Agent

Prelude

Release on: 2025-09-03

• Please refer to the 7.70.0 tag on integrations-core for the list of changes on the Core Checks

Upgrade Notes

• GPUM: GPU workload metadata and autodiscovery are now enabled by default. If you want to disable this functionality, add "nvml" to the autoconfig_exclude_features configuration flag in your datadog.yaml file.
• Upgraded the WDK version used by ddnpm.
• Upgrade ddprocmon to version 1.3.0. This version contains minor cosmetic changes to the ABI.
• None. This change is backward-compatible and does not modify behavior unless users explicitly set new environment variables.

New Features

• Windows: Add certificate chain validation to Windows Certificate Store integration.
• Added agent remote-config reset command to reset the remote configuration state which can be useful in troubleshooting issues related to remote configuration.
• GPUM: Add new configuration flag gpu.enabled to agent config to control the GPU Core-Check component, disabled by default.
• Adds Cisco IPsec VPN tunnel metrics and options data collection to SNMP. This can be enabled and disabled using the collect_vpn config.
• Support Cloud Run Jobs in serverless-init.
• The Windows Agent install script (Install-Datadog.ps1) now supports setting host tags and enabling logs collection at install time via the DD_TAGS and DD_LOGS_ENABLED environment variables. This allows users to configure tags and logs without post-install edits to datadog.yaml.

Enhancement Notes

• Add cluster_arn, service_arn, and task_arn tags to the ECS tasks.
• Add container_instance_arn as a host tag when running on ECS EC2.
• Add a script to manage IIS .NET APM instrumentation on Windows
• Add macOS build for AArch64/ARM64.
• Added garbage collection of the tmp directory to the datadog-installer.
• The Datadog autoscaler now can apply vertical scaling recommendations to sidecar init containers.
• Agents are now built with Go 1.24.6.
• Expose additional cgroup memory metrics: active_anon, inactive_anon, active_file, inactive_file, unevictable, shmem, file_mapped, file_dirty, file_writeback, page_tables, refault_anon, refault_file. Metrics are disbaled be default. To enable them please modify conf.d/container.d/conf.yaml and flip extended_memory_metrics to true.
• Add enhanced metrics for task started, ended, and duration for Cloud Run Jobs.
• gpu: Renamed gpu.core.usage and gpu.memory.usage to gpu.process.core/memory.usage
• If a path is specified in the URL in logs_config.logs_dd_url, use it as a prefix.
• Add the datadog.snmp.requests metric, which tracks the number of SNMP requests sent by the Agent to devices.
• Added detection and replacement of tags: blocks in datadog.yaml to ensure only top-level, uncommented, and non-nested tags are modified during install.
• Add thumbprints and serial numbers as tags for certificates and CRLs to the Windows certificate store integration.
• Remote Agent management on Windows now automatically retries when the MSI returns error 1601 (ERROR_INSTALL_SERVICE_FAILURE).
• Remote Agent Management on Windows now automatically retries when an MSI custom action returns error 1601 (ERROR_INSTALL_SERVICE_FAILURE) and causes Error 1719. The Windows Installer Service could not be accessed.

Deprecation Notes

• Remove deprecated serializer options.
  • enable_stream_payload_serialization
  • enable_service_checks_stream_payload_serialization
  • enable_events_stream_payload_serialization
  • enable_sketch_stream_payload_serialization

Bug Fixes

• Fix a panic in Docker file log parsing when received messages are null equivalent.
• When the Autodiscovery scheduler gets blocked, it triggers a health check failure (reported in the agent health command).
• When the prometheus_scrape.enabled option is set to true and the pod has a prometheus.io/port annotation, containers must expose that port in their spec in order for the Agent to schedule openmetrics checks.
• Fixes an issue where pending ECS tasks would temporarily remove the global tags ecs_cluster_name and cluster_arn.
• Fix bug causing integrations configured via docker labels or K8s annotations to be overwritten in the agent so logs from integrations now work when configured using via docker labels or K8s annotations.
• Fix noisy error logs when scrubbing large pod container commands.
• Remote updates on Windows now quotes and escapes MSI property values. This fixes an issue that caused remote updates to fail when the Agent username contains whitespace, for example DDAGENTUSER_NAME=NT AUTHORITY\SYSTEM.
• Remote Agent updates on Windows now handle computer name updates done after the initial install of the Agent. This allows the use of golden images to deploy remotely updatable hosts.
• gpu: Java programs using deeplearning4j are now properly detected and monitored
• gpu: fix container-to-device assignment for Docker containers
• Data Streams Messages: Fix Unscheduling of kafka_consumer integrations that was leading to duplicate integrations running in parallel.
• The Kafka messages feature is not working in some cases when using K8s labels because config is translated to yaml.
• Load check tags onto the orchestrator check configured to be dispatched and run as a cluster check
• APM: Fix issue where the probabilistic sampler would drop all OTLP traffic.
• Kubernetes logs source timestamps are now validated. Invalid timestamps will not be passed.
• The Agent no longer logs an error each time it fails to connect to system-probe through \\.\pipe\dd_system_probe. This log message was added in Agent 7.63.0. The error is still logged if it persists long enough, see [#35790](#35790). Each failed connection attempt still logs the error at debug level to aid troubleshooting.
• Fixed an issue where DD_TAGS and DD_LOGS_ENABLED were silently ignored during install, requiring manual configuration edits.
• Fixed an issue where metrics in the Windows certificate store integration are merged when multiple certificates have the same subject.
• Fixed an issue where the Windows Event Log tailer could lose historical events (the 'amnesia bug') when it started with no prior bookmark and no events were processed immediately. The Agent now correctly seeds an initial bookmark from the most recent event upon startup, ensuring no events are missed on subsequent runs.

Other Notes

• Add handling for dbm-health events in the event platform forwarder. These events are used by Database Monitoring integrations to report on Agent status.
• Add a new metric counter to the Agent telemetry for transactions.errors. This will help to monitor the rate of empty or rejected sketch payloads sent to the backend.

Datadog Cluster Agent

Prelude

Released on: 2025-09-03 Pinned to datadog-agent v7.70.0: CHANGELOG.

Enhancement Notes

• Created a new configuration option orchestrator_explorer.custom_resources.max_count that will allow customers to increase the number of custom resources they can collect in the Cluster Agent.
• Time zones are now collected for CronJobs in the orchestrator check
• Collect EndpointSlices by default.

Bug Fixes

• Correctly set the VerticalPodAutoscaler mode to "Auto" when not provided
• Fixed an issue that caused some Datadog Cluster Agent replicas to restart during the first install.
• When using leases for leader election with the leader_election_default_resource option, the Datadog Cluster Agent no longer creates an unused config map.
• Fix the cronjob.on_schedule_check service check of the kubernetes_state check for CronJobs that have a specific time zone.