S3GeoTiffRDD.scala

/*
 * Copyright 2016 Azavea
 *
 * Licensed 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 geotrellis.spark.io.s3

import geotrellis.proj4._
import geotrellis.raster._
import geotrellis.raster.io.geotiff.tags.TiffTags
import geotrellis.spark._
import geotrellis.raster.io.geotiff.reader.GeoTiffReader
import geotrellis.spark.io.{GeoTiffInfoReader, RasterReader}
import geotrellis.spark.io.s3.util.S3RangeReader
import geotrellis.util.{LazyLogging, StreamingByteReader}
import geotrellis.vector._
import org.apache.hadoop.conf.Configuration
import org.apache.spark.SparkContext
import org.apache.spark.rdd.RDD
import org.apache.spark.storage.StorageLevel
import com.amazonaws.services.s3.model._
import java.net.URI
import java.nio.ByteBuffer

import com.typesafe.config.ConfigFactory

/**
 * The S3GeoTiffRDD object allows for the creation of whole or windowed RDD[(K, V)]s from files on S3.
 */
object S3GeoTiffRDD extends LazyLogging {
  final val GEOTIFF_TIME_TAG_DEFAULT = "TIFFTAG_DATETIME"
  final val GEOTIFF_TIME_FORMAT_DEFAULT = "yyyy:MM:dd HH:mm:ss"
  lazy val windowSize: Option[Int] = try {
    Some(ConfigFactory.load().getInt("geotrellis.s3.rdd.read.windowSize"))
  } catch {
    case _: Throwable =>
      logger.warn("geotrellis.s3.rdd.read.windowSize is not set in .conf file.")
      None
  }

  /**
    * This case class contains the various parameters one can set when reading RDDs from S3 using Spark.
    *
    * TODO: Add persistLevel option
    *
    * @param tiffExtensions     Read all file with an extension contained in the given list.
    * @param crs            Override CRS of the input files. If [[None]], the reader will use the file's original CRS.
    * @param timeTag        Name of tiff tag containing the timestamp for the tile.
    * @param timeFormat     Pattern for [[java.time.format.DateTimeFormatter]] to parse timeTag.
    * @param maxTileSize    Maximum allowed size of each tiles in output RDD.
    *                       May result in a one input GeoTiff being split amongst multiple records if it exceeds this size.
    *                       If no maximum tile size is specific, then each file file is read fully.
    *                       1024 by defaut.
    * @param numPartitions  How many partitions Spark should create when it repartitions the data.
    * @param partitionBytes Desired partition size in bytes, at least one item per partition will be assigned.
                            This option is incompatible with the maxTileSize option.
    *                       128 Mb by default.
    * @param chunkSize      How many bytes should be read in at a time.
    * @param delimiter      Delimiter to use for S3 objet listings. See
    * @param getS3Client    A function to instantiate an S3Client. Must be serializable.
    */
  case class Options(
    tiffExtensions: Seq[String] = Seq(".tif", ".TIF", ".tiff", ".TIFF"),
    crs: Option[CRS] = None,
    timeTag: String = GEOTIFF_TIME_TAG_DEFAULT,
    timeFormat: String = GEOTIFF_TIME_FORMAT_DEFAULT,
    maxTileSize: Option[Int] = None,
    numPartitions: Option[Int] = None,
    partitionBytes: Option[Long] = Some(128l * 1024 * 1024),
    chunkSize: Option[Int] = None,
    delimiter: Option[String] = None,
    getS3Client: () => S3Client = () => S3Client.DEFAULT
  ) extends RasterReader.Options

  val contingencyTileSize = 512

  object Options {
    def DEFAULT = Options()
  }

  /**
   * Create Configuration for [[S3InputFormat]] based on parameters and options.
   * Important: won't pass partitionBytes into hadoop configuration if numPartition options is set.
   *
   * @param bucket   Name of the bucket on S3 where the files are kept.
   * @param prefix   Prefix of all of the keys on S3 that are to be read in.
   * @param options  An instance of [[Options]] that contains any user defined or default settings.
   */
  private def configuration(bucket: String, prefix: String, options: S3GeoTiffRDD.Options)(implicit sc: SparkContext): Configuration = {
    if(options.numPartitions.isDefined && options.partitionBytes.isDefined)
      logger.warn("Both numPartitions and partitionBytes options are set. " +
        "Only numPartitions would be passed into hadoop configuration.")

    val conf = sc.hadoopConfiguration
    S3InputFormat.setBucket(conf, bucket)
    S3InputFormat.setPrefix(conf, prefix)
    S3InputFormat.setExtensions(conf, options.tiffExtensions)
    S3InputFormat.setCreateS3Client(conf, options.getS3Client)
    options.numPartitions
      .fold(S3InputFormat.removePartitionCount(conf)) { n =>
        S3InputFormat.setPartitionCount(conf, n)
        S3InputFormat.removePartitionBytes(conf)
      }
    if(options.numPartitions.isEmpty)
      options.partitionBytes
        .fold(S3InputFormat.removePartitionBytes(conf))(S3InputFormat.setPartitionBytes(conf, _))
    options.delimiter.fold(S3InputFormat.removeDelimiter(conf))(S3InputFormat.setDelimiter(conf, _))
    conf
  }

  /**
    * A helper function to get the maximum (linear) tile size.
    */
  private def getMaxSize(options: Options) = {
    (options.maxTileSize, windowSize) match {
      case (Some(maxTileSize), Some(windowSize)) => math.min(maxTileSize, windowSize)
      case (Some(maxTileSize), None) => maxTileSize
      case (None, Some(windowSize)) => windowSize
      case _ => {
        val size = Options.DEFAULT.maxTileSize match {
          case Some(maxTileSize) => maxTileSize
          case None => contingencyTileSize
        }
        logger.warn(s"Neither maxTileSize nor windowSize was given, defaulting to $size.")
        size
      }
    }
  }

  /**
    * Creates a RDD[(K, V)] whose K and V  on the type of the GeoTiff that is going to be read in.
    *
    * @param  bucket    Name of the bucket on S3 where the files are kept.
    * @param  prefix    Prefix of all of the keys on S3 that are to be read in.
    * @param  uriToKey  Function to transform input key basing on the URI information.
    * @param  options   An instance of [[Options]] that contains any user defined or default settings.
    * @param  geometry  An optional geometry to filter by.  If this is provided, it is assumed that all GeoTiffs are in the same CRS, and that this geometry is in that CRS.
    */
  def apply[I, K, V](
    bucket: String, prefix: String,
    uriToKey: (URI, I) => K,
    options: Options,
    geometry: Option[Geometry]
  )(implicit sc: SparkContext, rr: RasterReader[Options, (I, V)]): RDD[(K, V)] = {

    val conf = configuration(bucket, prefix, options)
    lazy val sourceGeoTiffInfo = S3GeoTiffInfoReader(bucket, prefix, options)

    (options.maxTileSize, options.partitionBytes) match {
      case (_, Some(partitionBytes)) => {
        val maxSize = getMaxSize(options)

        val windows: RDD[(String, Array[GridBounds])] =
          sourceGeoTiffInfo.windowsByBytes(
            partitionBytes,
            maxSize,
            geometry
          )

        windows.persist()

        val windowCount = windows.count.toInt

        logger.info(s"Repartition into ${windowCount} partitions.")

        val repartition =
          if (windowCount > windows.partitions.length) windows.repartition(windowCount)
          else windows

        val result = repartition.flatMap { case (path, windowBounds) =>
          rr.readWindows(windowBounds, sourceGeoTiffInfo.getGeoTiffInfo(path), options).map { case (k, v) =>
            uriToKey(new URI(path), k) -> v
          }
        }

        windows.unpersist()
        result
      }
      case (Some(_), _) =>
        val objectRequestsToDimensions: RDD[(GetObjectRequest, (Int, Int))] =
          sc.newAPIHadoopRDD(
            conf,
            classOf[TiffTagsS3InputFormat],
            classOf[GetObjectRequest],
            classOf[TiffTags]
          ).mapValues { tiffTags => (tiffTags.cols, tiffTags.rows) }

        apply[I, K, V](objectRequestsToDimensions, uriToKey, options, sourceGeoTiffInfo)

      case _ =>
        sc.newAPIHadoopRDD(
          conf,
          classOf[BytesS3InputFormat],
          classOf[String],
          classOf[Array[Byte]]
        ).mapPartitions(
          _.map { case (key, bytes) =>
            val (k, v) = rr.readFully(ByteBuffer.wrap(bytes), options)
            uriToKey(new URI(key), k) -> v
          },
          preservesPartitioning = true
        )
    }
  }

  /**
    * Creates a RDD[(K, V)] whose K and V  on the type of the GeoTiff that is going to be read in.
    *
    * @param  bucket    Name of the bucket on S3 where the files are kept.
    * @param  prefix    Prefix of all of the keys on S3 that are to be read in.
    * @param  uriToKey  Function to transform input key basing on the URI information.
    * @param  options   An instance of [[Options]] that contains any user defined or default settings.
    * @param  geometry  An optional geometry to filter by.  If this is provided, it is assumed that all GeoTiffs are in the same CRS, and that this geometry is in that CRS.
    */
  def apply[I, K, V](
    bucket: String, prefix: String,
    uriToKey: (URI, I) => K,
    options: Options
  )(implicit sc: SparkContext, rr: RasterReader[Options, (I, V)]): RDD[(K, V)] = {
    apply(bucket, prefix, uriToKey, options, None)
  }

  /**
    * Creates a RDD[(K, V)] whose K and V  on the type of the GeoTiff that is going to be read in.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param options  An instance of [[Options]] that contains any user defined or default settings.
    */
  def apply[K, V](bucket: String, prefix: String, options: Options)
                 (implicit sc: SparkContext, rr: RasterReader[Options, (K, V)]): RDD[(K, V)] =
    apply[K, K, V](bucket, prefix, (_: URI, key: K) => key, options)

  /**
    * Creates a RDD[(K, V)] whose K and V depends on the type of the GeoTiff that is going to be read in.
    *
    * @param objectRequestsToDimensions A RDD of GetObjectRequest of a given GeoTiff and its cols and rows as a (Int, Int).
    * @param uriToKey function to transform input key basing on the URI information.
    * @param options An instance of [[Options]] that contains any user defined or default settings.
    */
  def apply[I, K, V](objectRequestsToDimensions: RDD[(GetObjectRequest, (Int, Int))], uriToKey: (URI, I) => K, options: Options, sourceGeoTiffInfo: => GeoTiffInfoReader)
    (implicit rr: RasterReader[Options, (I, V)]): RDD[(K, V)] = {

    val windows =
      objectRequestsToDimensions
        .flatMap { case (objectRequest, (cols, rows)) =>
          val bucket = objectRequest.getBucketName
          val key = objectRequest.getKey
          val layout = sourceGeoTiffInfo.getGeoTiffInfo(s"s3://$bucket/$key").segmentLayout.tileLayout
          val maxSize = getMaxSize(options)

          RasterReader
            .listWindows(cols, rows, maxSize, layout.tileCols, layout.tileRows)
            .map((objectRequest, _))
        }

    // Windowed reading may have produced unbalanced partitions due to files of differing size
    val repartitioned =
      options.numPartitions match {
        case Some(p) =>
          logger.info(s"repartition into $p partitions.")
          windows.repartition(p)
        case None =>
          options.partitionBytes match {
            case Some(byteCount) =>
              sourceGeoTiffInfo.estimatePartitionsNumber(byteCount, options.maxTileSize) match {
                case Some(numPartitions) if numPartitions != windows.partitions.length =>
                  logger.info(s"repartition into $numPartitions partitions.")
                  windows.repartition(numPartitions)
                case _ => windows
              }
            case _ =>
              windows
          }
      }

    repartitioned.map { case (objectRequest: GetObjectRequest, pixelWindow: GridBounds) =>
      val reader = options.chunkSize match {
        case Some(chunkSize) =>
          StreamingByteReader(S3RangeReader(objectRequest, options.getS3Client()), chunkSize)
        case None =>
          StreamingByteReader(S3RangeReader(objectRequest, options.getS3Client()))
      }

      val (k, v) = rr.readWindow(reader, pixelWindow, options)

      uriToKey(new URI(s"s3://${objectRequest.getBucketName}/${objectRequest.getKey}"), k) -> v
    }
  }

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband GeoTiffs.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket Name of the bucket on S3 where the files are kept.
    * @param prefix Prefix of all of the keys on S3 that are to be read in.
    * @param uriToKey function to transform input key basing on the URI information.
    */
  def singleband[I, K](bucket: String, prefix: String, uriToKey: (URI, I) => K, options: Options)(implicit sc: SparkContext, rr: RasterReader[Options, (I, Tile)]): RDD[(K, Tile)] =
    apply[I, K, Tile](bucket, prefix, uriToKey, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband GeoTiffs.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket Name of the bucket on S3 where the files are kept.
    * @param prefix Prefix of all of the keys on S3 that are to be read in.
    */
  def singleband[K](bucket: String, prefix: String, options: Options)(implicit sc: SparkContext, rr: RasterReader[Options, (K, Tile)]): RDD[(K, Tile)] =
    apply[K, Tile](bucket, prefix, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband GeoTiffs.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket Name of the bucket on S3 where the files are kept.
    * @param prefix Prefix of all of the keys on S3 that are to be read in.
    * @param uriToKey function to transform input key basing on the URI information.
    */
  def multiband[I, K](bucket: String, prefix: String, uriToKey: (URI, I) => K, options: Options)(implicit sc: SparkContext, rr: RasterReader[Options, (I, MultibandTile)]): RDD[(K, MultibandTile)] =
    apply[I, K, MultibandTile](bucket, prefix, uriToKey, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband GeoTiffs.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket Name of the bucket on S3 where the files are kept.
    * @param prefix Prefix of all of the keys on S3 that are to be read in.
    */
  def multiband[K](bucket: String, prefix: String, options: Options)(implicit sc: SparkContext, rr: RasterReader[Options, (K, MultibandTile)]): RDD[(K, MultibandTile)] =
    apply[K, MultibandTile](bucket, prefix, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband GeoTiffs.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket Name of the bucket on S3 where the files are kept.
    * @param prefix Prefix of all of the keys on S3 that are to be read in.
    */
  def spatial(bucket: String, prefix: String)(implicit sc: SparkContext): RDD[(ProjectedExtent, Tile)] =
    spatial(bucket, prefix, Options.DEFAULT)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband tiles.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param options  An instance of [[Options]] that contains any user defined or default settings.
    */
  def spatial(bucket: String, prefix: String, options: Options)(implicit sc: SparkContext): RDD[(ProjectedExtent, Tile)] =
    singleband[ProjectedExtent](bucket, prefix, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband tiles.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param uriToKey function to transform input key basing on the URI information.
    * @param options  An instance of [[Options]] that contains any user defined or default settings.
    */
  def spatial(bucket: String, prefix: String, uriToKey: (URI, ProjectedExtent) => ProjectedExtent, options: Options)(implicit sc: SparkContext): RDD[(ProjectedExtent, Tile)] =
    singleband[ProjectedExtent, ProjectedExtent](bucket, prefix, uriToKey, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband tiles.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    */
  def spatialMultiband(bucket: String, prefix: String)(implicit sc: SparkContext): RDD[(ProjectedExtent, MultibandTile)] =
    spatialMultiband(bucket, prefix, Options.DEFAULT)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband tiles.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param options  An instance of [[Options]] that contains any user defined or default settings.
    */
  def spatialMultiband(bucket: String, prefix: String, options: Options)(implicit sc: SparkContext): RDD[(ProjectedExtent, MultibandTile)] =
    multiband[ProjectedExtent](bucket, prefix, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband tiles.
    * If a GeoTiff contains multiple bands, only the first will be read.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param uriToKey function to transform input key basing on the URI information.
    * @param options  An instance of [[Options]] that contains any user defined or default settings.
    */
  def spatialMultiband(bucket: String, prefix: String, uriToKey: (URI, ProjectedExtent) => ProjectedExtent, options: Options)(implicit sc: SparkContext): RDD[(ProjectedExtent, MultibandTile)] =
    multiband[ProjectedExtent, ProjectedExtent](bucket, prefix, uriToKey, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband tiles.
    * Will parse a timestamp from the default tiff tags to associate with each file.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    */
  def temporal(bucket: String, prefix: String)(implicit sc: SparkContext): RDD[(TemporalProjectedExtent, Tile)] =
    temporal(bucket, prefix, Options.DEFAULT)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband tiles.
    * Will parse a timestamp from a tiff tags specified in options to associate with each tile.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param options  Options for the reading process. Including the timestamp tiff tag and its pattern.
    */
  def temporal(bucket: String, prefix: String, options: Options)(implicit sc: SparkContext): RDD[(TemporalProjectedExtent, Tile)] =
    singleband[TemporalProjectedExtent](bucket, prefix, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as singleband tiles.
    * Will parse a timestamp from a tiff tags specified in options to associate with each tile.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param uriToKey function to transform input key basing on the URI information.
    * @param options  Options for the reading process. Including the timestamp tiff tag and its pattern.
    */
  def temporal(bucket: String, prefix: String, uriToKey: (URI, TemporalProjectedExtent) => TemporalProjectedExtent, options: Options)(implicit sc: SparkContext): RDD[(TemporalProjectedExtent, Tile)] =
    singleband[TemporalProjectedExtent, TemporalProjectedExtent](bucket, prefix, uriToKey, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband tiles.
    * Will parse a timestamp from a tiff tags specified in options to associate with each tile.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    */
  def temporalMultiband(bucket: String, prefix: String)(implicit sc: SparkContext): RDD[(TemporalProjectedExtent, MultibandTile)] =
    temporalMultiband(bucket, prefix, Options.DEFAULT)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband tiles.
    * Will parse a timestamp from a tiff tags specified in options to associate with each tile.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param options  Options for the reading process. Including the timestamp tiff tag and its pattern.
    */
  def temporalMultiband(bucket: String, prefix: String, options: Options)(implicit sc: SparkContext): RDD[(TemporalProjectedExtent, MultibandTile)] =
    multiband[TemporalProjectedExtent](bucket, prefix, options)

  /**
    * Creates RDD that will read all GeoTiffs in the given bucket and prefix as multiband tiles.
    * Will parse a timestamp from a tiff tags specified in options to associate with each tile.
    *
    * @param bucket   Name of the bucket on S3 where the files are kept.
    * @param prefix   Prefix of all of the keys on S3 that are to be read in.
    * @param uriToKey function to transform input key basing on the URI information.
    * @param options  Options for the reading process. Including the timestamp tiff tag and its pattern.
    */
  def temporalMultiband(bucket: String, prefix: String, uriToKey: (URI, TemporalProjectedExtent) => TemporalProjectedExtent, options: Options)(implicit sc: SparkContext): RDD[(TemporalProjectedExtent, MultibandTile)] =
    multiband[TemporalProjectedExtent, TemporalProjectedExtent](bucket, prefix, uriToKey, options)
}