Provider Resources

Terraform Provider Resources Implementation Guide

Overview

This guide covers developing Terraform Provider resources and data sources. Resources represent infrastructure objects that Terraform manages through Create, Read, Update, and Delete (CRUD) operations.

Use the [Plugin Framework](https://developer.hashicorp.com/terraform/plugin/framework) for all net-new resources and data sources. Plugin SDKv2 is for maintaining resources that already exist on it; do not write new code against it. A provider can serve both during migration by muxing (terraform-plugin-mux), so adopting the Framework never requires a big-bang rewrite. To tell which mode an existing provider is in, check go.mod: terraform-plugin-mux present means it serves both SDKv2 and Framework code; only terraform-plugin-sdk/v2 means SDKv2-only; only terraform-plugin-framework means Framework-only. Be cautious about migrating existing SDKv2 resources: the Framework distinguishes null from zero values, so naive migrations change behavior for existing users (use the provider-framework-migration skill, if available).

References (load when needed):

• references/design-principles.md — what should (and should not) become a resource; data source semantics; relationship and async-task modeling
• references/retries-and-waiters.md — eventual consistency, retry patterns, and status/wait function structure

File Structure

Most providers keep every resource in a single package:

internal/provider/
├── provider.go                  # Provider schema + Configure
├── widget_resource.go           # Resource implementation
├── widget_resource_test.go      # Acceptance tests
├── widget_data_source.go        # Data source (if applicable)
└── widget_data_source_test.go

Large multi-service providers (e.g. terraform-provider-aws) split into internal/service/<service>/ packages instead, with an idiomatic file taxonomy worth adopting once a package grows: consts.go, find.go (finders), status.go (status functions), wait.go (waiters), sweep.go (test sweepers), exports_test.go.

Documentation lives in docs/ and is generated with tfplugindocs:

docs/
├── resources/<name>.md          # generated; optional <name>.md.tmpl template
└── data-sources/<name>.md

(Hand-written website/docs/r/*.html.markdown trees exist in some older, large providers — follow the target repo's convention when editing one.)

Resource Structure

A Framework resource is a struct holding the API client, with interface assertions making the implemented behaviors explicit:

var (
    _ resource.Resource                = &widgetResource{}
    _ resource.ResourceWithConfigure   = &widgetResource{}
    _ resource.ResourceWithImportState = &widgetResource{}
)

func NewWidgetResource() resource.Resource {
    return &widgetResource{}
}

type widgetResource struct {
    client *examplecloud.Client
}

func (r *widgetResource) Metadata(_ context.Context, req resource.MetadataRequest, resp *resource.MetadataResponse) {
    resp.TypeName = req.ProviderTypeName + "_widget"
}

// Configure receives the client the provider built in its own Configure.
func (r *widgetResource) Configure(_ context.Context, req resource.ConfigureRequest, resp *resource.ConfigureResponse) {
    if req.ProviderData == nil {
        return // provider not yet configured (e.g. validation phase)
    }
    client, ok := req.ProviderData.(*examplecloud.Client)
    if !ok {
        resp.Diagnostics.AddError(
            "Unexpected Resource Configure Type",
            fmt.Sprintf("Expected *examplecloud.Client, got: %T.", req.ProviderData),
        )
        return
    }
    r.client = client
}

func (r *widgetResource) Schema(ctx context.Context, req resource.SchemaRequest, resp *resource.SchemaResponse) {
    resp.Schema = schema.Schema{
        Attributes: map[string]schema.Attribute{
            "name": schema.StringAttribute{
                Required: true,
                PlanModifiers: []planmodifier.String{
                    stringplanmodifier.RequiresReplace(),
                },
                Validators: []validator.String{
                    stringvalidator.LengthBetween(1, 255),
                },
            },
            "id": schema.StringAttribute{
                Computed: true,
                PlanModifiers: []planmodifier.String{
                    stringplanmodifier.UseStateForUnknown(),
                },
            },
        },
    }
}

How the provider's Configure produces that client — schema, credential resolution, validation — is covered by the provider-configuration skill (if available).

On `id`: SDKv2 required a magic id attribute; the Framework does not. If the API has its own identifier, expose it under its real meaning and do not add a second, redundant id. Only keep id when it is the API's identifier (as above).

CRUD Operations

Create

func (r *widgetResource) Create(ctx context.Context, req resource.CreateRequest, resp *resource.CreateResponse) {
    var data widgetResourceModel
    resp.Diagnostics.Append(req.Plan.Get(ctx, &data)...)
    if resp.Diagnostics.HasError() {
        return
    }

    input := &examplecloud.CreateWidgetInput{
        Name: data.Name.ValueStringPointer(),
    }

    output, err := r.client.CreateWidget(ctx, input)
    if err != nil {
        resp.Diagnostics.AddError(
            "Error creating Widget",
            fmt.Sprintf("creating Widget (%s): %s", data.Name.ValueString(), err),
        )
        return
    }

    data.ID = types.StringPointerValue(output.ID)

    // For eventually consistent APIs, wait for the resource to be usable
    // before returning — see references/retries-and-waiters.md.

    resp.Diagnostics.Append(resp.State.Set(ctx, &data)...)
}

Read

Read must handle out-of-band deletion by removing the resource from state so the next plan recreates it, rather than erroring forever:

func (r *widgetResource) Read(ctx context.Context, req resource.ReadRequest, resp *resource.ReadResponse) {
    var data widgetResourceModel
    resp.Diagnostics.Append(req.State.Get(ctx, &data)...)
    if resp.Diagnostics.HasError() {
        return
    }

    output, err := findWidgetByID(ctx, r.client, data.ID.ValueString())
    if isNotFound(err) {
        tflog.Warn(ctx, "Widget not found, removing from state", map[string]any{"id": data.ID.ValueString()})
        resp.State.RemoveResource(ctx)
        return
    }
    if err != nil {
        resp.Diagnostics.AddError(
            "Error reading Widget",
            fmt.Sprintf("reading Widget (%s): %s", data.ID.ValueString(), err),
        )
        return
    }

    data.Name = types.StringPointerValue(output.Name)

    resp.Diagnostics.Append(resp.State.Set(ctx, &data)...)
}

Update

Only call the API for attributes that actually changed; compare plan against state:

func (r *widgetResource) Update(ctx context.Context, req resource.UpdateRequest, resp *resource.UpdateResponse) {
    var plan, state widgetResourceModel
    resp.Diagnostics.Append(req.Plan.Get(ctx, &plan)...)
    resp.Diagnostics.Append(req.State.Get(ctx, &state)...)
    if resp.Diagnostics.HasError() {
        return
    }

    if !plan.Description.Equal(state.Description) {
        input := &examplecloud.UpdateWidgetInput{
            ID:          plan.ID.ValueStringPointer(),
            Description: plan.Description.ValueStringPointer(),
        }
        if _, err := r.client.UpdateWidget(ctx, input); err != nil {
            resp.Diagnostics.AddError(
                "Error updating Widget",
                fmt.Sprintf("updating Widget (%s): %s", plan.ID.ValueString(), err),
            )
            return
        }
    }

    resp.Diagnostics.Append(resp.State.Set(ctx, &plan)...)
}

Delete

Treat "already gone" as success — the desired end state is reached: