AZ-400 last-minute revision

One page to read before the exam, updated November 2026

Take a mock exam

Read this the night before or on exam morning. It assumes you’ve studied the notes. It’s here to refresh, not to teach.

Exam day

  • Microsoft doesn’t publish a fixed question count or duration for AZ-400, so plan on a mix of formats (multiple choice, drag and drop, yes/no case-study sets, and some YAML-reading questions) and check the time shown at the start. Some sections can’t be revisited once you move on, so answer each before leaving it.
  • Passing score is 700 out of 1,000. There is no penalty for guessing.
  • Domain weights (skills measured from July 27, 2026): processes and communications 10-15%, source control 10-15%, build and release pipelines 50-55%, security and compliance 10-15%, instrumentation 5-10%.
  • The certification renews annually through a free online assessment.

How to read an AZ-400 question

Read the last sentence first, then look for the constraint:

The question says Pick
Without storing secrets / no credential rotation Managed identity, or workload identity federation (OIDC)
Least administrative effort Built-in feature (policy, check, template) over custom scripts
Must reach on-premises or private network Self-hosted agent or runner (or VNet-injected pool)
Prevent deployment from non-main branches Branch control check, environment or service connection check
Enforce across many pipelines Extends template + Required template check
Decouple release from deployment Feature flags
Instant rollback Blue-green, slot swap
Limit blast radius Canary or rings
Audit / who approved Environment approvals and deployment history

Elimination tricks

  • Task groups and classic release pipelines are only right in classic scenarios; new designs use YAML templates and multi-stage pipelines.
  • A personal access token tied to a person is rarely the best answer for automation.
  • Secrets in YAML, repository files or plain pipeline variables on fork builds are wrong.
  • Answers that disable a control “temporarily” (bypass policies, grant admin) are usually wrong.

Core syntax to have fresh

Azure Pipelines

  • Hierarchy: pipeline > stages > jobs > steps. Stages sequential, jobs parallel by default; dependsOn changes it.
  • Job types: agent, container, server, deployment (targets an environment; strategies runOnce, rolling, canary).
  • Variable syntax: $(x) macro (runtime), ${{ }} template (compile time), $[ ] runtime expression.
  • Templates: template: for step/job/stage/variable, extends: to enforce structure. Task groups are classic only.
  • Trigger keys: trigger, pr (GitHub and Bitbucket only), schedules (UTC cron), resources.pipelines, resources.webhooks.
  • Output between jobs: isOutput=true then dependencies.<job>.outputs['step.var'].

GitHub Actions

  • on: (push, pull_request, schedule, workflow_dispatch, workflow_call, workflow_run, repository_dispatch).
  • needs: orders jobs. permissions: limits GITHUB_TOKEN. concurrency: prevents overlap. environment: brings protection rules.
  • OIDC login needs permissions: id-token: write.
  • Reusable workflow = whole job set (workflow_call). Composite action = steps inside a job.
  • Code scanning upload needs security-events: write.

Deployment strategies

Strategy One-line summary
Rolling Update in batches; mixed versions meanwhile
Blue-green Two environments, switch traffic, instant way back, double cost
Canary Small percentage first, widen on good metrics
Ring Named audiences in order, validate between
Feature flag Code deployed dark, enabled at runtime (App Configuration)
A/B Compare variants on a business metric
Slot swap App Service staging slot swapped into production; sticky settings stay

Identity and secrets

Need Choose
Azure resource calling Azure Managed identity (system-assigned for 1:1; user-assigned to share or pre-create)
Pipeline outside Azure to Azure Service connection or azure/login with federated credential
Workflow within its repo GITHUB_TOKEN with least permissions
Cross-repo automation GitHub App
Read-only exec or PO in Azure DevOps Stakeholder
External person on GitHub Outside collaborator
Certificate file in pipeline Secure file
Secret rotated centrally Key Vault (RBAC, soft delete, purge protection)

Scanning

Class Tool
Dependencies Dependabot, dependency review
Your code CodeQL
Secrets Secret scanning + push protection
Licences Dependency review deny-licenses
Images Defender for Containers, Trivy
IaC MSDO (Template Analyzer, Checkov), PSRule
Central view Defender for Cloud DevOps security

Metrics and monitoring

Term Meaning
Lead time Request to production
Cycle time Work started to done
MTTR Failure to recovery
Change failure rate Deployments causing incidents
Insights VM (guest, map), Container (AKS), Storage, Network
App Insights Requests, dependencies, exceptions; Application Map, Live Metrics, availability tests

KQL: where, project, extend, summarize ... by bin(timestamp, 5m), top 10 by x desc, render timechart, ago(1d), join, let. sort by is descending by default; == is case-sensitive; has is faster than contains.

Numbers worth knowing

Thing Number
Azure DevOps free users (Basic) First 5
Microsoft-hosted agent job limit (private, paid) 360 minutes
GitHub artifact retention 1-90 days, default 90
GitHub Actions cache 10 GB per repository, evicted after 7 days unused
Key Vault soft-delete retention 7-90 days
Metrics store retention 93 days
GitHub objects over 100 MB Rejected, use LFS