Session

The never-ending innovation that powers the Linux networking community makes today’s datapaths increasingly diverse and complex. This makes observing packets as they traverse the many layers, protocols, and drivers more challenging. Consequently, feature development, construction of complex data paths, and resolution of technical issues can become significantly more difficult and time-consuming.

In order to address this challenge and speed up network problem diagnosis, Retis [1] combines the flexibility of eBPF, the popularity of libpcap filtering with extensible built-in and custom post-processing facilities to track packets, gather contextual information from the areas they traverse (e.g: nft, conntrack, openvswitch, …) and allow deep analysis of collected data.

In this tutorial, we discuss the main challenges that were found and how they led to the main architectural decisions of Retis. Then we cover the most important features supported by Retis and we examine the key implementation strategies. Also, we show a demonstration of how Retis can be used to gain insights into internal packet processing which can be very useful for troubleshooting and development. Finally, we discuss the main limitations that have been found and explore future improvements.

[1] https://github.com/retis-org/retis/