Agent

Prelude

Released on: 2026-02-23

• Please refer to the 7.76.0 tag on integrations-core for the list of changes on the Core Checks

Upgrade Notes

• DDOT now submits Fleet Automation metadata through the upstream datadogextension, which is enabled by default. As a result, your DDOT configuration will now appear under the OTel Collector tab. If you configured otelcollector.converter.features, you may need to add the datadog feature to enable Fleet Automation, as DDOT Fleet Automation metadata is no longer submitted through the ddflareextension.

New Features

• Allow users to filter agent check instances using a new --instance-id parameter, which filters by the instance hash found in the agent status.

• Add privateactionrunner binary in Agent artifacts to allow running actions using the Agent, and enable running it on Linux. The binary is disabled by default. To enable it, set privateactionrunner.enabled: true in your configuration file.

• Integration check failures are now automatically reported to the Agent Health Platform component when enabled via health_platform.enabled: true. This provides structured health issue tracking with:

  • Detailed error context including check name, error message, and configuration source
  • Actionable remediation steps for debugging check failures
  • Automatic issue resolution when checks recover
  • Integration with the health platform telemetry and reporting system

  This feature helps users proactively identify and troubleshoot integration issues across their fleet.

• The Agent Profiling check now supports automatic Agent termination after flare generation when memory or CPU thresholds are exceeded. This feature is useful in resource-constrained environments where the Agent needs to be restarted after generating diagnostic information.

  Enable this feature by setting terminate_agent_on_threshold: true in the Agent Profiling check configuration. When enabled, the Agent uses its established shutdown mechanism to trigger graceful shutdown after successfully generating a flare, ensuring proper cleanup before exit.

  Warning: This feature will cause the Agent to exit. This feature is disabled by default and should be used with caution.

• Experimental support the ConfigSync HTTP endpoints over unix sockets with agent_ipc.use_socket: true (defaults to false).

• Implements the flare command for the otel-agent binary. Now you can run otel-agent flare directly in the otel-agent container to get OTel flares.

• Adds system info metadata collection for macOS end-user devices.

• Adds system info metadata collection for Windows end-user devices.

• Added GPU runtime discovery support for ECS EC2 environments. The Datadog Agent can now detect GPU device UUIDs assigned to containers by extracting the NVIDIA_VISIBLE_DEVICES environment variable from the Docker container configuration. This enables GPU-to-container mapping for GPU metrics without requiring the Kubernetes PodResources API, which is not available in ECS environments.

• After falling back to TCP, the Logs Agent periodically retries to establish HTTP and upgrades the connection once HTTP connectivity is available.

• Container logs now include a LogSource tag indicating whether each log message originated from stdout or stderr. This applies to logs parsed via Docker and Kubernetes CRI runtimes.

• Added paging file metrics to the Windows memory check for pagefile.sys usage.

Enhancement Notes

• Add a new global_view_db variable to AWS Autodisovery templates. By default this is the value of the datadoghq.com/global_view_db tag on the instance or cluster.

• Add NotReady endpoint processing to be on par with EndpointSlices processing.

• The agentprofiling check now retries flare generation 2 times with exponential backoff (1 minute after first failure, 5 minutes after second failure) when flare creation or sending fails. This improves reliability when encountering transient failures during flare generation.

• Adds a kubernetes_kube_service_new_behavior flag (default false) to alter kube_service tag behavior. If the flag is set to true, kube_service tag is attached unconditionally. Previously, the tag was only attached when the Kubernetes service has the status Ready.

• APM: Add custom protobuf encoder for trace writer v1 with string compaction to reduce payload size.

• Extended the autodiscovery secret resolver to support refreshing secrets.

• Agents are now built with Go 1.25.7.

• The datadog-installer setup command now prints human-readable errors instead of mixing JSON and text.

• Added GPUDeviceIDs field to the workloadmeta Container entity to store GPU device UUIDs. This field is populated by the Docker collector in ECS environments from the NVIDIA_VISIBLE_DEVICES environment variable (e.g., GPU-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx).

• The GPU collector now uses GPUDeviceIDs from workloadmeta as the primary source for GPU-to-container mapping in ECS, with fallback to procfs for regular Docker environments and PodResources API for Kubernetes.

• GPU: add new tag gpu_type to the GPU metrics to identify the type of GPU (e.g., a100, h100).

• Improve eBPF conntracker support by using alternate probes when the primary probe is unavailable, enabling compatibility with GKE Autopilot and other environments running Google COS.

• The logs.dropped metric now tracks dropped logs for both TCP and HTTP log transports. Previously, this metric was only available when using TCP transport. Customers can now monitor dropped logs with a single unified metric regardless of which transport protocol is configured, making it easier to detect and troubleshoot log delivery issues.

• The logs agent now supports using start_position: beginning and start_position: forceBeginning with wildcard file paths. Previously, configurations like path: /var/log/*.log with start_position: beginning would fail validation. The agent's fingerprinting system when enabled prevents duplicate log reads during file rotation, making this combination safe to use.

• Site config URLs are now lowercased for consistent handling.

• APM: Add tags databricks_job_id, databricks_job_run_id, databricks_task_run_id, config.spark_app_startTime, config.spark_databricks_job_parentRunId to the default list of tags that are known to not be credit card numbers so they are skipped by the credit card obfuscator.

• Add option to switch on/off Infra-Attribute-Processor for traces in the OTLP ingest pipeline.
  otlp_config:
  traces:
  infra_attributes:
  enabled: false

  These settings can be configured in the Agent config file or by using the environment variables.

• The Datadog Agent now collects AWS Spot preemption events (requires IMDS access) as Datadog events.

• Added network_config.dns_monitoring_ports, which is a list of DNS ports Cloud Network Monitoring will use to monitor DNS traffic on.

• Automatically tag, but don't aggregate, multiline logs. Logs are tagged with the number of other logs they could potentially be aggregated with.

• Update the histogram helpers API in the pkg/opentelemetry-mapping-go/otlp/metrics package. The API now accepts accept pointers to the OTLP data points, and returns blank DDSketches when the pointer is nil.

• Update image resolution attempt telemetry to include the tag specified in the configuration, and remove the registry and digest_resolution tags.

• Windows: Add a new flare artifact agent_loaded_modules.json listing loaded DLLs with metadata (full path, timestamp, size, perms) and version info (CompanyName, ProductName, OriginalFilename, FileVersion, ProductVersion, InternalName). Keeps <flavor>_open_files.txt for compatibility.

Deprecation Notes

• The command agent diagnose show-metadata inventory-otel has been removed. To display DDOT metadata, you can query the datadog extension endpoint: http://localhost:9875/metadata.

Bug Fixes

• Properly scrub sensitive information from Kubernetes pod specifications in agent flares. Environment variables with sensitive names are now redacted.
• Fixed a bug where long Kubernetes event bundles were being truncated by dogweb.
• APM: Fix a bug where the Agent would log a warning when the DD_APM_MODE environment variable was unset.
• Properly parse the image_tag tag when defining a container spec that uses both an image tag and a digest like nginx:1.23@sha256:xxx.
• Updates tag enrichment logic to retry on failed tag resolution attempts. This regression was introduced in #41587 on Agent v7.73+. Impacts origin detection on cgroup v2 runtimes with DogStatsD, which led to tags not being enriched, even if origin detection was possible by using other methods like container ID from socket or ExternalData.
• Fixed a regression in the Go-native disk check (diskv2) where a failure in IO counter collection (e.g. ERROR_INVALID_FUNCTION from DeviceIoControl on Windows Server 2016) caused all disk metrics to be discarded, including successfully collected partition/usage metrics such as system.disk.total, system.disk.used, and system.disk.free. IO counter collection is now best-effort: known errors such as ERROR_INVALID_FUNCTION are logged at debug level, while unexpected errors are logged as warnings. Neither prevent partition metrics from being reported.
• Fleet installer: ensure the DD_LOGS_ENABLED environment variable is honored again when running setup scripts, so Windows installs using the new installer flow properly. Sets logs_enabled in datadog.yaml.
• Fixes a bug introduced in 7.73.0 that can cause a remote Agent update through Fleet Automation to fail to restore the previous version if the MSI fails and the C:\Windows\SystemTemp\datadog-installer\rollback\InstallOciPackages.json file is present.
• Fix Flux API groups, split fluxcd.io into source.toolkit.fluxcd.io and kustomize.toolkit.fluxcd.io.
• Fixes repetitive 'Could not make file tailer' warning logs when short lived pods are terminated and the Agent attempts to create a file tailer for the deleted containers in a pod. Now the Agent will not create container services for pods that have been deleted and no-longer have containers to tail.
• GPU: MIG devices and parents are now reporting correct core and memory limits.
• GPUm: fix gpu.memory.limit being duplicated in Hopper devices
• Fixed the logs.sent metric for the HTTP log transport to no longer increment when logs are dropped due to non-retryable errors. This ensures more accurate reporting of successfully delivered logs.
• Fix WLAN check failure on macOS systems.
• Fix datadog.agent.check_ready to always include the check_name tag value for Python checks.
• Rename kubernetes_kube_service_new_behavior to kubernetes_kube_service_ignore_readiness to better reflect the behavior.
• Prevent a deadlock from occurring in the otel-agent when its internal telemetry Prometheus endpoint is scraped.
• [oracle] Updates the oracle.d/conf.yaml.example file to include all supported sql obfuscator options. [DBM] Bump go-sqllexer to v0.1.12:
  • Fixes a normalization bug for Oracle queries with positional bind parameters.
  • Fixes a memory leak in the go-sqllexer package.

Other Notes

• Add metrics origins for battery integration.
• Remove procps-ng and associated tools from Agent packages.

Datadog Cluster Agent

Prelude

Released on: 2026-02-23 Pinned to datadog-agent v7.76.0: CHANGELOG.

New Features

• APM: Add apm_config.instrumentation.injection_mode configuration option to control APM library injection method. Possible values are auto (default), init_container, and csi. The auto mode automatically selects the best injection mode (currently uses init containers). The init_container mode is the legacy method that copies APM libraries into pods using init containers. The csi mode mounts APM libraries directly into pods using the Datadog CSI driver. It is experimental and requires Cluster Agent 7.76+ and the Datadog CSI driver.
• APM: Add CSI-based library injection as an alternative to init containers (experimental). This provides faster pod startup and reduced storage overhead.
• Reduced memory usage of compliance checks on large clusters

Enhancement Notes

• Reduced memory usage when pod collection is enabled in the Cluster Agent.

Bug Fixes

• When injection fails for Single Step Instrumentation due to constrained resources, we add an annotation to the pod with a reason for the error. This annotation now matches all other annotations the webhook writes to a pod spec by prefixing the annotation with internal. The full annotation is now: internal.apm.datadoghq.com/injection-error

Other Notes

• Refactor the auto-instrumentation webhook's injectTracers function to use a modular, explicit mutation pattern. This improves code readability and maintainability. Edge case behavior may differ slightly, but overall functionality remains unchanged.