In contrast with the first day which was focused on DevOps culture , the second and last day of DevOps Days Kiel conference covered the whole CAMS spectrum . Accounting for human issues when dealing with system failure (Culture), using Docker as an artifact in deployment pipelines (Automation), collecting operational data (Measurement) and bringing together academic research and industry practices for better performance engineering in DevOps (Sharing) were some of the topics covered.
Jorge Salamero Sanz , evangelist at Server Density, talked about the importance of treating human issues at the same level as system issues to reduce stress and improve system recoverability. Regularly runningwar games (failure injection and recovery exercises in production replicas) at Server Density has increased trust not only in the systems, but also among peers on each other’s ability to resolve issues without panicking, especially in complex failure modes. Considering incident types when designing on-call rotation is another way to ensure everyone feels comfortable, regardless of the kind of issue they might face. Finally, having a standard incident response process, checklists and quick access to known issues and fixes also helped reduce human errors, according to Salamero.
Andre van Hoorn , researcher at the University of Stuttgart, and Felix Willnecker , researcher at fortiss, initiated a discussion on how to evolve the performance engineering discipline to bridge the gap between DevOps (automation and speed) and traditional (mature but time consuming) tooling and methods . Performance concerns they want to address include system architecture (ability to scale according to demand), performance anti-patterns, and infrastructure and tools impact.
Philipp Krenn , developer advocate at Elastic, introduced the new Beats data shippers in the Elastic (ex-ELK) stack . Beats run as native binaries (built with Go), thus significantly speeding up data collection (when compared with Logstash). However, they do not provide the data enrichment capabilities of Logstash, thus Krenn recommends keeping a couple Logstash instances but switching to Beats for all other plain data collection needs. Beats come in different forms (and can be custom made), such as Filebeat (typical log files forwarding), Topbeat (collect CPU, memory and disk usage data, in similar fashion to Unix’s "top" command) or Packetbeat (collecting data on network traffic).
Baruch Sadogursky , developer advocate at JFrog, walked the audience through some challenges of using Docker images as artifacts in a build pipeline, such as managing dependencies, building immutable binaries or maintaining traceability with other artifacts and the source code.
Bianca Heinemann , consultant at IBM, alerted to potential hidden costs that might overshadow short term benefits in DevOps. For instance, adopting open source tools but not investigating their inter-operability, or going all in for cloud without planning for the IaaS support effort required.