testing_new_config.go It is possible for a testStep to be both empty and non-empty plan post apply

[Philip-Jonany]

SDK version

Latest

I'm implementing a switchover, where a primary becomes the replica, and the replica becomes the primary in one apply. This causes some issues in testing. After this operation is applied, a refresh or plan is needed to verify that plan is not empty. Currently, acceptance tests say that after applying this config, the plan is both empty and non-empty.

Here's a summary of https://github.com/hashicorp/terraform-plugin-sdk/blob/main/helper/resource/testing_new_config.go
Line 70: Apply config
Line 117: WorkingDir.CreatePlan()

Line 133:

if !planIsEmpty(plan) && !step.ExpectNonEmptyPlan {
    check if plan was not empty
}

At this time, plan is "not empty", we just need a refresh and it will be an empty plan.
...
line 149: we do a refresh. Now, the plan is empty. (we should've done this before we do the plan check in line 133)
line 161: we do another plan after the refresh.

Now, here is why we see "schrodinger's plan"; plan is "both empty and non-empty"
line 188:

else if step.ExpectNonEmptyPlan && planIsEmpty(plan) {
    check and expect that plan is empty
}

The plan was non-empty before refresh in line 149, and is now empty.

Terraform Configuration Files

Initially:

resource "google_sql_database_instance" "original-primary" {
  name                = "%s"
  region              = "us-east7"
  database_version    = "SQLSERVER_2019_ENTERPRISE"
  deletion_protection = false

  root_password = "sqlserver1"
  settings {
    tier              = "db-perf-optimized-N-2"
    edition           = "ENTERPRISE_PLUS"
  }
}

resource "google_sql_database_instance" "original-replica" {
  name                 = "%s"
  region               = "us-west2"
  database_version     = "SQLSERVER_2019_ENTERPRISE"
  master_instance_name = google_sql_database_instance.original-primary.name
  deletion_protection  = false
  root_password = "sqlserver1"
  replica_configuration {
    cascadable_replica = true
  }

  settings {
    tier              = "db-perf-optimized-N-2"
    edition           = "ENTERPRISE_PLUS"
  }
}

`, primaryName, replicaName)
...

In the next config step, we apply this config:

resource "google_sql_database_instance" "original-primary" {
  name                = "%s"
  region              = "us-east7"
  database_version    = "SQLSERVER_2019_ENTERPRISE"
  deletion_protection = false
  root_password = "sqlserver1"
  instance_type = "READ_REPLICA_INSTANCE"
  master_instance_name = "%s"
  replica_configuration {
    cascadable_replica = true
  }
  replica_names = []
  settings {
    tier              = "db-perf-optimized-N-2"
    edition           = "ENTERPRISE_PLUS"
  }
}

resource "google_sql_database_instance" "original-replica" {
  name                 = "%s"
  region               = "us-west2"
  database_version     = "SQLSERVER_2019_ENTERPRISE"
  deletion_protection  = false
  root_password = "sqlserver1"
  instance_type = "CLOUD_SQL_INSTANCE"
  replica_names = [google_sql_database_instance.original-primary.name]
  settings {
    tier              = "db-perf-optimized-N-2"
    edition           = "ENTERPRISE_PLUS"
  }
}
`, primaryName, replicaName, replicaName)

After applying this config step, a refresh is needed before we do checks.
I have:
{
Config: googleSqlDatabaseInstance_switchover(primaryName, replicaName),
ExpectNonEmptyPlan: true, // I commend and uncomment, and the error says that this step is both empty and non-empty.
},

Expected Behavior

When I commend and uncomment ExpectNonEmptyPlan: true, and the error should say that the step is either empty and non-empty, not BOTH.

Actual Behavior

The error says that the config test step results in an empty, and also non-empty plan.

Steps to Reproduce

Run this test with my pull request (switchover implementation)
GoogleCloudPlatform/magic-modules#11850

References

GoogleCloudPlatform/magic-modules#11850

[rileykarson]

/cc @slevenick