Couchbase vs Presto

To provide employees with the critical need of interactive querying, we’ve worked with Presto, an open-source distributed SQL query engine, over the years. Operating Presto at Pinterest’s scale has involved resolving quite a few challenges like, supporting deeply nested and huge thrift schemas, slow/ bad worker detection and remediation, auto-scaling cluster, graceful cluster shutdown and impersonation support for ldap authenticator.

Our infrastructure is built on top of Amazon EC2 and we leverage Amazon S3 for storing our data. This separates compute and storage layers, and allows multiple compute clusters to share the S3 data.

We have hundreds of petabytes of data and tens of thousands of Apache Hive tables. Our Presto clusters are comprised of a fleet of 450 r4.8xl EC2 instances. Presto clusters together have over 100 TBs of memory and 14K vcpu cores. Within Pinterest, we have close to more than 1,000 monthly active users (out of total 1,600+ Pinterest employees) using Presto, who run about 400K queries on these clusters per month.

Each query submitted to Presto cluster is logged to a Kafka topic via Singer. Singer is a logging agent built at Pinterest and we talked about it in a previous post. Each query is logged when it is submitted and when it finishes. When a Presto cluster crashes, we will have query submitted events without corresponding query finished events. These events enable us to capture the effect of cluster crashes over time.

Each Presto cluster at Pinterest has workers on a mix of dedicated AWS EC2 instances and Kubernetes pods. Kubernetes platform provides us with the capability to add and remove workers from a Presto cluster very quickly. The best-case latency on bringing up a new worker on Kubernetes is less than a minute. However, when the Kubernetes cluster itself is out of resources and needs to scale up, it can take up to ten minutes. Some other advantages of deploying on Kubernetes platform is that our Presto deployment becomes agnostic of cloud vendor, instance types, OS, etc.

#BigData #AWS #DataScience #DataEngineering

After using couchbase for over 4 years, we migrated to MongoDB and that was the best decision ever! I'm very disappointed with Couchbase's technical performance. Even though we received enterprise support and were a listed Couchbase Partner, the experience was horrible. With every contact, the sales team was trying to get me on a $7k+ license for access to features all other open source NoSQL databases get for free.

Here's why you should not use Couchbase

Full-text search Queries The full-text search often returns a different number of results if you run the same query multiple types

N1QL queries Configuring the indexes correctly is next to impossible. It's poorly documented and nobody seems to know what to do, even the Couchbase support engineers have no clue what they are doing.

Community support I posted several problems on the forum and I never once received a useful answer

Enterprise support It's very expensive. $7k+. The team constantly tried to get me to buy even though the community edition wasn't working great

Autonomous Operator It's actually just a poorly configured Kubernetes role that no matter what I did, I couldn't get it to work. The support team was useless. Same lack of documentation. If you do get it to work, you need 6 servers at least to meet their minimum requirements.

Couchbase cloud Typical for Couchbase, the user experience is awful and I could never get it to work.

Minimum requirements The minimum requirements in production are 6 servers. On AWS the calculated monthly cost would be ~$600. We achieved better performance using a $16 MongoDB instance on the Mongo Atlas Cloud

writing queries is a nightmare While N1QL is similar to SQL and it's easier to write because of the familiarity, that isn't entirely true. The "smart index" that Couchbase advertises is not smart at all. Creating an index with 5 fields, and only using 4 of them won't result in Couchbase using the same index, so you have to create a new one.

Couchbase UI The UI that comes with every database deployment is full of bugs, barely functional and the developer experience is poor. When I asked Couchbase about it, they basically said they don't care because real developers use SQL directly from code

Consumes too much RAM Couchbase is shipped with a smaller Memcached instance to handle the in-memory cache. Memcached ends up using 8 GB of RAM for 5000 documents! I'm not kidding! We had less than 5000 docs on a Couchbase instance and less than 20 indexes and RAM consumption was always over 8 GB

Memory allocations are useless I asked the Couchbase team a question: If a bucket has 1 GB allocated, what happens when I have more than 1GB stored? Does it overflow? Does it cache somewhere? Do I get an error? I always received the same answer: If you buy the Couchbase enterprise then we can guide you.

We implemented our first large scale EPR application from naologic.com using CouchDB .

Very fast, replication works great, doesn't consume much RAM, queries are blazing fast but we found a problem: the queries were very hard to write, it took a long time to figure out the API, we had to go and write our own @nodejs library to make it work properly.

It lost most of its support. Since then, we migrated to Couchbase and the learning curve was steep but all worth it. Memcached indexing out of the box, full text search works great.