Why does Twitter not use Cassandra to store tweets)

Address: http://highscalability.com/blog/2010/7/11/so-why-is-twitter-really-not-using-cassandra-to-store-tweets.html

At present, the focus of the discussion is that Cassandra, as the main nosql product, has been stripped of gorgeous clothes. First, Twitter has abandoned Cassandra. It is said that Facebook has abandoned Cassandra as a place where Cassandra was born. Nosql has failed.CommunityBut cassandra has failed? There are obstacles to development, but there is no failure.

(Note: This article contains a NSFW flowchart. Some people are very sensitive to bad things. If you are such a person, please do not continue to read it! There is no naked figure, no other things, only a few words. However, these are the flowcharts that I have always been most interested in, so they are worth it !)

Does twitter really give up Cassandra? According to the post posted by Twitter: Cassandra at Twitter today, they use Cassandra to store geographic information and analysis. However, Twitter will not use Cassandra to store tweets. Why? Twitter explained: "This is a strategic change. We will continue to maintain our original mysql-based storage. We believe that there is no time for large-scale data migration to a new technology ."

Twitter is busy with other emergencies, and has no time to migrate data originally stored in MySQL to Cassandra.

Now is the best time to share a flowchart that I got several years ago. This may be like the thought process before Twitter makes a decision.

Flow chart for project demo-making

 

I laugh every time I read this. This is the fact that every company has a decision on how to place resources. You need to build, buy, re-build, scale, and suspend your application. An engineer may feel that it is correct to try again. But the best business decisions may be completely different.

Joel Spolsky once issued a statement saying that you should never rewrite yourCode. Remember Netscape? There are so many companies that have been on the Internet but have disappeared. Because they have made this critical timing error, they have rewritten the code from the beginning.

Twitter does not want to repeat the same mistakes. Does this mean that Cassandra and nosql fail? No! I think this is just a brilliant project plan. This actually provides multiple platforms for different situations.

Is Twitter running enough? Brad McCarty gave a good example in Twitter implements more features; ignores its broken platform: Twitter should first set up its own infrastructure before migration. But now the problem is that Twitter has devices, but it does not have the core tweets engine. Till now, Twitter is still running very well. Therefore, based on this situation, it is more reasonable to keep the engine of tweets working. Although it is ugly, it is more reasonable to start migrating some new trivial functions?

These are unavoidable. Cassandra has always lived under the care of Facebook and early adopters. Until recently, Cassandra left home to enter this complex world where users have a wide variety of needs. This will make Cassandra mature. Cassandra needs to design new features that need to be added, rebuild the architecture for existing features, and tell people what Cassandra is good. These problems have not been solved until now. In addition, for developers, switching to a new technology platform means they need to learn, learn, and overcome various obstacles. This process is difficult. So we need to consider the problem about Cassandra and some products around us. This is the problem of switching to a new technology. We are like a mouse in this space.

does Facebook really give up Cassandra? No! Facebook still has 150 Cassandra cloud nodes for search, supporting 0.5 billion of users and TB of data, and these numbers are growing every day. Now, it is interesting to take a look at the speed of Cassandra's popularity in other projects or groups. When will there be a technology that implements Cassandra? Will he continue to develop? We do not know, but we can know that Facebook will continue to use Cassandra in the future.