A Terraform plan is more than a list of proposed changes: it is a decision record for what infrastructure may be created, modified, replaced, or destroyed. This repeatable Terraform plan review workflow helps you identify risky changes, investigate drift, resolve state problems, and capture enough evidence for a safe approval before running apply.
Overview
Review the plan from three perspectives: intent, impact, and state confidence. First, confirm that the plan matches the change request. Next, understand the effect on availability, data, access, cost, and dependencies. Finally, establish that Terraform is comparing the intended configuration with the correct workspace, variables, provider versions, and state.
Run the plan from a clean, reproducible environment whenever possible. Confirm the working directory, selected workspace or equivalent environment configuration, backend, variable files, and credentials before interpreting the output. A plan generated against the wrong account or state can look technically valid while describing the wrong environment.
For a reviewable artifact, save the plan rather than relying only on terminal output:
terraform plan -out=tfplan
terraform show tfplan
Keep the saved plan with the change evidence used by your team, subject to your organization’s handling rules. A plan can contain sensitive values or details about infrastructure. If automated checks consume plan data, generate machine-readable output deliberately, for example with terraform show -json tfplan, and protect the resulting file.
Use the same disciplined approach for related delivery work. For example, a plan may be produced by a CI workflow, so a GitHub Actions debugging workflow can help separate an infrastructure problem from a failed runner, secret, matrix, or deployment step.
Checklist by scenario
Routine update or additive change
- Match each changed resource to a ticket, commit, module change, or documented operational reason.
- Check whether new resources use the intended account, region, network, tags, naming pattern, and environment.
- Review dependencies and outputs. A new resource can affect routing, permissions, monitoring, DNS, or downstream modules even when those resources are not directly changed.
- Look for unexpected provider or module changes mixed into the plan. Separate dependency updates from the requested infrastructure change when possible.
- Confirm that unknown values will be resolved safely during apply and are not being used to hide an important decision.
Plan contains a destroy or replacement
Treat every destroy and replacement as a stop-and-explain event. A replacement usually means Terraform cannot update the existing object in place, while a destroy may be intentional or may indicate a changed address, identifier, provider behavior, or state problem.
- Identify the exact resource address and the reason for replacement or deletion.
- Determine whether the object contains data, serves live traffic, grants access, or anchors other resources.
- Check whether a changed immutable argument, renamed resource, altered key, or changed module path is responsible.
- Review ordering and downtime implications. “Create before destroy” behavior, where supported and configured, does not remove every dependency or data-loss risk.
- Verify backups, migration steps, rollback options, and an explicit owner for the decision.
Do not approve a Terraform plan destroy warning simply because the total number of changes is small. One deletion can matter more than dozens of harmless additions.
Unexpected changes suggest drift
Drift occurs when real infrastructure no longer matches the configuration and state assumptions used by Terraform. It can result from a console edit, another automation system, a provider-normalized value, or an incomplete state operation.
- Confirm that the plan is using the expected state and credentials.
- Compare the changed attributes with recent operational activity and cloud-provider audit information.
- Decide whether the real-world change should be preserved or reverted.
- If it should be preserved, update configuration or use an appropriate import or adoption workflow rather than forcing an unrelated change.
- If it should be reverted, verify that the configuration expresses the desired final state and that the reversal is approved.
A refresh-only plan can help examine differences without proposing the ordinary configuration changes, but use it as an investigation aid, not as a substitute for understanding the resource and provider behavior.
State errors, locks, or missing objects
State problems require caution because state records Terraform’s relationship with real infrastructure. Start by preserving the error output and identifying the backend, workspace, state version, and operation that produced it.
- For a lock error, check whether another legitimate plan or apply is running.
- Do not force-unlock until the lock owner and operation have been investigated and the team agrees it is stale.
- For a missing resource, determine whether it was deleted outside Terraform, moved, renamed, or represented by a changed address.
- For an import need, map the real object to the correct resource address and add matching configuration before accepting the result.
- Do not edit state files manually as a first response. Prefer documented state commands, backups, and peer review.
When a resource address changes because of a refactor, investigate state movement explicitly. A plan showing destroy-and-create may be a state-address issue rather than an intended replacement.
Provider, module, or variable changes are present
Review the inputs that shape the plan, not just the resources at the bottom of the output. Check provider constraints and lock-file changes, module sources and versions, variable files, environment variables, sensitive values, and workspace-specific defaults. A provider upgrade may alter validation or normalization, while a module change may affect many consumers through outputs or shared defaults.
Where policy-as-code is available, run checks for prohibited regions, public exposure, missing ownership metadata, excessive permissions, or unsafe deletion. Policy checks should complement human review; a passing policy result does not prove that the plan matches the requested operational change.
What to double-check
- Scope: Is the plan for the correct environment, account, region, workspace, and backend?
- Change count: Are additions, updates, replacements, and destructions plausible for this request?
- High-risk resources: Are databases, storage, identity, network boundaries, DNS, encryption, and production workloads involved?
- Replacement triggers: Which argument or address causes recreation, and is that behavior documented?
- Unknown values: Which values are deferred until apply, and could they change routing, permissions, names, or dependencies?
- Outputs: Will module outputs, connection details, or downstream consumers change?
- Concurrency: Is another operation using the same state, and is locking functioning as expected?
- Evidence: Does the review record the command, commit, plan artifact, reviewer, assumptions, and approval decision?
Keep the review focused on infrastructure behavior. Application-level symptoms can require separate investigation; the local-to-cloud debugging workflow provides a useful way to trace issues across code, configuration, deployment, and runtime layers.
Common mistakes
- Approving by resource count: A small plan can still delete a critical object.
- Applying immediately after planning: The environment, state, or external resources may change between commands. Use a saved plan where the workflow requires the reviewed artifact to be the applied artifact.
- Ignoring the first unexpected change: Drift and state-address errors often produce a cascade of secondary changes.
- Using broad variable files: Unintended values can target the wrong environment or alter replacement-sensitive arguments.
- Force-unlocking reflexively: A lock may represent an active operation. Investigate before removing it.
- Treating import as completion: Imported state still needs configuration that accurately describes the real object and future ownership.
- Reviewing only the diff: Provider, module, backend, dependency, and output changes may explain the most important risk.
When to revisit
Use this checklist before every non-trivial plan, especially for production changes, provider or module upgrades, state migrations, imports, and plans containing replacement or destruction. Revisit the workflow after Terraform, OpenTofu, provider, backend, CI, or policy tooling changes because command behavior and plan interpretation can evolve.
It is also worth reviewing after an unexpected plan, a drift incident, a failed apply, a recovered state lock, or a change in cloud-account structure. Before seasonal planning cycles or major platform work, test the approval path in a non-production environment and confirm that plan artifacts, locking, policy checks, and rollback evidence are still handled as expected.
For each review, finish with a clear decision:
- Approve: The plan matches intent, risks are understood, and evidence is complete.
- Revise: Configuration, variables, provider versions, or state mapping need correction before a new plan.
- Investigate: Drift, locking, unknown values, or destructive actions cannot yet be explained.
- Reject: The plan targets the wrong scope or creates unacceptable operational risk.
A consistent Terraform plan review turns infrastructure changes into inspectable decisions rather than last-minute terminal output. Save the plan, explain every destructive or replacement action, resolve state uncertainty first, and record why the final decision was made.