* Speed up hudi e2e

* clean up spark version

* fix cluster capacity counting for hostless/filtered vms

* changelog

* fix multi cluster case, improve testing fixtures

* improve test

* address review: use ClusterCapacity for hostless VMs, extract helper, fix test style

- Replace dict[str, tuple[int, int]] with dict[str, ClusterCapacity] for
  _hostless_vm_capacity_by_cluster to align with existing pattern
- Extract _extract_vm_capacity helper to eliminate duplication between
  _accumulate_vm_capacity and _report_vm_capacity_metrics
- Add debug log when a hostless VM has no cluster ID
- Add comment documenting non-batch mode limitation for hostless VMs
- Convert TestMultiClusterHostlessVMCapacity class to plain test functions
- Fix misleading comment: OFF VM has a host, it is not a hostless VM
- Import HOST_NAME from constants.py instead of redefining locally
- Parametrize test_exclude_vm_by_id for both batch modes

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* address review: cleanup code, separation of concerns, and flat unit tests

* change health_check failure log message to error instead of warning

* more refactoring and code cleanup

* fix unused HOST_NAME import in test_resource_filters

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* add exclude_filtered_resources_from_cluster_capacity property + regenerate models

* separate cluster capacity computation from metrics collection + add exclude_filtered_resources_from_cluster_capacity + cleanup + added tests

* address review + more cleanup + improve tests

* Update nutanix/README.md

Co-authored-by: Rosa Trieu <107086888+rtrieu@users.noreply.github.com>

* Address review feedback: reset state before health check, consolidate VM capacity extraction

- Move reset_state() before health check to guarantee clean state on every run
- Consolidate _extract_vm_capacity to return all fields, eliminating duplicate
  parsing in _report_vm_capacity_metrics
- Fix trailing whitespace and extra blank lines

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Clarify VM collection mode decoupling in docstrings

Make it explicit that _vms_by_host is populated by either batch or
non-batch mode, and that hostless VMs (the "" key) are only present
in batch mode.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Include hostless VMs in cluster capacity for all collection modes

Add _get_hostless_vms to lazily fetch and cache hostless VMs (the "" key)
in non-batch mode. Split _report_cluster_capacity_metrics into two clear
loops: hosted VMs (scoped by host_ids) and hostless VMs (scoped by
cluster_id). Parameterize hostless VM tests on batch_vm_collection.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Revert _get_hostless_vms: hostless VMs only available in batch mode

Fetching all VMs to extract hostless ones defeats the purpose of
non-batch mode. Hostless VMs are only captured via _build_vms_by_host_cache
(batch mode), which is the expected trade-off.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Rosa Trieu <107086888+rtrieu@users.noreply.github.com>

* port changelogs

* port snmp changes

* Update __about__.py

* Delete .github/workflows/scripts/port_release_to_master.py

* kafka: Add topic partition size metric

Add kafka.log.partition.size metric from the kafka.log:type=Log,name=Size
JMX bean, tagged with topic and partition. Reports the on-disk size in
bytes of each topic partition log per broker.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* kafka: Rename excluded version tag to request_version for RequestsPerSec beans

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…essages (#23281)

* [kafka_consumer] Send heartbeat via data streams messages when cluster monitoring is enabled

Report context count and limit on every check run so we can tell if a customer is hitting the context ceiling.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Add changelog entry for cluster monitoring heartbeat

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Fall back to admin client metadata for cluster_id in heartbeat

When highwater offset collection is skipped (context limit exceeded),
cluster_id is empty. Use the cluster metadata from request_metadata_update()
which runs unconditionally.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Derive cluster_id internally in heartbeat method

Always read cluster_id from admin client metadata instead of accepting
it as a parameter. Simpler and removes dependency on caller state.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Use getattr for _cluster_metadata access to handle mocked clients

Tests mock self.client without _cluster_metadata attribute, causing
AttributeError. Use getattr with a default to be safe.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Accept cluster_id from caller, use _auto_detected_cluster_id for original

The check() method already resolves cluster_id (with override) and stores
the auto-detected value in config._auto_detected_cluster_id. Reuse that
rather than trying to read _cluster_metadata directly, which breaks with
mocked clients in tests.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…tions (#23282)

* bump datadog-checks-base package

* Add changelog

* Add notable changes to changelog