6.7. Analytic Querying

GeoMesa provides advanced query capabilities through GeoTools query hints. You can use these hints to control various aspects of query processing, or to trigger distributed analytic processing. See Setting Query Hints for details on setting query hints.

6.7.1. Feature Sampling

Instead of returning all features for a query, GeoMesa can use statistical sampling to return a certain percentage of results. This can be useful when rendering maps, or when there are too many features to be meaningful.

Features can either be sampled absolutely, or sampled by a certain attribute. For example, given a series of points in a track, you may wish to sample by the track identifier so that no tracks are completely sampled out.

The sampling value should be a float in the range (0, 1), which represents the fractional value of features that will be returned. Due to distributed processing, the actual count returned is not guaranteed to equal the requested percentage - however, there will never be less features than requested. For example, if you sample 5 features at 10%, you will get back anywhere from 1 to 5 features, depending on how your data is distributed in the cluster.

Key Type GeoServer Conversion
QueryHints.SAMPLING Float any float
QueryHints.SAMPLE_BY String - attribute name (optional) any string
import org.locationtech.geomesa.index.conf.QueryHints;

 // returns 10% of features, threaded by 'track' attribute
query.getHints().put(QueryHints.SAMPLING(), new Float(0.1));
query.getHints().put(QueryHints.SAMPLE_BY(), "track");
import org.locationtech.geomesa.index.conf.QueryHints

// returns 10% of features, threaded by 'track' attribute
query.getHints.put(QueryHints.SAMPLING, 0.1f)
query.getHints().put(QueryHints.SAMPLE_BY, "track");

6.7.2. Density Query

To populate heatmaps or other pre-rendered maps, GeoMesa can use server-side aggregation to map features to pixels. This results in much less network traffic, and subsequently much faster queries.

The result from a density query is an encoded iterator of (x, y, count), where x and y refer to the coordinates for the center of a pixel. In GeoServer, you can use the WPS DensityProcess to create a heatmap from the query result. See Heatmaps for more information.

Key Type GeoServer Conversion
QueryHints.DENSITY_BBOX ReferencedEnvelope Use WPS
QueryHints.DENSITY_WEIGHT String
QueryHints.DENSITY_WIDTH Integer
QueryHints.DENSITY_HEIGHT Integer
import org.geotools.data.Transaction
import org.geotools.geometry.jts.ReferencedEnvelope.ReferencedEnvelope
import org.geotools.referencing.CRS
import org.locationtech.geomesa.accumulo.iterators.KryoLazyDensityIterator
import org.locationtech.geomesa.index.conf.QueryHints

val bounds = new ReferencedEnvelope(-120.0, -110.0, 45.0, 55.0, CRS.decode("EPSG:4326"))
query.getHints.put(QueryHints.DENSITY_BBOX, bounds)
query.getHints.put(QueryHints.DENSITY_WIDTH, 500)
query.getHints.put(QueryHints.DENSITY_HEIGHT, 500)

val reader = dataStore.getFeatureReader(query, Transaction.AUTO_COMMIT)

val decode = KryoLazyDensityIterator.decodeResult(bounds, 500, 500)

while (reader.hasNext) {
    val pts = decode(reader.next())
    while (pts.hasNext) {
        val (x, y, weight) = pts.next()
        // do something with the cell

6.7.3. Arrow Encoding

GeoMesa supports returning features as Apache Arrow encoded vectors. This provides an optimized columnar memory layout for fast processing and interoperability with other systems.

The result of an Arrow query will be an iterator of SimpleFeatures, where the first attribute of each will be a byte array. Concatenated together, the byte arrays will form an Arrow file, in the Arrow streaming format (i.e. no footer).

In GeoServer you can use the ArrowConversionProcess. Otherwise, the encoding is controlled through the following query hints:

Key Type GeoServer Conversion
QueryHints.ARROW_ENCODE Boolean Use WPS
QueryHints.ARROW_INCLUDE_FID Boolean (optional)
QueryHints.ARROW_SORT_FIELD String (optional)
QueryHints.ARROW_SORT_REVERSE Boolean (optional)
QueryHints.ARROW_DICTIONARY_FIELDS String (optional)
QueryHints.ARROW_DICTIONARY_VALUES String (optional)
QueryHints.ARROW_DICTIONARY_CACHED Boolean (optional)
QueryHints.ARROW_MULTI_FILE Boolean (optional)
QueryHints.ARROW_DOUBLE_PASS Boolean (optional)
QueryHints.ARROW_BATCH_SIZE Integer (optional)


Arrow conversion requires jackson-core-2.6.x. Some versions of GeoServer ship with an older version, jackson-core-2.5.0.jar. After installing the GeoMesa GeoServer plugin, be sure to delete the older JAR from GeoServer’s WEB-INF/lib folder. Explanation of Hints ARROW_ENCODE

This hint is used to trigger an Arrow query. ARROW_INCLUDE_FID

This hint controls whether to include the feature ID as an Arrow vector or not. The default is to include it. ARROW_SORT_FIELD

This hint allows for sorting the results by a particular attribute. Only attribute names are supported, not arbitrary CQL. ARROW_SORT_REVERSE

This hint is used to flip sort order from normal (ascending) to reverse (descending). ARROW_DICTIONARY_FIELDS

This hint indicates which simple feature attributes should be dictionary encoded. It should be a comma-separated list of attribute names. ARROW_DICTIONARY_VALUES

This hint indicates known dictionary values to use for encoding each field. This allows for specifying a known dictionary up front, which means the dictionary doesn’t have to be computed. Values which are not indicated in the dictionary will be grouped under ‘other’.

The hint should be an encoded map of attribute names to attribute values. The hint should be encoded in comma-separated values format, where each line indicates a different attribute. The first item in each line is the attribute name, and the subsequent items are dictionary values. Standard CSV escaping can be used. The function org.locationtech.geomesa.utils.text.StringSerialization.encodeSeqMap can be used to encode a map of values.

import org.locationtech.geomesa.index.conf.QueryHints
import org.locationtech.geomesa.utils.text.StringSerialization.encodeSeqMap

val dictionaries1 =

// equivalent to dictionaries1
val dictionaries2 = encodSeqMap(Map("name" -> Array("Harry", "Hermione", "Severus"), "age" -> Array(20, 25, 30)))

query.getHints.put(QueryHints.ARROW_DICTIONARY_VALUES, dictionaries1) ARROW_DICTIONARY_CACHED

This hint indicates that cached statistics (top-k) will be used for dictionaries, if available. Otherwise, dictionaries will be computed based on the data returned, which may be slower. ARROW_MULTI_FILE

This hint will cause multiple logical Arrow files to be returned, instead of a single file. This will generally be faster, as no client-side merging needs to be done. However, any sorting will only be applied per file, not globally. Also, the end result tends to be larger (in bytes), as metadata and dictionary values may be repeated in different logical files. ARROW_DOUBLE_PASS

This hint will cause any dictionaries to be computed first, through a separate scan. A second scan will construct the Arrow files. This is the behavior of the initial GeoMesa Arrow implementation, and is only included for back compatibility. ARROW_BATCH_SIZE

This hint will restrict the number of features included in each Arrow record batch. An Arrow file contains a series of record batches -limiting the max size of each batch can allow memory-constrained systems to operate more easily. Example Query

import org.geotools.data.Transaction
import org.geotools.geometry.jts.ReferencedEnvelope.ReferencedEnvelope
import org.geotools.referencing.CRS
import org.locationtech.geomesa.accumulo.iterators.KryoLazyDensityIterator
import org.locationtech.geomesa.index.conf.QueryHints

query.getHints.put(QueryHints.ARROW_ENCODE, java.lang.Boolean.TRUE)

val reader = dataStore.getFeatureReader(query, Transaction.AUTO_COMMIT)
val os = new ByteArrayOutputStream()

while (reader.hasNext) {

// use ArrowStreamReader or other Arrow libraries to process bytes