Skip to content

Latest commit

 

History

History
693 lines (553 loc) · 18.2 KB

CHANGELOG.md

File metadata and controls

693 lines (553 loc) · 18.2 KB
Raw

28.3.0

  • Rather than dropping data points, the Datadog backend will coerce non-numeric values resulting from aggregation to numeric. NaN is converted to -1, +Inf to float64 maximum, and -Inf to float64 minimum.

28.2.1

  • Document how semver is used in the codebase.

28.2.0

  • Changes how parser.bad_lines_seen and backend.retried metrics are sent. They are now only sent on change.

28.1.0

  • The Datadog backend will discard values that result in NaN or ±Inf since they can not be used with our JSON processing.

28.0.0

  • Internal pipeline refactor removes two internal metrics

27.0.0

  • Simplify internal tracking of the source address of a metric. The current rules are:

If ignore-host is set, it will be set to the host tag if present, otherwise blank. If ignore-host is not set, it will be set to the sending IP If the cloud provider is enabled, it will attempt to perform a lookup of this value to find a new value (instance ID, pod ID, etc) If the tag handler matches a drop-host filter, it will be removed Backends use it inconsistently

This really is simpler than it was before.

26.0.0

  • Adds new options flush-offset and flush-aligned to align the flush interval to a configurable multiple. See README.md for details.
  • Adds documentation for many of the server options.
  • Internal library refactoring

25.0.0

  • Adds InfluxDB as a backend, see BACKENDS.md for details.
  • Some logs have been changed to be slightly more machine readable.

24.0.1

  • Add pprof catch-all route to match the rest of pprof's available endpoints

24.0.0

  • Update to aggregator channel metrics. see METRICS.md for details.

23.0.0

  • Added experimental support for adding http headers based on metrics tags
  • Fixed a bug breaking internal metrics from the backend and cloud provider

22.0.0

  • Refactor pprof endpoints to sit under /debug/pprof

21.0.0

  • Breaks out expiry interval in to per metric type expiry intervals. See README.md for details.

20.3.2

  • Use the current timestamp when sending to NewRelic backend, rather than the last updated timestamp of the metric, which may be old.

20.3.1

  • Added a new internal metric called backend.series.sent which contains the number of flushed timeseries over time.

20.3.0

  • Added experimental support for generating latency histograms for timers, more details in README.md.

20.2.0

  • Support for response-header-timeout for transport section

20.1.2

  • Fixes a bug where aggregator.metricmaps_received are the same for for every worker

20.1.1

  • Fixes a bug where no cloud provider would crash on startup
  • Minor documentation fix

20.1.0

  • Minor internal refactoring
  • Support for regex matching in filtering, added by ermeaney

20.0.0

  • Build with 1.13.6
  • Refactor dependencies, re-run make setup if build failures are encountered
  • Adds support for Kubernetes as a cloud provider for enriching metrics. See CLOUDPRIVDERS.md for details.

19.1.0

  • Adds support for New Relics new dimensional metrics pipeline. See BACKENDS.md.
  • Test fixes

19.0.1

  • Internal refactor to de-dupe 8 getSubVipers

19.0.0

  • BREAKING: Refactor how http clients are configured. See TRANSPORT.md for details. Some configuration options are in a new location, however defaults are generally unchanged.

18.1.0

  • Build with Go 1.13.0
  • Base docker images on alpine 3.10 and ubuntu 18.04

18.0.0

  • Support for custom-headers for http_forwarder_v2

17.2.0

  • Support log-raw-metric configuration

17.1.2

  • Remove support for disabling retries in the forwarder
  • Document how to disable retries using timeout values

17.1.1

  • Adding support to disable retries within the gostatsd forwarder

17.1.0

  • NewRelic event support added by Kav91

17.0.1

  • Moved noisy log line from info to debug

17.0.0

16.0.0

  • Build with Go 1.12.3 for real.

15.0.2

  • Fixed events in HTTP style

15.0.1

  • Update to Alpine 3.9

15.0.0

  • Build with Go 1.12.3
  • Changed linter to golangci-lint

14.1.0

  • Fixes internal metrics for the dispatch_aggregator channel stats, and aggregator.* metrics.

14.0.0

  • More internal pipeline refactoring, to optimize the http ingestion path.

13.0.1

  • Adjust log level about forwarding failures and stop logging retries.

13.0.0

  • Removed v1 ingestion
  • Refactor internal pipeline
  • Added hostname flag to support override for internal metrics. Defaults to system hostname. Use this flag when running within docker to make host identification easier.

12.0.2

  • Fix bug with metrics going through the consolidator not having tags applied properly

12.0.1

  • Emit v2 ingestion http metrics.

12.0.0

  • Add metric consolidation in the HTTP path. Minor configuration changes, specifically metrics-per-batch is replaced with consolidator-slots. Defaults to the number of parsers.
  • Introduces v2 ingestion. Production use should be ok.

11.0.0

  • Add http forwarding and ingestion. See Configuring the server mode and Configuring HTTP servers in README.md for details. Not recommended for production use at this time. Healthcheck is ok to use
  • Introduces v1 ingestion

10.0.1

  • Fix linter failures

10.0.0

  • New Relic backend supports submitting directly to Insights Collector, refer to README.md for details.

9.1.0

  • New Relic backend added by Kav91, see README.md for options and details

9.0.0

  • Added support for sampling timers to be compatible with original statsd.

8.0.0

  • Fixed --statser-type didn't apply
  • Added the ability to filter tags and metrics, see FILTERING.md for details

7.2.0

  • New Cloudwatch backend contirubted by JorgenEvens
    • Backend name is cloudwatch
    • Contains the single option namespace, defaulting to StatsD
  • Duplicate tags are filtered out so they are aggregated correctly.

7.1.0

  • Datadog user-agent has changed to gostatsd by default. Can now be configured.
  • New Datadog option: user-agent, configures the user agent supplied to Datadog. Use python-requests/2.6.0 CPython/2.7.10 for old behavior.

7.0.1

  • Roll back change to configuration, due to spf13/viper#380. Documentation is still valid.

7.0.0

  • Fix a bug in the cache provider where transient failures were replacing good cache
  • Started passing around a logger, not used everywhere yet
  • Documentation fixes

6.2.0

  • Added enable-http2 flag for Datadog backend to control HTTP/2 support, defaults to false

6.1.2

  • Build with Go 1.10.2

6.1.1

  • Fixed a bug making the service not work on Windows.

6.1.0

  • Add new flag --statser-type to make internal metric destination configurable. Defaults to internal, also supports logging and null

6.0.1

  • Fixes rate limiter on bad lines. A value <= 0 will disable entirely.

6.0.0

  • Parses histogram metrics as timers
  • Log bad lines with rate limit
  • Add new flag --bad-lines-per-minute, controls the rate limit on logging lines which fail to parse. Defaults to 0. Supports floats.

5.4.1

  • Memory/GC optimisation with buffers being reused in the Datadog backend
  • Metrics in the cloud provider cache should now be correct

5.4.0

  • BREAKING: The way "per second" rates are calculated has changed from "value / configured flush interval" to "value / actual flush interval".

5.3.0

  • Uses the default EC2 credentials
  • Timer sub-metrics now have a configuration option for opt-out. See README.md for details.

5.2.2

  • Build -race on ubuntu:16.04 instead of 16.10

5.2.1

  • Performance work, round 3

5.2.0

  • Performance work, round 2
  • Add new flag --estimated-tags, pre-allocates the Tags array. Defaults to 4.

5.1.1

  • Fix index out of range error and tag corruption in AWS CP

5.1.0

  • Performance work

5.0.0

  • More metrics rework. Internal metrics are now in-phase with flushing.
  • Better distribution of metrics between aggregators when received from multiple hosts.
  • New Datadog option: max_requests, the maximum number of metric http requests that can be made by the Datadog backend.
  • BREAKING: Additional and renamed metrics in the flusher, see METRICS.md
  • BREAKING: Heartbeat changes:
    • --heartbeat-interval changed to --heartbeat-enabled.
    • Heartbeat is in-phase with the flush.
    • Heartbeat sends a value of 1, so a sum aggregation can be applied.

4.1.1

  • New internal metrics around the cloud provider cache and AWS cloud provider, see METRICS.md for details.

4.1.0

  • Add new flag --conn-per-reader adding support for a separate connection per reader (requires system support for reusing addresses)
  • Refactor MetricDispatcher into BackendHandler satisfying separate MetricHandler and EventHandler interfaces
  • Add TagHandler for static tags

4.0.0

  • BREAKING: Independent scaling of datagram reading and parsing
    • Separate MetricReceiver into DatagramReceiver and DatagramParser
    • Allow independent scaling of parser workers with new flag --max-parsers. Defaults to number of CPU cores.
    • Change default of --max-readers flag to min(8, number of CPU cores)

3.1.1

  • GC optimization with buffer reuse

3.1.0

  • New docker image suffix, version-syms, includes symbol table
  • New flag --receive-batch-size to set datagram batch size. Defaults 50.
  • Batch reading should improve performance by using recvmmsg(2). This has additional memory considerations documented in README.md, and can tuned by tracking the avg_packets_in_batch metric and adjusting as necessary.

3.0.1

  • New flag --heartbeat-interval sends heartbeat metrics on an interval, tagged by version and commit. Defaults 0. Set to 0 to disable.

3.0.0

  • BREAKING: use space instead of comma to specify multiple values for the following parameters: backends, percent-threshold, default-tags and internal-tags.
  • BREAKING: Removed Datadog dual_stack option in favor of explicit network selection.
  • New Datadog option: network allows control of the network protocol used, typical values are tcp, tcp4, or tcp6. Defaults tcp. See Dial for further information.

2.4.2

  • New Datadog option: dual_stack allows control of RFC-6555 "Happy Eyeballs" for IPv6 control. Defaults false.

2.4.1

  • No functional changes over previous version. Release tag to trigger build process.

2.4.0

  • Build with Go 1.9
  • Add support for compression in Datadog payload.
  • New Datadog option: compress_payload allows compression of Datadog payload. Defaults true.
  • Add staged shutdown
  • Update logrus import path

2.3.0

  • New flag --internal-tags configures tags on internal metrics (default none)
  • New flag --internal-namespace configures namespace on internal metrics (default "statsd")
  • BREAKING: Significant internal metric changes, including new names. See METRICS.md for details

2.2.0

  • New flag --ignore-host prevents capturing of source IP address. Hostname can be provided by client via a host: tag.

2.1.0

  • Handle EC2 InvalidInstanceID.NotFound error gracefully
  • Build with Go 1.8
  • Make cloud handler cache configurable
  • Batch AWS Describe Instances call
  • Fix a deadlock in Cloud Handler
  • Minor internals refactoring

2.0.1

  • Do not log an error if source instance was not found

2.0.0

  • BREAKING: Renamed aws.http_timeout into aws.client_timeout for consistency
  • BREAKING: Renamed datadog.timeout into datadog.client_timeout for consistency
  • Tweaked some timeouts on HTTP clients

1.0.0

  • BREAKING: Renamed parameter maxCloudRequests into max-cloud-requests for consistency
  • BREAKING: Renamed parameter burstCloudRequests into burst-cloud-requests for consistency
  • Fix a bunch of linting issues
  • Run tests concurrently
  • Configure various timeouts on HTTP clients
  • Update dependencies
  • Big internals refactoring and cleanup

0.15.1

  • Fix bug in Graphite backend introduced in 0.15.0 (#75)

0.15.0

  • Fix bug where max queue size parameter was not applied properly
  • Use context in more places
  • Stricter TLS configuration
  • Support TCP transport and write timeouts in statsd backend
  • Reuse UDP/TCP sockets to reduce number of DNS lookups in statsd and graphite backends
  • Reuse memory buffers in more cases
  • Update dependencies

0.14.11

  • Go 1.7
  • Config option to disable sending tags to statsdaemon backend

0.14.10

  • Fix NPE if cloud provider is not specified
  • Minor internal cleanups

0.14.9

  • Some additional tweaks to flushing code
  • Minor refactorings

0.14.8

  • Fix bug in rate calculation for Datadog

0.14.7

  • Fix bug introduced in 0.14.6 in Datadog backend when invalid hostname was sent for metrics
  • Set tags for own metrics (#55)
  • Add expvar support
  • Minor internals refactoring and optimization

0.14.6

  • Limit max concurrent events (#24)
  • Memory consumption optimizations

0.14.5

  • Improved and reworked tags support; Unicode characters are preserved now.
  • Minor internals refactoring and optimization

0.14.4

  • Send start and stop events (#21)
  • Linux binary is now built inside of Docker container rather than on the host

0.14.3

  • Fix data race in Datadog backend (#44)
  • Minor internals refactoring

0.14.2

  • Fix batching support in Datadog backend

0.14.1

  • Better Graphite support (#35)
  • Minor internals refactoring

0.14.0

  • Update to Alpine 3.4
  • Cap request size in Datadog backend (#27)
  • Set timeouts on Dials and tcp sends (#23)
  • Reuse HTTP connections in Datadog backend

0.13.4

  • Async rate limited cloud provider lookups (#22, #3)
  • Internals refactoring

0.13.3

  • Add configurable CPU profiler endpoint

0.13.2

  • Increase default Datadog timeouts to reduce number of errors in the logs

0.13.1

  • Log intermediate errors in Datadog backend
  • Consistently set timeouts for AWS SDK service clients
  • Update all dependencies
  • Fix Datadog backend retry error #18
  • Various internal improvements

0.13.0

  • Fix goroutine start bug in dispatcher - versions 0.12.6, 0.12.7 do not work properly
  • Datadog events support

0.12.7

  • Remove deprecated -f flag passed to Docker tag command
  • Rename num_stats back to numStats to be compatible with original statsd

0.12.6

  • null backend to do benchmarking
  • Internals refactoring

0.12.5

  • Implement negative lookup cache (#8)
  • Read configuration from environment, flags and config

0.12.4

  • Do not multiply number of metric workers

0.12.3

  • Do not replace dash and underscore in metric names and tags

0.12.2

  • Fix handling of NaN specific case
  • Minor refactorings for linter

0.12.1

  • Use pointer to metric instead of passing by value
  • Optimise (5-10x) performance of line parser

0.12.0

  • Revert dropping message and block instead
  • Use different values for number of worker to read from socket, process messages and metrics
  • Minor fixes and performance improvements
  • Calculate per second counters since last flush time instead of interval

0.11.2

  • Process internal stats as standard metrics
  • Add benchmarks
  • Improve performance for buffering data out of the socket
  • Drop messages and metrics instead of blocking when overloaded

0.11.1

  • Normalize tags consistently but don't force lower case for metric names

0.11.0

  • Add support for cloud plugins to retrieve host information e.g. instance id, aws tags, etc.

0.10.4

  • Add load testing tool

0.10.3

  • Performance improvements for receiver and aggregator

0.10.2

  • Use goroutines to read net.PacketConn instead of buffered channel

0.10.1

  • Graphite: replace dots by underscores in metric name (tags)
  • Limit concurrency by using buffered channels for incoming messages
  • Discard empty tags
  • Datadog: add retries on post errors

0.10.0

  • Add support for default tags

0.9.1

  • statsd backend: ensure not going over the udp datagram max size on metrics send
  • Datadog: normalise tags to always be of form "key:value"

0.9.0

  • Reset counters, gauges, timers, etc. after an expiration delay
  • Datadog: add interval to metric payload
  • Datadog: use source ip address as hostname

0.8.5

  • Remove extraneous dot in metric names for stdout and graphite backends when no tag
  • Add more and improve internal statsd stats

0.8.4

  • Datadog: set dogstatsd version and user-agent headers
  • Datadog: use rate type for per second metric

0.8.3

  • Fix issue with tags overwriting metrics without tags

0.8.2

  • Fix undesired overwriting of metrics

0.8.1

  • Datadog: rollback change to metric names for preserving backward compatibility

0.8.0

  • Add more timers aggregations e.g. mean, standard deviation, etc.

0.7.1

  • Improve graphite backend reliability by re-opening connection on each metrics send
  • Fix value of metrics displayed in web UI
  • Improve web console UI look'n'feel

0.7.0

  • Add statsd backend

0.6.0

  • Add support for global metrics namespace
  • Datadog: remove api url from config

0.5.0

  • Add support for "Set" metric type

0.4.2

  • Datadog: send number of metrics received
  • Use appropriate log level for errors

0.4.1

  • Remove logging the url on each flush in datadog backend
  • Use alpine base image for docker instead of scratch to avoid ca certs root error

0.4.0

  • Add datadog backend

0.3.1

  • Fix reset of metrics

0.3.0

  • Implement tags handling: use tags in metric names

0.2.0

  • Implement support for pluggable backends
  • Add basic stdout backend
  • Configure backends via toml, yaml or json configuration files
  • Add support for tags and sample rate

0.1.0

  • Initial release