May 2019 Summaries

The current approach to automated testing often leaves developers scrambling to diagnose issues, as test failures only reveal that something went wrong without providing context on what happened in the application. Surfacing JavaScript errors from Selenium tests by integrating with Sentry's JavaScript SDK can provide valuable insights into exactly where and why problems occurred, allowing developers to quickly identify and fix issues before deployment. This approach enables the display of error information in test reports, facilitating easier debugging and reducing the need for manual reproduction and debugging efforts.

Sentry is introducing several key features to its native crash reporting system, including support for symbol servers, native handling of PDBs, and a full SDK for C and C++. Symbol servers provide convenient integration with debuggers, allowing Sentry to automatically download debug information as needed. The new feature also simplifies uploads by eliminating the need to convert or upload PDBs, resolving inline function calls, and cleaning up stack trace views. Additionally, Sentry is releasing its native SDK, called Sentrypad, which provides a high-level SDK experience with a familiar face, easy integration, and a convenient C API for configuring the SDK. The new features aim to enhance the overall crash reporting experience at Sentry, making it more accessible and user-friendly for developers working on native applications.

Sentry has recently started supporting RAM bundles for React Native projects, which are a new format of application packaging that optimizes load times and improves the debugging experience by enabling on-demand loading of individual app modules. To support this format, Sentry's `sentry-cli` and `react-native-sentry` SDKs have been updated, allowing users to upload their RAM bundles along with source maps for better symbolicated stack traces. The new format is designed to reduce the size of the bundle and improve performance by allowing React Native runtime to load specific modules only when needed. Additionally, Sentry's `upload-sourcemaps` command now automatically detects and parses Indexed RAM bundles and their source maps, making it easier for users to upload their RAM bundles and take advantage of the improved debugging experience.

Sentry has introduced a new search infrastructure called Snuba, which is built on top of ClickHouse, an open-source column-oriented database management system. Snuba powers Sentry's primary storage and query service for event data and provides features such as search, graphs, issue detail pages, rule processing queries, and more. The need for Snuba arose from Sentry's growth and the increasing complexity of its data model, which required a flat event model that can be queried ad-hoc without denormalization. ClickHouse was chosen over other OLAP systems due to its open-source nature, operational simplicity, row-level sorting, real-time querying capabilities, and PREWHERE clause. Snuba abstracts ClickHouse away from Sentry and provides a rich query interface through a Flask web service, which hides underlying complexity from application developers. The system uses Kafka for event processing, batch writes to ClickHouse, and Redis for caching individual query results.

The Sentry team has released version 9.1, which includes a range of bug fixes, improvements, and new features. The major change is that Sentry will no longer support multiple abstractions, limiting it to only supporting service providers used by the project itself. This means that users will be required to use Postgres as their database instead of other options like SQL databases or Riak. The team is also introducing a new search infrastructure called Snuba, which is powered by Clickhouse and will provide more powerful search capabilities. Additionally, Sentry will be moving from Python 2 to Python 3, but the transition to Python 3 has not yet been completed. The team is also offering a single tenant solution for compliance concerns, allowing customers to offload data isolation risks to Sentry. Looking ahead, future versions of Sentry will include performance monitoring features and support for analyzing crash dumps from more runtimes.

Sentry developed an experimentation system using Facebook's PlanOut library, which allows for easy and complex experiment writing, negligible performance impact, and measurable results. The system requires installing the SDK, sending an error, and defining a method `assign()` to determine user assignment to variants. Sentry ran an experiment where users were shown either a "See Sample Event" button or a more subtle link, with 12% improvement in activation rate for the button variant compared to the control group. The system logs exposure and events, integrates with the frontend using AJAX requests, and enables analysis of click-through rates and other metrics.

The Sentry Integration Platform has been launched, allowing developers to build workflows on top of Sentry. The new integration is with Clubhouse, a project management tool that focuses on enabling collaborative work among team members. With the integration, errors and features can be tracked in both platforms, helping teams prioritize and fix bugs faster. Users can create new Clubhouse Stories from Sentry Issues, link existing ones, and move associated issues to complete when a story is done in Clubhouse. This integration aims to reduce toggling between tools, allowing developers to spend more time on other tasks.

Welcome Zac Propersi Sentry joins the team with a new hire, Zac Propersi, who brings his unique set of skills to the application engineering team. With strengths in motivating teams through interpretive dance and finding patterns in chaos, Zac is expected to enhance the team's performance. His weaknesses include kryptonite and spicy food, but these may not be significant enough to hinder his success. The new hire adds a fresh perspective to the Sentry team.