Cockroachdb vs kudu

General
#1

I’m currently working on a project involving storing data with simple relationships among tables (no complex join required). Although it is desirable to have somewhat strong fault tolerant attributes. Some tables could have records in the order of say 100M~200M. I’m currently using cockroach db. The ease of deployment and the level of community support really help me expedite the development. The ease of deploying/administrating the cluster is also a great feature.

However, some colleague of mine were complaining that we can only get 1000 qps with cockroach db. And due to it’s intrinsic design of not being ‘column based’ it cannot be our long term solution. And they are suggesting using something like kudu. Not knowing the internal design of cockroach db I cannot really say if they are making a good assertion or not. I’ll be very appreciated if anyone can provide me with some insights.

Thanks.

#2 
joel,

Here is an insight, born of experience, with little knowledge of anything, let alone kudu.

There are all always new kids comming along with their "better solutions".

From what you say it seems you have a working solution with cockroach already.

So, it's up to them to demonstrate the superiority of whatever it is they propose.

As always it comes down to what you want to do. What is important to your application.

Cockroach is built to be a reliable, distributed, SQL engine. Consistancy, ACID, fault tolerance and all that.

Which no doubt entails a performance hit over other non-SQL solutions like kudu.

If your application values some kind of performance metric over correctness then maybe other solutions will do.

One does not need to look so deeply into the internal design to make a guess at this.
#3

Hi Joe,

involving storing data with simple relationships among tables (no complex join required)

The kudu faq [1] says it doesn’t support secondary indexes, unless the simple relationships can all be met with the primary key, kudu won’t be all that great for your project

This is the first time I hear about kudu, but their FAQ page has a good amount of info on their current limitations, it should help you identify reasons why it may not be a good fit for your project.

Regarding the 1000 qps, have you had a chance to see if you could improve it in your setup?

[1] https://kudu.apache.org/faq.html#does-kudu-support-secondary-indexes

Regards,