Bug: Janusgraph Filter queries highly inefficient at medium to large scale

[etiennedi]

Reproduce

1. Import 50k to 100k things/actions of different classes
2. Perform simple GQL queries for those things/action classes

Expected behavior

At the above scale all queries should be fast, even list queries which aren't the strength of the NoSQL based setup

Actual Behavior

Depending on which class is used, queries are very slow, for example "Get 100 of Thing Foo" takes around 20 seconds

Analysis

Setting Janusgraph to debug mode, we can see that the queries to the index backend (ES) aren't restrictive at all:

They should be much more narrowed down, instead it seems janus is iterating over every single thing out there.