TL;DR: The DevOps lifecycle is five phases on one loop: plan, develop, deploy, operate, and feed back, with what operations learns feeding the next round of planning.
- The count is editorial: IBM draws eight phases on an infinity loop, Microsoft compresses them into four, AWS skips numbering and lists six practices.
- Version control has a winner: nearly 95 percent of developers named Git their primary system in 2022. Subversion remains the centralized legacy, still maintained.
- Tool lists carry dates. Travis CI ended free open-source builds in December 2020; GitHub Actions has shipped CI inside the largest code host since November 13, 2019.
- Continuous delivery leaves the release decision with a human gate; continuous deployment ships every passing change automatically, into production.
- DORA closes the loop with four metrics, two for velocity and two for stability, plus reliability joining in 2021.
Most tutorials on the DevOps lifecycle reach for the same drawing: five boxes circling a loop. Open the vendor pages and the count stops agreeing: eight phases at IBM, four at Microsoft, and no number at all at AWS, which lists practices instead. This article is the written half of the Whiteboard Architect lesson The DevOps Lifecycle: Five Phases, One Honest Loop. Watch the episode for the boards drawn in real time, or read on: every claim below sits beside the page it was checked against.
Board 2 of 14: the loop before any phase is named, with the vendors’ counts waiting in the red box.
The DevOps lifecycle, drawn by three vendors
The disagreement is the easiest part to verify. IBM’s Think page names eight key phases: plan, code, build, test, release, deploy, operate and monitor, drawn as an infinity loop, development activities left, operations right. Microsoft Learn makes the same journey four phases: plan, develop, deliver, operate. AWS drops the count and describes six practices instead, from continuous integration to communication and collaboration.
Five phases is the map this series drew, and the honest part is saying so. The part all three maps share is the shape. IBM’s page has every phase informing and overlapping the others, AWS describes engineers working across the whole application lifecycle instead of one silo, and both return what operations learns to the next plan. Cut that return arrow and you are left with a conveyor: work moves one way, nothing comes back; lesson one of this series covers the sequential model that produced those handoff walls.
How do the three vendor maps differ, side by side?| Source | Shape | What it names |
|---|---|---|
| IBM Think | Infinity loop, development left, operations right | Eight phases: plan, code, build, test, release, deploy, operate, monitor |
| Microsoft Learn | Four-phase lifecycle | Plan, develop, deliver, operate |
| AWS | No numbered phases, six practices | Continuous integration, continuous delivery, microservices, infrastructure as code (with configuration and policy as code), monitoring and logging, communication and collaboration |
IBM’s page, updated in June 2026, adds two framings of its own. DevSecOps appears as what IBM calls the ninth phase, security shifted left into development and right into runtime. A second section walks the 7 C’s, seven continuous processes running from continuous development to continuous operations.
Plan: security starts at the backlog
Planning decides what gets built; the definition worth keeping puts governance, security, and compliance inside this phase instead of bolting them on later. A requirement inherits its security obligations here, far cheaper than inheriting them in an incident review.
One architecture decision belongs at this table: whether the application ships as a monolith or as microservices. AWS defines the microservices approach as one application built from a set of small services, each one running in its own process, each dedicated to one purpose, speaking to each other over lightweight interfaces, usually an HTTP API. That split decides team boundaries and deployment units, and how much of the loop each service rides on its own schedule.
What does Microsoft put inside the plan phase?Microsoft’s DevOps page is the source of the four-phase map, and its plan phase is the busiest of the four. Teams define and describe the features and capabilities to build, track that work from backlogs to portfolios on Scrum and Kanban boards, and fold governance, security, and compliance into the same conversation rather than a later review board. For interviews, the security clause is the differentiator: security belongs at planning time, when a requirement can still change cheaply, not at release time, when the fix is a new cycle of its own.
Develop: one codebase, reviewed, merged daily
Development happens in a single shared codebase held in version control, and that question is settled. Git, created by Linus Torvalds in 2005 and distributed by design, is the system nearly 95 percent of developers named as their primary one in 2022, per its Wikipedia survey figures. Apache Subversion, centralized and older than Git, is still maintained by the Apache Software Foundation, with 1.14 still in LTS support and 1.15 the latest release since October 2026.
The branching model is deliberately flat: feature branches carry new work and fixes, merges happen through reviewed pull requests, and main stays current and releasable. Microsoft’s branching guidance, the same document that calls the pull-request review critical for quality, puts the optimum at two reviewers per change.
Continuous integration adds the next layer. Developers merge their work into the shared repository no less than once a day, and each merge triggers an automated build and test run, catching defects while they are small and cheap.
The tools for this practice turned over once already. Travis CI, the pioneer that built open-source projects at no charge, stopped in December 2020, while CI moved into the code host itself. GitHub, a Microsoft subsidiary since 2018 and the largest code host in the world, launched Actions on November 13, 2019. Jenkins and CircleCI remain current alongside both, and the tests underneath run in levels, from unit checks to a browser; the deep dive below breaks them apart.
Which test level catches which defect?| Level | What it exercises | The usual tool |
|---|---|---|
| Unit | One function or class in isolation | The xUnit family, JUnit being the classic member |
| Integration | Unit-tested modules combined, their interactions exercised | Built up from unit-tested pieces |
| Browser | The web-facing surface in a real browser | Selenium driving the page |
Unit tests run first in a CI pipeline because they fail in seconds and cost seconds. Integration tests catch contract mismatches between modules that each pass alone. The browser level is where a broken selector or a layout regression finally shows up, which is why a pipeline that skips it looks green while the product is broken on screen.
What happened to Travis CI’s free open-source builds?Travis CI built open-source projects at no charge before any competitor did, which is why a decade of tutorials name it. In November 2020 it announced the shutdown of travis-ci.org by December 31, 2020, with accounts migrating to travis-ci.com. The company pledged to keep open-source accounts completely free on the new domain; projects then reported their build jobs stalling.
Open-source access ended up as a non-renewable sign-on bonus of 10K credits over a 30-day period, meant for evaluating the paid tiers. The service had been acquired by Idera in January 2019 and remains commercial. If a course or interview guide still lists Travis as the CI tool, that material predates this collapse.
Deploy: always hold a releasable build
The deploy phase has one target property: a build you can put in front of users at any moment. Continuous delivery delivers it, in AWS’s description: each change is built, tested, and prepared for release automatically, in short cycles, so you always hold a version you could ship.
Two sibling terms describe different stops on one track, and interviews test the difference. With continuous delivery, releasing stays a decision: the build is ready, and a human gate decides when it goes out. With continuous deployment the gate is gone: every passing change lands in production by itself, per the continuous delivery entry that separates the two.
Board 6 of 14: the pipeline from build to deploy, with the MAY gate drawn between a green build and production.
The machinery behind both is a pipeline that builds, tests, and deploys across environments, with automated gates carrying the artifact from one stage to the next.
The pipeline lands on infrastructure that deserves the same discipline. Infrastructure as code provisions and manages servers, networks, and databases from code, versioned like code, which makes the work repeatable and reviewable and lets development environments match production. Terraform and Ansible are the two names to know here. Their partner, configuration management, keeps a recorded state for every server and database and pushes changes out in an orderly way, which stops configuration drift from piling up.
What is configuration drift, and what stops it?Drift is the gap between the state you defined and the state a server is actually in. A hotfix applied by hand to one box, a firewall rule opened during an incident, a package updated outside the pipeline: each leaves a small divergence that survives every future deploy and reappears in every debug session. Configuration management closes the gap by recording the intended state of each system and rolling it out systematically. Infrastructure as code makes that intended state reviewable, because it is code in version control with a history. The two together deliver the property the deploy phase promised, an environment built from a definition rather than from memory.
Operate: telemetry before the ticket
Operate never ends: keep the production system running, watch it, fix it, with reliability, high availability, security, and zero downtime as standing requirements. The method is telemetry and alerting in Microsoft’s phrasing, and monitoring and logging in AWS’s. Metrics and logs show how performance reaches real users, root-cause analysis explains failures, and stack-wide visibility in real time spots trouble before the first support ticket arrives.
Incident management owns the breaks: the ITSM definition frames it as a fast return to normal service with business impact kept small, run through a lifecycle of detection, classification, response, containment, recovery, and a closing review that turns the incident into prevention work. Cutting mean time to recovery appears on the vendors’ benefit lists in its own right. Capacity planning owns the growth: estimating storage, hardware, software, and connection infrastructure before demand arrives.
How does an incident move through its lifecycle?| Stage | What happens |
|---|---|
| Detection | Monitoring, or a user report, flags the failure |
| Classification | Severity and impact decided, the right responders paged |
| Response | The team engages and works the incident |
| Containment | Damage limited, for instance traffic routed away from the bad version |
| Recovery | Service returns to normal |
| Post-incident review | Blameless write-up, and prevention work enters the backlog |
The review is the stage that feeds the loop: an incident that ends without a prevention item will repeat. The mean-time-to-recovery clock spans detection through recovery, which is why better monitoring shortens it before anyone fixes anything else.
Feed back: a retrospective and four metrics
The feedback phase bends the sequence back on itself. Each iteration closes with a retrospective asking three questions, per the iteration retrospective definition: what worked, what should improve, and how both travel into the next round.
Judging the loop needs numbers, and DORA’s four key metrics supply them: deployment frequency and lead time for changes on the velocity side, change failure rate and time to restore service on the stability side. A fifth metric, reliability, joined the set in 2021, per Google’s four keys write-up. Releasing in short cycles makes planning and risk easier, each change carries less stability impact, and the organization adapts to customers and competitors faster.
Board 10 of 14: the four metrics bracketed into velocity and stability, with reliability arriving separately in 2021.
How do DORA’s four metrics split, and what does each one ask?| Metric | The question it answers | Half |
|---|---|---|
| Deployment frequency | How often does the team ship? | Velocity |
| Lead time for changes | How long from commit to running in production? | Velocity |
| Change failure rate | How often do changes cause failures? | Stability |
| Time to restore service | How fast does the team recover? | Stability |
The split exists to catch the team that ships constantly and still breaks production weekly; averaging the four numbers would hide exactly that. The fifth metric, reliability, joined the set in 2021 and widened the stability half beyond outage recovery, covering whether the service meets its own reliability targets.
The five phases in one table
| Phase | What it decides | What to check |
|---|---|---|
| Plan | What gets built, and with which security and compliance requirements | Backlog to portfolio tracking, Scrum or Kanban, the microservices boundary |
| Develop | How code stays mergeable and reviewed | Git, pull requests with two reviewers, CI on every merge, unit and integration tests |
| Deploy | Whether you can release right now | Delivery gate versus automatic deployment, pipeline stages, Terraform or Ansible, drift |
| Operate | How fast you notice and recover | Metrics and logs, the incident lifecycle, capacity sized ahead of demand |
| Feed back | What changes in the next iteration | The retrospective, DORA’s four metrics, short cycles |
Why the vendors draw a loop
The companies selling this loop also document its payoff. AWS’s benefit list runs speed, rapid delivery, reliability, scale, security, and collaboration. Microsoft’s page adds faster time to market, adaptability, stability that survives change, and better recovery. Both credit the same engine: automation, collaboration, and continuous deployment.
The drawing stays circular only while every feedback path works: operations informing the next plan, monitoring feeding the backlog, retrospectives changing behavior. How the loop earned its shape is lesson two of this series.
Four traps in lifecycle tutorials
| Trap | The correction |
|---|---|
| Presenting five phases as the industry model | The count is editorial: eight at IBM, four at Microsoft, practices at AWS. Say whose map you are using. |
| Listing Git and SVN as an even choice | Git was the primary choice of almost 95 percent of developers in 2022; Subversion is the centralized legacy, maintained but minority. |
| Listing Jenkins, CircleCI, and Travis CI as peers | Travis ended free open-source builds in December 2020, and CI has shipped inside GitHub since November 2019. That list is a 2018 snapshot. |
| Using delivery and deployment as synonyms | Delivery keeps a decision gate, so you may ship. Deployment removes it, so you do ship. |
Board 14 of 14: the recap checklist, with the sticky note carrying the three fuels: automation, collaboration, data.
A tool list is a snapshot of the year it was written; DevOps tooling turned over twice in the last decade. The next lesson follows one feature through all five phases, the loop taken end to end; the interview guide below will test whether it stuck.
The five-phase loop optimizes for a short distance between a change and the lesson it teaches, and pays for it with discipline at every stop: security planned in, every merge reviewed, every release kept one gate away, production watched, each iteration measured. Keep the vendor counts on the table and the return arrow intact, and the map holds up anywhere the series takes it.
Continue learning
- Why DevOps? Where the Traditional SDLC Breaks: lesson one of this series, on the handoff walls that made this loop necessary.
- A Short History of DevOps: The Checked Timeline: lesson two, on where the movement’s ideas and its name came from.
- DevOps Interview Questions and Answers: Topic Guide: test yourself on the vocabulary this loop produced.
Disclosure: this article was drafted with AI from the script of my video lesson, then checked against its sources before publishing.
This lesson first appeared on iqraa.tech.
More whiteboard lessons on the Whiteboard Architect YouTube channel.







