CentroidCalculator needs protection against very small polygons

[iverase]

I was working with one of my datasets on the geoshape doc values branch and some of the documents are now eroding out when indexing. After having a closer look, this polygons are very small (probably lines) and they only have four points. for example:

POLYGON(((-98.1912833 50.9375,-98.1912812 50.9375,-98.1912832 50.9375002,-98.1912833 50.9375)))

Although they can be more complicated and the problematic polygon is a hole:

POLYGON((-4.385064 55.2259599,-4.385056 55.2259224,-4.3850466 55.2258994,-4.3849755 55.2258574,-4.3849339 55.2258589,-4.3847033 55.2258742,-4.3846805 55.2258818,-4.3846282 55.2259132,-4.3846215 55.2259247,-4.3846121 55.2259683,-4.3846147 55.2259798,-4.3846369 55.2260157,-4.3846472 55.2260241,-4.3846697 55.2260409,-4.3846952 55.2260562,-4.384765 55.22608,-4.3848199 55.2260861,-4.3848481 55.2260845,-4.3849245 55.2260761,-4.3849393 55.22607,-4.3849996 55.2260432,-4.3850131 55.2260364,-4.3850426 55.2259989,-4.385064 55.2259599),(-4.3850104 55.2259583,-4.385005 55.2259752,-4.384997 55.2259892,-4.3849339 55.2259981,-4.3849272 55.2259308,-4.3850016 55.2259262,-4.385005 55.2259377,-4.3850104 55.2259583),(-4.3849996 55.2259193,-4.3847502 55.2259331,-4.3847548 55.2258921,-4.3848012 55.2258895,-4.3849219 55.2258811,-4.3849514 55.2258818,-4.3849728 55.2258933,-4.3849996 55.2259193),(-4.3849917 55.2259984,-4.3849849 55.2260103,-4.3849771 55.2260192,-4.3849701 55.2260019,-4.3849917 55.2259984),(-4.3846608 55.2259374,-4.384663 55.2259316,-4.3846711 55.2259201,-4.3846992 55.225904,-4.384718 55.2258941,-4.3847434 55.2258927,-4.3847314 55.2259407,-4.3849098 55.2259316,-4.3849098 55.2259492,-4.3848843 55.2259515,-4.3849017 55.2260119,-4.3849567 55.226005,-4.3849701 55.2260272,-4.3849299 55.2260486,-4.3849192 55.2260295,-4.384883 55.2260188,-4.3848776 55.2260119,-4.3848441 55.2260149,-4.3848441 55.2260226,-4.3847864 55.2260241,-4.384722 55.2259652,-4.3847053 55.2259706,-4.384683 55.225954,-4.3846608 55.2259374),(-4.3846541 55.2259549,-4.384698 55.2259883,-4.3847173 55.2259828,-4.3847743 55.2260333,-4.3847891 55.2260356,-4.3848146 55.226031,-4.3848199 55.2260409,-4.3848387 55.2260417,-4.3848494 55.2260593,-4.3848092 55.2260616,-4.3847623 55.2260539,-4.3847341 55.2260432,-4.3847046 55.2260279,-4.3846738 55.2260062,-4.3846496 55.2259844,-4.3846429 55.2259737,-4.3846523 55.2259714,-4.384651 55.2259629,-4.3846541 55.2259549),(-4.3846608 55.2259374,-4.3846559 55.2259502,-4.3846541 55.2259549,-4.3846608 55.2259374))

The issue is that when calculating the area, this value is zero and makes the centroid calculator to generate NaN for the coordinates which errors out when trying to serialise it.

cc @talevy

[elasticmachine]

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

[talevy]

I'll dig into this, the problem seems to be when a triangle in a polygon has zero weight, its centroid degrades to infinity.

for example, the second polygon you shared has a total ring area of 0, while the weighted centroid sum for each coordinate is something that is close to zero at 1e-11.

[talevy]

Fixed in #52782