TSDB on geo_line

[wchaparro]

TSDB on geo_line. https://github.com/elastic/elasticsearch/issues/85274

Time series data on Geo line

Business Justification:

Building upon the foundation for time series data stored as sampled metrics in the TSDB project, there are geo based use cases which can take advantage of the improvements in storage, and new time based iterator aggregations to solve real world problems.

Use cases

• GPS tracking: A user wants to predict when a delivery will arrive at its destination. The delivery object location data points are Geo coordinates, id, timestamp. The user would want to know the direction the object is headed, the speed, and along the trajectory any acceleration or deceleration (as multiple modes of transportation may be involved). Aircraft tracking, truck fleet tracking, taxi tracking to predict demand

Similar, related use cases

• A rescue team wants to locate a person for rescue. They can use the GPS coordinates of the user's phone to identify last known location, direction, and speed. Using this information they can identify potential locations for rescue operations.
• a user would like to track tornado / hurricane paths / storm paths, understand speed, direction, acceleration in order to make continuous predictions.
• Traffic management: data collection from mobile devices, mobility patterns, predict flow of traffic, end goal to optimize traffic signal flow

Requirements

list of requirements

Dependencies

• [ ] #74660 Ingest phase

Design notes

design or architecture related notes

Concerns

• One of the aspects of this data type is that data points may not be not measured consistently. For example, you might get information consistently for a period of time from an asset, and then it will slow down, perhaps stop (for two days), and then will resume sending data again. Is this inconsistency in the rate of data a problem?

[elasticmachine]

Pinging @elastic/es-analytics-geo (Team:Analytics)

[giladgal]

TSDB is designed for documents that are samples from a continuous function of time, and asset tracking is exactly such data, with the lon and lat being the metrics. If you were to draw a graph of the longitude or of latitude as a function of time, your pencil would not leave the paper, and the documents in the index would be points on that line.

Re.

• One of the aspects of this data type is that data points may not be not measured consistently. For example, you might get information consistently for a period of time from an asset, and then it will slow down, perhaps stop (for two days), and then will resume sending data again. Is this inconsistency in the rate of data a problem?

At least in some situations this is a feature, e.g. AIS systems transmit more frequently when the assets changes location more frequently. In any case derivative information (e.g. direction, speed, and acceleration) can only be as accurate as the sampling rate supports.

Additional functionality that could be developed based on that includes downsampling to reduce index size, extrapolation to predict future path and interpolation to draw a path based on a series of known locations.

[wchaparro]

Examples: https://www.vendeeglobe.org/en/tracking-map https://maplibre.org/maplibre-gl-js-docs/example/live-geojson/ https://flightaware.com/

[craigtaverner]

The support for line simplification in geo_line in TSDB has been completed in https://github.com/elastic/elasticsearch/pull/94954.

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)