A specific team must own every service that handles customer traffic, and that team must have the authority to release changes to it. Without that choice the rest of the work becomes difficult. The pipeline becomes a faster way to produce artefacts that operations will not accept, incidents get diagnosed by staff who are not allowed to deploy the fix, and the first rollback request becomes a dispute about credentials.
What follows is an order of work, one step resting on the one before it. It suits a small team with a handful of services and no dedicated platform group. The order is the practice; tools are what you buy once the order is clear. Each step ends with a check, because a practice a team believes it has adopted and cannot demonstrate is a practice it has not adopted.
1. Name the owner of each service
Write down the team responsible for the pager when every service a customer touches breaks. Then add that same team to the list of people who can deploy it. The two facts belong together; separating them rebuilds the handoff that DevOps exists to remove. Teams commonly buy a pipeline before they can name who is on call for the service it deploys.
Atlassian's account of DevOps best practices treats culture as a load-bearing item. It notes that siloed ownership with minimal cross-team communication is the most common organisational structure, and that those barriers have to come down before any progress can be made. Starting with the permission list is the cheapest way to find out whether they have.
You have finished this step when the on-call rota, the deploy permissions and the service catalogue all name the same team, and a member of that team can push a change to production without asking another group for a credential.
2. Store the delivery pipeline in one repository.
Bring six kinds of file into one place, under version control, reviewed like code:
- the application source
- the pipeline definition
- the infrastructure definition
- the policy files
- the runbooks
- the dashboard definitions
Six kinds of file, one history. Infrastructure as code is the provisioning and managing of infrastructure with the same techniques used for application code, including version control and continuous integration, and the overview of DevOps from AWS extends the same treatment to configuration management and policy as code. A definition kept in a repository can be reviewed, diffed and rebuilt. A setting changed by hand in a console cannot.
The guidance in André de Sousa's DevOps best-practices repository is direct about the repository itself: keep a central repository, minimise the number of branches, commit to the baseline every day, and keep the pipeline definition in code. Those habits are what make the next step possible.
You have finished this step when a new starter can clone one repository and run the same pipeline command locally against a scratch environment.
3. Make the pipeline the only way to production.
Every change that reaches production goes through the pipeline. No console deploys, no exceptions during incidents, no hotfix applied by hand just this once. The same repository warns that teams often start with good discipline and then make exceptions under pressure, and that each exception teaches everyone that the pipeline is optional.
The approval meeting is a separate question from the pipeline. CloudZero's review of DevOps practices reports that DORA found organisations with formal external approval processes are 2.6 times more likely to be low performers. A pipeline provides evidence strong enough that no one asks for a sign-off meeting.
The exception is the thing to watch. A team that allows one manual hotfix a month will, within a quarter, have a manual hotfix process.
You have finished this step when the production deploy log for the last month shows a pipeline run identifier beside every deploy. Any entry without one is the first thing to fix.
4. Build a test suite that fails for the right reasons
Order the tests so the fastest run first, run them in parallel where the infrastructure allows, make the whole suite runnable locally with one command, and treat a flaky test as a defect in the suite. The repository's testing notes make the same points: run the fastest tests early, avoid false positives and false negatives, and read the reports attached to each build.
People avoid a suite that takes an hour to fail. A suite that fails for reasons unrelated to the change teaches users to ignore red builds, and an ignored red build protects no code.
You have finished this step when a deliberately broken commit fails the pipeline in minutes, and every quarantined test has a named owner and a date for its return.
5. Rebuild environments from code and scan the definitions
Define infrastructure in version-controlled files and stop changing it in the console. Then scan those files the way you scan application code. Wiz's guide to DevOps security recommends tools such as Checkov and Terrascan for exactly this. Salesforce's guide defines environment drift as the slow divergence between development, staging and production, and treats infrastructure as code as the way to close it.
The drift is the expensive part. An environment that was right in March and has been adjusted by hand ever since will pass a smoke test and fail on the first change that depends on the original setting. Defined infrastructure can be destroyed and rebuilt; remembered infrastructure is one resignation away from an outage.
You have finished this step when staging can be destroyed and rebuilt from the repository alone, and the rebuilt environment passes the same smoke test as the one it replaced.
6. Integrate security checks into the commit workflow.
Perform these checks before the merge, so a change has their results before any user approves it:
- secrets scanning
- dependency scanning for known vulnerable libraries
- static analysis of the code
- container image scanning
- licence checks against the policy the organisation has set
- policy checks on the infrastructure definitions
Salesforce's guide to DevOps practices recommends running SAST, dependency scans and secrets checks on every commit, applying least privilege, and masking sensitive data in sandbox environments. Wiz adds the habits that stop the problem at the keyboard, including pre-commit hooks and a .gitignore that keeps credentials out of the repository in the first place.
.env
*.pem
secrets.yaml
The aim is for a developer to find out about a problem while the change is still in their head.
You have finished this step when a commit containing a fake API key is rejected before merge, and a dependency with a known vulnerability blocks the build with a link to the advisory.
7. Instrument the service and the pipeline
Collect logs, metrics and traces from the application, and put the same kind of visibility on the pipeline itself. Atlassian's DevOps best practices recognises logs, traces and metrics as the three pillars of observability, and suggests monitoring the pipeline as well as the application: A failed automated step that goes unnoticed leaves a team worse off than doing the work by hand.
Alert on what users feel. A slow checkout, a login that loops, a payment that fails twice are the signals worth a page. A dashboard full of CPU graphs that no one looks at is decoration.
You have finished this step when a synthetic failure in staging raises an alert that specifies the service, the version and the failing check, and the pipeline dashboard shows the failed run without any user opening a terminal.
8. Rehearse the rollback before you need it
Rolling back has to be as routine as deploying. Keep the last few artefacts from the pipeline, give the deploy a way to return to any of them, and record which version is live. Pink Elephant's essential DevOps practices define blue-green deployment as two identical production environments, where traffic points to one while the other is rebuilt, and the switch occurs once the new one works. A feature flag does the same job for a smaller change: The code is deployed and the feature stays off until a user activates it.
A rollback that needs an operator to remember which server to log into will be improvised during an incident, and improvisation during an incident is how the manual exception in step 3 gets made.
You have finished this step when a deploy to staging is reversed by a pipeline run, and the service returns to the previous version with no hand edits.
9. Read four delivery numbers every week
Salesforce's guide sets out the four DORA metrics a team can collect with no new tooling: deployment frequency, lead time for changes, change failure rate and time to restore. Each one has a source already in place once steps 3 and 7 are done. Deployment frequency and lead time come from the pipeline. Change failure rate comes from counting the deploys that needed a rollback or a hotfix. Time to restore comes from the incident record.
The numbers belong to the team that owns the service. Read them on the same day each week and act on them before the next release. A number that gets collected and never discussed is a report, and reports do not change behaviour.
You have finished this step when the four numbers sit on one page the team looks at together, with a note of what changed since the last reading.
The sequence remains if a step is skipped
Step 1 is the one that cannot be deferred. A team that starts with tooling and leaves ownership until later builds a pipeline for artefacts no one has agreed to run. Steps 2 to 6 can run in almost any order once step 1 is settled, though each is easier after the one before it. Steps 7 to 9 are what turn a working pipeline into one a team can trust at three in the morning.
No part of this requires a platform group, a tool budget or a reorganisation. It needs a list of services with IDs beside them, a repository that contains the definitions, and a team willing to make the pipeline the only way through.
Sources
- DevOps Best Practices | Atlassian
- DevOps Best Practices: The Complete Guide to Faster, Safer Delivery | Salesforce UK
- What is DevOps? - DevOps Models Explained | Amazon Web Services
- 15 Essential Practices Of DevOps | Pink Elephant EMEA
- andredesousa/devops-best-practices | GitHub
- DevOps Best Practices: 12 Principles That Ship Faster And Cost Less | CloudZero
- What Is DevOps Security? Implement, Challenges, Best Practices | Wiz


