23. Combined Data Store Views

GeoMesa provides two generic options for combining multiple data stores into a single view. The first option is a ‘merged’ view, where each data store contain a different set of features and they are queried concurrently. This can be used, for example, to keep a fast, small cache of the most recent data, or to archive off older data into lower-cost storage. The second option is a ‘routed’ view, where each data store contains the same set of features and each query is sent to only one data store. This can be used, for example, to support secondary indexing.

In order to use a layer through the a data store view, the SimpleFeatureType must exist (and match) in all of the underlying stores. If a schema exists in some stores but not all of them, it will not show up in the combined view.

23.1. Installation

The GeoMesa data store views are available in the geomesa-index-api JAR, which is bundled by default with all GeoMesa data stores. The underlying stores being combined must be installed separately; see the relevant documentation for each store. Depending on the stores, you may need to resolve classpath conflicts between their dependencies.

23.2. Merged Data Store View

The GeoMesa Merged Data Store View provides a unified way to query multiple data stores concurrently. For example, you may want to store more recent data in an HBase data store instance, and older data in a FileSystem data store instance to reduce storage costs, but still provide a single layer to query both.

In comparison to the Lambda Data Store, the Merged Data Store View does not provide any management for transitioning features between data stores. All writes must be done through the underlying stores directly, and if the same features exist in multiple stores, they may be returned multiple times in a single query response.

23.2.1. Usage

The Merged Data Store View can be instantiated through the standard GeoTools DataStoreFinder or the GeoServer New Data Source page. It only requires a single parameter: geomesa.merged.stores.

geomesa.merged.stores should be a TypeSafe Config string that defines the parameters for each merged store. The config should have a top-level key of stores that is a list of objects, where each object is a set of key-value pairs corresponding to the parameters for a single data store.

For example, to merge a GeoMesa Accumulo data store with a PostGis data store, you could use the following config:

{
  "stores": [
    {
      "accumulo.zookeepers": "localhost",
      "accumulo.instance.id": "test",
      "accumulo.catalog": "test",
      "accumulo.user": "test",
      "accumulo.password": "test"
    },
    {
      "dbtype": "postgis",
      "host": "localhost",
      "port": "5432",
      "database": "test",
      "user": "test",
      "passwd": "test"
    }
  ]
}

23.2.2. Query Filtering

If the stores being merged have known characteristics, filters can be applied selectively to each store in order to speed up queries. The filter is specified along with the other store parameters, under the key geomesa.merged.store.filter, and should be an ECQL filter string.

The filter will be applied against any query, in addition to the query filter. This can be used to short-circuit queries that are not relevant for a particular store. For example, if one store contains features from the past 24 hours, and a second store contains features older than 24 hours, then you could configure them with time-based filters:

{
  "stores": [
    {
      "other.store.params": "go here...",
      "geomesa.merged.store.filter": "dtg >= currentDate('-P1D')"
    },
    {
      "other.store.params": "go here...",
      "geomesa.merged.store.filter": "dtg < currentDate('-P1D')"
    }
  ]
}

23.2.3. Store Configuration Provider

As an alternative to specifying geomesa.merged.stores, config loading can be delegated to a provider interface: org.locationtech.geomesa.index.view.MergedViewConfigLoader. Implementations of this class must be made available via Java SPI loading, with a special descriptor defined in META-INF/services. To use a config provider, use the parameter geomesa.merged.loader set to the full class name of the provider class. In GeoServer, available providers will be displayed in a selection list.

Note that you may combine the explicit configuration of geomesa.merged.stores with the delegated configuration of geomesa.merged.loader, in which case the two configurations will be merged.

23.3. Routed Data Store View

The GeoMesa Routed Data Store View provides a unified way to route queries to multiple data stores. For example, you may want to store a geospatial index in an HBase data store instance, and attribute indices in a FileSystem data store instance to reduce storage costs, but still provide a single layer to query both.

All writes must be done through the underlying stores directly. Although not required, generally the same features should exist in each store. If the features vary from store to store, inconsistencies may occur when querying.

23.3.1. Usage

The Routed Data Store View can be instantiated through the standard GeoTools DataStoreFinder or the GeoServer New Data Source page. It only requires a single parameter: geomesa.routed.stores.

geomesa.routed.stores should be a TypeSafe Config string that defines the parameters for each underlying store. The config should have a top-level key of stores that is a list of objects, where each object is a set of key-value pairs corresponding to the parameters for a single data store.

For example, to combine a GeoMesa Accumulo data store with a PostGis data store, you could use the following config:

{
  "stores": [
    {
      "accumulo.zookeepers": "localhost",
      "accumulo.instance.id": "test",
      "accumulo.catalog": "test",
      "accumulo.user": "test",
      "accumulo.password": "test"
    },
    {
      "dbtype": "postgis",
      "host": "localhost",
      "port": "5432",
      "database": "test",
      "user": "test",
      "passwd": "test"
    }
  ]
}

23.3.2. Query Routing

The routed view requires a way to route any incoming query to a particular data store. By default, queries can be routed based on the query filter attributes. For more advanced use cases, see below.

To route based on attributes, each store can be configured with the attribute combinations that it can handle efficiently. The attributes are specified under the key geomesa.route.attributes, alongside the other data store parameters. The attributes should be an array of groups of names that can be queried together. An empty attribute array corresponds to Filter.INCLUDE, and will be used as a fall-back if nothing else is matched. The special string "id" can be used to match a feature ID query. If a query does not match any of the configured attributes, then it will not return any results.

The following example will route queries such as INCLUDE, IN ('1', '2'), bbox(geom,...) and bbox(geom, ...) and dtg during ... to the first store, while queries such as name = ... and age > ... will be routed to the second store:

{
  "stores": [
    {
      "other.store.params": "go here...",
      "geomesa.route.attributes": [ [], [ "id" ], [ "geom" ], [ "dtg", "geom" ] ]
    },
    {
      "other.store.params": "go here...",
      "geomesa.route.attributes": [ [ "name" ], [ "age" ] ]
    }
  ]
}

23.3.2.1. Custom Routing

As an alternative to routing by attribute, routing can be delegated to a provider interface: org.locationtech.geomesa.index.view.RouteSelector. Implementations of this class must be made available via Java SPI loading, with a special descriptor defined in META-INF/services. To use a config provider, use the parameter geomesa.route.selector set to the full class name of the provider class. In GeoServer, available providers will be displayed in a selection list.