In my previous post, I described how to correctly install and use OpenTSDB. After some time, I decided to move on to other solution.
Before everything, we need to know one thing. Because of IoT, the demand for storing sensor data has increased dramatically. Many new projects emerged, some are good, some are bad. They are different in technologies used, how fast they are and what kind of features they support.
You can read the full list of all IoT timeseries databases that can be used for storing data of you Internet of Things projects or startup.
Problems of OpenTSDB
OpenTSDB is great, don’t get me wrong. But when you try to use is with some more complex projects and customer demands, you can quickly hit the wall. It’s mostly because it involves a lot of moving parts to make it work (Hadoop, HBase, ZooKeeper). If one of the parts fail, the whole thing fails. Sure, you can replicate each thing and make it more robust, but you will also spend more money. When you are starting, it’s a over optimization and waste of money (that you don’t have).
Aggregation of the data is another problem. It does support basic function like min, max, avg etc. I spent days investigating the problem why avg aggregation is not working correctly when I filter by multiple tags. It just didn’t want to work and I couldn’t find anything in the docs. I asked on Google group and after some time I got a reply that I must use another aggregation function and that even that doesn’t work 100% as I want it. Another problem is when I want to get just one value – for example avg of all values from X to now. Not possible!
No clients to talk with OpenTSDB is another problem for me. Sure, storing the data with socket API is super simple and can be easily integrated in every language. The HTTP API is another story. Sure, again it shouldn’t be a problem to implement my own client, but why waste time with this?
Development of the OpenTSDB is slow and it takes ages for new features to be integrated. One of them (one of the most important for me) is an ability to support time zones. It’s used when downsampling data to one day (or even more) so data is correctly grouped. There was some work, but until today it still wasn’t implemented. Too bad.
On the bright side, OpenTSDB is super fast. I was able to store and load data as super fast rate – loading 3 million records in few seconds is for me super fast. Try it with relational database and you will be quickly disappointed.
KairosDB to the rescue
I remember when I was doing a research, I noticed KairosDB but I didn’t spend too much time testing it. It just wasn’t appealing and I didn’t know how it actually works. Big mistake.
KairosDB uses Cassandra to store data (compared to HBase used with OpenTSDB) and it’s actually a rewritten and upgraded version of OpenTSDB. It has evolved into great project. It has many more features: many more (and fully working) aggregation methods, option to easily delete metric or datapoint, easy extensibility with plugins etc. It has great clients and has much more active community. I remember when I asked a question on OpenTSDB Google group and waited weeks for an answer (I’m not forcing anyone to provide the support, because after all, it’s an opensource project), while on KairosDB Google group I got it within a day.
Why is this important you might ask? Well, when you are catching deadlines and something goes wrong, responsive community is very important. Sometimes this kind of things can be a difference between success and a failure.