Skip to content

Latest commit

 

History

History
507 lines (370 loc) · 22.2 KB

CHANGELOG.md

File metadata and controls

507 lines (370 loc) · 22.2 KB
Raw

CHANGELOG

5.6.1 / 2023.09.07

  • [IMPROVEMENT] Add support for IPv6 UDP connection. #280 by @kazwolfe

5.6.0 / 2023.07.10

  • [FEATURE] Add the delay_serialization option, allowing users to delay expensive serialization until a more convenient time, such as after an HTTP request has completed. In multi-threaded mode, it causes serialization to happen inside the sender thread. #271 by @pudiva and @BlakeWilliams

  • [FEATURE] Also, support the sender_queue_size in single_thread mode, so that it can benefit from the new delay_serialization option. Messages are now queued (possibly unserialized) until sender_queue_size is reached or #flush is called. It may be set to Float::INFINITY, so that messages are indefinitely queued until an explicit #flush. #271 by @pudiva and @BlakeWilliams

  • [IMPROVEMENT] Add support of DD_DOGSTATSD_URL for configuration through environment variable. Valid formats are: udp://some-host, udp://some-host:port and unix:///path/to/unix.sock. DD_DOGSTATSD_URL has priority on other environment vars (DD_AGENT_HOST, DD_DOGSTATSD_PORT and DD_DOGSTATSD_SOCKET) but does not throw an error if others are set, values are overridden instead. #278 by @remeh

  • [BUGFIX] Fix NoMethodError when Datadog::Statsd is initialized without telemetry. #272 by @matthewshafer

5.5.0 / 2022.06.01

  • [FEATURE] Add distribution_time method to facilitate measuring timing of a yielded block. #248 by @jordan-brough

  • [BUGFIX] Stop flush timer before closing the queue #257 by @abicky

  • [IMPROVEMENT] Various warnings removed #258 by @abicky

  • [OTHER] Remove unused rack dependency #260 by @remeh

5.4.0 / 2022.03.01

  • [IMPROVEMENT] Add a pre_sampled option to metric methods #235 by @matthewshafer

  • [OTHER] Ruby versions earlier than 2.1.0 are no longer supported. Ruby-2.0 was EOL as of 2016-02-24.

  • [OTHER] Ruby versions 3.0 and 3.1 are now supported, including a fix for keyword arguments to StatsD#open. #240

5.3.3 / 2022.02.02

  • [IMPROVEMENT] Add option "buffer_flush_interval" to flush buffered metrics #231 by @abicky

  • [IMPROVEMENT] Add Sender.queue_size limits to limit number of buffered metrics #232 by @djmitche

  • [IMPROVEMENT] The client can now be configured to use UDS via the DD_DOGSTATSD_SOCKET environment variable. This variable does not take precedence over any explicit parameters passed to the Statsd constructor. [#227][] by @djmitche

5.3.2 / 2021.11.03

  • [OTHER] add a warning message for the v5.x update on install #222 by @djmitche

5.3.1 / 2021.10.21

  • [OTHER] restore connection opening behavior from before 5.3.0 (connections not opened on client instantiation but on the first write instead) #214 by @remeh

5.3.0 / 2021.10.06

  • [ENHANCEMENT] Automatically re-allocate resources (e.g. background thread) if dogstatsd-ruby is used in an application using forks #205 by @remeh

    This will help in scenarios where applications are not handling cleanup/re-creation of the dogstatsd-ruby instances in forked processes. If you are an user of v4.x versions of dogstatsd-ruby and want to migrate to v5.x, please make sure to go through this section of the README and through the migration guide.

  • [BUGFIX] Fix client telemetry in applications using forks #205 by @remeh

Please note that this version will emit a deprecation message if you are using ruby < 2.1: we plan to drop support for ruby 2.0 in a future minor release.

5.2.0 / 2021.07.01

  • [FEATURE] Add single_thread mode for users having issues with the companion thread. #194 by @remeh

You can use this mode to avoid spawning a companion thread while using v5.x versions:

# Import the library
require 'datadog/statsd'

# Create a DogStatsD client instance.
statsd = Datadog::Statsd.new('localhost', 8125, single_thread: true)
...
# release resources used by the client instance and flush last metrics
statsd.close()

Note that if you want to restore the behavior of v4.x versions, you can also configure the buffer to flush on every metric submission:

# Import the library
require 'datadog/statsd'

# Create a DogStatsD client instance using UDP
statsd = Datadog::Statsd.new('localhost', 8125, single_thread: true, buffer_max_pool_size: 1)

5.1.0 / 2021.06.17

  • [FEATURE] Flush last metrics on Statsd#close #180 by @kbogtob
  • [ENHANCEMENT] Do not try to flush where there is no available message_queue [#189][] by @remeh
  • [OTHER] Add pry to development dependencies and enable gem in Gemfile #192 by @ivoanjo
  • [OTHER] Expand Ruby Support to Rubies 2.6, 2.7, and 3.0 [#191][] by @laserlemon

Known issues

  • If the DogStatsD client is instantiated before a fork, the forked process won't copy the companion thread that the original client needs to flush and the client won't work properly. This issue will be addressed in an upcoming version. If you are concerned by this issue, please read this section of the README.

5.0.1 / 2021.04.09

  • [OTHER] Re-introduce a Statsd#batch method to help with compatibility with v4.x versions:
    • It is deprecated and will be removed in v6.x
    • It does not have the exact same behavior as the batch method from v4.x versions since the flush could now automatically occur while the batch block is executed if the buffer has been filled. #176 by @remeh
  • [BUGFIX] Safely close concurrent resources on Sender #175 by @marcotc

5.0.0 / 2021.04.07

API breaking changes

  1. This new major version uses automatic buffering with preemptive flushing, there is no need to manually batch the metrics together anymore. The preemptive flushing part means that just before the buffer gets full, a flush is triggered. However, manual flush is still possible with the Statsd#flush method and is necessary to synchronously send your metrics. The Statsd#batch method has been deprecated from the API.

  2. Every instance of the client will spawn a companion thread for the new flush mechanism: it is important to close every instance using the method Statsd#close.

  3. As of (1), the metrics are now buffered before being sent on the network, you have to use the Statsd#flush method to force their sending through the socket. Note that the companion thread will automatically flush the buffered metrics if the buffer gets full or when you are closing the instance.

  4. Statsd#initialize parameter max_buffer_bytes has been renamed to buffer_max_payload_size for consistency with the new automatic batch strategy. Please note the addition of buffer_max_pool_size to limit the maximum amount of messages to buffer.

What would have been written this way with the v4 API:

require 'datadog/statsd'

statsd = Datadog::Statsd.new('127.0.0.1', 8125)

statsd.batch do |s|
  s.increment('example_metric.increment', tags: ['environment:dev'])
  s.gauge('example_metric.gauge', 123, tags: ['environment:dev'])
end

...

statsd.close()

should be written this way with the v5 API:

require 'datadog/statsd'

statsd = Datadog::Statsd.new('127.0.0.1', 8125)

statsd.increment('example_metric.increment', tags: ['environment:dev'])
statsd.gauge('example_metric.gauge', 123, tags: ['environment:dev'])

# synchronous flush
statsd.flush(sync: true)

...

statsd.close()
  1. statsd.connection should not be used anymore to get the host, the port and the socket_path of the statsd connection, they are now available directly in the statsd object.

Commits

  • [IMPROVEMENT] Use asynchronous IO to avoid doing IO in the hot paths of the library users #151 by @kbogtob
  • [IMPROVEMENT] Automatic buffering/preemptive flushing for better performances #146 by @kbogtob

4.9.0 / 2021.03.23

A version 4.9.0 containing changes intended for 5.0.0 (with API breaking changes) has been released and was available on 2021-03-23. It has been removed on 2021-03-24 and is not available anymore: v4.8.x should be used for latest v4 version of the gem, and v5.x.x versions should be used to benefit from the latest performances improvements.

4.8.3 / 2021.02.22

  • [FEATURE] Add truncate_if_too_long option to the event call to truncate the event if it is too long instead of raising an error #161 by @kazu9su

4.8.2 / 2020.10.16

  • [IMPROVEMENT] The overhead of submitting metrics through dogstatsd-ruby has been reduced #155 #156 by @marcotc

4.8.1 / 2020.05.25

  • [BUGFIX] Send global tags even if no tags provided when using service check / event call #147 by @f3ndot

4.8.0 / 2020.04.20

  • [FEATURE] Add support of more environment variables for tagging #140 by @delner
  • [OTHER] Small optimizations [#139][] by @tenderlove
  • [BUGFIX] Properly close UDPSocket before creating a new one #143 by @zachmccormick
  • [OTHER] Refactor to make code more idiomatic #138 by @kbogtob
  • [OTHER] Refactor to translate unit tests to rspec #135 by @kbogtob
  • [OTHER] Bump rake requirement to >= 12.3.3 #137 by @remeh

4.7.0 / 2020.02.14

  • [FEATURE] Add configurable flush interval for the telemetry #132 by @hush-hush
  • [OTHER] Code structure and tests improvements #131 by @kbogtob

4.6.0 / 2020.01.20

  • [FEATURE] Adding telemetry to the dogstatsd client #128 by @hush-hush

4.5.0 / 2019.08.22

4.4.0 / 2019.07.15

  • [BUGFIX] Allow Integer date_happened and timestamp options #115
  • [OTHER] Update yard gem to 0.9.20 #114

4.3.0 / 2019.06.24

  • [FEATURE] Allow passing tags as a hash #107 by @jtzemp
  • [FEATURE] Added a setting for the global sample rate #110 by @claytono
  • [BUGFIX] Fix non-ascii event texts being truncated #112 by @devleoper
  • [BUGFIX] Display error if write fails due to a bad socket #97 by @abicky

4.2.0 / 2019.04.04

  • [FEATURE] Added environment vars support for host, port and entity id #109 by @ahmed-mez

4.1.0 / 2019.03.08

  • [FEATURE] Handle ENOTCONN #102 by @blaines
  • [IMPROVEMENT] Retry first before losing message when receiving ENOTCONN #104 by @blaines

4.0.0 / 2018.08.20

  • [IMPROVEMENT] Add .open for short-lived reporting that does not leave sockets around #96 by @grosser
  • [IMPROVEMENT] Extract batch logic into a class #95 by @grosser
  • [IMPROVEMENT] Extract connection for separation of concerns #94 by @grosser
  • [IMPROVEMENT] Fail fast on unknown options #93 by @grosser
  • [IMPROVEMENT] Always lazy connect #92 by @grosser
  • [IMPROVEMENT] Batch events and service checks too #88 by @grosser
  • [IMPROVEMENT] Remove bad argument after options #83 by @grosser
  • [IMPROVEMENT] Reduce object allocation and make all strings frozen on ruby 2.3+ #78 by @grosser

Breaking changes

  • Remove deprecated version method #91 by @grosser
  • port / host / tags / namespace can no longer be set on the instance to allow thread-safety #87 by @grosser
  • port / host / socket_path readers are now on statsd.connection
  • Make logger an instance var #90 by @grosser
  • Make format_service_check private #89 by @grosser
  • Improve code coverage / make format_event private #84 by @grosser
  • Set buffer size in bytes #86 by @grosser
  • max_buffer_size initializer argument removed and replaced with max_buffer_bytes (defaults to 8192)
  • max_buffer_size/max_buffer_size= methods removed

3.3.0 / 2018.02.04

  • [FEATURE] Add distribution support (beta). See #72.
  • [IMPROVEMENT] A ton of cleanups and refinements courtesy of @grosser. See #68, #69, #73, #74, #75, #76, #77.
  • [IMPROVEMENT] Unify tag handling in format_service_check. See #71 by @grosser.
  • [IMPROVEMENT] Use faster time method on ruby >= 2.1. See #70 by @grosser.

3.2.0 / 2017.12.21

  • [IMPROVEMENT] Add statsd sample rate compat. #67, @sj26

3.1.0 / 2017.11.23

3.0.0 / 2017.05.18

  • [FEATURE] Expose (socket) close method. #46, @ramfjord
  • [IMPROVEMENT] Retry once when send fails on a closed socket. #46, @ramfjord
  • [IMPROVEMENT] Use a instance variable to decide whether to batch or not. #47 @fimmtiu

Breaking changes

Host resolution

Host resolution was previously done every time a message was sent, it is now done once when Datadog::Statsd is initiliazed (resulting in a non-negligible performance improvement). #44, @AMekss

Datadog::Statsd.new(host, port) will now raise a SocketError if unable to resolve the host.

2.2.0 / 2017.01.12

  • [IMPROVEMENT] Fewer string allocations, improves memory usage, #40, @janester

2.1.0 / 2016.10.27

  • [FEATURE] Add an optional by parameter for #increment and #decrement, #33
  • [BUGFIX] #time: record on all block exits, #38 @nelhage
  • [IMPROVEMENT] Replace string literals with symbols or frozen strings, #37 @janester

2.0.0 / 2016.09.22

Breaking changes

Namespace

The Statsd is now namespaced under the Datadog module. #32 @djpate

To update:

  • require 'statsd' -> require 'datadog/statsd'
  • Statsd -> Datadog::Statsd

Tags

, is now stripped from tags to avoid unexpected behavior. #34 @adimitrov

Datadog::Statsd also validates that it receives an array of tags, and strips , and | from them.

1.6.0 / 2015.12.21

  • [IMPROVEMENT] Stop mutating input parameters, #22 @olefriis

1.5.0 / 2015.05.20

Notice

This release drops testing for Ruby 1.8.7. Future versions are likely to introduce backward incompatibilities with < Ruby 1.9.3.

  • [FEATURE] Add service checks support, #11
  • [FEATURE] Send time stat on failing block, #16 @gleseur
  • [BUGFIX] Add instance tags to Statsd.event, #14 @gleseur
  • [OTHER] Use send_stat instead of overriding Ruby send method, #17 @sensadrome
  • [OTHER] Changelog update

1.4.1 / 2014.09.29

  • [BUGFIX] Fixed bug in message separator when batching metrics

1.4.0 / 2014.06.13

  • [FEATURE] Added support for metrics batching

1.3.0 / 2014.03.27

  • [FEATURE] Added support for submitting events

1.2.0 / 2013.07.10

  • [FEATURE] Added global tags
  • [FEATURE] Added ability to set namespace and tags from Statsd#initialize

1.1.0 / 2012.09.21

  • [FEATURE] Added sets metrics

1.0.0 / 2012.06.14

  • Initial release