Linaro Connect Bangkok

Over the past years, we ran the kernel selftests as part of the LTS release testing. During that time, we learned things, fixed things and created a wish list of work we want to tackle. This session is about what we learned and where we are heading.

KernelCI generates daily an average of two thousand Linux Kernel builds, submitted to several labs across planet to mainly check if system booting works correctly. Therefore tons of data are created and displayed on a frontend interface for further use. Current KernelCI requires significant technical effort changing visualizations, including builds and jobs listings. The KernelCI New Generation seeks to apply state-of-the-art data collection and data visualization tools as an alternative frontend to KernelCI. Our focus is on an instance of ElasticSearch, Logstash, and Kibana collecting and displaying live KernelCI data. All the tooling used to this project will be available.

Test results triage is the most time-consuming step currently in Texas Instruments CI cycle. Embedded tests are more susceptible to failures caused by external factors due to test setup complexity. A typical embedded test contains multiple points of failure.
This work uses a neural network to classify test results automatically and to detect 'real' regressions. It helps scaling TI's validation cycle by automating the most time-consuming step.