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

For such a more realtime-focused, data-centered application like an exchange, it's not the frontend or backend that matter much. In fact for that, they can do away with any of the popular frameworks like React/Vue/Angular for the frontend and Go/Python for the backend. For example uniswap's frontend (although much simpler than binance) is built in React. The main interesting part here would be how they are able to handle updating data so quickly. In my opinion, they might be heavily reliant on realtime processing systems like Kafka+Kafka Streams, Apache Flink or Apache Spark Stream or similar. For more processing heavy but not so real-time processing, they might be relying on OLAP and/or warehousing tools like Cassandra/Redshift. They could have also optimized few high frequent queries using NoSQL stores like mongodb (for persistance) and in-memory cache like Redis (for further perfomance boost to get millisecond latencies).

We have a Kafka topic having events of type A and type B. We need to perform an inner join on both type of events using some common field (primary-key). The joined events to be inserted in Elasticsearch.

In usual cases, type A and type B events (with same key) observed to be close upto 15 minutes. But in some cases they may be far from each other, lets say 6 hours. Sometimes event of either of the types never come.

In all cases, we should be able to find joined events instantly after they are joined and not-joined events within 15 minutes.