Abstract Mattias Jansson, Noa Resare: We have lots of material to draw from talking about how we work with integration between operations and software development.
This talk will be around a narrative about how we have grown as a company and how our challenges has changed as the number of users has grown and the technical complexity of the service has increased.
We have learned lots of stuff about how we communicate efficiently and handle short term issues that arise in the production environment as well as how we do medium to long term planning in cooperation between the operations team and development teams.
Moving forward with new development at a fast pace while at the same time maintaining a high quality of service is a balancing act that requires communication and thinking hard about how the organization is set up.
We think that it would be possible to squeeze in one or two real world case studies about more or less catastrophic failures that we've had to deal with and how those experiences has shaped our processes.
High level outline:
- description of the spotify service, our organization and a very brief technical description of the backend systems
- the early days: devs did everything
- the conflict of interest between reliability and new features
- how the ops team got overloaded with short term panicky stuff and how we're using kanban weekly retrospective and planning meetings to try to work around that
- system owner responsibilities at operations, subsystem planning meetings
- weekly scalability meetings to plan resource allocation and propagate information about incidents
- how to keep devs happy: the #operations irc channel, quick response times via email
- keeping track of incidents and that gets aggregated to service availability numbers
- case study: cascading failures, analyzing subsystem dependencies one of the most important things we do. Strategies to be able to do partial deployments, and rolling back changes.