7.32.0

Prelude

Release on: 2021-11-09

• Please refer to the 7.32.0 tag on integrations-core for the list of changes on the Core Checks

Upgrade Notes

• APM: Change default profiling intake to use v2 endpoint.
• CSPM the check subcommand is now part of the security-agent
  compliance.

New Features

• On Kubernetes, add a kube_priority_class tag on metrics coming from pods with a priority class.
• Priority class name of pods are now collected and sent to the orchestration endpoint
• Autodiscovery can now resolve template variables and environment variables in log configurations.
• The Windows installer now offers US5 as a new site choice.
• APM: New telemetry was added to measure /v.*/traces endpoints latency and response size. These metrics are datadog.trace_agent.receiver.{rate_response_bytes,serve_traces_ms}.
• APM: Metrics are now available for Windows Pipes and UDS connections via datadog.trace_agent.receiver {uds_connections,pipe_connections}.
• Introduce a new configuration parameter container_env_as_tags to allow converting containerd containers' environment variables into tags.
• The "containerd" check is now supported on Windows.
• Add experimental support for writing agent-side CSPM compliance checks in Rego.
• Runtime security can now attach span/trace to event.
• Provides alternative implementation for process collection on Windows using performance counters.
• Add multi-line auto-sensing when tailing logs from file. It checks the 1000 first lines (or waits 30 seconds, whichever is first) when tailing for a list of known timestamp formats. If the number of matched lines is greater than the threshold it switches to the MultiLineHandler with the pattern matching the timestamp format. The pattern chosen is saved in the log config and is reused if the file rotates. Use the new global config parameter logs_config.auto_multi_line_detection to enable the feature for the whole agent, or the per log integration config parameter auto_multi_line_detection to enable the feature on a case by case basis.
• Added experimental support for OTLP metrics via experimental.otlp.{http_port,grpc_port} or their corresponding environment variables (DD_OTLP{HTTP,GRPC}_PORT).
• Created a new process discovery check. This is a lightweight check that runs every 4 hours by default, and collects process metadata, so that Datadog can suggest potential integrations for the user to enable.
• Added new executable readsecret_multiple_providers.sh that allows the agent to read secrets both from files and Kubernetes secrets. Please refer to the docs for more details.

Enhancement Notes

• KSM core check has a new labels_as_tags parameter to configure which pod labels should be used as datadog tag in an easier way than with the label_joins parameter.
• Add namespace to snmp listener config
• Remove network_devices from datadog.yaml configuration
• kubernetes state core check: add kubernetes_state.job.completion.succeeded and kubernetes_state.job.completion.failed metrics to report job completion as metrics in addition to the already existing service check.
• Add use_device_id_as_hostname in snmp check and snmp_listener configuration to use DeviceId as hostname for metrics and service checks
• APM: The maximum allowed tag value length has been increased to 25,000 bytes.
• Reduce memory usage when checks report new metrics every run. Most metrics are removed after two check runs without new samples. Rate, historate and monotonic count will be kept in memory for additional 25 hours after that. Number of check runs and the additional time can be changed with check_sampler_bucket_commits_count_expiry and check_sampler_stateful_metric_expiration_time. Metric expiration can be disabled entirely by setting check_sampler_expire_metrics to false.
• CSPM reports the agent version as part of the events
• Agents are now built with Go1.16. This will have one user-visible change: on Linux, the process-level RSS metric for agent processes will be reduced from earlier versions. This reflects a change in how memory usage is calculated, not a reduction in used memory, and is an artifact of the Go runtime switching from MADV_FREE to MADV_DONTNEED.
• Tag Kubernetes containers with image_id tag.
• Eliminates the need to synchronize state between regular and RT process collection.
• APM: Added a configuration option to set the API key separately for Live Debugger. It can be set via apm_config.debugger_api_key or DD_APM_DEBUGGER_API_KEY.
• Update EP forwarder config to use intake v2 for ndm metadata
• Remove the reason tag from the kubernetes_state.job.failed metric to reduce cardinality
• The runtime security module of system-probe is now powered by DataDog/ebpf-manager instead of DataDog/ebpf.
• Security Agent: use exponential backoff for log warning when the security agent fails to connect to the system probe.
• APM: OTLP traces now supports semantic conventions from version 1.5.0 of the OpenTelemetry specification.
• Show enabled autodiscovery sources in the agent status
• Add namespace to SNMP integration and SNMP Listener to disambiguate devices with same IP.
• Add snmp corecheck autodiscovery
• Enable SNMP device metadata collection by default
• Reduced CPU usage when origin detection is used.
• The Windows installer now prioritizes user name from the command line over stored registry entries

Bug Fixes

• Make sure DD_ENABLE_METADATA_COLLECTION="false" prevent all host metadata emission, including the initial one.
• Most checks are stripping tags with an empty value. KSM was missing this logic so that KSM specific metrics could have a tag with an empty value. They will now be stripped like for any other check.
• Fixed a regression that was preventing the Agent from retrying kubelet and docker connections in case of failure.
• Fix the cgroup collector to correctly pickup Cloud Foundry containers.
• Fix an issue where the orchestrator check would stop sending updates when run on as a cluster-check.
• Port python-tuf CVE fix on the embedded Python 2 see GHSA-wjw6-2cqr-j4qr.
• Fix some string logging in the Windows installer.
• The flare command now correctly copies agent logs located in subdirectories of the agent's root log directory.
• Kubernetes state core check: job.status.succeeded and job.status.failed gauges were not sent when equal 0. 0 values are now sent.
• Tag Namespace and PV and PVC metrics correctly with phase instead of pod_phase in the Kube State Metrics Core check.