskBusy.keyElements",
                    ParDo.of(new DoFn<T, KV<Integer, T>>() {

                        @ProcessElement
                        public void processElement(ProcessContext context) {
                            context.output(KV.of(0, context.element()));
                        }
                    }));
            // Apply actual transform that generates disk IO using state API
            return kvCollection.apply("diskBusy.generateIO", ParDo.of(new DoFn<KV<Integer, T>, T>() {

                private static final String DISK_BUSY = "diskBusy";

                @StateId(DISK_BUSY)
                private final StateSpec<ValueState<byte[]>> spec = StateSpecs.value(ByteArrayCoder.of());

                @ProcessElement
                public void processElement(ProcessContext c, @StateId(DISK_BUSY) ValueState<byte[]> state) {
                    long remain = bytes;
                    long now = System.currentTimeMillis();
                    while (remain > 0) {
                        long thisBytes = Math.min(remain, MAX_BUFFER_SIZE);
                        remain -= thisBytes;
                        byte[] arr = new byte[(int) thisBytes];
                        for (int i = 0; i < thisBytes; i++) {
                            arr[i] = (byte) now;
                        }
                        state.write(arr);
                        now = System.currentTimeMillis();
                    }
                    c.output(c.element().getValue());
                }
            }));
        }
    }

    /** Return a transform to write given number of bytes to durable store on every record. */
    public static <T> PTransform<PCollection<T>, PCollection<T>> diskBusy(final long bytes) {
        return new DiskBusyTransform<>(bytes);
    }

    /** Return a transform to cast each element to {@link KnownSize}. */
    private static <T extends KnownSize> ParDo.SingleOutput<T, KnownSize> castToKnownSize() {
        return ParDo.of(new DoFn<T, KnownSize>() {
            @ProcessElement
            public void processElement(ProcessContext c) {
                c.output(c.element());
            }
        });
    }

    private static class GenerateSideInputData extends PTransform<PBegin, PCollection<KV<Long, String>>> {

        private final NexmarkConfiguration config;

        private GenerateSideInputData(NexmarkConfiguration config) {
            this.config = config;
        }

        @Override
        public PCollection<KV<Long, String>> expand(PBegin input) {
            return input.apply(GenerateSequence.from(0).to(config.sideInputRowCount))
                    .apply(MapElements.via(new SimpleFunction<Long, KV<Long, String>>() {
                        @Override
                        public KV<Long, String> apply(Long input) {
                            return KV.of(input, String.valueOf(input));
                        }
                    }));
        }
    }

    /**
     * Write data to be read as a side input.
     *
     * <p>Contains pairs of a number and its string representation to model lookups of some enrichment
     * data by id.
     *
     * <p>Generated data covers the range {@code [0, sideInputRowCount)} so lookup joins on any
     * desired id field can be modeled by looking up {@code id % sideInputRowCount}.
     */
    public static PCollection<KV<Long, String>> prepareSideInput(Pipeline queryPipeline,
            NexmarkConfiguration config) {

        checkArgument(config.sideInputRowCount > 0, "Side input required but sideInputRowCount is not >0");

        PTransform<PBegin, PCollection<KV<Long, String>>> generateSideInputData = new GenerateSideInputData(config);

        switch (config.sideInputType) {
        case DIRECT:
            return queryPipeline.apply(generateSideInputData);
        case CSV:
            checkArgument(config.sideInputUrl != null,
                    "Side input type %s requires a URL but sideInputUrl not specified",
                    SideInputType.CSV.toString());

            checkArgument(config.sideInputNumShards > 0,
                    "Side input type %s requires explicit numShards but sideInputNumShards not specified",
                    SideInputType.CSV.toString());

            Pipeline tempPipeline = Pipeline.create();
            tempPipeline.apply(generateSideInputData)
                    .apply(MapElements.via(new SimpleFunction<KV<Long, String>, String>(
                            kv -> String.format("%s,%s", kv.getKey(), kv.getValue())) {
                    })).apply(TextIO.write().withNumShards(config.sideInputNumShards).to(config.sideInputUrl));
            tempPipeline.run().waitUntilFinish();

            return queryPipeline.apply(TextIO.read().from(config.sideInputUrl + "*"))
                    .apply(MapElements.via(new SimpleFunction<String, KV<Long, String>>(line -> {
                        List<String> cols = ImmutableList.copyOf(Splitter.on(",").split(line));
                        return KV.of(Long.valueOf(cols.get(0)), cols.get(1));
                    }) {
                    }));
        default:
            throw new IllegalArgumentException(
                    String.format("Unknown type of side input requested: %s", config.sideInputType));
        }
    }

    /** Frees any resources used to make the side input available. */
    public static void cleanUpSideInput(NexmarkConfiguration config) throws IOException {
        switch (config.sideInputType) {
        case DIRECT:
            break;
        case CSV:
            FileSystems.delete(FileSystems.match(config.sideInputUrl + "*").metadata().stream()
                    .map(metadata -> metadata.resourceId()).collect(Collectors.toList()));
            break;
        default:
            throw new IllegalArgumentException(
                    String.format("Unknown type of %s clean up requested", SideInputType.class.getSimpleName()));
        }
    }

    /**
     * A coder for instances of {@code T} cast up to {@link KnownSize}.
     *
     * @param <T> True type of object.
     */
    private static class CastingCoder<T extends KnownSize> extends CustomCoder<KnownSize> {
        private final Coder<T> trueCoder;

        public CastingCoder(Coder<T> trueCoder) {
            this.trueCoder = trueCoder;
        }

        @Override
        public void encode(KnownSize value, OutputStream outStream) throws CoderException, IOException {
            @SuppressWarnings("unchecked")
            T typedValue = (T) value;
            trueCoder.encode(typedValue, outStream);
        }

        @Override
        public KnownSize decode(InputStream inStream) throws CoderException, IOException {
            return trueCoder.decode(inStream);
        }
    }

    /** Return a coder for {@code KnownSize} that are known to be exactly of type {@code T}. */
    private static <T extends KnownSize> Coder<KnownSize> makeCastingCoder(Coder<T> trueCoder) {
        return new CastingCoder<>(trueCoder);
    }

    /** Return {@code elements} as {@code KnownSize}s. */
    public static <T extends KnownSize> PCollection<KnownSize> castToKnownSize(final String name,
            PCollection<T> elements) {
        return elements.apply(name + ".Forget", castToKnownSize()).setCoder(makeCastingCoder(elements.getCoder()));
    }

    // Do not instantiate.
    private NexmarkUtils() {
    }
}