0.11.0 (beta1)

Configuration system

This pre-release includes breaking changes for our configuration system. Our documentation will be updated in the stable release.

0.10.0

Distributed Sampling (beta)

New feature that propagates the sampling priority across services. This ensures traces are always consistent and complete when distributed tracing is used. This new functionality requires at least the Datadog Agent 5.19+. Frameworks and libraries with out-of-the-box propagation are: Rack, Rails, Sinatra, net/http and Faraday (#248, #245, #229, #249, #254, docs)

Improvements

• [core] add Datadog::Registry for better configuration API (#200)
• [core] add Configuration API for integrations (#203)
• [core] migrate http to new configuration API (#225)
• [core] make HTTPTransport compatible with multiple API versions (#228)
• [core] improve shutdown process (#253)

New configuration

Introduced a new experimental API that is available for some integrations. The API is still experimental and will be fully available in the next major release (0.11.0). In the meantime you can check some examples for the following integrations:

• Rails (#224)
• Sinatra (#226, #260, docs)
• Sidekiq still use the previous API, though some changes have been done to support the new API (#227)

Bugfixes

• [redis] Updates Redis integration tag/metric (#216)
• [core] handle integrations that don't implement #patch (#241)
• [resque] removing useless waits when a job ends, using a synchronous writer so the trace is flushed immediately before exiting the process (#252)

!! Breaking changes !!
[sinatra] introducing a new API that replaces the previous approach. This version changed the following:

• default_service has been renamed service_name
• you don't use datadog_tracer configuration object, but directly the new Datadog#configure

If you have a configuration like:

configure do
  settings.datadog_tracer.configure default_service: 'my-app', trace_agent_hostname: 'ddagent'
end

now it must be:

Datadog.configure do |c|
  c.use :sinatra, service_name: 'my-app', trace_agent_hostname: 'ddagent'
end

Read the full changeset

0.9.2

Bugfixes

• [rails] fixed error reporting when exceptions happen in the cache layer (#244)

Read the full changeset.

0.9.1

Improvements

• [rails] use direct instrumentation instead of Rails built-in instrumentation avoiding a level of indirection (#235)
• [core] remove debug logging when the Pin instance is retrieved, removing the time spent in this critical path (#233)
• [core] add an exponential back-off to our flushing strategy so that the thread will flush less often if the Trace Agent is not available (#239)
• [core] remove duplicate filter on Trace#trace (#234)

Bugfixes

• [resque] safe-guard if the Pin is nil (#223, #242 -- thanks @drewbailey)
• [resque] cleanup the current Context after the process fork to avoid useless Copy-on-Write (#231)

Read the full changeset

0.9.0

New Integrations

• Faraday with Distributed Tracing support (#185, #212, docs)
• AWS SDK client (#179, docs)
• SuckerPunch (#194, docs)
• MongoDB Driver (#177, docs)
• Dalli (#196, docs)
• Resque (#204, docs)

Improvements

• [core] add a Processing Pipeline to filter or update spans before they're flushed (#214, docs)
• [transport] services are flushed in the same loop of traces, made exception if the Hash is empty; the timeout is set every second like before (#201)
• [core] add Datadog.tracer.shutdown! method to force sending traces before exiting the process; it's useful when your process exits before the flushing thread finishes to send data to the Datadog Agent (#195)

Bugfixes

• [rails] the template instrumentation works even if unexpected errors are raised; before this led to incomplete / unsent traces (#191)
• [rails] improve safety of ActionController instrumentation, making it robust when Rails signals are not executed in the expected order (#197)
• [rails] use custom signals for cache instrumentation (#198)
• [rails] add the TraceMiddleware immediately instead of relying in another initializer (#208 -- thanks @Ferdy89 )

Read the full changeset

0.8.2

Minor changes

• Enable access to the context provider (#183)

Bugfixes

• Fix possible leaks across multiple requests, when all spans are not properly closed (#184)
• Fix enconding for error messages (#181)
• Fix documentation discrepancy (#180 -- thanks @knappe)

Read the full changeset

0.8.1

Minor changes

• Added an opt-out option to not instrument Rails if one wishes to not use it, with the help of @ejholmes (#156, #173)

Bugfixes

• fixed stack traces on Rails errors, with the help of @Ferdy89 (#170, #110)
• fixed instrumentation problem when ActiveRecord was not here (#166, contributed by @driv3r)
• documentation updates (#138, #164, contributed by @whithajess)

Read the full changeset

0.8.0

New features

• Experimental distributed tracing support. Distributed tracing allows you to connect your traces from different applications, and possibly different languages. This feature is disabled by default. Libraries supporting this are Rack and Net/HTTP (#141, #151 - docs, contributed by @whithajess, @ejholmes and @cabello)

Major changes

• Introduction of a context object, this is a deep refactoring which enables better future support of async behavior, and makes the Ruby library more similar to the Python implementation (#145)

Minor changes

• Jruby is back in our CI chain (#129)
• library is now tested against Ruby 2.4.1 (#159)
• Rails 3.0 support. As this is a deprecated Rails version, our support for this is "best effort". (#154, #157, contributed by @dorner)
• Adds a PID a as tag to root spans, so that you can know which process the trace belongs to (#147).

Breaking changes

• Span constructor does not set the start_time any more. The start time is now set when calling start_span from the Tracer. This should not have any impact if you either use Tracer.trace or use out-of-the box integrations. The only side-effect is if you were working on a custom integration.
• the finish_at method of the Span object, which was marked as deprecated in 0.7.2, has been completely removed. It can be safely replaced with finish, which takes an optional argument.

For example:

span.finish_at(timestamp)

becomes:

span.finish(timestamp)

And if you don't need to give a specific timestamp, this defaults to now:

span.finish()

Bugfixes

• documentation updates (#136, #144, #146, #152, with the help of @emaiax, @cabello and @tobypinder)
• don't treat http 4xx return codes as errors (#103. #162, with the help of @sj26)
• catch Exception and not only StandardError when it makes sense to do so (#120, #158, #160, suggested by @Ferdy89).
• limit span and trace IDs to 63 bits (#161)

Special thanks to all contributors, who made this release possible.

Read the full changeset

0.7.2

Bugfixes

• [tracer] ensuring submitted spans belong to the same trace, this fixes a possible data loss with message dropping trace reason: trace id mismatch when using threads (#133)
• [doc] added missing comma for Rack example, thanks @elyara for fixing this (#130)

Read the full changeset

0.7.1

New features

• it's now possible to configure Sidekiq service on a worker level, contributed by @krasnoukhov (#124, #125)

Minor changes

• a header X-Datadog-Trace-Count is now sent when posting payloads to agent, it contains the number of traces in the payload (#123)

Read the full changeset