A eulogy for RedisGraph

Last month, Redis announced that RedisGraph, the graph database it introduced only a scant five years ago, is going to be phased out. New licenses are no longer available, and support for existing customers will expire in early 2025.

This is happening despite the fact that, all things considered, RedisGraph was quite an impressive offering. As we have previously discussed, it used a novel and, to the best of my knowledge, unique way of representing and storing graph data that gave it some pretty significant performance advantages for a pretty significant number of use cases. At the time, the company seemed quite gung-ho about pushing the new product – as I recall, it even got into a small spat with TigerGraph about benchmark results.

This begs the question – why abandon it? Well, the end-of-life announcement from Redis mentions two key factors. Firstly, that graph databases require extensive and graph-specific technical knowledge to make good use of, which doesn’t fit with Redis’ “simple and joyful” vision of how its products are “meant to be”. Although the company is not wrong about the traditional barrier to entry for graph, there are a number of vendors in the space that positively pride themselves on addressing this very issue. Consequently, this says more about RedisGraph than it does about graphs as a whole.

Moreover, to me this seems like a way for Redis to save face by positioning its decision to discontinue RedisGraph as one that is fundamentally beneficial to its overall userbase, when in reality it is actively harmful to a small subset of it – namely, the customers that bought into RedisGraph. Talking about how much they care about developers and “building things that developers love” smacks of hypocrisy when they are leaving a small but nonetheless existent number of developers in the lurch with this move.

Secondly, the announcement claims that it was proving to be a greater investment than anticipated to build, sell, and support RedisGraph than Redis expected. Presumably, it was also not seeing the return it wanted on that investment, as the company admits it was expecting “exponential growth” in the graph space. My suspicion is that Redis was taken in by the longstanding hype surrounding graph databases, pushed by many graph vendors (and, to be brutally honest, analysts). For years, the narrative has been that graph is just about to “arrive” and become explosively popular, usually due to the inherent advantages it offers over, say, RDBMS. This has not happened, and it will probably never happen: graph databases will almost certainly remain a relatively niche storage technology deployed in use cases that particularly benefit from graph.

And that’s fine. Indeed, much of the graph space seems to have realised this, and vendors are increasingly selling their graphs on the strength of the use cases they can address rather than the inherent properties of graph databases. This is very much a good thing. Others, like Redis, have simply exited the space, though usually with less fanfare. Regardless, there are still plenty of graph vendors – including several recent entrants to the space – to go around.

If you want to read more of my thoughts on the state of the graph space, please look forward to our upcoming Graph Database Market Update, which will contain – among other things – a more detailed discussion of these points as they apply to the entire space, not just Redis.