C++ Developer for the YDB String Tables Team

YDB is a distributed, fault-tolerant, open-source relational database that combines high availability, scalability, and performance. Alongside its serverless technology stack, YDB enables the creation of data storage and processing systems for the most complex and demanding applications.

YDB is also a platform for building infrastructure components. We offer users not only reliable transactions on tables but also ready-made solutions based on the engine itself: persistent queues, federated queries, network block storage for virtual machines (Yandex Network Block Store), and more.

What tasks await you

Distributed Transactions YDB ensures strict consistency through distributed transactions. They can span data located in different tables or within a single table across multiple nodes. The number of nodes can reach tens of thousands. YDB uses deterministic transactions as stages of executing user transactions with additional orchestration and locking.

Partitioning A string table in the database can be sharded by ranges of primary key values. Each table shard is responsible for its own range of primary keys. The key ranges served by different shards do not overlap. Different shards of a table can be served by different servers of the distributed database (including those located in different locations) and can also independently move between servers for rebalancing or to maintain shard operability in case of server or network equipment failures.

Primary and Secondary Indexes YDB automatically creates an index on the primary key, so selections with a condition on the primary key are always performed efficiently, affecting only the required rows. A selection with a condition imposed on one or more non-key columns typically leads to a full table scan. To make such selections efficient, secondary indexes must be used. The current version of YDB implements synchronous and asynchronous global secondary indexes.

Vector Search The k-nearest neighbors (k-NN) search is an optimization problem that involves finding the k closest points to a query point. This can be useful in various applications such as image classification, recommendation systems, and much more. Solving the k-NN problem is divided into two major subclasses of methods: exact and approximate.

Full-Text Search Full-text indexes are an experimental type of secondary index that provides fast text search capabilities. The main idea of a full-text index is to store a mapping of terms to rows that contain those terms.

More about backend at Yandex — in the channel Yandex for Backend

We expect that you

• Have at least three years of development experience in C++
• Know everything that is typically asked in interviews: classic algorithms, data structures, etc.
• Understand key aspects of building fault-tolerant distributed systems
• Have read the classic books by Date and Kleppmann :)