20.7. Index Configuration

When creating a new feature type using the GeoMesa FileSystem data store (FSDS), there are several required options that must be specified. Through the command line tools, these options are exposed as flags. If using the GeoTools data store API, the options must be set as user data before calling createSchema.

20.7.1. Configuring the File Encoding

The FSDS currently supports three options for file encoding: orc, parquet, and converter. Both ORC and Parquet support reading and writing, while converter is a synthetic, read-only format that supports querying a wide variety of data files using the GeoMesa Convert API.

File encoding can be specified through the user data key geomesa.fs.encoding:

import org.locationtech.geomesa.fs.storage.common.interop.ConfigurationUtils;

SimpleFeatureType sft = ...
// use the utility method
ConfigurationUtils.setEncoding(sft, "parquet");
// or set directly in the user data
sft.getUserData().put("geomesa.fs.encoding", "parquet");
import org.locationtech.geomesa.fs.storage.common.RichSimpleFeatureType

val sft: SimpleFeatureType = ???
// use the implicit method from RichSimpleFeatureType
sft.setEncoding("parquet")
// or set directly in the user data
sft.getUserData.put("geomesa.fs.encoding", "parquet");

20.7.2. Configuring the Partition Scheme

Partition schemes define how data files are laid out in folders on the file system. Schemes are defined by a well-known name, and an optional map of configuration values. See Partition Schemes for more details.

Partition schemes can be specified through the user data key geomesa.fs.scheme:

import org.locationtech.geomesa.fs.storage.common.interop.ConfigurationUtils;
import java.util.Collections;

SimpleFeatureType sft = ...
// use the utility method
ConfigurationUtils.setScheme(sft, "daily", Collections.singletonMap("dtg-attribute", "dtg"));
// or set directly in the user data as JSON
sft.getUserData().put("geomesa.fs.scheme",
    "{ \"name\": \"daily\", \"options\": { \"dtg-attribute\": \"dtg\" } }");
import org.locationtech.geomesa.fs.storage.common.RichSimpleFeatureType

val sft: SimpleFeatureType = ???
// use the implicit method from RichSimpleFeatureType
sft.setScheme("daily", Map("dtg-attribute" -> "dtg"))
// or set directly in the user data
sft.getUserData.put("geomesa.fs.scheme",
    """{ "name": "daily", "options": { "dtg-attribute": "dtg" } }""")

20.7.3. Configuring Leaf Storage

Leaf storage controls the final layout of files and folders. When using leaf storage (which is enabled by default), the last component of the partition path is used as a prefix to the data file name, instead of as a separate folder. This can result in less directory overhead for filesystems such as S3.

As an example, a partition scheme of yyyy/MM/dd would produce a partition path like 2016/01/01. With leaf storage, the data files for that partition would be 2016/01/01_<datafile>.parquet. If leaf storage is disabled, the data files would be 2016/01/01/<datafile>.parquet, creating an extra level of directories.

Leaf storage can be specified through the user data key geomesa.fs.leaf-storage:

import org.locationtech.geomesa.fs.storage.common.interop.ConfigurationUtils;

SimpleFeatureType sft = ...
// use the utility method
ConfigurationUtils.setLeafStorage(sft, false);
// or set directly in the user data as a string
sft.getUserData().put("geomesa.fs.leaf-storage", "false");
import org.locationtech.geomesa.fs.storage.common.RichSimpleFeatureType

val sft: SimpleFeatureType = ???
// use the implicit method from RichSimpleFeatureType
sft.setLeafStorage(false)
// or set directly in the user data as a string
sft.getUserData.put("geomesa.fs.leaf-storage", "false")

20.7.4. Configuring Target File Size

By default data files can grow to unlimited size as more data is written and files are compacted. This may lead to poor performance, if a file becomes too large. To manage this, a target file size can be configured through the user data key geomesa.fs.file-size:

import org.locationtech.geomesa.fs.storage.common.interop.ConfigurationUtils;

SimpleFeatureType sft = ...
// use the utility method
ConfigurationUtils.setTargetFileSize(sft, false);
// or set directly in the user data as a string
sft.getUserData().put("geomesa.fs.file-size", "1GB");
import org.locationtech.geomesa.fs.storage.common.RichSimpleFeatureType

val sft: SimpleFeatureType = ???
// use the implicit method from RichSimpleFeatureType
sft.setTargetFileSize("1GB")
// or set directly in the user data as a string
sft.getUserData.put("geomesa.fs.file-size", "1GB")

Note that target file size can also be specified in some operations, which will override any default configured in the feature type. See compact and ingest for details. See geomesa.fs.size.threshold for controlling the file size error margin.

20.7.5. Configuring Metadata Persistence

The FSDS keeps metadata on partitions and data files, to avoid repeatedly interrogating the file system. By default, metadata information is stored as a change log in the file system, which does not require any additional infrastructure. For more advanced use-cases, the FSDS also supports persisting metadata in a relational database using JDBC. For more information, see FileSystem Metadata.

Metadata persistence can be specified through the user data key geomesa.fs.metadata:

import org.locationtech.geomesa.fs.storage.common.interop.ConfigurationUtils;
import java.util.Collections;

SimpleFeatureType sft = ...
// use the utility method
Map<String, String> options = Collections.singletonMap("jdbc.url", "jdbc:postgresql://localhost/geomesa");
ConfigurationUtils.setMetadata(sft, "jdbc", options);
// or set directly in the user data as JSON
sft.getUserData().put("geomesa.fs.metadata",
    "{ \"name\": \"jdbc\", \"options\": { \"jdbc.url\": \"jdbc:postgresql://localhost/geomesa\" } }");
import org.locationtech.geomesa.fs.storage.common.RichSimpleFeatureType

val sft: SimpleFeatureType = ???
// use the implicit method from RichSimpleFeatureType
sft.setMetadata("jdbc", Map("jdbc.url" -> "jdbc:postgresql://localhost/geomesa"))
// or set directly in the user data as JSON
sft.getUserData.put("geomesa.fs.metadata",
    """{ "name": "jdbc", "options": { "jdbc.url": "jdbc:postgresql://localhost/geomesa" } }""")

20.7.6. Configuring Custom Observer Callbacks

The FSDS provides a mechanism to add custom handling during file writing. Users can implement observer factories, which will be invoked for each new file that is created. Observer factories must extend the trait FileSystemObserverFactory:

package org.locationtech.geomesa.fs.storage.common.observer

trait FileSystemObserverFactory extends Closeable {

  /**
   * Called once after instantiating the factory
   *
   * @param conf hadoop configuration
   * @param root root path
   * @param sft simple feature type
   */
  def init(conf: Configuration, root: Path, sft: SimpleFeatureType): Unit

  /**
   * Create an observer for the given path
   *
   * @param path file path being written
   * @return
   */
  def apply(path: Path): FileSystemObserver
}

Note

Observer factories must have a default no-arg constructor in order to be instantiated by the framework.

Observers can be specified through the user data key geomesa.fs.observers:

import org.locationtech.geomesa.fs.storage.common.interop.ConfigurationUtils;
import java.util.Arrays;
import java.util.Collections;
import java.util.List;

SimpleFeatureType sft = ...
List<String> factories =
  Arrays.asList("com.example.MyCustomObserverFactory", "com.example.MySecondObserverFactory");
// use the static utility method
ConfigurationUtils.setObservers(sft, factories);
// or set directly in the user data as a comma-delimited string
sft.getUserData().put("geomesa.fs.observers", String.join(",", factories));
import org.locationtech.geomesa.fs.storage.common.RichSimpleFeatureType

val sft: SimpleFeatureType = ???
val factories = Seq("com.example.MyCustomObserverFactory", "com.example.MySecondObserverFactory")
// use the implicit method from RichSimpleFeatureType
sft.setObservers(factories)
// or set directly in the user data as a comma-delimited string
sft.getUserData.put("geomesa.fs.observers", factories.mkString(","))