Provider Framework Migration

Migrating from Plugin SDKv2 to the Plugin Framework

The Plugin Framework is required for net-new resources and data sources; SDKv2 is maintenance-only. Migration is per-resource and incremental: a muxed provider serves SDKv2 and Framework implementations side by side, so you never need a big-bang rewrite. This skill covers the mux setup, the per-resource workflow, and the behavioral traps that turn a mechanical translation into a silent breaking change.

Reference (load when needed):

• references/schema-mapping.md — the full SDKv2 → Framework translation table with code pairs

Official guide: Framework migration (opens in a new tab).

Decide Whether to Migrate at All

Migration has real risk and little user-visible payoff, so triage first:

• Do not migrate complex or heavily-used resources without a driving need (a Framework-only feature, a bug that SDKv2 cannot fix). The two SDKs differ behaviorally — most importantly around null versus zero values — and those differences surface as breaking changes for existing users. This is the standing policy in large providers like terraform-provider-aws.
• Simple resources migrate safely: flat schemas, no DiffSuppressFunc, no CustomizeDiff, no StateFunc, no complex nested blocks.
• New capabilities never require migrating old code — mux and write the new resource in the Framework alongside the old ones.

To tell what mode a provider is in, check go.mod: terraform-plugin-mux present means it already serves both; only terraform-plugin-sdk/v2 means SDKv2-only (mux setup is your first step); only terraform-plugin-framework means the migration is done.

Step 1: Mux the Provider

Combine both plugin servers in main.go. Serving protocol version 6 requires upgrading the SDKv2 server with tf5to6server (protocol 6 needs Terraform CLI >= 1.0; if you must support 0.12+, mux at protocol 5 with tf6to5server/tf5muxserver instead — but the Framework provider then cannot use protocol-6-only features like nested attributes):

package main

import (
    "context"
    "flag"
    "log"

    "github.com/hashicorp/terraform-plugin-framework/providerserver"
    "

Mux requirements that bite in practice:

• Provider schemas must match exactly across both plugins — same provider-level attributes, same types, same descriptions. Keep one source of truth for the provider configuration and mirror it.
• Each resource and data source may exist in only one of the two plugins. Migration’s final step is deleting the SDKv2 registration.
• If publishing to the Registry with protocol 6, set "metadata": {"protocol_versions": ["6.0"]} in terraform-registry-manifest.json.

Step 2: Baseline Before You Touch Anything

The migrated resource must be indistinguishable to users. Prove it with tests that exist before the migration:

1. Ensure the resource has passing acceptance coverage: _basic with an import step (ImportStateVerify: true), _disappears, and per-attribute update tests. If coverage is missing, write it against the SDKv2 implementation first — these tests are the migration’s acceptance criteria and must pass unchanged afterward.
2. Note behaviors tests don’t capture: attribute defaults, what happens when optional attributes are omitted (null vs ""/0/false is about to matter), and any DiffSuppressFunc/StateFunc normalization.

Step 3: Port the Resource

Translate schema and CRUD using the mapping table in references/schema-mapping.md. The rules that prevent breaking changes:

• Blocks stay blocks. An SDKv2 Elem: &schema.Resource{...} written as block { ... } syntax in user configs must become a Framework Block (schema.ListNestedBlock/SetNestedBlock) — converting it to a nested attribute changes the HCL syntax users must write, which is a breaking change. Nested attributes are for new schema only.
• Null is not zero. SDKv2 d.Get("name") returned "" for unset; the Framework model gives you types.String that distinguishes null, unknown, and "". Everywhere the old code checked == "" or relied on GetOk, decide explicitly what null means, and make sure you send the API the same thing SDKv2 sent (usually: omit the field when null).
• Keep the id attribute. Net-new Framework resources may omit a redundant id, but a migrated resource must keep its exact schema — removing or renaming attributes breaks existing state and configs.
• State must round-trip. The Framework reads the state SDKv2 wrote. If every attribute keeps its name and type, no state upgrade is needed. If the old schema stored a value the new types package normalizes differently, you need a StateUpgrader — treat that as a signal the resource may be in the do-not-migrate bucket.

Step 4: Move the Registration

Register the resource in the Framework provider’s Resources() and delete it from the SDKv2 provider’s ResourcesMap in the same commit — mux errors on duplicates.

Step 5: Verify

1. The pre-existing acceptance tests pass without modification — especially ImportStateVerify, which diffs imported state against stored state and catches most null-vs-zero regressions.
2. Add a state-compatibility step: apply a config with the last released (SDKv2) provider version, then plan with the migrated build — the plan must be empty. In terraform-plugin-testing this is a two-step test using ExternalProviders for the old version, then ProtoV6ProviderFactories with ConfigPlanChecks asserting an empty plan. The provider-test-patterns skill (if available) documents the pattern.
3. terraform plan against a real pre-migration state file shows no diff.

Checklist

• Not done: Resource is simple enough to migrate (no complex diff customization), or there’s a driving need
• Not done: Mux serves both plugins; provider-level schemas identical in both
• Not done: Acceptance tests existed before migration and pass unchanged after
• Not done: Blocks remained blocks; attribute names and types unchanged; id kept
• Not done: Null/omitted semantics preserved (API receives what SDKv2 sent)
• Not done: SDKv2 registration removed in the same change
• Not done: Empty-plan verified against state written by the previous release
• Not done: Changelog entry added, if the repo tracks release notes

Related Skills

Use the provider-resources skill (if available) for Framework CRUD, finder, and waiter patterns in the ported code, and provider-test-patterns for the regression and version-upgrade test patterns.