test(cluster): unit tests for cluster_getter

[mkoura]

What this does

Adds unit tests for the previously untested ClusterGetter
functionality, building on the test_cluster_getter.py foundation
from #3567. 40 new tests covering:

• _make_instances_order - light tests iterate the two tail instances
  first (fixed order), heavy tests last, head order randomization,
  single-instance edge case
• _init_use_resources - CLUSTER always added, locked resources
  filtered out of "use", filter objects passed through, one-shot
  iterator input
• _test_needs_respin - custom scripts and the non-initial marked
  test exception
• _update_marked_tests - the staleness matrix: stale mark cleaned up
  with "respin after mark" conversion, marks kept while the instance
  is respun, fresh idle mark kept, running mark refreshed (not
  refreshed while fresh so polling stays write-free, and never cleaned
  no matter how old), the newest record deciding staleness for the
  whole mark, mixed staleness of multiple marks on one instance, and
  no database writes when no marks exist
• _resolve_resources_availability - lock/use conflicts in all
  directions, locked-implies-in-use dedup, OneOf filter resolution,
  resolved resources recorded only on success
• _init_prio / _wait_for_prio - waiting on a priority test and all
  three no-wait exceptions
• _respun_by_other_worker, _is_already_running, _finish_respin
  two-phase contract, _create_test_status_records (plain test,
  marked cleanup, unmarked cleanup, (setup) suffix stripping),
  _check_dead_fraction thresholds and the _fail_on_dead_clusters
  strict check window

Not covered (needs a live cluster): _respin, _is_healthy,
_cluster_needs_respin and the get_cluster_instance loop itself.

Latent contract violation caught while writing the tests

test_one_shot_iterator exposed that _init_use_resources violates
its own Iterable contract: it iterates use_resources twice, so a
one-shot iterator is exhausted after the first pass and every resource
filter (e.g. OneOf) is silently dropped. Latent on the production
path - the caller materialized the argument - but reachable by any
direct call with an iterator. The first commit moves the
materialization into the method, drops the caller's now-redundant
conversion and re-wrap, and makes the return type an explicit list.

Verification

• All 90 framework tests pass, make lint clean, each commit tested
  individually.
• Reviewed in a review-fix loop until convergence; the test suite was
  mutation-tested by the reviewers (e.g. max -> min in the
  staleness fold, dropped light-test condition, unconditional mark
  refresh - all caught by the tests).

[Copilot]

🟢 Ready to approve

The functional change is small and clearly correct, and it is backed by substantial, targeted unit test coverage for the affected and related behaviors.

This review doesn't count toward merge requirements. Sign up for the private preview to control whether Copilot approvals count.

Pull request overview

This PR expands unit-test coverage for the cluster scheduler’s ClusterGetter internals, adding broad behavioral checks around instance selection, resource initialization/resolution, mark staleness handling, respin/priority flows, and dead-cluster thresholds. It also fixes a latent contract violation in _init_use_resources by materializing use_resources internally so one-shot iterators aren’t accidentally exhausted across multiple passes.

Changes:

• Add extensive unit tests for key ClusterGetter helpers (ordering, resource init/availability resolution, marked-test staleness, respin/priority helpers, status record creation, dead-cluster checks).
• Fix _init_use_resources to eagerly materialize use_resources, preventing loss of filter objects (e.g., OneOf) when use_resources is a one-shot iterator.
• Simplify get_cluster_instance by removing redundant use_resources materialization and using the new _init_use_resources return value directly.

File summaries

File	Description
framework_tests/test_cluster_getter.py	Adds a large suite of focused unit tests covering ClusterGetter internal behaviors and edge cases.
cardano_node_tests/cluster_management/cluster_getter.py	Fixes _init_use_resources iterator contract and removes redundant list conversion in get_cluster_instance.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Low

We're testing this review assessment. Please use 👍 or 👎 to tell us if it's correct.