* feat(nutanix): add state tags and disk_status

Introduce resource state tags so capacity-planning workflows can filter
out hosts in maintenance, disconnected hosts, powered-off VMs, degraded
disks, and clusters in special operation modes:

- ntnx_maintenance_state, ntnx_connection_state on host.*
- ntnx_power_state on vm.* (was previously only on vm.status)
- ntnx_operation_mode on cluster.*
- ntnx_disk_status on host.storage_* (worst-status aggregation)

Disk status is sourced once per check from the cluster-wide
api/clustermgmt/v4.0/config/disks endpoint and cached by node ID.
_report_stats gains an extra_tags_by_key parameter so disk_status is
scoped to host.storage_* metrics only.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(nutanix): cover disk_status and state tag extraction

Add focused unit tests for the new state-tag and disk_status paths:

- test_disk_status: parametrized aggregation matrix (NORMAL, degraded
  states, $UNKNOWN/$REDACTED, mixed, forward-compat enum values),
  cache-build defensiveness (skips disks missing nodeExtId), and an
  end-to-end check that a degraded disk flows ntnx_disk_status:degraded
  onto storage_* metrics. Disks-endpoint failure is asserted to leave
  storage metrics emitted without the tag.
- test_tag_extraction: defensive checks for missing hypervisor block,
  missing config.operationMode, empty-string field values, and verbatim
  PAUSED power-state preservation.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore(nutanix): record fixtures by resource selectively

Add --resources / -r flag to the fixture recorder accepting a comma-
separated or repeated list of resources to refresh, plus --list to
enumerate available names. When a dependent resource is requested
without its prerequisites (e.g., host_stats without clusters/hosts),
the prerequisites are fetched in memory but their fixtures are not
overwritten.

Adds record_disks() targeting api/clustermgmt/v4.0/config/disks.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* feat(nutanix): expose state tags in Capacity Planning

Consolidate the Capacity Planning section into three unified tables —
Clusters, Hosts, VMs — each grouped by their state tags so capacity
planners can filter inline by maintenance/connection/power/operation
state. Add a Storage Capacity by Host table grouped by ntnx_disk_status
to surface degraded storage. The intro note now summarizes the inputs
and the available state tags.

Datadog lowercases tag values at ingestion, so all tag-value filters in
the dashboard use lowercase (e.g., ntnx_connection_state:connected).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(nutanix): tighten disk_status helpers

Minor compression in the disk_status path with no behavior change:

- _aggregate_disk_status: drop the redundant empty-set guard and
  inline the degraded_states constant — set ops on empty sets work,
  the constant was used once.
- _report_stats: collapse the 4-line param docstring to one line.
- _process_single_host: inline the disk_status_extra_tags local;
  it was used exactly once.
- _get_disk_status_storage_tags: fold the storage-keys tuple onto
  one line.

Net -22 lines in infrastructure_monitor.py; all 136 unit tests pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(nutanix): guard disks API consumers against malformed entries

The new /api/clustermgmt/v4.0/config/disks consumers (_build_disks_by_
host_cache, _aggregate_disk_status) iterate over response items and
call .get(...) on each — if Nutanix ever returns a null entry or a
non-dict, that would raise AttributeError. Add isinstance(d, dict)
filters at both sites and parametrized test coverage for None /
strings / ints mixed into the disk list.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(nutanix): emit vm.disk_capacity_bytes from config

The metric was mapped from diskCapacityBytes in VM_STATS_METRICS, but
the Nutanix v4 stats endpoint never returns that field — so the metric
was effectively never emitted (the existing test list parked it under
VM_STATS_METRICS_OPTIONAL, which silently let it slip).

Source it instead from the VM config: sum vm.disks[].backingInfo.disk
SizeBytes. Same approach as memory.allocated_bytes from memorySizeBytes.
This makes the metric reliably available for capacity-planning queries
and the dashboard's per-VM Disk Allocated column now populates.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(nutanix): correct host power metric name in optional test list

HOST_STATS_METRICS_OPTIONAL had nutanix.host.power.consumption.instant_watt
(dots) but the canonical metric — emitted by the integration and matching
the cluster equivalent — is nutanix.host.power_consumption_instant_watt
(underscores). The typo made the OPTIONAL entry dead: when the metric ever
lands in a power-meter-equipped environment, assert_all_metrics_covered()
would fail because the OPTIONAL bucket entry didn't actually cover it.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(nutanix): add per-metric state-tag coverage guards

Six tests that introspect the aggregator after a check run and assert
every new state tag lands on every applicable metric:

- ntnx_maintenance_state on every host metric for hosts whose source
  data has the field
- ntnx_connection_state on every host metric (both fixture hosts have it)
- ntnx_power_state on every VM metric
- ntnx_operation_mode on every cluster metric
- ntnx_disk_status on every host.storage_* metric
- ntnx_disk_status NOT present on non-storage host metrics (scoping guard)

If someone removes a tag emission from one of the _extract_*_tags methods,
the failure message names every metric that lost the tag rather than
failing at the first bundled assertion in test_hosts/test_vms/test_clusters
and stopping there.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix changelog pr number

* fix(nutanix): address review feedback for state tags

Bundle of changes driven by the PR review:

- Lowercase state-tag values (ntnx_connection_state, ntnx_operation_mode,
  ntnx_power_state) at emission so the code matches Datadog's
  ingestion-time normalization explicitly rather than relying on it
  implicitly. ntnx_maintenance_state was already lowercase from the API.
- Always emit ntnx_power_state, falling back to "unknown" when the
  source field is missing/empty. Previously the tag silently dropped
  for those VMs, breaking dashboards/monitors that group by power_state.
- Hoist DEGRADED_DISK_STATUSES and HOST_STORAGE_STAT_KEYS to
  datadog_checks/nutanix/metrics.py as module-level frozensets, and
  derive HOST_STORAGE_METRICS in tests/metrics.py from them — eliminates
  the three independent enumerations that could drift if a fifth
  storage stat is added.
- Log the disk count alongside the host bucket count when caching
  ("Cached %d disks across %d hosts") for easier triage.
- Move the InfrastructureMonitor stub fixture used by tag-logic unit
  tests into tests/conftest.py with the union of attributes both files
  need; remove the divergent local fixtures.
- Add a parametrized unit test for _extract_vm_disk_capacity_bytes
  covering missing disks, missing backingInfo/diskSizeBytes, and
  non-dict entries — previously only happy-path e2e coverage existed.

156 unit tests pass; 1 dropped review finding (functional reviewer's
"silent failure" claim about the dashboard) was correctly contested
by the integrations reviewer — Datadog auto-lowercases tag values at
ingestion. Lowercasing at emission is purely a clarity improvement.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(nutanix): guard state-tag .lower() against non-string values

The previous commit introduced .lower() on the four state tags
(maintenance, connection, operation_mode, power) to align emission
with Datadog's lowercased ingestion. If the Nutanix API ever returns
a non-string value (e.g. an int, bool, or list — implausible per the
v4 schema, but possible if the API misbehaves), .lower() would raise
AttributeError. The previous walrus-only pattern would have just
formatted the unusual value via __str__ — no crash.

Add a tiny module-level _norm_state(value) helper:
- Returns value.lower() when value is a non-empty str.
- Returns None for missing, empty, or non-string values.

Threading it through the four sites preserves the walrus :=
pattern and keeps each emission a single line. For VM powerState
the fallback to "unknown" is unchanged.

Adds parametrized regression tests covering int/bool/list inputs
on host, cluster, and VM tag extraction. 162 unit tests pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* docs(nutanix): rewrite changelog entries for customer audience

- Drop 23578.fixed.1: the always-emit ntnx_power_state behavior is
  part of the tag's first shipped contract (added in this same PR);
  not a fix to a previously-released feature.
- Rewrite 23578.added to lead with the user benefit (capacity-planning
  filtering) and structure the tags as bullets.
- Trim 23578.fixed of internal field names; surface only what users
  observe ("the metric now reports values where it didn't before").

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore(nutanix): split vm.disk_capacity_bytes fix into separate PR

Move the vm.disk_capacity_bytes fix out of this PR; it ships separately
in PR #23583. This PR is now state-tags-only.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(nutanix): drop private-method tests in favor of check-level coverage

Remove test_tag_extraction.py and test_state_tag_coverage.py, and trim
test_disk_status.py to only the integration-style tests that exercise
the check end-to-end. Coverage of the new state tags is preserved
through HOST_TAGS/CLUSTER_TAGS/PCVM_TAGS in the existing per-entity
tests. Also drop the now-unused monitor fixture from conftest, simplify
_norm_state, and remove a redundant defensive check in
_aggregate_disk_status.

* docs(nutanix): rewrite state-tags changelog for users

* docs(nutanix): collapse state-tags changelog into one line

* docs(nutanix): frame state-tags changelog as a new feature

* fix(nutanix): always emit state tags with unknown fallback

Make ntnx_maintenance_state, ntnx_connection_state, ntnx_operation_mode,
and ntnx_disk_status emit consistently like ntnx_power_state already
does — always present, falling back to "unknown" when the source field
is missing. Also normalize the spec-defined sentinel values (\$UNKNOWN,
\$REDACTED, UNDETERMINED) to "unknown" so they don't surface as ugly
tag values like ntnx_operation_mode:\$unknown.

Without this, dashboards and monitors filtering on these tags silently
drop entities whose source field is missing — the same failure mode
the power-state fallback was originally added to prevent.

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>