Definition of DevOps: a term born of usage
The word DevOps contracts development and operations. It refers to a way of working in which the teams that write software and those that keep it running share objectives, tools and responsibility for the service provided. DevOps spread through practice: its definition varies from one author to another.
Two public reference points exist. The term is associated with the devopsdays conference series, the first of which was held in Ghent, Belgium, in 2009, according to the organisation’s website. A standard also bears the name, IEEE 2675-2021 (“IEEE Standard for DevOps”): according to its public summary, it specifies principles and processes for building, packaging and deploying systems and applications in a reliable and secure way. The word was in circulation before this standard.
The silos the approach seeks to remove
In a siloed organisation, the development team is judged on the changes it delivers, the operations team on the stability of production. The production release becomes a handover: accumulated changes are passed, with an installation procedure, to people who did not write the code. The larger the batch, the riskier its installation, and the greater the temptation to space releases further apart.
DevOps turns the problem around. Changes are small, and therefore simpler to understand and to roll back. They go through an automated pipeline, and the team that wrote the code follows it all the way into production.
Culture, technical practices, measurement: what the word covers
What usage files under the word falls into three families.
- Culture: development and operations are jointly accountable for the service, with the same indicators and an incident analysis focused on causes rather than on people.
- Technical practices: version control of everything that describes the system, continuous integration, which builds and tests each change as soon as it is proposed, continuous delivery, which makes deployment a reliable operation that can be performed on demand, and infrastructure described as code.
- Measurement: the DORA research programme (dora.dev), unrelated to the European regulation with the same acronym, publishes throughput and instability metrics for software delivery.
The DORA capability catalogue, moreover, places technical practices (continuous integration, deployment automation) side by side with organisational traits (culture, working in small batches).
DevOps, tooling, job title, SRE, agile: not to be confused
DevOps is not a toolchain. Continuous integration and deployment are means to it: installed in an organisation that has remained siloed, they speed up the handover without removing it.
Nor is it a job title, or a team inserted between development and operations, which would risk forming a third silo.
Three neighbouring notions have their own entry. SRE is a codified way of managing reliability, through service level objectives. Platform engineering entrusts a self-service internal platform to a dedicated team. DevSecOps extends the approach to security. Agile methods, finally, concern the design and development of the product in short iterations; DevOps extends this logic all the way to operations.
An example: the same fix, with and without DevOps
Take the contract management application of a mutual insurer, delivered in batches several months apart. A minor fix waits for the next batch. Installation takes place at night, carried out by operations, following a document. At the first incident, each team looks for the cause on the other side.
In a DevOps organisation, the developer proposes the fix in the code repository. The pipeline builds it, runs the tests and deploys it to an environment recreated from its versioned description. It is then opened to a fraction of users, then to all. The team, which brings together development and operations skills, follows its dashboards and can revert to the previous version. The fix did not wait for any batch.
Building such a pipeline, and then an internal platform as teams multiply, is the subject of our page DevOps and platform engineering.