v3.0.2 fixes a spurious + wait = false diff on kubectl_manifest after upgrading from 2.x or migrating from gavinbunney/kubectl (#412, #413). It's a drop-in upgrade from 3.0.x.
Fixes
- No more
+ wait = falseon every resource after an upgrade. 2.x and gavinbunney/kubectl stored an unsetwaitas null. 3.0.x kept that null, so every such resource planned a change, and applying it re-applied the manifest, overwriting fields listed inignore_fields. The provider now fills its default into state in three places: the state upgrade from 2.x, the gavinbunneymoved {}path, and refresh. The same fix covers the other apply options with a default (force_new,server_side_apply,field_managerand so on). State that 3.0.0 or 3.0.1 already rewrote is fixed on the next refresh. With-refresh=false, the diff can still appear once. Thanks to @tjorri for the report and diagnosis.
Other changes
- New CI smoke test that upgrades a real 2.4.1 state through the built provider.
- Built with Go 1.27.4 (security fix) and golang.org/x/crypto v0.59.0 (security fix). Dependency updates: go-cty v1.20.0, backoff v7, yaml.v3.