Skip to content

Latest commit

 

History

History
3461 lines (2114 loc) · 123 KB

CHANGELOG.rst

File metadata and controls

3461 lines (2114 loc) · 123 KB
Raw

Release Notes

6.14.1

Prelude

Release on: 2019-09-26

Bug Fixes

  • Disable debug log lines for the 'hostname' command since it's directly called by some Agent components. Fixes hostname resolution issues for APM and Live Process.

6.14.0

Prelude

Release on: 2019-09-16

Upgrade Notes

  • The GPG key used to sign the Agent RPM packages has been rotated. See the dedicated Agent documentation page to know how to make sure that the new Agent RPM packages can be installed on hosts.
  • Update to the configuration of the systemd check: unit_names is now required and only matching units will be monitored, unit_regexes configuration has been removed.
  • Several metrics sent by the systemd check have been renamed. The integration is now stable.

New Features

  • APM: add support for container tagging. It can be used with any client tracer that supports it.
  • APM: Incoming TCP connections are now measured in the datadog.trace_agent.receiver.tcp_connections metrics with a "status" tag having values: "accepted", "rejected", "timedout" and "errored".
  • Allows the user to blacklist source and destination connections by passing IPs or CIDRs as well as port numbers.
  • Docker label autodiscovery configurations are now polled more often by default.
  • The Agent can now expose runner stats via the CLC Runner API Server, a remotely-accessible authenticated API server. The Cluster Agent can use these stats to optimize dispatching cluster level checks. The CLC Runner API Server is disabled by default, it must be enabled in the Agent configuration, also the cluster agent must be enabled since it's the only client of the server. By default, the server listens on 5005 and its host address must be set to the Agent Pod IP using the Kubernetes downward API.
  • [preview] Checks can now send histogram buckets to the agent to be sent as distribution metrics.
  • In macOS datadog-agent is now able to start/stop process-agent.
  • The Agent now includes a Python 3 runtime to run checks. By default, the Python 2 runtime is used. See the dedicated Agent documentation page for details on how to configure the Agent to use the Python 3 runtime and how to migrate checks from Python 2 to Python 3.
  • High-level RTLoader memory usage statistics exposed as expvars on the agent.
  • Adding tracemalloc_debug configuration setting (Python3 only). Enables Tracemalloc memory profiling on Python3. Enabling this option will override the number of check runners to 1 to guarantee sequential execution of checks.
  • For NTP check, add the option use_local_defined_servers. When use_local_defined_servers is true, use the ntp servers defined in the current host otherwise use the hosts defined in the configuration.

Enhancement Notes

  • Show configuration source for each check's instance in the "status" and the "configcheck" commands.
  • Add a new invoke task, rtloader.generate-doc which generates Doxygen documentation for the rtloader directory and warns about documentation errors or warnings.
  • Allow the check command to display and/or store memory profiling data.
  • For Windows, add a message when the user cannot perform the action in the systray.
  • APM: The datadog.trace_agent.normalizer.traces_dropped metric now has a new reason payload_too_large which was confusingly merged with decoding_error.
  • APM: Bind apm_config.replace_tags parameter to DD_APM_REPLACE_TAGS environment variable. It accepts a JSON formatted string of the form [{"name":"tag_name","pattern":"pattern","repl":"repl_str"}]
  • The default collection interval for host metadata has been reduced from 4 hours to 30 min.
  • Collection interval for the default metadata providers ('host', 'agent_checks' and 'resources') can now be configured using the 'metadata_providers' configuration entry.
  • Agent commands now honor the DD_LOG_LEVEL env variable if set.
  • Distributions: Distribution payloads are now compressed before being sent to Datadog if the agent is built with either zlib or zstd.
  • Configuration files for core checks in cmd/agent/dist/conf.d/ have been migated to the new configuration file norm. https://docs.datadoghq.com/developers/integrations/new_check_howto/#configuration-file
  • When a valid command is passed to the agent but the command fails, don't display the help usage message.
  • Add private_socket configuration to the systemd check. Defaults to /run/systemd/private (or /host/run/systemd/private when using Docker Agent).
  • Warnings returned by the Python 3 linter for custom checks are now logged in the Agent at the 'debug' level.
  • Make NTP check less verbose when a host can't be reached. Warn only after 10 consecutive errors.
  • Added detection of a network ID which will be used to improve destination resolution of network connections.
  • Windows events will now display a full text message instead of a JSON object. When available, the agent will now enrich the events with status, human readable task name and opcode.
  • On Windows, adds system.mem.pagefile.* stats, previously available only in Agent 5.

Deprecation Notes

  • The --log-level argument in agent check and agent jmx commands has been deprecated in favor of the DD_LOG_LEVEL env variable.

Bug Fixes

  • APM: The datadog.trace_agent.receiver.payload_refused metric now has language tags like its peer metrics.
  • The agent jmx command now correctly takes into account the options in the init_config section of the JMXFetch integration configs
  • Escape message when using JSON log format. This, for example, fixes multiline JSON payload when logging a Exception from Python.
  • Fix a bug, when a check have its init configuration before that all the tagger collector report tags.
  • Fix spikes for system.io.avg_q_sz metrics on Linux when the kernel counter was wrapping back to 0.
  • Fix system.io.* metrics on Linux that were off by 1 when the kernel counters were wrapping back to 0.
  • Fixed placeholder value for the marathon entry point to match the new configuration file layout.
  • Fix a tagger goroutine race issue when adding a new entry in the tagger.Store and requesting an entry in another goroutine.
  • Fix files descriptor leak when tailing a logs file with file rotation and the tailer is stuck for instance because of lost connectivity with the logs intake endpoint.
  • The parameter jmx is not supported with the command check, the jmx command should be used instead.
  • Fixed NTP timeout not being used from the configuration.
  • On Windows, correctly configure the config file if the path includes a space.
  • When uninstalling the agent, remove ddagentuser home folder.
  • APM: Fix incorrect traces_dropped and spans_malformed metric counts.
  • On Windows, "ddagentuser" (the user context under which the Agent runs), is now added to the "Event Log Readers" group, granting access to Security event logs.

Other Notes

  • The Windows agent no longer depends on the Windows WMI service. If the WMI service stops for any reason, the Windows agent will no longer stop with it. However, any integrations that do use WMI (wmi_check and win32_event_log) will not be able to function until the WMI service restarts.
  • Ignore the containerd startup script and the kubeconfig as part of the host metadata on GKE.
  • JMXFetch upgraded to 0.31.0
  • On Windows, during an uninstall, if the user context for the datadog agent is a domain user, the user will no longer be deleted even when the user was created by the corresponding install.

6.13.0

Prelude

Release on: 2019-07-24

Upgrade Notes

  • The port option in the NTP check configuration is now parsed as an integer instead of a string.

New Features

  • APM: add support for Unix Domain Sockets by means of the apm_config.receiver_socket configuration. It is off by default. When set, it must point to a valid sock file.
  • APM: API emitted metrics now have a lang_vendor tag when the Datadog-Meta-Lang-Vendor HTTP header is sent by clients.
  • APM: Resource-based rate limiting in the API can now be completely disabled by setting apm_config.max_memory and/or apm_config.max_cpu_percent to the value 0.
  • Add support for environment variables in checks' config files using the format "%%env_XXXX%%".
  • Add new systemd integration to monitor systemd itself and the units managed by systemd.
  • The total number of bytes received by dogstatsd is now reported by the dogstatsd-udp/Bytes and dogstatsd-uds/Bytes expvar.
  • Adds the ability to use DD_TAGS to set global tags in Fargate.
  • Added a support for the new pod log directory pattern introduced in version 1.14 of Kubernetes to make sure the agent keeps on collecting logs after upgrade of a Kubernetes cluster.

Enhancement Notes

  • Add a kube_cronjob tag in the tagger. It applies to container metrics, autodiscovery metrics and logs.
  • Change the prefix of entity IDs to make it easier to query the tagger without knowing what the container runtime is.
  • APM: reduce memory usage in high traffic by up to 10x.
  • APM: Services are no longer aggreagated in the agent, nor written to the Datadog API. Instead, they are now automatically extracted on the backend based on the received traces.
  • APM: The default interval at which the agent watches its resource usage has been reduced from 20s to 10s.
  • APM: Improved processing concurrency and as a result, CPU usage decreased by 20% in some scenarios.
  • APM: Queued sender was rewritten to improve performance around scenarios where network problems are present.
  • APM: Code clean up around configuration and writer.
  • The datadog-agent version command now prints the version of Golang the agent was compiled with.
  • Display Go version in output of status command
  • Upgraded JMXFetch to 0.30.0. See https://github.com/DataDog/jmxfetch/releases/tag/0.30.0
  • APM: the trace agent now lets through a wider variety of traces, automatically correcting some malformed traces instead of dropping them. The following fields are now replaced with reasonable defaults if invalid or empty and truncated if exceeding max length: span.service, span.name, span.resource, span.type. span.duration=0 is now allowed. Missing span start date now defaults to duration - now. The datadog.trace_agent.receiver.traces_dropped metric is now tagged with a reason tag explaining the reason it was dropped. There is a new datadog.trace_agent.receiver.spans_malformed metric also tagged by reason explaining how the span was malformed.
  • Refactored permissions check in the integration command.
  • Support Python 3 for the integration command.

Deprecation Notes

  • APM: The presampler has been rebranded as a "rate limiter" to avoid confusing it with other sampling mechanisms.
  • APM: The "datadog.trace_agent.presampler_rate" metric has been deprecated in favor of "datadog.trace_agent.receiver.ratelimit".

Security Issues

  • On Windows, quote the service name when registering service. Mitigates CVE-2014-5455. Note that since the Agent is not running as admin, even a successful attack would not give admin rights as specified in the CVE.

Bug Fixes

  • Fix the tagger behavior returning None when no tags are present for the kubelet and fargate integration.
  • APM: metrics generated by the processing function (such as *.traces_priority) now contain language specific tags.
  • APM: Memory spikes when retry queue grows have been fixed.
  • Fix 'vcruntime140.dll is being held in use by the following process'
  • System-probe s6 services: ensure that the system-probe binary is bundled before trying to run it / stop it. This is to ensure that the s6-services definitions will be backward compatible with older builds that didn't have the system-probe yet.
  • Fix a bug in the log scanning logic of the JMXFetch wrapper that would make JMXFetch hang if it logged a very large log entry
  • Fixed an issue where logs collected from kubernetes using '/var/log/pods' would show up with a wrong format '{"log":"x","stream":"y","time":"z"}' on the logs explorer when using docker as container runtime.
  • Fix TLS connection handshake that hang forever making the whole logs pipeline to be stucked resulting in logs not being tailed and file descriptor not being closed.
  • On Windows, fixes bug in which Agent can't start if the Go runtime can't determine the ddagentuser's profile directory. This information isn't used, so shouldn't cause a failure
  • The External Metrics Setter no longer stops trying to get metrics after 3 failed attempts. Instead, it will retry indefinitely.
  • Removes an unused duplicate copy of the system-probe binary from the Linux packages
  • The NTP check now properly uses the port configuration option.

Other Notes

  • Logs informing about check runs and payload submission are now displayed once every 500 events instead of every 20 events.

6.12.2

Prelude

Release on: 2019-07-03

This release is only available on Windows and contains all the changes introduced in 6.12.0 and 6.12.1.

6.12.1

Prelude

Release on: 2019-06-28

This release is not available on Windows.

Bug Fixes

  • Fixed a bug in the kubelet and fargate integrations preventing the collection of the kubernetes.cpu.* and kubernetes.memory.* metrics.

6.12.0

Known Issues

Some metrics from the kubelet and fargate integrations (kubernetes.cpu.* and kubernetes.memory.*) are missing for certain configurations. A fix will be released in v6.12.1. Meanwhile if downgrading to 6.11.3 is not an option we recommend using the runtime metrics (ex: docker.cpu.*, docker.mem.*, containerd.cpu.*, ...).

Prelude

Release on: 2019-06-26

This release is not available on Windows.

Upgrade Notes

  • APM: Log throttling is now automatically enabled by default when log_level differs from debug. A maximum of no more than 10 error messages every 10 seconds will be displayed. If you had it enabled before,

    it can now be removed from the config file.

  • On Windows, the path of the embedded python.exe binary has changed from %ProgramFiles%\Datadog\Datadog Agent\embedded\python.exe to %ProgramFiles%\Datadog\Datadog Agent\embedded2\python.exe. If you use this path from your provisioning scripts, please update it accordingly. Note: on Windows, to call the embedded pip directly, please use %ProgramFiles%\Datadog\Datadog Agent\embedded2\python.exe -m pip.

  • Logs: Breaking Change for Kubernetes log collection - In the version 6.11.2 logic was added in the Agent to first look for K8s container files if /var/log/pods was not available and then to go for the Docker socket. This created some permission issues as /var/log/pods can be a symlink in some configuration and the Agent also needed access to the symlink directory.

    This logic is reverted to its prior behaviour which prioritise the Docker socket for container log collection. It is still possible to force the agent to go for the K8s log files even if the Docker socket is mounted by using the logs_config.k8s_container_use_file' or `DD_LOGS_CONFIG_K8S_CONTAINER_USE_FILE. parameter. This is recommended when more than 10 containers are running on the same pod.

New Features

  • A count named datadog.agent.started is now sent with a value of 1 when the agent starts.
  • APM: Maximum allowed CPU percentage usage is now configurable via DD_APM_MAX_CPU_PERCENT.
  • Node Agent can now perform checks on kubernetes service endpoints. It consumes the check configs from the Cluster Agent API via the endpointschecks config provider. Versions 1.3.0+ of the Cluster Agent are required for this feature.
  • Logs can now be collected from init and stopped containers (possibly short-lived).
  • Allow tracking pod labels and annotations value change to update labels/annotations_as_tags. Make the explicit tagging feature dynamic (introduced in DataDog#3024).

Enhancement Notes

  • APM: the writer will now flush based on an estimated number of bytes in accumulated buffer size, as opposed to a maximum number of spans, resulting in better traffic and payload size management.
  • APM: traces are not dropped anymore because or rate limiting due to performance issues. Instead, the trace is kept in a queue awaiting to be processed.
  • Logs docker container ID when parse invalid docker log in DEBUG level.
  • Set the User-Agent string to include the agent name and version string.
  • Adds host tags in the Hostname section of the agent status command and the status tab of the GUI.
  • Expose the number of logs processed and sent to the agent status
  • Added a warning message on agent status command and status gui tab when ntp offset is too large and may result in metrics ignored by Datadog.
  • APM: minor improvements to CPU performance.
  • APM: improved trace writer performance by introducing concurrent writing.
  • APM: the stats writer now writes concurrently to the Datadog API, improving resource usage and processing speed of the trace-agent.
  • Extends the docker check to accommodate the kernel memory usage metric. This metric shows the cgroup current kernel memory allocation.
  • Ask confirmation before overwriting the output file while using the dogstatsd-stats command.
  • Do not ship autotools within the Agent package.
  • The datadog-agent integration subcommand is now capable of installing prereleases of official integration wheels
  • Upgraded JMXFetch to 0.29.1. See https://github.com/DataDog/jmxfetch/releases/tag/0.28.0, https://github.com/DataDog/jmxfetch/releases/tag/0.29.0 and https://github.com/DataDog/jmxfetch/releases/tag/0.29.1
  • Added validity checks to NTP responses
  • Allow the '--check_period' flag of jmxfetch to be overriden by the DD_JMX_CHECK_PERIOD environment variable.
  • Ship integrations and their dependencies on Python 3 in Omnibus.
  • Added a warning about unknown keys in datadog.yaml.

Deprecation Notes

  • APM: the yaml setting apm_config.trace_writer.max_spans_per_payload is no longer in use; writes are now based solely on accumulated byte size.

Bug Fixes

  • Updated the DataDog/gopsutil library to include changes related to excessive DEBUG logging in the process agent
  • The computeMem is only called in the check when we ensure that it does not get passed with an empty pointer. But if someone was to reuse it without checking for the nil pointer it could cause a segfault. This PR moves the nil checking logic inside the function to ensure it is safe.
  • APM: Fixed a bug where normalize tag would not truncate tags correctly in some situations.
  • APM: Fixed a small issue with normalizing tags that contained the unicode replacement character.
  • APM: fixed a bug where modulo operators caused SQL obfuscation to fail.
  • Fix issue on process agent for DD_PROCESS_AGENT_ENABLED where 'false' did not turn off process/container collection.
  • Fix an error when adding a custom check config through the GUI when the folder where the config will reside does not exist yet.
  • APM: on macOS, trace-agent is now enabled by default, and, similarly to other platforms, can be enabled/disabled with the apm_config.enabled config setting or the DD_APM_ENABLED env var
  • Fix a bug where when the log agent is mis-configured, it temporarily hog on resources after being killed
  • Fix a potential crash when doing a configcheck while the agent was not properly initialized yet.
  • Fix a crash that could occur when having trouble connecting to the Kubelet.
  • Fix nil pointer access for container without memory cgroups.
  • Improved credentials scrubbing logic.
  • The datadog-agent integration show subcommand now properly accepts only Datadog integrations as argument
  • Fix incorrectly reported IO metrics when OS counters wrap in Linux.
  • Fixed JMXFetch process not being terminated on Windows in certain cases.
  • Empty logs could appear when collecting Docker logs in addition to the actual container logs. This was due to the way the Agent handles the header Docker adds to the logs. The process has been changed to make sure that no empty logs are generated.
  • Fix bug when docker container terminate the last logs are missing and partially recovered from restart.
  • Properly move configuration files for wheels installed locally via the integration command.
  • Reduced memory usage of the flare command
  • Use a custom patch for a costly regex in PyYAML, see yaml/pyyaml#301.
  • On Windows, restore the system.mem.pagefile.pct_free metric

Other Notes

  • The 'integration freeze' cli subcommand now only displays datadog packages instead of the complete result of the 'pip freeze' command.
  • The Secrets Management feature is no longer in beta.

6.11.3

Prelude

Release on: 2019-06-04

Upgrade Notes

  • Upgrade JMXFetch to 0.27.1

Bug Fixes

  • APM: fixed a bug where secrets in environment variables were ignored.

6.11.2

Prelude

Release on: 2019-05-23

Enhancement Notes

  • Add option cf_os_hostname_aliasing to send the OS hostname as an alias when using the BOSH agent on Cloud Foundry.

Bug Fixes

  • Fixes problem in which Windows Agent wouldn't install on non-English machines due to assumption that "Performance Monitor Users" didn't need to be localized.
  • Windows Installer is now more resilient to missing domain controller.

6.11.1

Important: 6.11.1 is not marked as latest for Windows: we are investigating some cases where 6.11.0 and 6.11.1 are not installing correctly on Windows. Downloading datadog-agent-6-latest.amd64.msi will give you version 6.10.1.

Prelude

Release on: 2019-05-06

Upgrade Notes

  • Change the prioritization between the two logic that we have to collect logs on Kubernetes. Now attempt first to collect logs on '/var/log/pods' and fallback to using the docker socket if the initialization failed.

Bug Fixes

  • Fix a bug where short image name wouldn't be properly set on old docker versions
  • Properly handle docker container logs in multiline mode in case of infrequence log messages, log file rotations or agent restart

6.11.0

Prelude

Release on: 2019-04-17

Upgrade Notes

  • APM: move flush notifications from level "INFO" to "DEBUG"
  • APM: logging format has been changed to match the format of the core agent.
  • Metrics coming through dogstatsd with the following internal prefixes: activemq, activemq_58, cassandra, jvm, presto, solr, tomcat, kafka, datadog.trace_agent, datadog.process, datadog.agent, datadog.dogstatsd are no longer affected by the statsd_metric_namespace option.
  • Removed the internal ability to send logs to a specific logset at agent level.
  • On Windows, the Datadog Agent now runs as a non-privileged user (ddagentuser by default) rather than LOCAL_SYSTEM. Please refer to our dedicated docs for more information
  • The Windows installer will no longer allow direct downgrades; if a downgrade is required, the user must uninstall the newer version and install the older version.

New Features

  • Secrets beta feature is now available on windows allowing users to pull secrets from secret management services.
  • APM: JSON logging is now supported using the log_format_json: true setting.
  • Collect container thread count and thread limit
  • JMXFetch upgraded to 0.27.0. See 0.27.0 for more details.
  • The agent now ignores pod that exited more than 15 minutes ago to reduce its resource footprint when pods are not garbage-collected. This is configurable with the kubernetes_pod_expiration_duration option.
  • Now support CRI-O container runtime for log collection on Kubernetes.
  • Automatically add a "dirname" tag representing the directory of logs tailed from a wildcard path.

Enhancement Notes

  • AutoDiscovery can now monitor unready pods. It looks for a new pod annotation "ad.datadoghq.com/tolerate-unready" which, if set to true will make AutoDiscovery monitor that pod regardless of its readiness state.
  • Add the ability for the datadog-agent check command to have Python checks start an interactive debugging session.
  • Change the logging format to include the name of the logging agent instead of appending it in the agent container logs.
  • Add /metrics to the bare endpoints the agent can access. This is required to support querying endpoints protected by RBAC, by kube-rbac-proxy for instance.
  • APM: errors reported by the receiver's HTTP server are now shown in the logs.
  • APM: slightly improved normalization error logs.
  • On Windows, allows Agent to be installed to nonstandard directories. Uses APPLICATIONDATADIRECTORY to set the root of the configuration file tree, and PROJECTLOCATION to set the root of the binary tree. Please refer to the docs for more details
  • In order to decrease the number of API DCA requests, The Agent now uses a different API endpoint to call the DCA's API only once in order to retrieve the Pods metadata.
  • Host metadata payloads are now zlib-compressed
  • Log file size and number of rotation is now configurable.
  • Add a command dogstatsd-stats to the agent to get basic stats about the processed metrics.
  • Support JSON arrays within environment variables, in addition to space separated values.
  • On Google Compute Engine, the Agent now reports <instance_name>.<project_id> as a host alias instead of <hostname_prefix>.<prefix_id>, which improves the uniqueness and relevance of the host alias when the GCE instance has a custom hostname.
  • The import command doesn't stop anymore when there is no conf.d or auto_conf directory.
  • Kubernetes event collection timeout can now be configured.
  • Improve status page by splitting errors and warnings from the Logs agent
  • Secrets are no longer decrypted in agent command when it's not needed (commands like hostname, launchgui, configuration ...). This reduce the number of times the 'secret_command_backend' executable will be called.
  • Improved memory efficiency on hosts sending very high numbers of metrics.
  • Resolve once the DNS name given by docker and try the associated IP to reach the kubelet. Prioritize HTTPS over HTTP to connect to kubelet. Prioritize communication using IPs over hostnames to spare DNS servers accross the cluster.

Deprecation Notes

  • Removal of largely unused go SNMP check. SNMP support still provided by the python variant.

Bug Fixes

  • Fix an auto-discovery annotation value parsing limitation in version 6 compared to version 5. Now, ad.datadoghq.com/*.instances annotation key supports value like [[{"foo":"bar1"}, {"foo":"bar2"}], {"name":"bar3"}]
  • The agent container will now output valid JSON when using JSON log format.
  • APM: Multiple value "Content-Type" headers are now parsed correctly for media type in the HTTP receiver.
  • APM: always reply with correct Content-Type in API responses.
  • APM: when a span's resource is empty, the error "Resource can not be empty" will be returned instead of the wrong "Resource is invalid UTF-8".
  • APM: sensitive information is now scrubbed from logs.
  • APM: Fix issue with --version flag when API key is unset.
  • APM: Ensure UTF-8 characters are not cut mid-way when truncating span fields.
  • Metrics coming through dogstatsd with the following internal prefixes: activemq, activemq_58, cassandra, jvm`, ``presto, solr, tomcat, kafka, datadog.trace_agent, datadog.process, datadog.agent, datadog.dogstatsd are no longer affected by the statsd_metric_namespace option.
  • Fixes ec2 tags collection when datadog agent is deployed into a kubernetes cluster along with kube2iam.
  • Fixes bug in which upgrading from agent5 doesn't correctly import the configuration
  • Fix a race condition in gohai that could make the Agent crash while collecting the host's filesystem metadata
  • Hostnames containing characters that are invalid for a filename no longer prevent the agent from generating a flare.
  • Allow macOS users to invoke the datadog-agent integration command as root since the installation directory is owned by root.
  • Change to a randomized exponential backoff in case of connection failure
  • Ignore empty logs_dd_url to fall back on default config for logs agent.
  • Detect and handle Docker logs with only header and empty content
  • To mitigate issues with the hostname detection on AKS, hostnames gathered from the metadata endpoints of AWS, GCE, Azure, and Alibaba cloud are no longer considered valid if their length exceeds 255 characters.

Other Notes

  • Bump embedded Python to 2.7.16

6.10.2

Prelude

Release on: 2019-03-20

Bug Fixes

  • Fix a race condition in Autodiscovery leading to some checks not being unscheduled on container exit

6.10.1

Prelude

Release on: 2019-03-07

Bug Fixes

  • APM: Mixing cases in apm_config.analyzed_spans and apm_config.analyzed_rate_by_service entries is now allowed. Service names and operation names will be treated as case insensitive.
  • Refactor the ContainerdUtil so that each call to the containerd api has a dedicated timeout.

6.10.0

Prelude

Release on: 2019-02-28

  • Please refer to the 6.10.0 tag on integrations-core for the list of changes on the Core Checks.
  • Please refer to the 6.10.0 tag on process-agent for the list of changes on the Process Agent.
  • Starting with this release, the changes on the Trace Agent are listed in the present release notes.

Security Notes

  • The Agent now defaults to aliasing yaml.load and yaml.dump to yaml.safe_load and yaml.safe_dump for ALL checks as a defense-in-depth measure against CVE-2017-18342. The Datadog Agent does not use the vulnerable code directly. The effort to patch the PyYAML library guards against the accidental unsafe use of this library by custom checks and transitive dependencies. Specifically, the kubernetes client library v8.0.1 calls the unsafe yaml.load function, but the fix provided forces the use of yaml.safe_load by default. In this release of the Agent, kubernetes client library v8.0.1 is only used by the new kube_controller_manager integration. If for any reason you encounter problems with your custom checks, please reach out to support.

New Features

  • Introduce pod and container tagging through annotations.
  • Docker images are now signed with Content Trust to ensure their integrity when pulling
  • Dogstatsd can now inject extra tags on a metric when a special entity tag is provided
  • datadog-agent integration install command allows to install a check from a locally available wheel (.whl file) with the added parameter --local-wheel.
  • JMXFetch upgraded to 0.26.1: introduces concurrent metric collection across JMXFetch check instances. Concurrent collection should avoid stalling healthy instances in the event of issues (networking, system) in any of the remaining instances configured. A timeout of jmx_collection_timeout (default 60s) is enforced on the whole metric collection run. See 0.25.0, 0.26.0 and 0.26.1.
  • Added the possibility to define global logs processing rules in datadog.yaml that will be applied to all logs, in addition to integration logs processing rules when defined.

Enhancement Notes

  • Consider static pods as ready, even though their status is never updated in the pod list. This creates the risk of running checks against pods that are not actually ready, but this is necessary to make autodiscovery work on static pods (which are used in standard kops deployments for example).
  • Adds the device mapper logical volume name as a tag in the system.io infos.
  • Extends the docker check to accommodate the failed memory count metric. This metric increments every time a cgroup hits its memory limit
  • Add a --json flag to the check command that will output all aggregator data as JSON.
  • [tagger] Add pod phase to kubelet collector
  • The Agent logs now contains the relative file path (including the package) instead of only the filename.
  • Each corecheck could now send custom tags using the tags field in its configuration file.
  • ECS: running the agent in awsvpc mode is now supported, provided it runs in a security group that can reach both the containers to monitor and the host via its private IP on port 51678
  • The performance of the Agent under DogStatsD load has been improved.
  • Improve memory usage when metrics, service checks or events contain many tags.
  • APM: improve performance of NormalizeTag function.
  • Use dedicated datadog_checks_downloader to securely download integrations wheels when using the datadog-agent integration install command.
  • A warning is now displayed in the status when the connection to the log endpoint cannot be established
  • When shutting the agent down, cancel ongoing python subprocess so they can exit as cleanly and gracefully as possible.
  • Add of a "secrets" command to show information about decrypted secrets. We now also track the configuration's name where each secrets was found.
  • Secrets are now resolved in environment variables.
  • In order to ensure compatibility with systemd < 229, StartLimitBurst and StartLimitInterval have been moved to the Service section of the service files.
  • Files are not tailed in reverse lexicographical order w.r.t their file names then dir name. If you have files /1/2017.log, /1/2018.log, /2/2018.log and logs_config.open_files_limit == 2, then you will tail /2/2018.log and /1/2018.log.
  • Include .yml configuration files in the flare.

Bug Fixes

  • Fix an issue where some auto-discovered integrations would not get rescheduled when the template was not containing variables
  • Autodiscovery now removes children configurations when removing templates
  • Fix the display of unresolved configs in the verbose output of the configcheck command
  • Fix custom command line port configuration on configcheck and tagger-list CLI commands.
  • When the secrets feature is enabled, fix bug preventing the additional_endpoints config option from being read correctly
  • Fix "status" command JSON output to exclude non JSON header. The output of the command is now a valid JSON payload.
  • APM: Fix a potential memory leak problem when the trace agent is stopped.
  • Fixed a bug where logs forwarded by UDP would not be split because of missing line feed character at the end of a datagram. Now adding a line feed character at the end of each frame is deprecated because it is automatically added by the agent on read operations.
  • Fix an issue where some kubernetes tags would not be properly removed.

Other Notes

  • The Agent is now compiled with Go 1.11.5
  • Custom checks default on safe pyyaml methods.

6.9.0

Prelude

Release on: 2019-01-22

Upgrade Notes

  • On EC2 hosts that were upgraded from Agent 5.x using the install script or that have the hostname_fqdn option enabled if your hostname currently begins with ip- or domU- (default EC2 hostnames) your hostname will change to the EC2 instance ID. Example: ip-10-1-1-1.ec2.internal => i-1234567890abcdef0. This is an effort to fix a bug in the hostname resolution that was introduced in the version 6.3 of the Agent.
  • Kubernetes logs integration is now automatically enabled if it can find /var/log/pods. If logs_config.container_collect_all is not enabled, only pods with Datadog logs annotation will be collected. If logs_config.container_collect_all is enabled, logs for all pods (matching ac_exclude and ac_include filters if applicable) will be collected.

New Features

  • Introduce a way to configure the cardinality level of tags that the tagger should return. This is split between two options - one for checks and one for dogstatsd. The three cardinality levels are High, Orchestrator, and Low. Checks get Low and Orchestrator-level tags by default Dogstatsd benefits from Low-card tags only by default.
  • You can add extra listeners and config providers via the DD_EXTRA_LISTENERS and DD_EXTRA_CONFIG_PROVIDERS enviroment variables. They will be added on top of the ones defined in the listeners and config_providers section of the datadog.yaml configuration file.
  • Adding native containerd check, based on the containerd socket.
  • You can now see an extra instance id when displaying the Agent status depending on the check. If the instance contains an attribute name or namespace, it will be displayed next to the instance id.
  • Added a new container_cgroup_prefix option to fix some cases where system slices were detected as containers.
  • Add datadog-agent integration show [package] command to show information about an installed integration.

Enhancement Notes

  • AutoDiscovery can now monitor unready pods. It looks for a new pod annotation "ad.datadoghq.com/tolerate-unready" which, if set to true will make AutoDiscovery monitor that pod regardless of its readiness state.
  • Add debug information about the secrets feature to the flare.
  • On the check command, add a pause of 1sec between the 2 check runs when --check-rate is set. Allows some checks to gather more meaningful metric samples.
  • Docker disk IO metrics are now tagged by device
  • Introduces an expvar reporting the number of dogstatsd packets per second processed if dogstatsd_stats_enable is enabled.
  • Add an Endpoints section in the GUI status page and the CLI status command, listing all endpoints used by the agent and their api keys.
  • Expose number of packets received for each dogstatsd listeners through expvar
  • Better descriptions of the install and freeze subcommands of the datadog-agent integration command.
  • In the flare, try to redact api keys from other services.
  • Support the site config option in the log agent.
  • Add ability for Python checks to submit trace logs.

Bug Fixes

  • datadog/dogstatsd image: gohai metadata collection is now disabled by default

  • If dogstatsd_stats_enable is indeed enabled, we should consume and report on the generated stats. Fixes stagnant channel and misleading debug statement.

  • Fix a hostname resolution bug on EC2 hosts which could happen when the hostname_fqdn option was enabled, and which made the Agent use a non-unique FQDN instead of the EC2 instance ID.

  • Fix a bug with parsing of trace.ignore in the import command.

  • Fixes bug in windows core checks where adding/removing devices isn't caught, so only devices present on startup are monitored.

  • Fix bug of the datadog-agent integration install command that prevented moving configuration files when the conf.d folder is a mounted directory.

  • The datadog-agent integration install command creates the configuration folder for an integration with the correct permissions so that the configuration files can be copied.

  • On windows, fixes downgrades. Fix won't be apparent for an additional release, since the core fix occurs on install.

  • On Windows, further fixes when installation drive isn't c:. Fixes problem where logs was effectively hardcoded to use c: for programdata Fixes installation problem where process & trace service were using c:programdata... to find datadog.yaml regardless of installation dir

    If upgrading from a prior version, the configuration file (datadog.yaml) may have incorrect data. It will be necessary to manually update those entries. For example confd_path: c:programdatadatadogconf.d will have to be changed to confd_path: d:programdatadatadogconf.d etc.

  • Removed the command arguments from the flare's container list to avoid collecting sensitive information

  • Fix a rare crash caused by a nil map dereference in the gohai library

  • Reintroducing JMXFetch process lifecycle management on Linux. Adding JMXFetch healthcheck for docker environments.

  • Fix warning about unknown setting "StartLimitIntervalSecs" in the agent service file with systemd version <=229.

Other Notes

6.8.3

Prelude

Release on: 2018-12-27

Bug Fixes

  • Fix a bug that could send the last log multiple times when a container was not writing new logs
  • Fix a panic that could occur if a container doesn't send logs for more than 30 sec and the timestamp of the last received log is corrupted

6.8.2

Prelude

Release on: 2018-12-19

Bug Fixes

  • Fix a panic that could occur when a container stopped while the agent was reading logs from it.

6.8.1

Prelude

This is a container only release that fixes a bug introduced in 6.8.0 that was impacting the kubernetes integration.

Release on: 2018-12-17

Bug Fixes

  • Fixes the default kubelet check configuration that was preventing the kubernetes integration from working properly

6.8.0

Prelude

Please note that a critical bug has been identified in this release that would prevent the kubernetes integration from collecting kubelet metrics on containerized agents. The severity of the issue has led us to remove the 6.8.0 images on dockerhub and to make the latest tag point to the 6.7.0 release. If you have upgraded to this version of the containerized agent we recommend you downgrade to 6.7.0. Linux packages are not affected.

Release on: 2018-12-13

The Datadog Agent now automatically look for the container short image name to set the default value for the log source and service. The source is especially important as it triggers the automatic configuration of your platform with integration pipeline and facets. The Datadog Agent autodiscovery can still be used to override the default source and service with pod annotations or container labels.

Upgrade Notes

  • The agent now requires a cluster agent version 1.0+ to establish a valid connection
  • JMX garbage collection metrics jvm.gc.cms.count and jvm.gc.parnew.time were renamed to jvm.gc.minor_collection_count, jvm.gc.major_collection_count, jvm.gc.minor_collection_time, jvm.gc.major_collection_time in 6.6 to be more meaningful. To ensure backward compatibility the change was reverted in this release and the new names put behind a config option. If you started relying on these new names please enable the new_gc_metrics option in your jmx configurations. An example can be found here. This flag will be enabled by default in a future major release.

New Features

  • Enable docker config provider if docker.sock exists
  • The new command datadog-agent config prints the runtime config of the agent.
  • Adds eBPF-based network collection component called network-tracer.
  • Add diagnosis to the agent for connectivity to the cluster agent
  • datadog-agent integration install command prevents a user from downgrading an integration to a version older than the one shipped by default in the agent.
  • Adding kerberos support with libkrb5.
  • datadog-agent integration install command moves configuration files present in the data directory of the wheel upon successful installation

Enhancement Notes

  • Adding a default location on Windows for the file storing pointers to make sure we never lose nor duplicate any logs
  • Add an option to the agent check command to run the check n times
  • Set service and source to the docker short image name when container_collect_all flag is enabled and no label or annotation is defined
  • Docker: the datadog/dogstatsd image now ships a healthcheck
  • Improved consistency of the ECS and Fargate tagging
  • Improve logging when python checks use invalid types for tags
  • Added a region tag to Fargate containers, indicating the AWS region they run in
  • Adds system.cpu.interrupt, and system.mem.committed, system.mem.paged, system.mem.nonpaged, system.mem.cached metrics on Windows
  • Add permissions.log file to the flare archive.
  • Add an agent go-routine dump to the flare as reported by the built-in pprof runtime profiling interface.
  • The agent can now expose its healthcheck on a dedicated http port. The Kubernetes daemonset uses this by defaut, on port 5555.
  • It's possible now to have different poll intervals for each autodiscovery configuration providers
  • Improve Windows Event parsing. Event.EventData.Data fields are parsed as one JSON object. Event.EventData.Binary field is parsed to its string value
  • Rename the Windows Event "#text" field to "value". This fixes the facet creation of those fields
  • Add a status.log and a config-check.log with a basic message in the flare if the agent is not running or is unreachable.
  • Added support for wildcards to DD_KUBERNETES_POD_LABELS_AS_TAGS. For example, DD_KUBERNETES_POD_LABELS_AS_TAGS='{"*":"kube_%%label%%"}' will all pod labels as tags to your metrics with tags names prefixed by kube_.

Deprecation Notes

  • Removed support for logs_config.tcp_forward_port as it's no longer needed for other integrations.

Bug Fixes

  • Configure error log when failing to run docker inspect to read as debug instead, as this log is duplicated by the tagger.
  • Fix a bug where datadog-agent integration users could not test the --in-toto flag due to a filesystem permission issue.
  • The cluster agent client init now fails as expected if the cluster agent URL is not valid
  • Print correct error when the datadog-agent integration command fails after installing an integration
  • Fix build failure on 32bit armv7
  • Fix a bug with Docker logging driver where logs would not be tailed after a log rotation when the option --log-opt max-file=1 was set.
  • Display the correct timezone name in the status page.
  • On Windows, the agent now properly computes the location of ProgramData for configuration files instead of using hardcoded values

Other Notes

  • JMXFetch upgraded to 0.23.0. See https://github.com/DataDog/jmxfetch/releases/tag/0.23.0
  • On linux, use the cgo dns resolver instead of the golang one. The will make the agent use glibc to resolve hostnames and should give more predictable results.
  • Starting with this Agent release, all the Datadog integrations that are installed with the datadog-agent integration install command are reset to their default versions when the Agent is upgraded. This guarantees the integrity of the embedded python environment after the upgrade.
  • The datadog-agent integration command is now in Beta.

6.7.0

Prelude

Release on: 2018-11-21

This release only ships changes to the trace-agent.

This release focuses on simplifying Trace Search configuration. APM Events can now be configured at the tracer level. Tracers will get updated in the near future to expose this option.

6.6.0

Prelude

Release on: 2018-10-25

Known Issues

  • JMX garbage collection metrics jvm.gc.parnew.time and jvm.gc.cms.count got renamed to jvm.gc.minor_collection_time and jvm.gc.major_collection_count on some JMX integrations. Since this change on the name of these 2 metrics may affect your dashboards and monitors, these metrics will also be sent under their older names in a later version of the Agent.

New Features

  • Disk check support for the puppy agent on unix-like systems
  • Support for the upcoming cluster-agent cluster-level checks feature, via the clusterchecks config provider
  • Add a new CRI core check that will send metrics about resource usage of your containers via the Container Runtime Interface.
  • Support SysVinit on Debian note: some warnings can appear if you enable/disable the agent manually on a systemd system. They can be safely ignored
  • The datadog-agent integration install command will now check for compatibility with datadog-checks-base shipped with the agent. In case of mismatch, it will try to rollback to the previously installed integration version and exit with a failure.
  • Add --in-toto flag to datadog-agent integration command to enable in-toto
  • Add --verbose flag to datadog-agent integration command to enable verbose logging on pip and TUF
  • Docker image: running with a read-only root filesystem is now supported

Enhancement Notes

  • Add a setting to configure the interval at which configs should be polled for autodiscovery.
  • Support a new config option, site, that allows setting the Datadog site to which the Agent should send data. dd_url is still supported and, when set, overrides site.
  • Display a warning in the agent status when too many logs are being tailed and the agent is not tailing them all. This happens with wildcards in path of the tailed files
  • Dogstatsd supports removing the hostname on events and services checks as it did with metrics, by adding an empty host: tag
  • Added new dogstatsd_tags variable which can be used to specify additional tags to append to all metrics received by dogstatsd.
  • dogstatsd cleans up stale UNIX socket on startup.
  • The ecs-agent's docker container name can now be set via the ecs_agent_container_name option or the DD_ECS_AGENT_CONTAINER_NAME envvar for autodetection.
  • EKS pause containers are ignored by default
  • All python and go checks support the new empty_default_hostname option to send metrics with no hostname. This is used for cluster-level checks
  • All go checks now support the min_collection_interval option, as python check already do
  • Added a kubelet_wait_on_missing_container option to handle hosts where the kubelet's podlist is slow to update, leading to missing tags or failing Autodiscovery. Set it to 1 for a 1 second maximum wait
  • Add an option to enable protobuf communication with the Kubernetes apiserver
  • datadog-agent integration command will not pull any of the integration's dependencies
  • More accurate tag extraction logic for Docker Swarm
  • Added new command line properties to the Windows installer which allow for setting site specific configuration.

Bug Fixes

  • Fix an issue preventing the exit logs of the agent from displaying the correct filename.
  • Fix bug that occurs when checks labels/annotation are misconfigured and would prevent the logs of the container to be tailed
  • Fix an issue causing the agent to stop when systemd-journald service is stopped or fails
  • Fix deadlock when an config item under logs is invalid
  • Fix system.mem.pct_usable implementation on Linux 3.14+ to match Datadog Agent 5
  • Fix a potential race in the autodiscovery where a service would be removed before its config could be resolved (causing the agent to crash)
  • Fixes crash on Windows when the agent encounters a malformed performance counter database
  • Fixes config.Digest that was not stable depending on the oder of tags in the instance. It also did not take into account LogsConfig, this is fixed as well.
  • Fix an issue where the log agent would prevent files from being log rotated on Windows
  • Correctly pass the agent's proxy settings to pip when using the datadog-agent integration command with TUF enabled.
  • Recover from errors when connection to the docker socket is lost to continue tailing containers.
  • When installing / updating wheels using the datadog-agent integration command, we replace the PyPI index with our own by default, in order to prevent accidental installation of Datadog or even third-party packages from PyPI.
  • Remove some undocumented power user options to the datadog-agent integration command to prevent accidental misconfiguration that may reduce security guarantees.

Other Notes

  • JMXFetch upgraded to 0.21.0; Adds support for rmi registry connection over SSL and client authentication.
  • Use autodiscovery in log-agent kubernetes integration

6.5.2

Prelude

Release on: 2018-09-20

Bug Fixes

  • Fix a crash in the logs package that could occur when a docker tailer initialization failed.

6.5.1

Prelude

Release on: 2018-09-17

Bug Fixes

  • Fix possible deadlocks that could occur when new docker sources and services are pushed and:
    • The docker socket is closed at agent setup
    • The docker socket is not mounted
    • The kubernetes integration is enabled
  • Fix a deadlock that could occur when the logs-agent is enabled and the configuration parameter 'logs_config.container_collect_all' or the environment variable 'DD_LOGS_CONFIG_CONTAINER_COLLECT_ALL' are set to true.

6.5.0

Prelude

Released on: 2018-09-13

Please note that a critical bug identified in this release affecting container log collection when the container_collect_all was set, would lead to an agent deadlock. The severity of the issue has led us to remove the packages for the affected platforms (Linux and Docker). If you have upgraded to this version, on Linux or Docker we recommend you downgrade to 6.4.2.

New Features

  • Autodiscovery: the docker and kubelet listeners will retry on error, to support starting the agent before your container runtime (host install)
  • Bump the default number of check runners to 4. This has some concurrency implications as we will now run multiple checks in parallel.
  • Kubernetes: to avoid hostname collisions between clusters, a new cluster_name option is available. It will be added as a suffix to the host alias detected from the kubelet in order to make these aliases unique across different clusters.
  • Docker image: handle docker/kubernetes secret files with a helper script
  • The Node Agent can rely on the Datadog Cluster Agent to collect Node Labels.
  • Improved ECS fargate tagging:
    • Honor the docker_labels_as_tags option to extract custom tags
    • Make the cluster_name tag shorter
    • Add the short_image and container_id tags
    • Remove some noisy tags
    • Fix a lifecycle issue that caused missing tags
  • The live containers view can now retrieve containers directly from the kubelet, in order to support containerd and crio
  • Kubernetes events: setting event host tags to the related hosts, instead of the host collecting the events.
  • Added dedicated configuration parameters to send logs to a proxy by TCP. Note that 'logs_config.dd_url', 'logs_config.dd_port' and 'logs_config.dev_mode_no_ssl' are deprecated and will be unvailable soon, use the new parameters 'logs_config.logs_dd_url' and 'logs_config.logs_no_ssl' instead.
  • Added the possibility to send logs to Datadog using the port 443.

Enhancement Notes

  • Add more environment variables to the flare whitelist
  • When dd_url is set to app.datadoghq.eu, the infra Agent also sends data to versioned endpoints (similar to app.datadoghq.com)
  • Make all numbers on the status page more human readable (using unit and SI prefix when appropriate)
  • Display hostname provider and errors on the status page
  • Kubelet Autodiscovery: reduce logging when no change is detected
  • On Windows, the hostname_fqdn flag will now be honored, and the host reported by Datadog will be the fully qualified hostname.
  • Enable all configuration options to be set with env vars
  • Tags generated from GCE metadata may now be omitted by using collect_gce_tags configuration option.
  • Introduction of a new bucketed scheduler to enable multiple check workers to increase concurrency while spreading the load over the collection interval.
  • The 'status' command and 'status' page (in the GUI) now displays errors raised by the '__init__' method of a Python check.
  • Exclude the rancher pause container in the agent
  • On status page, allow users to know which instance of a check matches which yaml instance in configcheck page
  • The file_handle check reports 4 new metrics for feature parity with agent 5
  • The ntp check will now query multiple servers by default to be more resilient to servers returning wrong offsets. A now config option hosts is now available in the ntp check configuration file to allow users to change the list of ntp servers.
  • Tags and sources in the tagger-list command are now sorted to ease troubleshooting.
  • To allow concurrent execution of subprocess calls from python, we now save the thread state and release the GIL to unblock the interpreter . We can reaquire the GIL and restore the thread state when the subprocess call returns.
  • Add a new configuration option, named tag_value_split_separator, allowing the specified list of raw tags to have its value split by a given separator. Only applies to host tags, tags coming from container integrations. Does not apply to tags on dogstatsd metrics, and tags collected by other integrations.

Upgrade Notes

  • Autodiscovery now enforces the ac_exclude and ac_include filtering options for all listeners. Please double-check your exclusion patterns before upgrading and add inclusion patterns if some autodiscovered containers match these.

  • The introduction of multiple runners for checks implies check instances may now run concurrently. This should help the agent make better use of resources, in particular it will help prevent or reduce the side-effects of slow checks delaying the execution of all other checks.

    The change will affect custom checks not enforcing thread safety as they may, depending on the schedule, access unsynchronized structures concurrently with the corresponding data race ensuing. If you wish to run checks in a fully sequential fashion, you may set the check_runners option in your datadog.yaml config or via the DD_CHECK_RUNNERS to 1. Also, please feel free to reach out to us if you need more information or help with the new multiple runner/concurrency model.

    For more details please read the technical note in the datadog.yaml.

  • Prometheus custom checks are now limited to 2000 metrics by default to provide users control over the maximum number of custom metrics sent in the case of configuration errors or input changes. This limit can be changed with the max_returned_metrics option in the check configuration.

Bug Fixes

  • All Autodiscovery listeners now enforce the ac_exclude and ac_include filtering options, as described in the documentation.
  • Fixed "logs_config.frame_size" override that would not be taken into account.
  • collect io metrics for drives with path only (like: C:C0) on Windows
  • Fix API_KEY validation for 'additional_endpoints' by using their respective endpoint instead of the main one all the time.
  • Fix port ordering for the %%port_%% Autodiscovery tag on the docker listener
  • Fix missing ECS tags under some conditions
  • Change the name of the agent expvar from aggregator/ServiceCheckFlushed) to aggregator/ServiceCheckFlushed
  • Fix an issue where logs wouldn't be ingested if the API key contains a trailing new line
  • Setting the log level of the check subcommand using the -l flag was not setting the log level of python integrations.
  • Display embedded Python version in the status page instead of the version from the system Python.
  • Fixes a bug causing kube_service tags to be missing when kubernetes_map_services_on_ip is false.
  • The ntp check now handles negative offsets if the host time is in the future.
  • Fix a possible index out of range panic in Dogstatsd origin detection
  • Fix a verbose debug log caused by rescheduling services with no checks associated with them.

Other Notes

  • JMXFetch upgraded to 0.20.2; ships updated FasterXML.
  • Remove noisy and useless debug log line from contextResolver

6.4.2

Prelude

Release on: 2018-08-13

Enhancement Notes

  • The flare command does not collect the agent container's environment variables anymore

Bug Fixes

  • Fixes an issue with docker tailing on restart of monitored containers. Previously, at each container restart the agent would re submit all logs. Now, on restart we use tracked offsets properly, and as a result submit only new logs

6.4.1

Prelude

Release on: 2018-08-01

New Features

  • Create packaging for google cloud launcher integration.
  • Add options to exclude specific payloads from being sent to Datadog. In some environments, some of the gathered information is considered too sensitive to be sent to Datadog (i.e. IP addresses in events or service checks). This feature adds to option to exclude specific payload types from being sent to the backend.
  • Collect container disk metrics less often in the docker check, decreasing its effect on performance when enabled.
  • Autodiscovery now supports the %%hostname%% tag on the docker listener This tag will resolve to the containers' hostname value if present in the container inspect. It is useful if the container IP is not available or erroneous.
  • Dogstatsd origin detection now supports container tagging for Kubernetes clusters running containerd or cri-o, in addition to the existing docker support
  • This release ships full support of Kubernetes 1.3+
  • OpenShift ClusterResourceQuotas metrics are now collected by the kube_apiserver check, under the openshift.clusterquota.* and openshift.appliedclusterquota.* names.
  • Display the version for Python checks on the status page.

Enhancement Notes

  • Adding DD_EXPVAR_PORT to the configuration environment variables.
  • On Windows, Specifically log to both the log file and the event viewer what initiated an agent shutdown. Also logs specific startup errors to both the log file and event viewer.
  • The embedded Python has been bumped from 2.7.14 to 2.7.15
  • Agent expvar metrics now have default values. Metrics like the number of packets dropped by the agent or errors were previously not reported until a first event occurred. This should make it easier to use the expvar configuration agent_stats.yaml.
  • Proxy settings can be configured through the environment variables DD_PROXY_HTTP, DD_PROXY_HTTPS and DD_PROXY_NO_PROXY. These environment variables take precedence over the proxy options configured in datadog.yaml, and behave exactly the same way as these options. The standard HTTP_PROXY, HTTPS_PROXY and NO_PROXY are still honored but have known side effects on integrations, for simplicity we recommended using the new environment variables. For more information, please refer to our proxy docs
  • Update to distribution metrics algorithm with improved accuracy
  • Added ECS pause containers to the default docker exclusion list
  • Adding logging for when the agent fails to detect the origin of a packet in dogstatsd socket mode because of namespace issues.
  • The skip_ssl_validation configuration option can now be set through the related DD_SKIP_SSL_VALIDATION env var
  • The Agent will log failed healthchecks on query and during exit
  • On Windows, provides installation parameter to set the cmd_port, the port on which the agent command interface runs. To be used if the default (5001) is already used by another program.
  • The kube_service tag is now collected on Kubernetes 1.3.x versions. The matching uses a new logic. If it were to fail, reverting to the previous logic is possible by setting the kubernetes_map_services_on_ip option to true.
  • The Kubernetes event collection timeout is now configurable
  • Logs Agent: Added SOCKS5 proxy support. Use logs_config: socks5_proxy_address: fqdn.example.com:port to set the proxy.
  • The diagnose output is now sorted by the diagnosis name
  • Adding the status of the DCA (If enabled) in the Agent status command.

Upgrade Notes

  • If the environment variables that can be used to configure a proxy (DD_PROXY_HTTP, DD_PROXY_HTTPS, DD_PROXY_NO_PROXY, HTTP_PROXY, HTTPS_PROXY and NO_PROXY) are present with an empty value (e.g. HTTP_PROXY=""), the Agent now uses this empty value instead of ignoring it and using lower-precedence options.

Deprecation Notes

  • Begin deprecating "Agent start" command. It is being replaced by "run". The "start" command will continue to function, with a deprecation notice

Security Issues

  • 'app_key' value from the configuration is now redacted when creating a flare with the agent.

Bug Fixes

  • Fixes presence of invalid UTF-8 characters when docker log message is greater than 16Kb
  • Fix a possible agent crash due to a race condition in the auto discovery.
  • Fixed an issue with jmxfetch not being killed on agent exit.
  • Errors logged before the agent initialized the log module are now printed on STDERR instead of being silenced.
  • Detect and handle Docker messages without header.
  • Fixes installation, packaging scripts for OpenSUSE LEAP and greater.
  • In the event of being unable to lock the dd-agent user (eg. dd-agent is an LDAP user) during installation, do not fail; print relevant warning.
  • The leader election process is now restarted if the leader stops leading.
  • Avoid Linux package installation failures when both the initctl and systemctl commands are present but upstart is used as the init system

Other Notes

  • The system information collected from gohai no longer includes network information when the agent is running in a container since the network information is for the the container and not the host itself.
  • The ntp check now runs every 15 minutes by default to avoid over-loading the NTP server pools
  • Added new command "run" to the agent. This command replaces the "start" command, to reduce ambiguity with the service lifecycle commands

6.3.3

Prelude

Release on: 2018-07-17

Enhancements

  • Add 'system.mem.buffered' metric on linux system.

Bug Fixes

  • Fix the IO check behavior on unix based on 'iostat' tool:
    • Most metrics are an average time, so we don't need to divide again by 'delta' (ex: number of read/time doing read operations)
    • time is based on the millisecond and not the second
  • Kubernetes API Server's polling frequency is now customisable.
  • Use as expected the configuration value of kubernetes_metadata_tag_update_freq, introduce a kubernetes_apiserver_client_timeout configuration option.
  • Fix a bug that led the agent to panic in some cases if the log_level configuration option was set to error.

6.3.2

Prelude

Released on: 2018-07-05

Bug Fixes

  • The service mapper now groups the mappings of pods to services by namespace. This prevents kube_service tags from being erroneously applied to metrics for a pod not targeted by a service but has the same name as a pod in a different namespace targeted by that service.
  • Fix a bug in dogstatsd metrics parsing where the Agent would leave the host tag empty instead of applying its hostname on metrics with a tag metadata field but no tags (i.e. the tags field is only one # character). Regression introduced in 6.3.0
  • Replace invalid utf-8 characters by the standard replacement char.

6.3.1

Prelude

Release on: 2018-06-27

Upgrade Notes

  • JMXFetch upgraded to 0.20.1; ships tagging bugfixes.

Bug Fixes

  • Fixes panic when the agent receives an unsupported pattern in a log processing rule
  • Fixes problem in 6.3.0 in which agent wouldn't start on Windows Server 2008r2.
  • Provide the actual JMX check name as check_name in configurations provided to JMXFetch via the agent API. This addresses a regression in 6.3.0 that broke the instance: tag. Due to the nature of the regression, and the fix, this will cause churn on the tag potentially affecting dashboards and monitors.

6.3.0

Prelude

Release on: 2018-06-20

New Features

  • Add docker memory soft limit metric.
  • Added a host tag for docker swarm node role.
  • The import command now support multiple dd_url and API keys.
  • Add an option to set the read buffer size for dogstatsd socket on POSIX system (SO_RCVBUF).
  • Add support for port names in template vars for autodiscovery.
  • Add a new "tagger-list" command that outputs the tagger content of a running agent.
  • Adding Azure pause containers to the default image exclusion list
  • Add flag histogram_copy_to_distribution to send histogram metric values as distributions automatically. Note that the distributions feature is in beta. An additional flag histogram_copy_to_distribution_prefix modifies the existing histogram metric name by adding a prefix, e.g. dist., to better distinguish between these values.
  • Add docker & swarm information to host metadata
  • "[BETA] Encrypted passwords in configurations can now be fetched from a secrets manager."
  • Add docker ps -a output to the flare.
  • Introduces a new redacting writer that will make sure anything written into the flare is scrubbed from credentials and sensitive information.
  • The agent now supports setting/overriding proxy URLs through environment variables (HTTP_PROXY, HTTPS_PROXY and NO_PROXY).
  • Created a new journald integration to collect logs from systemd. It's only available on debian distributions for now.
  • Add kubelet version to container metadata.
  • Adds support for windows event logs collection
  • Allow overriding procfs path. Should allow to collect relevant host metrics in containerized environments. The override will affect python checks and will result in psutil using the overriding path.
  • The fowarder will now spaw specific workers per domain to avoid slow down when one domain is down.
  • ALPHA - Adding new tooling to securely upgrade integration packages/wheels from our private TUF repository. Please note any third party dependencies will still be downloaded from PyPI with no additional security validation.

Upgrade Notes

  • If your Agent is configured to use a web proxy through the proxy config option or one of the *_PROXY environment variables, and the configured proxy URL starts with the https:// scheme, the Agent will now attempt to connect to your proxy using HTTPS, whereas it would previously connect to your proxy using HTTP. If you have a working proxy configuration, please make sure your proxy URL(s) start with the http:// scheme before upgrading to v6.3+. This has no impact on the security of the data sent to Datadog, since the payloads are always secured with HTTPS between your Agents and Datadog whatever proxy configuration you may use.
  • Docker image: we moved the default configuration from the docker image's default environment variables to the datadog-*.yaml files. This allows users to easily mount a custom datadog.yaml configuration file to set all options. If you already did so, you will need to update your datadog.yaml to include these new defaults. If you only used envvars, no change is needed.
  • The agent now supports the environment variables "HTTP_PROXY", "HTTPS_PROXY" and "NO_PROXY". If set these variables will override the setting in datadog.yaml.
  • Moves away from the community library for the kubernetes client in favor of the official one.

Deprecation Notes

Bug Fixes

  • Default config agent_stats.yaml used to collect go_expvar metrics from the Agent has been updated.
  • Take into account empty hosts on metrics coming from dogstatsd, instead of ignoring them and applying the Agent's hostname.
  • Decrease epsilon and increase incoming buffer size for improved accuracy of distribution metrics.
  • Better handling of docker return values to avoid errors
  • Fix log format when no log file is specified which cause the log date to not be correctly displayed.
  • Configurations of unscheduled checks are now properly removed from the configcheck command display.
  • The agent would send the source twice when protobuf enabled (default), once in the source field and once in tags. As a result, we would see the source twice in the app. This PR fixes it, by sending it only in the source field.
  • Fix a bug on windows where the io check was reporting metrics for the C: drive only.
  • Multiple config files can now be used for the same JMX based integration
  • The auto-discovery mechanism can now properly discover multiple configs for one JMX based integration
  • The JMXFetch process is now managed properly when JMXFetch configs are unscheduled through auto-discovery
  • Fix a possible panic in the kubernetes event watcher.
  • Fix panics within the agent when using non thread safe method from Viper library (Unmarshall).
  • On RHEL/SUSE, stop the Agent properly in the pre-install RPM script on systems where /lib is not a symlink to /usr/lib.
  • To match the behavior of Agent 5, a flag has been introduced to make the agent use hostname -f on unix-based systems before trying os.Hostname(). This flag is turned off by default for 6.3 and will be enabled by default in 6.4. The import command used to upgrade from the Agent5 to the Agent6 will enable this flag in the config.
  • Align docker agent's kubernetes liveness probe timeout with docker healthcheck (5s) to avoid too many container restarts.
  • Fix kube_service tagging of kubernetes network metrics
  • Fixed parsing issue with logs processing rules in autodiscovery.
  • Prevent logs agent from submitting protocol buffer payloads with invalid UTF-8.
  • Fixes JMXFetch on Windows when the custom_jar_paths and/or tools_jar_path options are set, by using a semicolon as the path separator on Windows.
  • Prevent an empty response body from being marked as a "successful call to the GCE metadata api". Fixes a bug where hostnames became an empty string when using docker swarm and a non GCE environment.
  • Config option specified in syslog_pem if syslog logging is enabled with TLS should be a path to the certificate, not a textual certificate in the configuration.
  • Changes the hostname used for Docker events to be the hostname of the agent.
  • Removes use of gopsutil on Windows. Gopsutil relies heavily on WMI; because the go runtime doesn't lock goroutines to system threads, the COM layer can have difficulties initializing. Solves the problem where metadata and various system checks can't initialize properly

Other Notes

  • The agent is now compiled with Go 1.10.2
  • The datadog/agent docker image now runs two collector runners by default
  • The DEB and RPM packages now create the dd-agent user with no login shell (/sbin/nologin or /usr/sbin/nologin). The packages do not modify the login shell of the dd-agent user if it already exists.
  • The scripts of the Linux packages now don't exit with errors when no supported init system is detected, and only print warnings instead
  • On the status and check command outputs, rename checks' Metrics to Metric Samples to reflect that the number represents the number of samples submitted by the check, not the number of metrics after aggregation.
  • Scrub all logging output from credentials. Should prevent leakage of credentials in logs from 3rd-party code or code added in the future.

6.2.1

2018-05-23

Prelude

Known Issues

  • If the kubelet is not configured with TLS auth, the agent will fail to communicate with the API when it should still try HTTP.

Bug Fixes

  • Fix collection of host tags pulled from GCP project (project: and numeric_project_id: tags) and GCP instance attributes.
  • A bug was preventing some jmx configuration options to be set from the jmx checks configs.
  • The RPM packages now write systemd service files to /usr/lib/systemd/system/ (recommended path on RHEL/SUSE) instead of /lib/systemd/system/

6.2.0

2018-05-11

Prelude

Enhancements

  • Introduce new docker cpu shares gauge.
  • Add ability to configure the namespace in which the resources related to the kubernetes check are created.
  • The kubelet check now honors container filtering options
  • Adding Datadog Cluster Agent client in Node Agent. Adding support for TLS in the Datadog Cluster Agent API.
  • Docker: set a default 5 seconds timeout on all docker requests to mitigate possible docker daemon freezes
  • Connection to the ECS agent should be more resilient
  • Add agent5-like JMXFetch helper commands to help with JMXFetch troubleshooting.
  • The agent has been tested on Kubernetes 1.4 & OpenShift 3.4. Refer to https://github.com/DataDog/datadog-agent/blob/master/Dockerfiles/agent/README.md for installation instructions
  • Extract creator tags from kubernetes legacy created-by annotation if the new ownerReferences field is not found
  • The agent import command now handles converting options from the legacy kubernetes.yaml file, for agents running on the host
  • The memory corecheck sends 2 new metrics on Linux: system.mem.commit_limit and system.mem.committed_as
  • Added the possibility to filter docker containers by name for log collection.
  • Added a support for docker labels to enrich logs metadata.
  • Logs Agent: add a filename tag to messages with the name of the file being tailed.
  • Shipping protobuf C++ implementation for the protobuf package, this should help us be more performant when parsing larger/binary protobuf messages in relevant integrations.
  • Enable to set collect_ec2_tags from environment variable DD_COLLECT_EC2_TAGS
  • The configcheck command now display checks in alphabetical orders and are no longer grouped by configuration provider
  • Add average check run time to datadog-agent status and to the GUI.
  • Consider every configuration having autodiscovery identifier a template
  • Implement a circuit breaker and use jittered, truncated exponential backoff for network error retries.
  • Change logs agent configuration to use protocol buffers encoding and endpoint by default.

Known Issues

  • Kubernetes 1.3 & OpenShift 3.3 are currently not fully supported: docker and kubelet integrations work OK, but apiserver communication (event collection, kube_service tagging) is not implemented

Deprecation Notes

  • Removing python PDH code bundled with the agent in favor of code already included in the integrations-core` repository and bundled with datadog_checks_base wheel. This provides a single source of truth for the python PDH logic.

Bug Fixes

  • Fix a possible race condition in AutoDiscovery where configuration is identical on container churn and considered as duplicate before being de-scheduled.
  • It is now possible to save logs only configuration in the GUI without getting an error message.
  • Docker network metrics are now tagged by interface name as a fallback if a docker network name cannot be determined (affects some Swarm stack deployments)
  • Dogstatsd now support listening on an IPv6 address when using bind_host config option.
  • The agent now fetches a hostname alias from kubernetes when possible. It fixes some duplicated host issues that could happen when metrics were using kubernetes host names, as the kubernetes_state integration
  • Fix case issues in tag extraction for docker/kubernetes container tags and kubernetes host tags
  • Fixes initialization of performance counter (Windows) to be able to better cope with missing counter strings, and non-english locales
  • Bind the kubelet_tls_verify as an environment variable.
  • Docker image: fix entrypoint bug causing the kubernetes_apiserver check to not be enabled
  • Fixed an issue with collecting logs bigger than 4096 chars on windows.
  • Fixes a misleading log line on windows for logs file tailing
  • Fixed a concurrent issue in the logs auditor causing the agent to crash.
  • Fix an issue for docker image name filtering when images contain a tag.
  • On Windows, changes the configuration for Process Agent and Trace Agent services to be manual-start. There is no impact if the services are configured to be active; however, if they're disabled, will stop the behavior where they're briefly started then stopped, which creates excessive Windows service alert.
  • API key validation logic was ignoring proxy settings, leading to situations where the agent reported that it was "Unable to validate API key" in the GUI.
  • Fix EC2 tags collection when multiple marketplaces are set.
  • Fixes collection of host tags from GCE metadata
  • Fix Go checks errors not being displayed in the status page.
  • Sanitize logged Datadog URLs when proxies are configured.
  • Fix a race condition in the kubernetes service tagging logic
  • Fix a possible panic when docker cannot inspect a container

Other Notes

  • In the metrics aggregator, log readable context information (metric name, host, tags) instead of the raw context key to help troubleshooting
  • Remove executable permission bits from systemd/upstart/launchd service definition files.
  • Improved the flare credential removing logic to work in a few edge cases that where not accounted for previously.
  • Make file tailing a little less verbose. We avoid logging at every iteration the different issues we encountered, instead we log them at first run only. The status command shows the up-to-date information, and can be used at anytime to troubleshoot such issues
  • Adds collection of PDH counter information to the flare; saves the step of always asking the customer for this information.
  • Improve logging for the metamap, avoid spammy error when no cluster level metadata is found.

6.1.4

2018-04-19

Prelude

Our development staff observed that a local, unprivileged user had the ability to make an HTTP request to the /agent/check-config endpoint on the agent process that listens on localhost. This request would result in the local-users' ability to read Agent integration configurations. This issue was patched by enforcing authentication via a session token. Please upgrade your agent accordingly.

Security Issues

  • The /agent/check-config endpoint has been patched to enforce authentication of the caller via a bearer session token.

6.1.3

2018-04-16

Prelude

Bug Fixes

  • Fix a bug where the docker_network tag incorrectly appeared on non-network docker metrics and autodiscovery tags
  • Fix the use of "docker restart" with the agent image

6.1.2

2018-04-05

Bug Fixes

  • Fix some edge cases where flare could contain secrets if the secrets where encapsulated in quotes.

6.1.1

2018-03-29

Bug Fixes

  • Fix a crash in the docker check when collecting sizes on an image with no repository tags.
  • Fixes bug on Windows where, if configuration options are specified on the installation command line, invalid proxy options are set.
  • Removed the read timeout for UDP connections causing the agent to stop forwarding logs after one minute of nonactivity.
  • Updating the data type of the CPU of the task and the metadata name for Version to Revision.

Other Notes

  • Add environment variable DD_ENABLE_GOHAI for setting option enable_gohai when running in a container.

6.1.0

2018-03-23

New Features

  • Add Agent Version to flare form
  • Add the DD_CHECK_RUNNERS environment variable binding
  • Add the status command to the DCA.
  • Docker check: ignore the new exec_die event type by default
  • Extract the swarm_namespace tag for docker swarm containers, in addition to the already present swarm_service tag.
  • Allow configuration of the enabled-state of process, logs, and apm to be specified on the installation command line for Windows.
  • Add a jmx_use_cgroup_memory_limit option to set jmxfetch to use cgroup memory limits when calculating its heap size. It is enabled by default in the docker image.
  • Add option to extract kubernetes pod annotations as tags, similar to labels
  • Added an environment variable DD_LOGS_CONFIG_CONTAINER_COLLECT_ALL to enable logs tailing on all containers.
  • Adding the 'bind_host' option to configure the interface to bind by dogstatsd and JMX.
  • Support setting tags as a YAML array in the logs agent integration configuration

Bug Fixes

  • Fix docker memory metrics parsing from cgroup files
  • Fix docker.mem.in_use metric computation
  • When using the import script, change the group owner of configuration files to the dd-agent user.
  • Fix a false positive in the collector-queue healthcheck
  • The old docker_daemon check is now properly converted in the "import" command by default
  • Docker check: fix event filtering for exec events
  • Improve docker monitoring when the system is under a very high load. The agent might still temporarily miss a healthcheck, but will be able to run already scheduled checks, and recover once the spike ends
  • Fixes the container startup on Fargate, where we tried and remove the same file twice, failing hard (stopping) on the second attempt.
  • Fix flare failing on zipping individual components
  • Fixed an issue where the import script would put an empty histogram aggregates and percentiles in datadog.yaml if they didn't exist in datadog.conf.
  • Fix the build for platforms not supporting Gohai.
  • Fixes flaw where Windows Performance counters were not properly initialized on non EN-US versions of windows
  • Menu in system tray reports wrong version (6.0.0) for all versions of Agent. This fixes the system tray menu to report the correct version.
  • Fixing clear passwords in "config-check.log" when sending a flare.
  • Allow network proxy settings set on the Windows installation command line to be set in the registry, where they'll be translated to the configuration
  • Accept now short names for docker image in logs configuration file and added to the possibility to filter containers by image name with Kubernetes.
  • Fixes an issue that would prevent the agent from stopping when it was tailing logs of a container that had no logs.
  • fixes an issue with wildcard tailing of logs files on windows
  • Allow Linux package uninstallation to proceed without errors even on platforms that aren't supported by the Agent
  • Fixes agent to run on Server "Core" versions
  • Changes default precision of pdh-based counters from int to float. Fixes bug where fidelity of some counters is quite low, especially counters with values between 0 and 1.
  • Adds back the removed system.mem.usable metric for Agents running on Windows.
  • Avoid multiple initializations of the tagger subsystem

Other Notes

  • Normalize support of nested config options defined with env vars.
  • Make the check-rate command more visible when running "check` to get a list of metrics.

6.0.3

2018-03-12

Prelude

  • This release also includes bugfixes to the process agent. See diff.

Bug Fixes

  • Fixed the issue preventing from having docker tags when collecting logs from containers.
  • Fix docker metrics collection on Moby Linux hosts (default Swarm AMI)

6.0.2

2018-03-07

Critical Issues

  • Packaging issue in 6.0.1 resulted in the release of nightly builds for trace-agent and process-agent. 6.0.2 ships the stable intended versions.

6.0.1

2018-03-07

Enhancements

  • Add information about Log Agent checks to the GUI General Status page.

Bug Fixes

  • Run the service mapper on all the agents running the apiserver check. Exit before running the rest of the check if the agent is not the leader.
  • Fixing docker network metrics collection for the docker check and the process agent on some network configurations.
  • Replaces the system.mem.free metric with gopsutil's 'available' and splits the windows and linux memory checks. Previously this reported with a value of 0 and system.mem.used was reporting the same as system.mem.total
  • ".pdh" suffix was added to system.io metrics on windows for side-by-side testing when changed the collection mechanism, and inadvertently left.
  • Fix bug where global tags for PDH based python checks are not read correctly from the configuration yaml.
  • IE does not support String.prototype.endsWith, add implementation to the string prototype to enable the functionality.
  • remove .pdh suffix from system.io.wkb_s, system.io_w_s, system.io.rkb_s, system.io.r_s, system.io.avg_q_sz
  • Fix GUI for JMX checks, they are now manageable from the web UI.
  • Fix the launch of JMXFetch on windows and make multiplatform treatment of the launch more robust.

6.0.0

2018-02-27

Bug Fixes

  • Fixes bug in agent hostname command, whereby the configuration library wasn't initialized. This caused agent hostname to use the default computed hostname, rather than the entry in the configuration file

6.0.0-rc.4

2018-02-23

Enhancements

  • Change the kubernetes leader election system to use configmaps instead of endpoints. This allows a simpler migration from Agent5, as Agent6 will not require additional permissions.
  • Adds in the proc.queue_length and proc.count metrics with the windows version of the Agent.

Bug Fixes

  • Process agent service should pass the configuration file argument to the executable when launching - otherwise service will always come up on reboots.
  • Add the windows icon to the Infrastructure List for Agents installed on Windows machines.
  • Fix Docker container --pid=host operations. Previous RCs can cause host system instabilities and should not be run in pid host mode.
  • Windows: set correct default value for apm config to enabled, so that the trace agent is started by default
  • Removes deprecated process_agent_enabled flag
  • metrics.yaml is not a "configurable" file - it provides default metrics for checks and shouldn't be altered. Removed from the GUI configuration file list.
  • Windows: gopsutil calls to the CPU module require COM threading model to be in multi-threaded mode, to guarantee it's safe to make those calls we load the python checks setting the right COM concurrency mode first. Once loaded we clear the concurrency mode and python checks that might use COM will set it as they need.
  • Windows: make stop/restart of DatadogAgent service stop/restart dependent services accordingly
  • Windows: Prevent system tray icon from being displayed more than once
  • Windows: Make default start behavior of process-agent consistent with Linux Agent
  • Windows: Fix the item launching the web-based GUI in the systray icon menu
  • Windows: Process agent service now passes the configuration file argument to the executable when launching - otherwise service will always come up on reboots.

Other Notes

  • Windows: Added developer documentation regarding the caveats of the COM concurrency model and its implications moving forward. The current state affects auto-discovery and dynamic scheduling of checks.

6.0.0-rc.3

2018-02-22

Enhancements

  • Adds windows systray icon. System tray icon includes menu options for starting/stopping/restarting services, creating a flare, and launching the browser-based GUI.
  • allow auth token path to be set in the config file
  • Implementation for disabling checks from the web UI
  • Agent restart message on UI, clears after restart.
  • Add SSL support & label joins for the prometheus check

Bug Fixes

  • Fix the command-line flag parsing regression caused by a transitive dependency importing the glog library. agent flags should now behave as in beta9.
  • GUI had broken after the introduction of integrations as wheels this PR ensures we collect the full list of available integrations so we can enable the corresponding configurations from the UI.
  • Fix an issue preventing logs-agent to tail container logs when docker API version is prior to 1.25
  • Fix line miss issue that could happen when tailing new files found when scanning
  • On windows Automatic services would fail to start across reboots due to a known go issue on 1.9.2: golang/go#23479 We now start windows services as delayed start automatic services (ie. they now start automatically after all other automatic services).

Other Notes

  • The OSX build of the agent does not include the containers integrations as they are only supported on Linux for now. The Windows build already excluded them since beta1
  • The auth_token file, used to store the api authentication token, is now only readable/writable by the user running the agent instead of inheriting datadog.yaml permissions.

6.0.0-rc.2

2018-02-20

New Features

  • Add namespace configuration for metric names for dogstatsd
  • Rework autodiscovery label names to be consistent, still support the previous names
  • Ships updated integrations from integrations-core 6.0.0-rc.2, including new kubelet check
  • Add envvar bindings for docker/kubernetes custom tag extraction features

Upgrade Notes

  • Normal installations: APM now listens to localhost only by default, you need to set apm_config.apm_non_local_traffic = true to enable listening on the network
  • Docker image: APM is now disabled by default, you need to set DD_APM_ENABLED=true to run the trace agent. It listens on all interfaces by default when running, you can set DD_APM_NON_LOCAL_TRAFFIC=false to only listen on localhost

Bug Fixes

  • Don't try to match containers by image name if they provide an AD template via docker labels or pod annotations. This avoid scheduling double checks.
  • Fix handling of the %%host%% and %%port%% autodiscovery tags
  • The aggregator now discards metric samples with NaN values. Also solves a serializing error on metric payloads.
  • Fixes bug whereby device tag was (correctly) removed from tags list, but device field was only added to the metric on the first run.
  • Fix an issue unscheduling checks discovered through auto-discovery
  • Upstart would indefinitely respawn trace and process agents even when exiting with a zero status code. We now explicitly define exit code 0 as a valid exit code to prevent respawn when the agents are disabled.
  • Fix cases where empty host tags in the Agent datadog.yaml configuration caused the host metadata payload parsing to fail in the backend.
  • Fix resources metadata collector so that its payload is correctly parsed in the backend even when empty.
  • Make sure we don't get stuck if the API server does not return events.
  • make tagger more resilient to malformed docker events
  • Removing vsphere and sqlserver from the blacklist. The former is available on all platforms, sqlserver is currently windows-only.

Other Notes

  • The apm.yaml.default config file was removed on linux and the trace-agent.conf.example was removed on every other platform.
  • Only enable the resources metadata collector on Linux by default, to match Agent 5's behavior.

6.0.0-rc.1

2018-02-16

Prelude

The execution of the main agent, trace-agent (APM), and process-agent processes is now orchestrated using systemd/upstart facilities on Linux.

On Linux and macOS, the trace-agent and process-agent now read their configuration from the main datadog.yaml file (located, by default on Linux, at /etc/datadog-agent/datadog.yaml).

Changes implementation of IOStats in Windows to use the Performance Data Helper API, rather than WMI.

New Features

  • Introducing the Datadog Process Agent for Windows
  • Make the trace-agent read its configuration from datadog.yaml.
  • Add a HTTP header containing the agent version to every transaction sent by the agent.
  • Add a configcheck command to the agent API and CLI, it prints current loaded & resolved configurations for each provider. The output of the command is also added to a new config-check.log into the flare.
  • Add option to output logs in JSON format
  • Add a pod entity to the tagger.
  • Introducing the service mapper strategy for the DCA. We periodically hit the API server, to get the list of pods, nodes and services. Then we proceed to match which endpoint (i.e. pod) is covered by which service. We create a map of pod name to service names, cache it and expose a public method. This method is called when the Service Mapper endpoint of the DCA API is hit. We also query the cache instead of the API Server if a cache miss happens, to separate the concerns.
  • The kubelet request /pods is now cached to avoid stressing the kubelet API and for better performances
  • Create a cluster agent util to query the Datadog Cluster Agent (DCA) API.
  • Convert the format of histogram_aggregate when importing agent5 configs to agent6
  • Add support for histogram_percentile when importing config options from agent5 to agent6
  • Adds in a datadog.agent.running metric that showcases a value of 1 if the Agent is currently reporting to Datadog.
  • Use the S6 light init in the docker image to start the process-agent and the trace-agent. This allows to remove the process orchestration logic from the infra-agent
  • Add short_image tag to docker tagger collector
  • Try to connect to Docker even /var/run/docker.sock is absent, to honor the $DOCKER_HOST environment variable.
  • Get the docker server version in the debug logs for troubleshooting
  • Add ECS & ECS Fargate metadata API connectivity diagnose
  • Add optional forwarding of dogstatsd packets to another UDP server.
  • Don't block the aggregator if the forwarder input queue is full (we now drop transaction).
  • Add healthcheck to the forwarder. The agent will be unhealthy if the apikey is invalid or if transactions are dropped because all workers are busy.
  • Make the number of workers used by the forwarder configurable and set default to 1.
  • The agent has internal healthchecks on all subsystems. The result is exposed via the agent health command and used in the docker image as a healthcheck. The /probe.sh wrapper is provided for compatibility with agent5 and future-proofing.
  • Adds the ability to import the Trace configuration options from datadog.conf to their respective datadog.yaml fields with the import command.
  • Add jmx_custom_jars option to make sure they are loaded by jmxfetch
  • Added ability to retrieve the hostname from the Kubernetes kubelet API.
  • Support the new kubelet integration by providing the kubelet url and credentials to it
  • Add source_component and namespace tags to Kubernetes events.
  • Support the Kubernetes service name tags on containers. The tag used is kube_service. The default listener for kubernetes is now the kubelet instead of docker.
  • Implementation of the Leader Election for the node agent.
  • Add more details to the information displayed by the logs agent status.
  • Log lines starting with the special character < were considered rfc5424 formatted, and any further formatting was skipped. This commit updates the detection rule to match logs starting with `<pri>version `, to reduce false positives
  • Support tagging on Nomad 0.6.0+ clusters
  • adds a core check and the core logic to query the kubernetes API server and format the events into Datadog events. Also adds the Control Plane status check.
  • The process-agent uses the datadog.yaml file for activation and additional config.
  • Support the new kubelet integration by providing the container tags to it
  • Set configurable timeout for IPC api server
  • Add support for collect_ec2_tags configuration option.
  • Support the legacy default_integration_http_timeout configuration option.
  • handle the "warning" value to log_level, translating it to "warn"
  • Use systemd/upstart facilities on linux to orchestrate the agents execution. Rate limit process restarts such that 5 failures in a 10 second span will result in no further restart attempts.

Upgrade Notes

  • Increased the number of versions of Docker API that logs-agent support from 1.25 to 1.18
  • If you run a Nomad agent older than 0.6.0, the nomad_group tag will be absent until you upgrade your orchestrator.

Deprecation Notes

  • Changed the attribute name to enable log collection from YAML configuration file from "log_enabled" to "logs_enabled", "log_enabled" is still supported.

Bug Fixes

  • Properly listen for events emitted on OSes like CentOS 6 so the Agent starts on reboot
  • Relieves CPU consumption of the WMI service by using PDH rather than WMI
  • Updating custom error used to insure the collection of the token in the configmap datadogtoken.
  • Fix docker event reconnection logic to gracefully wait if docker daemon is unresponsive
  • The checks packaged with the new wheels method are in the default python package site, already included in sys.path. We therefore removed this path them from the locations that are appended to the default python sys.path
  • Strip hostname if running inside ECS Fargate and disable core checks (not relevant without host infos) Fix hostname caching consistency
  • The agent/dogstatsd docker image now ships the appropriate binary
  • Fix the uploaded file name of the flare archive.
  • Fixed the structure of the Flare archive on all platforms.
  • Fix the import proxy build conversion
  • Fix the agent stop command
  • Fixes a bug that caused the GUI to create a flare without a status file
  • Fix DataDog#1159 where error was not explicit when the check had invalid configuration or code
  • Fix to evaluate whether the DCA can query resources (events, services, nodes, pods) before running core component or checks. Logic allows for the DCA to run components independently if they are configured and we can query the associated resources.
  • Only load and log the kube_apiserver check if KUBERNETES=yes is used.
  • The collector will no longer block when the number of runners is lower than the number of long running check. We now start a new runner for those checks.
  • Modify JMXFetch jar permissions to allow the agent to read it on osx.
  • Deleted linux-specific configuration files on macosx to avoid polluting the logs and the web ui.
  • Fix dogstatsd unix socket rights so every user could write to it.

Other Notes

  • The resources metadata provider is now enabled by default in order to populate the "process treemap" visualization continuously (on host dashboards)
  • Decrease verbosity of urllib3's logger (used by python checks through the requests module)
  • Document the exclusion of dockercloud containers.
  • The flare now includes a dump of whitelisted environment variables that can have an impact on agent behaviour. If no whitelisted envvar is found, the envvars.log file is not created in the flare.
  • Upgraded Go runtime to 1.9.4 on Linux builds
  • The linux packages now own the custom check directory (/etc/datadog-agent/checks.d/) and the log directory (/var/log/datadog/)
  • Add an automated install script for osx.

6.0.0-beta.9

2018-01-26

Prelude

In this release, the order of precedence of the Custom Checks has changed. This may affect your custom checks. Please refer to the Upgrade Notes section for more details.

This release includes support for Datadog Logs for Windows.

New Features

  • In this release, the Datadog Log feature is supported on all supported versions of windows.
  • Support AD on Rancher 1.x by getting the container port through the image's exposed ports as a fallback mechanism
  • APM / Process / Log agents can be enabled/disabled via the DD_*_ENABLED environment variables, see the agent docker image readme for details
  • Add a dd-agent user in the docker image to prepare for running root-less
  • Config parsing errors are now displayed in the output of the 'status' command and on the web ui.
  • The DD_TAGS environment variable allows to set host tags, in addition to the tags option in datadog.yaml
  • Added a section to the agent status to report live information about logs-agent
  • Set the default "procfs_path" configuration to /host/proc when containerized to allow the network check to collect the host's network metrics. This can be overridden with the the DD_PROCFS_PATH envvar.
  • Series for a common metric name will no longer be split among multiple transactions/payload. This guarantee that every point for a time T and a metric M will be bundled together when push to the backend. This allows some optimization on the backend side.
  • Add a service listener for ECS Fargate, and a config provider. Also add the concept of ECSContainer and make it compatible with Docker containers so that the process agent can handle them.

Known Issues

  • Having a separate type of container for ECS is not ideal, we will need to change how we represent containers to make it more generic. We also need to improve hostname handling, this will come in a follow up PR.

Upgrade Notes

  • Custom checks (located by default on Linux in /etc/datadog-agent/checks.d/) now have a lower precedence than the checks that are bundled with the Agent. This means that a custom check with the same name as a bundled check will now be ignored, and the bundled check will be loaded instead. If you want to override a bundled check with a custom check, please use a new name for your custom check, and use that new name for the related yaml configuration file.
  • Tags in the DD_TAGS environment variable are now separated by spaces instead of commas in agent5

Bug Fixes

  • Added a support for multi-line tailing with docker
  • Fix the extraction of the environment variables in the pkg/tagger/collectors/docker_extract.go when the variable is like "KEY="
  • Fix a nil-pointer segfault in docker event processing when an event is ignored
  • Make yum revalidate the cache before installing the rpm package when using the install script.
  • Added some missing spaces in logs
  • Fixed build pipeline commenting flaky tests from logs tailer
  • Fixed a bug on file tailing causing the logs-agent to reprocess multiple times the same data when restarted because of a wrong file offset management when lines are trimmed.
  • Be more lenient when filtering containers using labels.
  • Fixes an issue when pulling tags from a yaml config file with any integration that uses the PDHBaseCheck class

Other Notes

  • Updated the shipped CA certs to latest
  • For CircleCI use builder images in the datadog dockerhub repo.

6.0.0-beta.8

2018-01-11

New Features

  • Logs-agent now runs as a goroutine in the main agent process
  • All docker event subscribers are multiplexed in one connection, reduces stress on the docker daemon
  • The Agent can find relevant listeners on its host by using the "auto" listener, for now only docker is supported
  • The Docker label AD provider now watches container events and only updates when containers start/die to save resources
  • Add a new option, force_tls_12, to the agent configuration to force the TLS version to 1.2 when contactin Datatog.
  • Reno and releasenotes are now mandatory. A test will fail if no releasenotes where added/updated to the PR. A 'noreno' label can be added to the PR to skip this test.

Bug Fixes

  • [logs] Fix an issue when the hostname was not provided in datadog.yaml: the logs-agent logic uses the same hostname as the main agent
  • [logs] Trim spaces from single lines
  • Fix missing fields in forwarder logging entries
  • Fix RancherOS cgroup mountpoint detection
  • [linux packaging] Fix missing dd-agent script after upgrade, the fix will take effect on a fresh install of '>= beta.8' or upgrade from '>= beta.8'
  • [logs] Do not send empty logs with multilines
  • [flare] Fix command on Windows by fixing path of collected log files
  • Fix path of logs collected by flare on Windows, was breaking flare command

Other Notes

  • Remove resversion handling from podwatcher, as it's unused
  • Refactor corecheck boilerplate in CheckBase
  • [flare] Rename config file dumped from memory