test: Serialize search tests, cap index waits, skip internal CI package

[EspenAlbert]

Why

Both search packages fail often against cloud-dev. The two packages share one three-node M10 execution cluster and every test runs in parallel, so index builds queue. searchindexapi makes it worse: its generated resource waits three hours for READY, STEADY per operation, unconditionally, which holds a job and can cross the 300 minute test deadline. That wait is what makes this a cost change and not only a flake fix.

The same wait is opt-in in the legacy searchindex resource behind wait_for_index_build_completion, and no test in that package ever set it, so neither its create nor its update wait had run.

What

Follows the search-test 1-pager, which carries the measurements and the rejected alternatives: One Pager: Search acceptance test fixes and WRITING-40266.

This PR carries all four test-side changes from the 1-pager.

• Skip searchindexapi in CI. acc.SkipTestForCI(t) is the first statement in all 9 tests of the package, above the acc.ClusterNameExecution call. The package stays in the autogen_slow ACCTEST_PACKAGES list, so it still runs locally and on demand and still detects upstream schema changes. Its 3 hour wait holds the job for hours, and the resource is internal: not registered in the provider, no customer docs page.
• Serial execution. 16 of the 18 tests in searchindex and all 9 in searchindexapi move from resource.ParallelTest to resource.Test. The other 2 legacy tests already ran serially. Each package runs in its own job, so each is serialized independently.
• Cover the build waits. TestAccSearchIndex_updatedToEmptyMappingsFields now sets wait_for_index_build_completion in both steps, so it is the first test to exercise the create and update READY, STEADY waits in the legacy package. No new test.
• Bound the failure path. Every test config in both packages gets a timeouts block with create, update, and delete at 60 minutes: 9 configs in each package. The tests inherit no timeout today and the resource default is 3 hours. The block caps one Terraform operation, so it does not bound the multi-step test total and does not change shipped defaults.
• Loader budget and reason. The sample-data load polls for 30 minutes instead of 15, and a FAILED load now reports the job's errorMessage, so the CI log names the cause instead of only the job ID and state. The fail-fast on FAILED is unchanged.

Type of change:

• Bug fix (non-breaking change which fixes an issue). Please, add the "bug" label to the PR.
• New feature (non-breaking change which adds functionality). Please, add the "enhancement" label to the PR. A migration guide must be created or updated if the new feature will go in a major version.
• Breaking change (fix or feature that would cause existing functionality to not work as expected). Please, add the "breaking change" label to the PR.
• This change requires a documentation update
• Documentation fix/enhancement

Required Checklist:

• I have signed the MongoDB CLA
• I have read the contributing guides
• I have checked that this change does not generate any credentials and that they are NOT accidentally logged anywhere.
• I have added tests that prove my fix is effective or that my feature works per HashiCorp requirements
• I have added any necessary documentation (if appropriate)
• I have run make fix and verified my code
• If changes include deprecations or removals I have added appropriate changelog entries.
• If changes include removal or addition of 3rd party GitHub actions, I updated our internal document. Reach out to the APIx Integration slack channel to get access to the internal document.

FollowUp/Next Steps

• The READY, STEADY misread stays open. The 1-pager reads it as the provider missing a terminal state rather than the build never finishing, and skipping searchindexapi removes the only CI reproduction. Reinstating those tests is the way back to a signal.
• No package-level M10 figure for searchindexapi on dev, so there is no full before-and-after for the serialization there. Five of nine tests have no clean M10 measurement.
• Helper changes are test-only. No shipped resource default changes.

[Copilot]

🟡 Changes recommended

internal/testutil/acc/atlas.go does not compile, and its cluster-tier override expands the stated PR scope.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Updates search acceptance tests to reduce contention, bound long-running operations, skip internal API tests in CI, and improve sample-load diagnostics.

Changes:

• Serializes search acceptance tests and adds 60-minute operation timeouts.
• Skips searchindexapi tests in CI.
• Extends sample-data polling and reports failure details.
• Adds configurable test cluster tiers.

File summaries

File	Summary
internal/testutil/acc/cluster.go	Adds configurable cluster-tier selection.
internal/testutil/acc/cluster_test.go	Tests cluster-tier resolution.
internal/testutil/acc/cloud_backup_collection_restore_fixture_test.go	Updates sample-load failure tests.
internal/testutil/acc/atlas.go	Applies the configured cluster tier.
internal/testutil/acc/advanced_cluster.go	Improves sample-data polling and error reporting.
internal/serviceapi/searchindexapi/resource_test.go	Serializes, skips CI execution, and adds timeouts.
internal/service/searchindex/resource_search_index_test.go	Serializes legacy tests, adds timeouts, and covers wait behavior.

Review details

• Files reviewed: 7/7 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[augmentcode]

🤖 Augment PR Summary

Summary: This PR reduces search-index acceptance-test flakiness and excessive CI runtime caused by queued index builds on a shared execution cluster.

Changes:

• Runs legacy searchindex acceptance tests serially instead of using resource.ParallelTest.
• Skips all internal-only searchindexapi acceptance tests when running in CI, while retaining local/on-demand coverage.
• Adds one-hour create, update, and delete Terraform timeouts to all search-index test configurations.
• Enables wait_for_index_build_completion in the mappings-fields update test to cover legacy create and update wait paths.
• Extends sample-data loading retries from 15 to 30 minutes.
• Includes Atlas's sample-data job error message when a load reaches FAILED, with unit-test coverage for that message.

Technical Notes: The timeout blocks only affect acceptance-test configurations; provider defaults are unchanged. The API resource's existing multi-hour READY/STEADY waits are bounded in local test configurations and avoided in routine CI.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. No suggestions at this time.

Comment augment review to trigger a new review at any time.