Node.js agent release notesRSS

Features

• Added default support for config files with a 'cjs' extension (newrelic.cjs) in addition to newrelic.js.

  Thank you to @Maddemacher for the contribution!

• Added ability to specify a custom config file name with the NEW_RELIC_CONFIG_FILENAME environment variable.

  Thank you to @Maddemacher for the contribution!

Improvements

• Bumped @newrelic/test-utilities to ^5.1.0.

• Replaced deprecated util.isArray with Array.isArray.

• Removed unused listenerCount method on Shim.

• Properly bootstraped husky as a prepare script.

• Removed commented-out console log from fastify instrumentation.

Bug Fixes

• Fixed issue when using the 'new_promise_tracking' feature flag where segment mapping may not get cleaned up for promises which never resolve but have all references removed (and thus get cleaned up by GC).

  Adds segment cleanup on 'destroy' when using 'new_promise_tracking' feature flag in addition to the existing 'promiseResolve' hook. Unfortunately, preventing leaks for this edge-case does come with additional overhead due to adding another hook. Memory gains from feature flag usage should still be worth the trade-off and reduced garbage collection may offset perf/CPU impacts or event still result in net gain, depending on the application.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

Features

• Added flaky_code and success_delay_ms handling of flaky grpc connections to infinite tracing.

• Logged all New Relic metadata env vars at startup.

Improvements

• Updated third party notices and manifest for husky and lint-staged.

• Updated redis versioned tests to use unique DB indexes per file to avoid collisions and flushing of in-progress tests.

• Pinned hapi 17 versioned tests to only minor/patch versions within 17.x.

• Bumped timeout for redis versioned tests.

• Wired up husky + lint staged to execute linting on all changed files in pre-commit hook.

• Added resources to README to highlight external modules that customers should be aware of and possibly use for their applications.

• Handled a proxy misconfiguration of collector and log an actionable warning message.

• Fixed images for improved reader experience.

  Thank you to @henryjw for the contribution.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

Fixes

• Fixed issue with new_promise_tracking feature flag functionality where segments for ended transactions would get propagated in certain cases by promises that had no continuations scheduled (via await or manually).

  If you are experiencing high overhead levels with your promise usage and the agent attached, we recommend testing your application with new_promise_tracking set to true to see if overhead is reduced. You'll also want to verify your data is still being captured correctly in case it falls into a known or unknown limitation of this approach.

  NOTE: With this new functionality turned on, chaining of promise continuations onto an already resolved promise across an async hop (scheduled timer) will result in state-loss. This is a less-common use-case but worth considering with your applications.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

New Features

• Added new feature-flag 'new_promise_tracking' which enables cleaning up of segment references on native promise resolve instead of destroy. Includes usage of async-await. This can be enabled via feature_flag: { new_promise_tracking: true } in the config file or NEW_RELIC_FEATURE_FLAG_NEW_PROMISE_TRACKING=1 in your ENV vars.

  Applications with heavy promise usage or high-throughput applications with some promise usage should see moderate to high reduction in memory usage and may see a slight reduction in CPU usage. A bump in throughput may also be noticed in some cases. Results will vary by application.

  If you are experiencing high overhead levels with your promise usage and the agent attached, we recommend testing your application with 'new_promise_tracking' set to true to see if overhead is reduced. You'll also want to verify your data is still being captured correctly in case it falls into a known or unknown limitation of this approach. NOTE: chaining of promise continuations onto an already resolved promise across an async hop (scheduled timer) will result in state-loss with this new functionality turned on. This is a less-common use-case but worth considering with your applications.

Fixes

• Fixed memory leak introduced when Infinite Tracing is enabled.

  When Infinite Tracing endpoints reconnected they would instantiate a new gRPC client prior to calling client.recordSpan(). It appears several objects created by grpc-js (ChannelImplementation and child objects, promises, etc.) are held in memory indefinitely due to scheduled timers even when the client is no-longer referenced and the associated stream closed. We now avoid this situation by only creating the client once and then reusing it to establish new stream connections.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

Improvements

• Bumped @grpc/grpc-js to ^1.2.11.

• Bumped @grpc/proto-loader to ^0.5.6.

Fixes

• Fixed issue where capturing axios request errors could result in a memory leak.

  The agent now clears error references on transaction end, which are not used for later processing. Errors returned from 'axios' requests contain a reference to the request object which deeper down has a handle to a promise in handleRequestError. The TraceSegment associated with that promise has a handle to the transaction, which through the error capture ultimately kept the promise in memory and prevented it from being destroyed to free-up the TraceSegment from the segment map. This change also has the benefit of freeing up some memory early for transactions held onto for transaction traces.

• Agent no longer propagates segments for promises via async-hooks when the transaction associated with the parentSegment has ended.

  This change reduces the amount of context tracking work needed for certain rare edge-case scenarios involving promises.

• Added active transaction check to wrappedResEnd to prevent unecessary work for ended transactions in the case of multiple Response.prototype.end() invocations.

• Dev-only sub-dependency bump of 'y18n' to clear npm audit warnings.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

Improvements

• The NEW_RELIC_NO_CONFIG_FILE environment variable is no longer needed to run the agent without a configuration file.

  • If a configuration file is used with agent configuration environment variables, the environment variables will override the corresponding configuration file settings.

• Added feature flag to allow disabling of certificate bundle usage.

  Deprecation Warning: The certificate bundle included by New Relic will be disabled by default and then fully removed in later major versions. We recommend testing with the certificate_bundle feature flag set to false to determine if you will need to modify your environment or setup your own appropriate bundle. Example configuration: feature_flag: { certificate_bundle: false }.

• Set distributed tracing to enabled in the newrelic.js template configuration file supplied with the agent.

• Added module root to shim.require() logging to aid debugging.

• Migrated from .npmignore to 'files' list in package.json to control which files are packaged.

  Thank you to @JamesPeiris for the initial nudge via PR to move in this direction.

• Converted remaining collector unit tests to use tap API.

• Added linting to scripts in /bin folder.

  Linting rules added are slightly more permissive than production rules and allow full ecma 8.

• Added new developer documentation to /docs folder.

  This information is ported over from private GHE wiki used prior to going open source. S/O @astorm for original versions of the function wrapping and module instrumentation docs.

Fixes

• Fixed bug where applications with multiple names on a dynamically named host (UUID like) would have instances consolidated, losing per-host breakdowns.

  Removed 'host' from agent 'identifier' override to prevent server safety mechanism from kicking in. Host will still be used to identify unique agent instances, so was unnecessary to include as part of the identifier. This also resulted in additional processing overhead on the backend. The identifier override is still kept in place with multiple application names to continue to allow uniquely identifying instances on the same host with multiple application names where the first name may be identical. For example app_name['myName', 'unique1'] and app_name['myName', 'unique2']. These names would consolidate down into a single instance on the same host without the identifier override.

• Fixed bug where truncated http (external) or datastore segments would generate generic spans instead of appropriate http or datastore spans.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

Improvements

• Removed requirement to configure application name when running in AWS Lambda (serverless mode).

  Currently New Relic doesn't leverage the application name for Lambda invocations. The agent now defaults the application name in serverless mode to remove the requirement of end-user configuration while handling cases if it were to be leveraged in the future.

• Improved clarity of logging between 'no log file' or disabled agent startup issues.

  • Logs no-config file error to initialized logger (stdout) in addition to existing console.error() logging.
  • Adds specific message to no config file separate from being disabled.

• Stopped binding/propagating segments via setImmediate for ended transactions.

• Bumped @grpc/grpc-js to ^1.2.7.

• Added support metric to be able to track usage of cert bundle via usage of custom certificates.

Fixes

• Fixed bug where agent would attempt to call the 'preconnect' endpoint on the redirect host returned by the previous 'preconnect' call when reconnecting to the New Relic servers.

  The 'preconnect' calls now always use the original agent configuration value. Subsequent endpoints (connect, harvest endpoints, etc.) will continue to leverage the new redirect host value returned by 'preconnect.' The original config values are no longer overridden.

• Fixed issue where a call to transaction.acceptDistributedTraceHeaders would throw an error when the headers parameter is a string.

• Removed index-bad-config test which tested a no-longer possible use-case.

• Removed license-key test logic from serverless-harvest test.

  Serverless mode does not require a license key as data transfer is handled by the integration.

• Removed aws-sdk versioned test filtering.

• Removed unused Travis CI scripts.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

Improvements

• Bumped @newrelic/aws-sdk to ^3.1.0.

Fixes

• Fixed bug where the agent failed to reconnect to Infinite Tracing gRPC streams on Status OK at higher log levels.

  • Node streams require all data be consumed for the end/status events to fire. We were only reading data at lower log levels where we'd use/log the data. This resulted in a failure to reconnect and 'ERR_STREAM_WRITE_AFTER_END' errors. The agent now always listens to the 'data' event, even if not logging, and will also reconnect (with 15 second delay) on any 'ERR_STREAM_WRITE_AFTER_END' error.

• Removed initial harvest send() call on streaming span event aggregator to prevent warning in logs.

Support statement:

• New Relic recommends that you upgrade the agent regularly to ensure that you're getting the latest features and performance benefits. Additionally, older releases will no longer be supported when they reach end-of-life.

Improvements

• Added workflow for publishing to npm when a v* tag is pushed to the repo.
• Fixes resolveMx test by using example.com for a valid exchange.

Fixes

• Upgrade @grpc/grpc-js to v1.2.5 to fix non-propagation of internal http2 errors Now allows minor and patch auto-updates.