Damir Maksyutov

Instrument classification of clients' portfolios was known to be a slow process on large portfolios. The process would regularly timeout and require clients to restart the product. Proper optimization or rewrite was long due.

Related C++ code was old and complex, and the team did not understand the implementation. Besides, no instrumentation in the code could help identify the algorithm bottlenecks.

Due to the complexity of the code and the approaching deadline, I focused on understanding the bottlenecks by investigating the actual stack traces of the running classification code rather than going through the code in detail.

After collecting enough samples and writing a helper processing program in Python, it was obvious that most of the algorithm bottlenecks were due to recalculating the same multi-dimensional arrays and graph traversals. It was clear that introducing a simple cache data structure would solve the problem.

After introducing the cache layer to the algorithm, testing, and rolling it out, it was shown to provide an average improvement of around 40% for the time spent in instrument classification and eliminate timeouts.

Developing complex applications in constantly evolving codebases can be a challenging task. Scheduling an overnight build against the latest code can result in unexpected build or test failures that require manual resolution, wasting time in the morning.

To improve developer experience in our team, I created a tool to remotely via iOS app control overnight build and validation for Xcode projects and perform automated code check-ins. Upon detecting a local failure on the latest code, it notified the developer and allowed them to choose another revision to perform the build.

The states were managed by a Node.js app on an Azure server and backed by MongoDB. With the iOS application written in Swift, XCode operations were performed by a macOS daemon.