Documentation is still a bit all over the place, at least for the .Net space. Some information is only apparent after diving into GitHub issues, it's not a total disaster but could definitely be better. Especially when some documentation recommends entirely incorrect dependencies or misleading support for some gremlin commands.
However the product itself is great, I've been using the gremlin interface almost exclusively. I will say however, that having used JanusGraph and other TinkerPop compatible products some of the more advanced gremlin features like branch and choose are absent which can make porting some applications problematic.
A further potential annoyance or benefit is the fixed usage of GraphSON for the documents in Cosmos which can make the results of your query appear radically different to what you would expect if you're familiar with other TinkerPop graphs.
They have "bounded staleness" as an option for consistency, but obviously it doesn't break the laws of physics, latency is latency and you still have to account for that somewhat but at least you have the bounds to work with.
Comments
Documentation is still a bit all over the place, at least for the .Net space. Some information is only apparent after diving into GitHub issues, it's not a total disaster but could definitely be better. Especially when some documentation recommends entirely incorrect dependencies or misleading support for some gremlin commands.
However the product itself is great, I've been using the gremlin interface almost exclusively. I will say however, that having used JanusGraph and other TinkerPop compatible products some of the more advanced gremlin features like branch and choose are absent which can make porting some applications problematic.
A further potential annoyance or benefit is the fixed usage of GraphSON for the documents in Cosmos which can make the results of your query appear radically different to what you would expect if you're familiar with other TinkerPop graphs.
I wonder how good is their geo-replication.
If I increment a counter in 2 regions/cross Atlantic, at a rate of 1000 RPS.
am I going to observe consistent count in both region on each read request?
I am not sure how they defeat laws of physics.
They have "bounded staleness" as an option for consistency, but obviously it doesn't break the laws of physics, latency is latency and you still have to account for that somewhat but at least you have the bounds to work with.