20.6. Partitioned PostGIS Index Configuration

GeoMesa exposes a variety of configuration options that can be used to customize and optimize a given installation. See Setting Schema Options for details on setting configuration parameters. Note that most of the general options for GeoMesa stores are not supported by the partitioned PostGIS store, except as specified below.

Note

Most configurations can be updated after a schema has been created. See below for details specific to each configuration.

20.6.1. Configuring the Default Date Attribute

The default date attribute is the attribute that will be used for sorting data into partitions. See Setting the Indexed Date Attribute for details on how to specify it.

The default date cannot be changed after the schema has been created.

20.6.2. Configuring Indices

Attributes in the feature type may be marked for indexing, which will create a B-tree index on the associated table column. See Attribute Index for details on how to specify indices.

After the schema has been created, additional indices can be added through CREATE INDEX statements on the parent partition tables. See Table Design for a description of the partition tables.

20.6.3. Configuring Partition Size

Each feature type can be configured for a particular partition size. The partition size is specified as the number of hours that each partition will cover, and must be a divisor of 24, i.e. 1, 2, 3, 4, 6, 8, 12 or 24.

The number of hours should be based on the expected volume of data and type of queries. Due to partition pruning, PostGIS will not need to scan partitions that fall outside a given query window.

Partition size is configured with the key pg.partitions.interval.hours.

SimpleFeatureType sft = ....;
sft.getUserData().put("pg.partitions.interval.hours", "12");

After the schema has been created, changes to the partition size can be made through the update-schema command. Changes will not be applied to any existing partitions. If the partition size is increased, any recent partitions that would overlap with the new partition size will need to be manually dropped and the data re-inserted in the write-ahead table in order to prevent partition range conflict errors.

20.6.4. Configuring Index Resolution

Each feature type can be configured with a number of pages per range. The partition tables use a BRIN index, which is a lossy index structure. The number of data pages stored in each index range controls how lossy, and how large the index becomes. By default, Postgres stores 128 pages in each range. Storing fewer pages will generally make the index more efficient, at the cost of requiring more space; however, the optimal number will depend on data characteristics and typical query patterns.

The number of pages is configured with the key pg.partitions.pages-per-range.

SimpleFeatureType sft = ....;
sft.getUserData().put("pg.partitions.pages-per-range", "64");

The index resolution cannot be changed after the schema has been created.

20.6.5. Configuring Data Age-Off

Each feature type can be configured to automatically drop partitions older than a certain threshold. This is accomplished by setting the maximum number of partitions to keep. The age of the data will depend on the number of hours in each partition (see above). For example, keeping 14 partitions where each partition is 12 hours will keep the last week’s worth of data.

If not specified, data will not be dropped automatically.

Age-off is configured with the key pg.partitions.max.

SimpleFeatureType sft = ....;
sft.getUserData().put("pg.partitions.max", "14");

After the schema has been created, changes to the age-off can be made through the update-schema command, or by directly updating the geomesa_userdata table in Postgres. Changes will take effect within the next 10 minutes.

20.6.6. Configuring Filter Optimizations

By default, GeoMesa will ignore filters that contain the entire world, i.e. they encompass all of [-180, 180] longitude and [-90 90] latitude. This may speed up such queries, but it may also produce incorrect results if there are geometries outside the world bounds, or if the data is not stored in EPSG:4326/WGS84.

This behavior can be configured through the key pg.partitions.filter.world. The default value is false, which will ignore whole world filters.

SimpleFeatureType sft = ....;
 // enable filtering on "whole world" queries
sft.getUserData().put("pg.partitions.filter.world", "true");

After the schema has been created, changes to the filter optimization can be made through the update-schema command, or by directly updating the geomesa_userdata table in Postgres. Clients must be restarted in order to pick up the change.

20.6.7. Configuring Tablespaces

Each feature type can be configured to use different tablespaces for the different partition tables. Since all the writes initially go to the write-ahead table, having it on a fast disk may be beneficial. Conversely, since the main partitions are written once and not generally updated, having them on slower storage may be acceptable.

Any configured tablespaces must already exist in the PostreSQL instance being used.

Tablespaces are configured with the keys pg.partitions.tablespace.wa, pg.partitions.tablespace.wa-partitions and pg.partitions.tablespace.main. See Table Design for details on the different tables.

SimpleFeatureType sft = ....;
sft.getUserData().put("pg.partitions.tablespace.wa", "fasttablespace");

After the schema has been created, changes to the configured tablespaces can be made through the update-schema command, or by directly updating the partition_tablespaces table in Postgres. Changes will not be applied to any existing partitions.

20.6.8. Configuring the Maintenance Schedule

Maintenance scripts are run every 10 minutes to move data between the write-ahead table and the partitioned tables. By default, the schedule is randomized to avoid all feature types running maintenance at the same time. To specify the exact minute that the scripts should run, use the key pg.partitions.cron.minute.

The scheduled minute must be between 0 and 8, inclusive. For example, setting the scheduled minute to 1 will cause the scripts to run at 00:01, 00:11, 00:21, 00:31, etc.

The write-ahead table gets rolled over on the 9th minute of each ten minute block. Thus, running maintenance at minute 0 will move data out of the write-ahead table the fastest. Since the write-ahead table must be read for each query, moving data out of it faster may improve performance.

SimpleFeatureType sft = ....;
sft.getUserData().put("pg.partitions.cron.minute", "0");

After the schema has been created, changes to the schedule can be made through the update-schema command.