Ruby agent release notesRSS

v9.2.2

Version 9.2.2 of the agent fixes a bug with the Transaction#finished? method.

• Bugfix: Transaction#finished? no longer throws a NoMethodError when initial_segment is nil

  This change adds a safe navigation operator to Transaction#finished? to prevent NoMethodErrors when a transaction does not have any segments. Our thanks goes to @JulienDefrance for reporting this issue. PR#1983

v9.2.1

Version 9.2.1 fixes a bug causing the agent to continue storing data on finished transactions, and a bug preventing errors from being expected.

• Bugfix: Finished transactions continue to store data on different threads

  Previously, when a new thread was spawned the agent would continue using the current transaction to record data on, even if this transaction had finished already in a different thread. Now the agent will only use the current transaction in the new thread if it is not yet finished. Thank you to @fcheung for reporting this bug and providing us with an extremely helpful reproduction to debug. PR#1969

• Bugfix: Expected Errors passed to notice_error are expected again

  A bug was introduced in 9.1.0 that caused to agent not to mark errors as expected if the error was passed in to notice_error using the expected: true parameter. This has been fixed and errors will now be marked as expected, as expected. Thank you very much to @eiskrenkov for finding this bug and contributing a fix for it! PR#1954

v9.2.0

Version 9.2.0 of the agent introduces some performance improvements for working with high numbers of nested actions, and deprecates instrumentation for the memcached and memcache-client gems (with dalli still being supported).

• Feature: Enhance performance for handling high numbers of nested actions

  With Issue#1910, community members @parkerfinch and @travisbell informed us of some CPU spikes and process hangs seen only when using the agent's thread instrumentation, which was enabled by default with v9.0. When thread instrumentation is enabled, instrumented actions taking place within threads are seen and reported on by the agent whereas they would have previously gone unnoticed. This is a great improvement to the agent's usefulness in an async context, and also makes it easier for higher numbers of nested actions to be observed. For example, if an instrumented background job framework (Sidekiq, Resque) kicks off a job that the agent notices and then that job in turn performs actions such as database queries that the agent also instruments, nested actions are seen. However, with very high (10,000+) numbers of actions nested within a single instrumented outer action, the agent would struggle to efficiently crunch through all of the collected data at the time when the outer action finished. The agent should now be much more efficient when any observed action with lots of nested actions is finished. Our performance testing was conducted with hundreds of thousands of nested actions taking place, and we hope that the benefits of thread tracing can now be enjoyed without any drawbacks. Thanks very much @parkerfinch and @travisbell! PR#1927

• Feature: Deprecate memcached and memcache-client instrumentation

  Instrumentation for the memcached and memcache-client libraries is deprecated and will be removed during the next major release.

v9.1.0

Version 9.1.0 of the agent delivers support for two new errors inbox features: error fingerprinting and user tracking, identifies the Amazon Timestream data store, removes Distributed Tracing warnings from agent logs when using Sidekiq, fixes bugs, and is tested against the recently released JRuby 9.4.2.0.

• Feature: Error fingerprinting - supply your own errors inbox group names

  Are your error occurrences grouped poorly? Set your own error fingerprint via a callback function. A new set_error_group_callback public API method has been added that will accept a user defined proc. The proc will be invoked for each noticed error and whenever it returns a string, that string will be used as the error group name for the error and will take precedence over any server-side grouping that takes place with the New Relic errors inbox. This gives users much greater control over the grouping of their errors.

  The customer defined proc will be expected to receive exactly one input argument, a hash. The hash contains the following:

  Key	Value
  :error	The Ruby error class instance. Offers #class, #message, and #backtrace
  :customAttributes	Any customer defined custom attributes for the current transaction
  :'request.uri'	The current request URI if available
  :'http.statusCode'	The HTTP status code (200, 404, etc.) if available
  :'http.method'	The HTTP method (GET, PUT, etc.) if available
  :'error.expected'	Whether (true) or not (false) the error was expected
  :'options'	The options hash passed to NewRelic::Agent.notice_error

  The callback only needs to be set once per initialization of the New Relic agent.

  Example usage:

  proc = proc { |hash| "Access" if hash[:'http.statusCode'] == 401 }
  NewRelic::Agent.set_error_group_callback(proc)

  Copia

• Feature: User tracking - associate errors with a user id

  You can now see the number of users impacted by an error group. Identify the end user with a new set_user_id public API method that will accept a string representation of a user id and associate that user id with the current transaction. Transactions and errors will then have a new enduser.id agent attribute associated with them. This will allow agent users to tag transactions and errors as belonging to given user ids in support of greater filtering and alerting capabilities.

• Identify Amazon Timestream when the amazon_timestream AR adapter is used

  When the agent sees the activerecord-amazon-timestream-adapter gem being used, it will now identify the data store as "Timestream". Thanks very much to @wagner for contributing this enhancement! PR#1872

• Bugfix: Remove Distributed Tracing related warnings from agent logs when headers are not present in Sidekiq

  Previously, the agent would log a warning to newrelic_agent.log every time it attempted to accept empty Distributed Tracing headers from Sidekiq jobs which could result in an excessive number of warnings. Now the agent will no longer create these warnings when using Sidekiq. PR#1834

• Bugfix: Log request headers in debug-level logs instead of human-readable Objects

  Previously, the agent sometimes received children of the NewRelic::Agent::HTTPClients::AbstractRequest class as an argument when NewRelic::Agent::Transaction::DistributedTracers#log_request_headers was called. This caused debug-level log messages that print the request headers to show human-readable Objects (ex. #<NewRelic::Agent::HTTPClients::HTTPClientRequest:0x00007fd0dda983e0>) instead of the request headers. Now, the hash of the request headers should always be logged. PR#1839

• Bugfix: Fix undefined method controller_path logged in Action Controller Instrumentation

  Previously, the agent could log an error when trying to determine the metric name in the Action Controller instrumentation if the controller class did not respond to controller_path. This has been resolved and the agent will no longer call this method unless the class responds to it. Thank you to @gsar for letting us know about this issue. PR#1844

• Bugfix: Fix Transaction#finish exception and decrease log level for related warning during async transactions

  Previously, the agent would raise a non-fatal error when a segment without a parent was unfinished when the transaction completed. This error was raised while constructing a warn-level log message. Now that Thread instrumentation is on by default, this log message emits more frequently and is less concerning. In cases where we see a Thread, Fiber, or concurrent-ruby segment in a transaction, the message will be degraded to a debug-level. Thanks to @NielsKSchjoedt for creating the issue and @boomer196 for testing solutions. PR#1876

• CI: Target JRuby 9.4.2.0

  The agent is now actively being tested against JRuby 9.4.2.0. NOTE that this release does not contain any non-CI related changes for JRuby. Old agent versions are still expected to work with newer JRubies and the newest agent version is still expected to work with older JRubies.

v9.0.0

Version 9.0.0 of the agent removes several deprecated configuration options and API methods, enables Thread tracing by default, adds Fiber instrumentation, removes support for Ruby versions 2.2 and 2.3, removes instrumentation for several deprecated gems, changes how the API method set_transaction_name works, and updates rails_defer_initialization to be an environment variable only configuration option.

• Remove deprecated configuration options

  The following configuration options have been removed and will no longer work. Please update all configs to use the replacements listed below. PR#1782

• Enable Thread instrumentation by default

  The configuration option instrumentation.thread.tracing is now enabled by default. This configuration allows the agent to properly monitor code occurring inside threads. In Ruby agent 9.0, instrumented code within threads will be recorded and associated with the current transaction when the thread was created.

  This may be a breaking change if you are currently using custom thread instrumentation. New transactions inside of threads will no longer be started if one already exists. PR#1767

• Add Fiber instrumentation

  Fiber instances are now automatically instrumented similarly to Thread instances. This can be configured using instrumentation.fiber. PR#1802

• Removed support for Ruby 2.2 and 2.3

  Ruby 2.2 and 2.3 are no longer supported by the Ruby agent. To continue using the latest Ruby Agent version, please update to Ruby 2.4.0 or above. PR#1778

• Removed deprecated instrumentation

  Instrumentation for the following gems had been previously deprecated and has now been removed. PR#1788

  • Acts As Solr
  • Authlogic
  • DataMapper
  • Rainbows
  • Sunspot

  Versions of the following technologies had been previously deprecated and are no longer supported.

  • Passenger: 2.2.x - 4.0.x
  • Puma: 2.0.x
  • Grape: 0.2.0
  • Padrino: 0.14.x
  • Rails: 3.2.x
  • Sinatra: 1.4.x, 1.5.x
  • Mongo: 1.8.x - 2.3.x
  • Sequel: 3.37.x, 4.0.x
  • Delayed_Job: 2.0.x - 4.0.x
  • Sidekiq: 4.2.x
  • Excon: below 0.55.0
  • HttpClient: 2.2.0 - 2.8.0
  • HttpRb: 0.9.9 - 2.2.1
  • Typhoeus: 0.5.3 - 1.2.x
  • Bunny: 2.0.x - 2.6.x
  • ActiveMerchant: 1.25.0 - 1.64.x

• Updated API method set_transaction_name

  When the method NewRelic::Agent.set_transaction_name is called, it will now always change the name and category of the currently running transaction to what is passed into the method. This is a change from previous agent versions.

  Previously, if set_transaction_name was called with a new transaction name and a new category that did not match the category already assigned to a transaction, neither the new name nor category would be saved to the transaction. If this method is being called in a situation in which it was previously ignored due to category differences, this will now change the name and category of the transaction. PR#1797

• Removed API method: NewRelic::Agent.disable_transaction_tracing

  The deprecated API method NewRelic::Agent.disable_transaction_tracing has been removed. Instead use either NewRelic::Agent#ignore_transaction to disable the recording of the current transaction or NewRelic::Agent.disable_all_tracing to yield a block without collecting any metrics or traces in any of the subsequent calls. PR#1792

• Renamed ActiveJob metrics

  Previously, ActiveJob was categorized as a message broker, which is inaccurate. We've updated the naming of ActiveJob traces from leading with MessageBroker/ActiveJob to simply leading with ActiveJob. PR#1811

• Code cleanup

  Thank you to community member @esquith for contributing some cleanup of orphaned constants in our code base. PR#1793 PR#1794 PR#1808

  Community member @fchatterji helped standardize how we reference NewRelic throughout our codebase PR#1795 and updated our README's community header PR#1815. Thanks fchatterji!

• Bugfix: Allow rails initialization to be deferred by environment variable

  The Ruby agent may force some Rails libraries to load on agent initialization, preventing some settings defined in config/initializers from being applied. Changing the initialization process to run after config/initializers, however, may break the configuration for other gems (ex. Roadie Rails).

  For those having troubles with agent initialization and Rails initializers, you can now pass the environment variable NEW_RELIC_DEFER_RAILS_INITIALIZATION=true to make the agent initialize after config/initializers are run. This config option can only be set using an environment variable and can't be set using YAML. PR#1791

  Thanks to @jdelStrother for bringing this issue to our attention and testing our fixes along the way. Issue#662

v8.16.0

Version 8.16.0 introduces more Ruby on Rails instrumentation (especially for Rails 6 and 7) for various Action*/Active* libraries whose actions produce Active Support notifications events.

• Add Various Ruby on Rails Library Instrumentations

  New instrumentation is now automatically provided by several Action*/Active* libaries that generate Active Support notifications. With each Ruby on Rails release, new the Rails libraries add new events and sometimes existing events have their payload parameters updated as well. The New Relic Ruby agent will now automatically process more of these events and parameters with New Relic segments created for each event. At a minimum, each segment gives timing information for the event. In several cases, all non-sensitive event payload parameters are also passed along in the segment.

  The agent now newly supports or has updated support for the following libraries:

  • Action Cable (for WebSockets) PR#1749
  • Action Controller (for the 'C' in MVC) PR#1744
  • Action Mailbox (for sending mail) PR#1740
  • Action Mailer (for routing mail) PR#1740
  • Active Job (for background jobs) PR#1742
  • Active Support (for caching operations) PR#1742

  The instrumentations for each of these libaries are all enabled by default, but can be independently disabled via configuration by using the following parameters:

  Configuration name	Default	Behavior
  disable_action_cable	false	If true, disables Action Cable instrumentation.
  disable_action_controller	false	If true, disables Action Controller instrumentation.
  disable_action_mailbox	false	If true, disables Action Mailbox instrumentation.
  disable_action_mailer	false	If true, disables Action Mailer instrumentation.
  disable_activejob	false	If true, disables Active Job instrumentation.
  disable_active_support	false	If true, disables Active Support instrumentation.

v8.15.0

Version 8.15.0 of the agent confirms compatibility with Ruby 3.2.0, adds instrumentation for concurrent-ruby, and confirms Sinatra 3 compatibility with Padrino 0.15.2. It also enables batching and compression for Infinite Tracing.

• Add Support for Ruby 3.2.0

  Following the 3.2.0 release of Ruby, the New Relic Ruby agent has confirmed compatibility with and now supports the official release of Ruby 3.2.0. PR#1715

• Add instrumentation for concurrent-ruby

  The agent now instruments the concurrent-ruby gem for versions 1.1.5 and above. This instrumentation provides visibility inside blocks passed to many of the gem's API methods, including Concurrent::Promises.future and Concurrent::Future.execute.

• Infinite Tracing: Use batching and compression

  For Infinite Tracing, which Ruby applications can leverage with the newrelic-infinite_tracing gem, payloads will now be batched and compressed to signficantly decrease the amount of outbound network traffic. PR#1723

• Add Support for Padrino 0.15.2 and Sinatra 3

  We've added testing to confirm Padrino 0.15.2 and Sinatra 3 are compatible with the Ruby agent. Thank you @nesquena for letting us know 0.15.2 was ready! PR#1712

v8.14.0

Version 8.14.0 of the agent restores desired Capistrano-based changelog lookup functionality when a deployment is performed, speeds up GUID generation, delivers support for instrumenting Rails custom event notifications, fixes potential compatibility issues with the RedisClient gem, and fixes bugs related to initialization in Rails.

• Deployment Recipe: Restore desired Capistrano-based changelog lookup behavior

  The New Relic Ruby agent offers a Capistrano recipe for recording app deployments. The recipe code was significantly cleaned up with PR#1498 which inadvertently changed the way the recipe handles the changelog for a deployment. Community member @arthurwozniak spotted and corrected this change in order to restore the desired changelog lookup functionality while retaining all of the previous cleanup. Thank you very much for your contribution, @arthurwozniak! PR#1653

• Speed up GUID generation

  The agent leverages random numbers in its GUID (globally unique identifier) generation and would previously always freshly calculate the result of 16^16 or 32^32 before generating a random number. Given that those 16^16 and 32^32 operations are expected, it makes sense to calculate their results up front and store them in constants to be referred to later. Doing so has resulted in a performance gain for the generation of GUIDs. Many thanks to @tungmq for contributing this optimisation and the benchmarks to support it! PR#1693

• Support for Rails ActiveSupport::Notifications for custom events

  When the new active_support_custom_events_names configuration parameter is set equal to an array of custom event names to subscribe to, the agent will now subscribe to each of the names specified and report instrumentation for the events when they take place. Creating custom events is simple and now reporting instrumentation for them to New Relic is simple as well. PR#1659

• Bugfix: Support older versions of the RedisClient gem, handle unknown Redis database index

  With version 8.13.0 of the agent, support was added for the redis-rb gem v5+ and the new RedisClient gem. With versions of RedisClient older than v0.11, the agent could cause the monitored application to crash when attempting to determine the Redis database index. Version 8.14.0 adds two related improvements. Firstly, support for RedisClient versions older than v0.11 has been added to get at the database index value. Secondly, the agent will no longer crash or impact the monitored application in the event that the database index cannot be obtained. Thank you very much to our community members @mbsmartee and @patatepartie for bringing this issue to our attention, for helping us determine how to best reproduce it, and for testing out the update. We appreciate your help! Issue#1650 PR#1673

• Bugfix: Defer agent startup in Rails until after application-defined initializers have run

  In Rails, the agent previously loaded before any application-defined initializers. This allowed initializers to reference the add_method_tracer API. However, this had the side-effect of forcing some framework libraries to load before initializers ran, preventing any configuration values related to these libraries from being applied. This fix provides an option to split initialization into two parts: load add_method_tracer before application-defined initializers and start the agent after application-defined initializers. This may cause other initializers to behave differently.

  If you'd like to use this feature, set defer_rails_initialization to true. It is false by default, but may become true by default in a future release.

  Furthermore, our Action View instrumentation was missing an ActiveSupport.on_load block around the code that loads our instrumentation.

  Thank you @jdelStrother for bringing this to our attention and collaborating with us on a fix. PR#1658

  Unfortunately, this bugfix is unreachable as written because the configuration value used to access the bugfix won't be applied until after initialization. Follow along for updates at Issue#662.

v8.13.1

Version 8.13.1 of the agent provides a bugfix for Redis v5.0 instrumentation.

• Fix NoMethodError when using Sidekiq v7.0 with Redis Client v0.11

  In some cases, the RedisClient object cannot directly access methods like db, port, or path. These methods are always available on the client.config object. This raised a NoMethodError in environments that used Sidekiq v7.0 and Redis Client v0.11. Thank you to fcheung and @stevenou for bringing this to our attention! Issue#1639