7.41.0

Agent

Prelude

Release on: 2022-12-12

• Please refer to the 7.41.0 tag on integrations-core for the list of changes on the Core Checks

Upgrade Notes

• Troubleshooting commands in the Agent CLI have been moved to the diagnose command. troubleshooting metadata_v5 command is now diagnose show-metadata v5 and troubleshooting metadata_inventory is diagnose show-metadata inventory.
• Journald launcher can now create multiple tailers on the same journal when config_id is specified. This change enables multiple configs to operate on the same journal which is useful for tagging different units. Note: This may have an impact on CPU usage.
• Upgrade tracer_agent debugger proxy to use logs intake API v2 for uploading snapshots
• The Agent now defaults to TLS 1.2 instead of TLS 1.0. The force_tls_12 configuration parameter has been removed since it's now the default behavior. To continue using TLS 1.0 or 1.1, you must set the min_tls_version configuration parameter to either tlsv1.0 or tlsv1.1.

New Features

• Added a required infrastructure to enable protocol classification for Network Performance Monitoring in the future. The protocol classification will allow us to label each connection with a L7 protocol. The features requires Linux kernel version 4.5 or greater.
• parse the snmp configuration from the agent and pass it to the integrated snmpwalk command in case the customer only provides an ip address
• The Agent can send its own configuration to Datadog to be displayed in the Agent Configuration section of the host detail panel. See https://docs.datadoghq.com/infrastructure/list/#agent-configuration for more information. The Agent configuration is scrubbed of any sensitive information and only contains configuration you’ve set using the configuration file or environment variables.
• Windows: Adds support for Windows Docker "Process Isolation" containers running on a Windows host.

Enhancement Notes

• APM: All spans can be sent through the error and rare samplers via custom feature flag error_rare_sample_tracer_drop. This can be useful if you want to run those samplers against traces that were not sampled by custom tracer sample rules. Note that even user manual drop spans may be kept if this feature flag is set.
• APM: The trace-agent will log failures to lookup CPU usage at error level instead of debug.
• Optionally poll Agent and Cluster Agent integration configuration files for changes after startup. This allows the Agent/Cluster Agent to pick up new integration configuration without a restart. This is enabled/disabled with the autoconf_config_files_poll boolean configuration variable. The polling interval is configured with the autoconf_config_files_poll_interval (default 60s). Note: Dynamic removal of logs configuration is currently not supported.
• Added telemetry for the "container-lifecycle" check.
• On Kubernetes, the "cluster name" can now be discovered by using the Node label ad.datadoghq.com/cluster-name or any other label key configured using to the configuration option: kubernetes_node_label_as_cluster_name
• Agents are now built with Go 1.18.8.
• Go PDH checks now all use the PdhAddEnglishCounter API to ensure proper localization support.
• Use the windows_counter_refresh_interval configuration option to limit how frequently the PDH object cache can be refreshed during counter initialization in golang. This replaces the previously hardcoded limit of 60 seconds.
• [netflow] Add disable port rollup config.
• The OTLP ingest endpoint now supports the same settings and protocol as the OpenTelemetry Collector OTLP receiver v0.61.0.
• The disable_file_logging setting is now respected in the process-agent.
• The process-agent check [check-name] command no longer outputs to the configured log file to reduce noise in the log file.
• Logs a warning when the process agent cannot read other processes due to misconfiguration.
• DogStatsD caches metric metadata for shorter periods of time, reducing memory usage when tags or metrics received are different across subsequent aggregation intervals.
• The agent CLI subcommands related to Windows services are now consistent in use of dashes in the command names (install-service, start-service, and so on). The names without dashes are supported as aliases.
• The Agent now uses the V2 API to submit series data to the Datadog intake by default. This can be reverted by setting use_v2_api.series to false.

Deprecation Notes

• APM: The Rare Sampler is now disabled by default. If you wish to enable it explicitly you can set apm_config.enable_rare_sampler or DD_APM_ENABLE_RARE_SAMPLER to true.

Bug Fixes

• APM: Don't include extra empty 'env' entries in sampling priority output shown by agent status command.

• APM: Fix panic when DD_PROMETHEUS_SCRAPE_CHECKS is set.

• APM: DogStatsD data can now be proxied through the "/dogstatsd/v1/proxy" endpoint and the new "/dogstatsd/v2/proxy" endpoint over UDS, with multiple payloads separated by newlines in a single request body. See https://docs.datadoghq.com/developers/dogstatsd#setup for configuration details.

• APM - remove extra error message from logs.

• Fixes an issue where cluster check metrics would be sometimes sent with the host tags.

• The containerd check no longer emits events related with pause containers when exclude_pause_container is set to true.

• Discard aberrant values (close to 18 EiB) in the container.memory.rss metric.

• Fix Cloud Foundry CAPI Metadata tags injection into application containers.

• Fix Trace Agent's CPU stats by reading correct PID in procfs

• Fix a potential panic when df outputs warnings or errors among its standard output.

• The OTLP ingest is now consistent with the Datadog exporter (v0.56+) when getting a hostname from OTLP resource attributes for metrics and traces.

• Make Agent write logs when SNMP trap listener starts and Agent receives invalid packets.

• Fixed a bug in the workloadmeta store. Subscribers that asked to receive only unset events mistakenly got set events on the first subscription for all the entities present in the store. This only affects the container_lifecycle check.

• Fix missing tags on the kubernetes_state.cronjob.complete service check.

• In kubernetes_state_core check, fix the labels_as_tags feature when the same Kubernetes label must be turned into different Datadog tags, depending on the resource:

  labels_as_tags:
  daemonset:
  first_owner: kube_daemonset_label_first_owner

  deployment:
  first_owner: kube_deployment_label_first_owner

• Normalize the EventID field in the output from the windowsevent log tailer. The type will now always be a string containing the event ID, the sometimes present qualifier value is retained in a new EventIDQualifier field.

• Fix an issue where the security agent would panic, sending on a close channel, if it received a signal when shutting down while all components were disabled.

• Fix tokenization of negative numeric values in the SQL obfuscator to remove extra characters prepended to the byte array.

Datadog Cluster Agent

New Features

• Add Namespace collection in the orchestrator check and enable it by default.

Enhancement Notes

• Improves performance of the Cluster Agent admission controller on large pods.

7.40.1

Release Notes

7.40.1

Prelude

Release on: 2022-11-09

• Please refer to the 7.40.1 tag on integrations-core for the list of changes on the Core Checks

Enhancement Notes

• Agents are now built with Go 1.18.8.

Bug Fixes

• Fix log collection on Kubernetes distributions using cri-o like OpenShift, which began failing in 7.40.0.

7.40.0

Agent

Prelude

Release on: 2022-11-02

• Please refer to the 7.40.0 tag on integrations-core for the list of changes on the Core Checks

Upgrade Notes

• Starting Agent 7.40, the Agent will fail to start when unable to determine hostname instead of silently using unrelevant hostname (usually, a container id). Hostname resolution is key to many features and failure to determine hostname means that the Agent is not configured properly. This change mostly affects Agents running in containerized environments as we cannot rely on OS hostname.

• Universal Service Monitoring now requires a Linux kernel version of 4.14 or greater.

New Features

• The Agent RPM package now supports Amazon Linux 2022 and Fedora 30+ without requiring the installation of the additional libxcrypt-compat system package.
• Add support for CAPI metadata and DCA tags collection in PCF containers.
• Add a username and password dialog window to the Windows Installer
• APM: DogStatsD data can now be proxied through the "/dogstatsd/v1/proxy" endpoint over UDP. See https://docs.datadoghq.com/developers/dogstatsd#setup for configuration details.
• Cloud Workload Security now has Agent version constraints for Macros in SECL expressions.
• Added the "helm_values_as_tags" configuration option in the Helm check. It allows users to collect helm values from a Helm release and use them as tags to attach to the metrics and events emitted by the Helm check.
• Enable the new DogStatsD no-aggregation pipeline, capable of processing metrics with timestamps. Set dogstatsd_no_aggregation_pipeline to false to disable it.
• Adds ability to identify the interpreter of a script inside a script via the shebang. Example rule would be exec.interpreter.file.name == ~"python*". This feature is currently limited to one layer of nested script. For example, a python script in a shell script will be caught, but a perl script inside a python script inside a shell script will not be caught.

Enhancement Notes

• JMXFetch now supports ZGC Cycles and ZGC Pauses beans support out of the box.

• Adds new aws.lambda.enhanced.post_runtime_duration metric for AWS Lambda functions. This gauge metric measures the elapsed milliseconds from when the function returns the response to when the extensions finishes. This includes performing activities like sending telemetry data to a preferred destination after the function's response is returned. Note that aws.lambda.enhanced.duration is equivalent to the sum of aws.lambda.enhanced.runtime_duration and aws.lambda.enhanced.post_runtime_duration.

• Add the flare command to the Cloud Foundry cluster agent to improve support experience.

• Add CreateContainerError and InvalidImageName to waiting reasons for kubernetes_state.container.status_report.count.waiting in the Kubernetes State Core check.

• [netflow] Ephemeral Port Rollup

• APM: A warning is now logged when the agent is under heavy load.

• APM: The "http.status_code" tag is now supported as a numeric value too when computing APM trace stats. If set as both a string and a numeric value, the numeric value takes precedence and the string value is ignored.

• APM: Add support for cgroup2 via UDS.

• A new config option, logs_config.file_wildcard_selection_mode, allows you to configure how log wildcard file matches are prioritized if the number of matches exceeds logs_config.open_files_limit.

  The option defaults to by_name which is the previous behavior. The new option is by_modification_time which prioritizes more recently modified files, but using it can result in slower performance compared to using by_name.

• Agents are now built with Go 1.18.7. This version of Go brings changes to the garbage collection runtime that may change the Agent's memory usage. In internal testing, the RSS of Agent processes showed a minor increase of a few MiB, while CPU usage remained consistent. Reducing the value of GOGC as described in the Go documentation was effective in reducing the memory usage at a modest cost in CPU usage.

• KSM Core check: Add the helm_chart tag automatically from the standard helm label helm.sh/chart.

• Helm check: Add a helm_chart tag, equivalent to the standard helm label helm.sh/chart (see https://helm.sh/docs/chart_best_practices/labels/).

• The OTLP ingest endpoint now supports the same settings and protocol as the OpenTelemetry Collector OTLP receiver v0.60.0. In particular, this drops support for consuming OTLP/JSON v0.15.0 or below payloads.

• Improve CCCache performance on cache miss, significantly reducing the number of API calls to the CAPI.

• Add more flags to increase control over the CCCache, such as refresh_on_cache_miss, sidecars_tags, and isolation_segments_tags flags under cluster_agent properties.

• Windows: Add a config option to control how often the agent refreshes performance counters.

• Introduces an unbundle_events config to the docker integration. When set to true, Docker events are no longer bundled together by image name, and instead generate separate Datadog events.

• Introduces an unbundle_events config to the kubernetes_apiserver integration. When set to true, Kubernetes events are no longer bundled together by InvolvedObject, and instead generate separate Datadog events.

• On Windows the Agent now uses high-resolution icon where possible. The smaller resolution icons have been resampled for better visibility.

Known Issues

• APM: OTLP Ingest: resource attributes such as service.name are correctly picked up by spans.
• APM: The "/dogstatsd/v1/proxy" endpoint can only accept a single payload at a time. This will be fixed in the v2 endpoint which will split payloads by newline.

Deprecation Notes

• The following Windows Agent container versions are removed: 1909, 2004, and 20H2.

Bug Fixes

• Add the device field to the MetricPayload to ensure the device tag is properly handled by the backend.
• APM: Revised support for tracer single span sampling. See datadog-agent/pull/13461.
• Fixed a problem that could trigger in the containerd collector when fetching containers from multiple namespaces.
• Fixed a crash when dogstatsd_metrics_stats_enable is true
• Fix a bug in Autodiscovery preventing the Agent to correctly schedule checks or logs configurations on newly created PODs during a StatefulSet rollout.
• The included aerospike Python package is now correctly built against the embedded OpenSSL and thus the Aerospike integration can be successfully used on RHEL/CentOS.
• Fix configresolver to continue parsing when a null value is found.
• Fixed issue with CPU count on MacOS
• The container CPU limit that is reported by docker and container checks on ECS was not defaulting to the task limit when no CPU limit is set at container level.
• Fix potential panic when removing a service that the log agent is currently tailing.
• On SUSE, fixes the permissions declared in the package list of the RPM package. This was causing package conflicts between the datadog-agent package and other packages with files in /usr/lib/systemd/system.
• Fixed a resource leak in the helm check.
• Fix golang performance counter initialization errors when counters are not available during agent/check init time. Checks now retry the counter initilization on each interval.
• [snmp] Cache snmp dynamic tags from devices

Other Notes

• JMXFetch upgraded to 0.47.1 https://github.com/DataDog/jmxfetch/releases/0.47.1
• The logs_config.cca_in_ad feature flag now defaults to true. This selects updated codepaths in Autodiscovery and the Logs Agent. No behavior change is expected. Please report any behavior that is "fixed" by setting this flag to false.

Datadog Cluster Agent

New Features

• Experimental: The Datadog Admission Controller can inject the Python APM library into Kubernetes containers for auto-instrumentation.
• The orchestrator check is now able to discover resources to collect based on API groups available in the Kubernetes cluster.

Enhancement Notes

• The admission controller now injects variables and volume mounts to init containers in addition to regular containers.
• Chunk orchestrator payloads by size and weight
• KSM Core check: Add the helm_chart tag automatically from the standard helm label helm.sh/chart.
• Helm check: Add a helm_chart tag, equivalent to the standard helm label helm.sh/chart (see https://helm.sh/docs/chart_best_practices/labels/).

Bug Fixes

• Fixed an edge case in the Admission Controller when mutateUnlabelled is enabled and configMode is set to socket. This combination could prevent the creation of new DaemonSet Agent pods.
• Fixed a resource leak in the helm check.

7.39.2

Prelude

Release on: 2022-10-12

Bug Fixes

• Cloud Workload Security uses a default Traffic Control filter priority of 10.
• Traffic Control filters priority and handle can be set from the configuration of Cloud Workload Security in system-probe.yaml.

7.39.1

Prelude

Release on: 2022-09-27

Security Notes

• Bump github.com/open-policy-agent/opa to v0.43.1 to patch CVE-2022-36085.

Other Notes

• Bump embedded Python3 to 3.8.14.
• Deactivated support of HTTP/2 in all non localhost endpoint used by Datadog Agent and Cluster Agent. (except endpoints)

7.39.0

Agent

Prelude

Release on: 2022-09-12

• Please refer to the 7.39.0 tag on integrations-core for the list of changes on the Core Checks

Upgrade Notes

• Starting with version 6.39.0, Agent 6 is no longer built for macOS. Only Agent 7 will be built for macOS going forward. macOS 10.14 and above are supported with Agent 7.39.0.

New Features

• Add an integrated snmpwalk command to perform a walk for all snmp versions based on the gosnmp library.
• APM: Add two options under the vector config prefix to send traces to Vector instead of Datadog. Set vector.traces.enabled to true. Set vector.traces.url to point to a Vector endpoint. This overrides the main endpoint. Additional endpoints remains fully functional.

Enhancement Notes

• Add the tagger-list command to the process-agent to ease tagging issue investigation.
• Update SNMP traps database with bit enumerations.
• Resolve SNMP trap variables with bit enumerations to their string representation.
• Logs: Support filtering on arbitrary journal log fields
• APM: The trace-agent version string has been made more consistent and is now available in different build environments.
• Delay starting the auto multi-line detection timeout until at least one log has been processed.
• The helm check has new configuration parameters:
  • extra_sync_timeout_seconds (default 120)
  • informers_resync_interval_minutes (default 10)
• Improves the labelsAsTags feature of the Kubernetes State Metrics core check by performing the transformations of characters ['/' , '-' , '.'] to underscores ['_'] within the Datadog agent. Previously users had to perform these conversions manually in order to discover the labels on their resources.
• The new min_tls_version configuration parameter allows configuration of the minimum TLS version used for connections to the Datadog intake. This replaces the force_tls_12 configuration parameter which only allowed the minimum to be set to tlsv1.2.
• The OTLP ingest endpoint now supports the same settings and protocol as the OpenTelemetry Collector OTLP receiver v0.56.0
• 'agent status' command output is now parseable as JSON directly from stdout. Before this change, the logger front-matter made it hard to parse 'status' output directly as JSON.
• Raise the default logs_config.open_files_limit to 200 on Windows and macOS. Raised to 500 for all other operating systems.
• Support disabling DatadogMetric autogeneration with the external_metrics_provider.enable_datadogmetric_autogen configuration option (enabled by default).

Deprecation Notes

• APM: The datadog.trace_agent.trace_writer.bytes_estimated metric has been removed. It was meant to be a metric used for debugging, without any user added value.
• APM: The trace-agent /info endpoint no longer reports "build_date".
• The force_tls_12 configuration parameter is deprecated, replaced by min_tls_version. If min_tls_version is not given, but force_tls_12 is true, then min_tls_version defaults to tlsv1.2.

Bug Fixes

• Traps variable OIDs that had the index as a suffix are now correctly resolved.
• Agent status command should always log at info level to allow full status output regardless of Agent log level settings.
• APM: The "datadog.trace_agent.otlp.spans" metric was incorrectly reporting span count. This release fixes that.
• Fix panic when Agent stops jmxfetch.
• Fixed a bug in Kubernetes Autodiscovery based on pod annotations: The Agent no longer skips valid configurations if other invalid configurations exist. Note: This regression was introduced in Agents 7.36.0 and 6.36.0
• Fix a bug in autodiscovery that would not unschedule some checks when check configuration contains secrets.
• Orchestrator check: make sure we don't return labels and annotations with a suffixed :
• Fixed a bug in the Docker check that affects the docker.containers.running metric. It was reporting wrong values in cases where multiple containers with different env, service, version, etc. tags were using the same image.
• Fixed a deadlock in the DogStatsD when running the capture (agent dogstatsd-capture). The Agent now flushes the captured messages properly when the capture stops.
• Fix parsing of init_config in AD annotations v2.
• The internal_profiling.period parameter is now taken into account by the agent.
• Fix duplicated check or logs configurations, targeting dead containers when containers are re-created by Docker Compose.
• Fix concurrent map access issues when using OTLP ingest.
• [orchestrator check] Fixes race condition during check startup.
• The Windows installer will now respect the DDAGENTUSER_PASSWORD option and update the services passwords when the user already exists.
• The KSM Core check now handles cron job schedules with time zones.
• The v5 metadata payload's filesystem information is now more robust against failures in the df command, such as when a mountpoint is stuck.
• Fixes a disk check issue in the Docker Agent where a disproportionate amount of automount request system logs would be produced by the host after each disk check run.
• [epforwarder] Update NetFlow EP forwarder default configs
• The Agent starts faster on a Windows Docker host with many containers running by fetching the containers in parallel.
• On Windows, NPM driver adds support for Receive Segment Coalescing. This works around a Windows bug which in some situations causes system probe to hang on startup

Cluster Agent

New Features

• Experimental: The Datadog Admission Controller can inject the Node and Java APM libraries into Kubernetes containers for auto-instrumentation.

Enhancement Notes

• When injecting env vars with the admission controller, env vars are now prepended instead of appended, meaning that Kubernetes [dependent environment variables](https://kubernetes.io/docs/tasks/inject-data-application/define-interdependent-environment-variables/) can now depend on these injected vars.
• The helm check has new configuration parameters:
  • extra_sync_timeout_seconds (default 120)
  • informers_resync_interval_minutes (default 10)
• Improves the labelsAsTags feature of the Kubernetes State Metrics core check by performing the transformations of characters ['/' , '-' , '.'] to underscores ['_'] within the Datadog agent. Previously users had to perform these conversions manually in order to discover the labels on their resources.

Bug Fixes

• Fix the DCA leader_election_is_leader metric that could sometimes report is_leader="false" on the leader instance
• Fixed an error when running datadog-cluster-agent status with DD_EXTERNAL_METRICS_PROVIDER_ENABLED=true and no app key set.
• The KSM Core check now handles cron job schedules with time zones.

Other Notes

• Align Cluster Agent version to Agent version. Cluster Agent will now be released with 7.x.y tags

7.38.2

Prelude

Release on: 2022-08-10

• Please refer to the 7.38.2 tag on integrations-core for the list of changes on the Core Checks

Bug Fixes

• Fixes a bug making the agent creating a lot of zombie (defunct) processes. This bug happened only with the docker images 7.38.x when the containerized agent was launched without hostPID: true.

7.38.1

Prelude

Release on: 2022-08-02

Bug Fixes

• Fixes CWS rules with 'process.file.name !=""' expression.

Datadog Cluster Agent 1.22.0

Prelude

Released on: 2022-07-26
Pinned to datadog-agent v7.38.0: CHANGELOG

New Features

• Enable collection of Ingresses by default in the orchestrator check.

7.38.0

Prelude

Release on: 2022-07-25

• Please refer to the 7.38.0 tag on integrations-core for the list of changes on the Core Checks

New Features

• Add NetFlow feature to listen to NetFlow traffic and forward them to Datadog.
• The CWS agent now supports filtering events depending on whether they are performed by a thread. A process is considered a thread if it's a child process that hasn't executed another program.
• Adds a diagnose datadog-connectivity command that displays information about connectivity issues between the Agent and Datadog intake.
• Adds support for tailing modes in the journald logs tailer.
• The CWS agent now supports writing rules on processes termination.
• Add support for new types of CI Visibility payloads to the Trace Agent, so features that until now were Agentless-only are available as well when using the Agent.

Enhancement Notes

• Tags configured with DD_TAGS or DD_EXTRA_TAGS in an EKS Fargate environment are now attached to OTLP metrics.
• Add NetFlow static enrichments (TCP flags, IP Protocol, EtherType, and more).
• Report lines matched by auto multiline detection as metrics and show on the status page.
• Add a containerd_exclude_namespaces configuration option for the Agent to ignore containers from specific containerd namespaces.
• The log_level of the agent is now appended to the flare archive name upon its creation.
• The metrics reported by KSM core now include the tags "kube_app_name", "kube_app_instance", and so on, if they're related to a Kubernetes entity that has a standard label like "app.kubernetes.io/name", "app.kubernetes.io/instance", etc.
• The Kubernetes State Metrics Core check now collects two ingress metrics: kubernetes_state.ingress.count and kubernetes_state.ingress.path.
• Move process chunking code to util package to avoid cycle import when using it in orchestrator check.
• APM: Add support for PostgreSQL JSON operators in the SQL obfuscate package.
• The OTLP ingest endpoint now supports the same settings and protocol as the OpenTelemetry Collector OTLP receiver v0.54.0 (OTLP v0.18.0).
• The Agent now embeds Python-3.8.13, an upgrade from Python-3.8.11.
• APM: Updated Rare Sampler default configuration values to sample traces more uniformly across environments and services.
• The OTLP ingest endpoint now supports Exponential Histograms with delta aggregation temporality.
• The Windows installer now supports grouped Managed Service Accounts.
• Enable https monitoring on arm64 with kernel >= 5.5.0.
• Add otlp_config.debug.loglevel to determine log level when the OTLP Agent receives metrics/traces for debugging use cases.

Deprecation Notes

• Deprecateotlp_config.metrics.instrumentation_library_metadata_as_tags in in favor of otlp_config.metrics.instrumentation_scope_metadata_as_tags.

Bug Fixes

• When enable_payloads.series or enable_payloads.sketches are set to false, don't log the error Cannot append a metric in a closed buffered channel.
• Restrict permissions for the entrypoint executables of the Dockerfiles.
• Revert docker.mem.in_use calculation to use RSS Memory instead of total memory.
• Add missing telemetry metrics for HTTP log bytes sent.
• Fix panic in container, containerd, and docker when container stats are temporarily not available
• Fix prometheus check Metrics parsing by not enforcing a list of strings.
• Fix potential deadlock when shutting down an Agent with a log TCP listener.
• APM: Fixed trace rare sampler's oversampling behavior. With this fix, the rare sampler will sample rare traces more accurately.
• Fix journald byte count on the status page.
• APM: Fixes an issue where certain (#> and #>>) PostgreSQL JSON operators were being interpreted as comments and removed by the obfuscate package.
• Scrubs HTTP Bearer tokens out of log output
• Fixed the triggered "svType != tvType; key=containerd_namespace, st=[]interface {}, tt=[]string, sv=[], tv=[]" error when using a secret backend reader.
• Fixed an issue that made the container check to show an error in the "agent status" output when it was working properly but there were no containers deployed.