Terraform Version, Provider Version and New Relic environment

• Terraform: v1.6.3
• Provider: v3.80.x (bug present since notification resources were introduced)
• New Relic: US and Staging regions

Expected Behavior

When newrelic_notification_channel (or newrelic_notification_destination) is created and the NR API returns a valid resource ID, the resource should be recorded in Terraform state — even if the response also contains errors.

Actual Behavior

The resource is created in New Relic but not recorded in Terraform state, producing an orphaned resource and an infinite create→fail loop on subsequent applies.

The error shown is:

UNKNOWN_ERROR: 

(empty description and details — the API returned null for both fields)

Root Cause

AiNotificationsChannelResponse can contain both a valid Channel (with ID) and entries in the Errors array simultaneously. The Create function checks errors before calling d.SetId():

// resource_newrelic_notifications_channel.go:111-116
errors := buildAiNotificationsErrors(channelResponse.Errors)
if len(errors) > 0 {
    return errors                            // ← exits here
}
d.SetId(channelResponse.Channel.ID)          // ← never reached

The same pattern exists in resource_newrelic_notifications_destination.go.

Steps to Reproduce

This occurs intermittently (the NR backend returns UNKNOWN_ERROR with null fields on some channel creations despite the channel being created). We've observed it consistently during alert policy migrations involving PagerDuty notification channels (PAGERDUTY_ACCOUNT_INTEGRATION).

1. Define a newrelic_notification_channel resource
2. Run terraform apply
3. If the NR API returns a response with both a valid channel ID and an UNKNOWN_ERROR:
   • Apply fails with UNKNOWN_ERROR:
   • Channel exists in NR (verifiable via NerdGraph aiNotifications.channels query)
   • Channel is not in Terraform state
4. Subsequent terraform apply attempts to create the channel again → same error → loop

Impact

• Orphaned resources accumulate in New Relic (channels exist but aren't managed by Terraform)
• Blocked workflows: terraform apply fails indefinitely until manual terraform import
• Affects both newrelic_notification_channel and newrelic_notification_destination

Suggested Fix

Move d.SetId() before the error check, guarded by a non-empty ID:

if channelResponse.Channel.ID != "" {
    d.SetId(channelResponse.Channel.ID)
}

errors := buildAiNotificationsErrors(channelResponse.Errors)
if len(errors) > 0 {
    return errors
}

Per the Terraform Plugin SDK documentation, SetId must be called with a non-empty value for the resource to be saved into state. Per the Plugin Framework Create Caveats, any response errors will cause Terraform to mark the resource as tainted — which is the correct behavior here: the resource exists but may need reconciliation.

Workaround

Query NerdGraph for orphaned channels and import them manually:

{
  actor {
    account(id: ACCOUNT_ID) {
      aiNotifications {
        channels(filters: {destinationId: "DESTINATION_ID"}) {
          entities { id name status }
        }
      }
    }
  }
}

terraform import 'newrelic_notification_channel.example' CHANNEL_ID