Improvements
- Added regression test for promise instrumentation and stack overflows.
Bug fixes
Fixed naming bug in Restify instrumentation regarding parameters to
next
.The instrumentation previously considered any truthy value passed to
next
to be an error. It is possible to pass a string or boolean tonext
in Restify to control further routing of the request. This would cause the middleware's mounting path to be erroneously appended to the transaction name.Fixed access to
bluebird.coroutine.addYieldHandler
.This was accidentally not copied by our instrumentation making access to the function fail. This has been resolved and tests expanded to ensure no other properties were missed.
New features
Added
transaction_tracer.hide_internals
configuration.This configuration controls the enumerability of the internal properties of the agent. Making these properties non-enumerable can have an impact on the performance of the agent. Disabling this option may decrease agent overhead.
Improvements
Refactored promise instrumentation.
This new instrumentation is far more performant than the previous and maintains a more sensible trace structure under a wider range of sequences.
Added concurrent environment scanning, limited to 2 reads at a time.
This improves the performance of dependency scanning at agent startup, allowing the agent to connect to our services more quickly.
Refactored instrumentation tests to run against wide range of module versions.
Instrumentation tests will be run against all supported major versions of every instrumented module. For releases, we will test against every supported minor version of the modules. This vastly improves our test coverage and should reduce the instances of regressions for specific versions of modules.
Added tests for all of bluebird's promise methods.
These tests ensure that we 100% instrument bluebird. Some gaps in instrumentation were found and fixed. Anyone using bluebird should upgrade.
Bug fixes
Fixed a crashing error in the hapi instrumentation.
When recording the execution of an extension listening to a server event (for example, 'onPreStart') the agent would crash due to the lack of a
raw
property on the first argument passed to the extension handler. The agent now checks the event before wrapping the extension handler and checks for the existence of theraw
property before attempting to dereference off of it.Fixed an incompatibility with the npm module
mimic-response
.The agent's HTTP instrumentation previously did not play well with the way
mimic-response
copied properties from anhttp.IncomingMessage
. This caused modules that relied on that process, such asgot
, to hang.Fixed naming rule testing tool to use same url scrubbing as the agent itself.
New features
Added hapi v17 instrumentation
Hapi v17 added support for promise-based middleware which broke transaction tracking in the agent. This caused issues in naming, as the agent will name the transaction after the path to the middleware that responded to a request.
Added instrumentation for
vision@5
Due to the way
vision
is mounted to the hapi server when using hapi v17.x, the agent's instrumentation would not pick up on the middleware being mounted. This new instrumentation now correctly times rendering done in thevision
middleware.Added
unwrapOnce
method to shim objectThis new method can be used to unwrap a single layer of instrumentation.
unwrapOnce
is useful in cases where multiple instrumentations wrap the same method and unwrapping of the top level is required.Added
isErrorWare
checks aroundnameState.appendPath
/nameState.popPath
calls to avoid doubling up paths in transaction namesPreviously, the agent would append its transaction name with the path fragment where an error handler middleware was mounted. The extraneous path fragment will now be omitted, and the transaction will be named properly after the middleware that threw the error.
Added support for
pg
6 on Node 5 or higher
Improvements
- Added
parent
property to webframework-shim segment description - Refactored existing hapi instrumentation for different
server.ext()
invocations - Refactored webframework-shim
_recordMiddleware
to construct different segment descriptions for callback- or promise-based middleware - Updated
node-postgres@^6
versioned tests to avoid deprecation warning on direct moduleconnect
andend
calls - Fixed running domain tests on Node 9.3.0.
- Improved logging for CAT headers and transaction name-state management.
- All
json-safe-stringify
calls now wrapped intry/catch
- Removed
lib/util/safe-json
Bug fixes
- Fixed creating supportability metric when mysql2 goes uninstrumented.
- Added a
segmentStack.pop
to the middlewareafter
in cases when an error is caught and there is no next handler - Fixed determining parents for middleware segments when transaction state is lost and reinstated
- Added check to
_recordMiddleware
to avoid prepending a slash if originalroute
is an array - Changed logic in http instrumentation to attach
response.status
to the transaction as a string - Updated
startWebTransaction
andstartBackgroundTransaction
to add nested transactions as segments to parent transactions
Notes
- Added Peter Svetlichny to the contributors list!
Improvements
- Optimized
NameState#getPath
. - Optimized
shim.record
. - Optimized
shim.recordMiddleware
. - Upgraded
eslint
to v4.
Bug fixes
- Fixed parsing SQL for queries containing newlines.
Improvements
- Added promise benchmarks to test non-async_hooks instrumentation.
- Added logging for external calls made outside of a transaction.
- Added logging for when
unhandledRejection
is noticed. - Improved performance of creating and merging metrics.
- Improved performance of
tracer.bindFunction
. - Removed try-catch around internal property setting on older versions of Node.
Bug fixes
- Moved
require
calls for vendor metadata to module-level.
New features
Instrumentation will now only modify the arity of wrapped functions when needed.
This can be controlled with the
matchArity
property on aWrapSpec
. Disabling arity matching has a significant, positive impact on the performance of instrumentation.
Improvements
Added benchmarks for shimmer methods.
Pinned hapi tests at v16 due to incompatibility in hapi v17 with Node.js versions <8.
The agent's parsed queries will now only hold onto the stack that the query was made at, instead of an error object instance.
Previously, the parsed query objects would hold onto an error instance, which would in turn hold onto references to all the functions in the stack when the error was created. This could cause memory issues if the functions were holding onto references to other pieces of data.
Bug fixes
Revert wrapping of
https
for Node^8.9.1
.The original cause for this problem was reverted by Node.
Improvements
- Added Node 9 to test suite.
- Removed problematic tests for ancient version of Hapi (7.1).
- Document purpose of
throw
in tracer to prevent developer confusion. - Added script for running agent micro benchmarks.
- Added benchmarks for all the
Shim
andTracer
methods.
Bug fixes
- Fixed a bug with Node >=8.9 that prevented https externals from being recorded.
Improvements
- Agent will attempt to reconnect to the collector forever after backing off to 5 minute delays.
Bug fixes
- Refactored environment scan to improve startup time and fix cyclical symlink resolving.
New features
- The agent will now support the Node 8
await
keyword by default.
Improvements
- Added benchmark for performance of function wrapping.
- Added Garbage Collection information to async_hooks benchmark.
- Improved trace-level logging for capturing queries.
- Added cases for omitting the agent with and without async hooks to the async hooks microbenchmark.
Bug fixes
- Pinned version of Mocha to 3.x due to the incompatibility of Mocha v4 and Node v0.10 and v0.12.
Bug fixes
Hapi handlers will now preserve the defaults associated with them.
Previously when wrapping handlers, the agent would drop the associated defaults on the ground, these are now properly forwarded to the wrapper. Big thanks to Sean Parmelee (@seanparmelee) for finding the root cause of this bug and reporting it!
Fixed starting the agent with an invalid process version.
Pinned
request
version for testing old versions of Node.Added tests for feature flags created at agent initialization.