Error: could not load provider for resource: could not find provider

Error: could not load provider for resource ... could not find provider urn:pulumi:...

TL;DR

Your state points at a provider version that is not installed anymore. Back up state with pulumi stack export, rewrite the stale provider URN to the installed version, re-import, then destroy or update.

The error

error: could not load provider for resource urn:pulumi:[stack]::[project]::eks:index:Cluster$aws:eks/cluster:Cluster::eksCluster: could not find provider urn:pulumi:[stack]::eks::pulumi:providers:aws::default_6_37_1::[id]

Fix it

1. Back up state before touching anything: pulumi stack export > stack.json and keep a copy somewhere safe.

• Success check: stack.json exists and is valid JSON.

1. Find which provider version IS installed: pulumi plugin ls and note the aws provider version (e.g. 6.40.0).

• Success check: you have a concrete installed version.

1. In a copy of the export, replace every occurrence of the stale provider URN segment (e.g. default_6_37_1) with the installed one (e.g. default_6_40_0), keeping the trailing id suffix consistent with the installed provider.

• Success check: no occurrences of the old version string remain.

1. pulumi stack import < stackmodified.json, then re-run pulumi destroy or pulumi up.

• Success check: the provider loads and the operation proceeds.

When to use this

You hit this when destroy/update fails at provider load after a CLI or provider upgrade changed which provider versions are installed.

When NOT to use this

Do not use this for "failed to download plugin" errors. That is a missing plugin; install it with pulumi plugin install resource aws [version] instead.

Compatibility

Pulumi CLI 3.x. The URN surgery approach works across provider versions.

Variants

• error: could not load provider for resource [urn]: could not find provider [urn] naming any provider
• The same failure on pulumi preview when the preview needs the provider

Root cause

State records the exact provider instance (name + version + id) each resource was created with. Upgrading the CLI does not rewrite those references, so if the old provider binary is gone, even a destroy cannot load what it needs.

Edge cases

• If the replacement still fails, the id suffix may also need to match an installed provider instance. Compare against pulumi plugin ls output carefully.
• --disable-integrity-checking does not bypass this; the provider binary itself is missing.