A Terraform plan review is the last practical checkpoint before infrastructure changes become real. This checklist helps reviewers identify destructive actions, unexpected replacements, dependency risks, security-sensitive changes, and missing operational safeguards before approving an apply.
Overview
A plan should be reviewed as a proposed change to a live system, not merely as a technical diff. Terraform and OpenTofu translate configuration, state, provider behavior, and input values into a set of actions. The result can include resources to create, update, destroy, or replace. A clean-looking code review does not necessarily mean the resulting infrastructure change is safe.
Reviewers should establish three things before approval:
- What will change? Identify every resource and attribute affected by the plan.
- Why will it change? Connect each action to an intentional configuration, variable, provider, or dependency change.
- What will happen afterward? Confirm service availability, security posture, cost expectations, and rollback or recovery options.
For team workflows, generate the plan from the same commit that will be applied, use a controlled state backend with locking, and preserve the reviewed plan output where your process requires an auditable approval record. The plan is time-sensitive: state, external resources, provider behavior, and input values can change between preview and apply.
Teams that need a broader foundation should also review Terraform state management, backends, locking, and team workflows before formalizing an approval process.
Checklist by scenario
Use the following checklist for every production plan, then apply the scenario-specific checks that match the change.
Copyable Terraform plan review checklist
- [ ] Confirm the repository, workspace or stack, account, subscription, project, and region are correct.
- [ ] Confirm the plan was generated from the intended commit and current variable set.
- [ ] Check that the state was read successfully and that no unexpected refresh or state errors occurred.
- [ ] Count resources to create, update, destroy, and replace.
- [ ] Explain every change in plain language and link it to an approved task or design decision.
- [ ] Investigate every destroy and replace action before approval.
- [ ] Review changes to identity, network exposure, encryption, logging, backups, and secrets.
- [ ] Check dependencies, ordering, references, and outputs consumed by other systems.
- [ ] Confirm lifecycle settings do not hide, delay, or amplify the intended behavior.
- [ ] Review policy, security, cost, and compliance checks and resolve exceptions.
- [ ] Confirm maintenance windows, service impact, monitoring, and rollback or recovery steps.
- [ ] Obtain the required peer or service-owner approval before apply.
- [ ] Apply through the controlled workflow rather than an unrelated local environment.
- [ ] Verify the resulting resources, application health, alerts, access, and state after apply.
Classify the change before discussing details
| Risk level | Typical plan actions | Review approach |
|---|---|---|
| Lower | Tags, descriptions, noncritical metadata, or an isolated development resource | Confirm scope, ownership, and that no computed dependency causes additional actions. |
| Moderate | Scaling values, instance types, policy adjustments, monitoring changes, or application-facing configuration | Check capacity, service behavior, cost impact, maintenance needs, and dependent systems. |
| High | Resource replacement, deletion, public exposure, identity changes, encryption changes, database work, or network routing changes | Require explicit impact analysis, owner approval, recovery planning, and post-apply verification. |
When the plan includes destroy or replace
Do not treat a replacement as a routine update. A replacement usually means the provider cannot change an attribute in place, so the resource may be destroyed and recreated, or created before the old object is removed depending on configuration and provider behavior. Review the exact action sequence and the provider documentation for the resource type.
Ask whether the object contains persistent data, serves traffic, holds an identity used elsewhere, or has an externally managed attachment. Check names, addresses, disks, snapshots, backup policies, and recovery points. A change to an immutable identifier, region, network placement, or other provider-controlled attribute can have replacement implications even when the configuration diff appears small.
When the plan changes security or access
Prioritize changes to IAM roles, policies, firewall rules, security groups, routes, load balancer listeners, storage permissions, encryption keys, and secret references. Compare the proposed access with the intended trust boundary: who can access the resource, from where, using which identity, and for how long?
Separate a reference to an existing secret from a value that Terraform may place in state or plan output. Review sensitive outputs and state handling, and never approve a plan that exposes credentials in a pull request or log. For a structured access review, see the Cloud IAM access review workflow.
When the plan affects Kubernetes or a platform layer
Look beyond the Terraform resource itself. A cluster, node pool, identity binding, storage class, ingress component, or autoscaling setting can affect many workloads. Check Kubernetes version compatibility, capacity, disruption behavior, resource requests, limits, and workload health checks. The Kubernetes cluster upgrade checklist and guidance on resource requests and limits can support this review.
What to double-check
Dependencies and replacement behavior
Trace references between resources instead of reviewing each block in isolation. A change to a subnet, security group, key, role, or module input may produce downstream actions that are easy to miss in a long plan. Pay attention to dependency changes caused by renamed resources, changed keys in a collection, altered module addresses, or modified data-source results.
Review lifecycle settings such as replacement ordering, ignored attributes, and protection against deletion. These settings can be useful safeguards, but they can also make the plan differ from a reviewer’s assumption. An ignored attribute may conceal drift; a deletion guard may cause an apply to fail rather than protect the wider change. Treat lifecycle configuration as part of the risk decision.
Inputs, providers, and state
Verify that environment variables, variable files, provider aliases, credentials, and region or account selectors point to the intended target. A correct plan in the wrong account is still a dangerous plan. Check provider version changes separately from resource changes because provider behavior and replacement rules can change the resulting actions.
Investigate unexpected drift rather than dismissing it as noise. Determine whether the resource was changed manually, whether a data source now returns different information, or whether state no longer represents the intended ownership boundary. If state is stale or locked incorrectly, stop and resolve that issue before approval.
Operational readiness
For significant changes, write down the expected user impact, duration, maintenance window, monitoring signals, and abort criteria. Confirm that backups or snapshots are recent enough for the data at risk and that recovery steps are realistic, not merely theoretical. After apply, verify both infrastructure and service behavior: health checks, logs, metrics, alerts, connectivity, permissions, and key user journeys.
If the change is part of a delivery pipeline, use the same controls for preview, policy checks, approval, apply, and verification. The CI/CD pipeline troubleshooting guide is useful when a controlled infrastructure deployment fails or requires a rollback.
Common mistakes
- Approving by resource count: “Two resources changed” says nothing about whether one is a production database or a harmless label.
- Ignoring replacements: A replacement can interrupt service, change an address, detach data, or invalidate a dependent reference.
- Reviewing code without reviewing the plan: Expressions, modules, provider defaults, and data sources can produce effects that are not obvious from a small diff.
- Applying an old plan: A plan generated before a merge, state change, credential change, or provider update may no longer describe the desired action.
- Trusting lifecycle protection as a complete safety plan: Preventing deletion may stop an apply, but it does not provide backups, failover, or a tested recovery path.
- Overlooking sensitive output: Plans and state can contain values or references that should not be copied into tickets, chat, or pull-request comments.
- Skipping post-apply checks: Successful Terraform execution confirms that the tool completed its actions; it does not prove that the application, access model, or monitoring is healthy.
- Using local exceptions as the normal path: Manual applies make the reviewer, credentials, inputs, and resulting audit trail harder to verify.
When to revisit
Use this checklist for every significant production pull request, but revisit the checklist itself at least before seasonal planning cycles and whenever the infrastructure workflow changes. It deserves a focused review after a Terraform or OpenTofu upgrade, provider upgrade, state-backend change, module redesign, CI/CD migration, or adoption of a new policy tool.
Also update the review process after a failed deployment, unexpected replacement, security incident, recovery exercise, or internal control audit. Capture the missed signal as a new checklist item or policy rule rather than relying on memory. For teams managing secrets alongside infrastructure, compare the workflow with this secrets management and rotation playbook.
Before the next apply, make the workflow concrete: generate a fresh plan, classify its actions, investigate every destroy or replace, run policy and security checks, obtain an accountable reviewer, and record the expected verification steps. After apply, compare the observed result with the approved plan and close the loop with evidence from state, monitoring, access checks, and application health. That repeated discipline turns a Terraform plan from a command output into a dependable infrastructure change-control checkpoint.