This release of the Python agent marks the official introduction of support for pypy. Instrumentation has also been updated to address issues arising from the release of Celery 3.1.
For a list of known issues with the Python agent see our online help article on the status of the Python agent.
- The agent has worked with and has already been used by customers with pypy for some time, but we have not officially acknowledged that we support pypy. We have now integrated pypy into our test procedures and now officially support this implementation of the Python language.
- Instrumentation for Celery has been updated to accommodate changes in Celery 3.1. The changes in Celery 3.1 resulted in no metrics being reported in prior versions of the agent.
- When collecting data on external web service calls, the agent now drops port 80/443 from the name of the host when used with the standard http/https protocol schemes. This ensures that a URL with or without the ports are seen as the same service in the external web services page in the UI.
- When using newrelic.agent.initialize() explicitly, if no arguments are provided, the values for the config file and environment arguments will now be read from the NEW_RELIC_CONFIG_FILE and NEW_RELIC_ENVIRONMENT arguments if specified.
- Flask instrumentation was not correctly mapping URLs related to a HTTP 404 response to a known framework or application handler function. Instead the web transaction was named after the URL, which could result in metric grouping issues if an application was hit with a large number of URLs which couldn't be mapped by the application. Such requests will now be mapped to flask.app:Flask.handle_http_exception.
- Pyramid instrumentation should no longer report as errors instances of exceptions derived from HTTPRedirection, raised to generate a HTTP redirect response.
- Improvements to the newrelic-admin wrapper script and agent bootstrapping procedure to better deal with a local sitecustomize.py file. Changes will also better handle the case where the newrelic-admin wrapper script was used around a Python script using a different Python installation than that which the newrelic package was installed in, avoiding possible errors when the wrapped script was run, due to the mismatch.
This release of the Python agent contains support for web applications and worker processes using 'gearman' for background task execution. The agent also switches to using the 'json' package from the standard library instead of a separate 'simplejson' package. This should result in potential reductions in base level memory usage by the agent and a reduction of CPU overhead for some environments.
For a list of known issues with the Python agent see our online help article on the status of the Python agent.
- Instrumentation now provided for the gearman package. The support covers use of both the gearman client and worker interfaces. In the case of workers, each task executed by the worker will be recorded as a background task and be displayed as such in the New Relic UI.
newrelic.agent.record_exception()can now be a callable in addition to being able to pass a sequence. When it is a callable, the callback will be called with the three arguments
tb, being the same values as returned by
sys.exc_info(). The callback should return
Trueif the exception is to be ignored.
Falseif the exception should never be ignored regardless of any other checks, and
Noneif subsequent checks and inbuilt rules should determine if the exception should be ignored. A callback would normally return either
- The bundled 'simplejson' package has now been removed and is no longer used. This was previously used due to the requirement to support Python 2.5, for which support was removed in version 2.0.0 of the agent. Instead of simplejson, the 'json' package contained in the Python standard library is now used instead. This should see the base level memory usage of the agent drop as a result. Further, in environments where a C compiler was not available and the C extension modules could not be compiled, there should now be a minimal reduction in the CPU overhead when the agent is uploading data to our backend. This is because the optimised C extension in the json package will always be available as part of the Python standard library and will always be able to be used.
- If a WSGI application was returning an iterable such as a generator, and an exception was raised when a specific part of the response content was yielded from the generator, the details of the exceptions were not being recorded.
- In a multithreaded web application where deferred module imports were being performed in secondary threads, agent registration could fail the first time due to concurrent changes made by the secondary threads to
sys.modules. Registration would succeed on a subsequent attempt. This bug was only occurring in Python 2 and was introduced in version 2.0.0 of the agent when Python 3 support was added.
- Custom parameters which were explicitly supplied to
newrelic.agent.record_exception()were being ignored and were not appearing in the error details page in the UI. Instead, only custom parameters added using
newrelic.agent.add_custom_parameters()against the web transaction or background task itself were being shown. This bug was introduced in version 1.13.0 of the agent when support for cross application tracing was added.
- The port number used in the URL for a web external call, was not being retained when the URL was added as a parameter against the web external node for transaction traces. The port would therefore not be displayed when drilling down into the details of a web external node in a transaction trace.
newrelic-admin validate-configcommand will now work for enterprise high security mode accounts, provided of course that no setting is otherwise specified in the agent configuration file which is in conflict with that mode. That is, features such as SSL would still need to be enabled by the
sslsetting and the
capture_paramssetting indicating whether URL query string parameters should be captured, also set to
- When using CherryPy, in addition to the explicit exception type 'NotFound' being ignored as an error, raising of a 'HTTPError' exception where the status is 404 will also be ignored. Similarly, instances of the 'HTTPError' exception will be ignored as an error when they are for a HTTP error in the 30X range.
This release of the Python agent is a bug fix release primarily to address a number of issues introduced in version 22.214.171.124 which have effected a small number of users.
This version of the agent is also the minimum recommended agent version required to be able to see percentiles and histogram charts. All reporting hosts for an application must be upgraded to the minimum agent version for the feature to be visible.
For a list of known issues with the Python agent see our online help article on the status of the Python agent.
- Fixed issues with instrumentation wrappers which caused failure of WSGI application integration for FASTCGI/SCGI/AJP using flup.
- Fixed issues with instrumentation wrappers which caused failure of wrappers for external web service calls in certain uses cases.
- Fixes issues with instrumentation wrappers which caused failure when queuing Celery tasks from within a web application using django-celery.
- When using the ErrorTrace context manager to capture details of exceptions within a certain context, None can now be passed for the transaction and an assertion failure will not be raised. This avoids the need to check explicitly to see if a web request is being monitored and avoid using ErrorTrace where there isn't.
This release of the Python agent is a hot fix version to address a Celery instrumentation issue introduced with version 126.96.36.199.
- Fixed bug with Celery instrumentation introduced with version 188.8.131.52 of the agent. The bug was causing a Python exception to be raised when the instrumentation was trying to derive the task name to use as the name of the background task.
This release of the Python agent is an incremental release and includes minor feature changes and enhancements.
- Use of HTTP proxies over SSL using the HTTP CONNECT tunnelling mechanism is now supported. Set the 'proxy_scheme' to 'http' to invoke this mode. This will become the default in a future version of the agent. If migrating from an older agent version and need to keep the existing behaviour going forward, set 'proxy_scheme' to 'https'. Alternatively, ensure you provide the scheme in the form of a URI as part of the 'proxy_host' setting.
- Added request queueing support to the instrumentation for the Tornado ASYNC framework. Request queueing times should start showing in the APM Overview page when a frontend web server such as Nginx, has been configured to add the appropriate headers. For more details see our documentation on setting up tracking of queueing time.
- The 'newrelic.agent.record_exception()' can now be called without actually passing it any exception details. In this case, the details of any current exception being handled will be used instead.
- Added a new mechanism for automatic discovery of third party instrumentation modules as an alternative to having to list them explicitly in the agent configuration file. Any such instrumentation module should register entry points under the 'newrelic.hooks' group in 'setup.py' for that package. The entries should be of the form 'target-module = instrumentation-module:function'. When the 'target-module' is imported, the function 'instrumentation-module:function' will be executed and passed the module. Any instrumentation module should then use functions provided by our agent API under 'newrelic.agent' to instrument the module as necessary.
- When using the CherryPy web framework, the NotFound, InternalRedirect and HTTPRedirect exceptions if raised are now ignored and not treated as errors.
- When using the Pyramid web framework, if a view could not be found to handle a request, it could cause the agents' Pyramid instrumentation to fail, causing an exception.
- When using the Pyramid web framework, the PredicateMismatch exception if raised when trying to resolve a URL to a view handler is now ignored and not treated as an error.
- Updated the instrumentation for the pywapi module to drop support for the now discontinued Google weather API.
- Overhauled the mechanisms used to wrap instrumentation around functions to be monitored. When the agent is installed, if the optional C extension module can be compiled and installed, then the new wrappers should have a reduced overhead compared to the pure Python versions of the wrappers otherwise used.
- Running thread profiles on coroutine (gevent/eventlet) based systems no longer cause a timeout in the UI. Although the UI now no longer times out, as explained in our status of the Python agent documentation, we still however do not generate any results when running thread profiling on coroutine based systems.
This release of the Python agent is a major new version and primarily focuses on the addition of support for Python 3.
- Python 3.3 or later is now supported. This includes support for popular web frameworks that have been ported to Python 3, such as CherryPy, Django, Flask, Pyramid and Tornado.
- Added new instrumentation to track as external service calls, requests made by the 'thrift' client module.
- Added new instrumentation to track memcache calls made by the 'bmemcached' client module.
- Python 2.5 is no longer supported. The minimum required Python version is now 2.6. If using Python 2.5, you should ensure that any requirement for the 'newrelic' module listed in a 'setup.py' or pip requirements file says 'newrelic<184.108.40.206'.
- Fixed broken instrumentation for 'httplib' module, which would cause an exception where the 'httplib' connect() method was being invoked via the Connection class type rather than an instance of the Connection class.
- Fixed issue where if the host system clock was wound backwards, then the response time of active requests could be set as being zero, resulting in a ZeroDivisionError when calculating thread utilization.
- Fixed issue where the use of SSL via a HTTP proxy by the agent to connect to the New Relic data collector to report data would fail.
- Fixed broken instrumentation for sqlite database client module which resulted in the executescript() API call failing.
- Add workaround for uWSGI issue where when using gevent mode of uWSGI, it would attempt to wait on all greenlets on process shutdown, instead of only on non daemon threads, specifically the request threads. The issue was causing processes to hang for 60 seconds on shutdown before the uWSGI master process killed the process. We are in discussion with uWSGI authors about a permanent fix to uWSGI.
This release of the Python agent includes a hot fix to address an issue whereby external web service calls made via HTTPS connections were failing. This issue was introduced with version 220.127.116.11 of the agent.
- The Cross Application Tracing feature added in version 18.104.22.168 of the agent was causing failure of external web service calls made over HTTPS connections when that call was made via the 'httplib' module. As the 'httplib' module is used internally in other HTTP client libraries such as the 'urllib', 'urllib2', 'urllib3' and 'requests' modules, they would also have been affected.
- Setting the 'cross_application_tracer.enabled' setting in the agent configuration file to 'false' to disable Cross Application Tracing would cause an exception when an external web service call was being made via the 'httplib' module.
This release of the Python agent adds support for Cross Application Tracing.
- Support for New Relic's Cross Application Tracing feature is now available for the Python agent. This feature enables correlation of transaction traces between different services in your stack. It works on external calls made between applications monitored by any New Relic agents (Java, .NET, Ruby and Python), which support this feature. At this time the Python agent supports this feature on outgoing calls when using the 'httplib' module or any other module which uses 'httplib', such as the 'urllib', 'urllib2', 'urllib3' and 'requests' modules. Any inbound calls from other agents which support the feature will also be handled.
- Added instrumentation for the weberror package. This provides visibility into potentially blocking operations such as sending of email for exceptions which are being reported.
- Added instrumentation for the umemcache package.
- Added instrumentation for the DBAPI2 compliant interface for the IBM DB2 ibm_db_dbi database client package.
- When running a WSGI application under the Tornado WSGI container, data reporting would stop if an unhandled exception managed to propagate back up to the Tornado WSGI container. This was due to a bug in Tornado in respect of its compliance with the WSGI specification. A workaround is provided to avoid the problems this caused to the Python agent. Tornado versions prior to 3.2 will still carry that bug however and we make no attempt to address the bug in Tornado.
- The Tornado instrumentation was causing template rendering to fail where a relative path was used to refer to the template and no template path had been specified for the RequestHandler instance or globally.
- The Tornado instrumentation was causing exceptions when using the @tornado.gen decorators under Tornado 2.X.
- When using Django, web transactions will now be named after individual view handlers when using class based views, rather than being named after the class itself.
- When using Pyramid, web transactions will now be named after individual view handlers when using view classes, rather than being named after the class itself.
- Celery instrumentation had stopped working correctly for Celery versions 2.5.3 through 2.5.5.
- No data was being reported where a monitored process was being shutdown within a couple of seconds of being started even if the agent was able to register. This could result in custom metrics in particular not being reported.
- SSL certificate validation was failing on older Debian systems due to the OpenSSL libraries not being able to process a couple of the certificates bundled with the Python agent. Those certificates were not required and have been removed.
- Calculation of a time string was failing on Windows due to using a strftime() formatter which only existed on UNIX systems.
This release of the Python agent removes the experimental label from our Tornado Web and Pyramid support, with support now being enabled by default.
For a list of known issues with the Python agent, see Status of the Python agent.
- The previously experimental support for the native Tornado Web API has been greatly expanded and is now enabled by default. This is in addition to the tornado.wsgi.WSGIContainer support we previously provided and includes the native Tornado Web API for constructing async applications. Running a limited subset of the native Tornado Web API within a separate WSGI server using the tornado.wsgi.WSGIApplication adapter is also now supported.
- The previously experimental support for the Pyramid web framework has been improved and is now enabled by default.
- When using a DBAPI2 compliant database client, calls to stored procedures via the callproc() method of the database cursor are now tracked. Previously, only calls made to stored procedures by explicitly using a CALL SQL statement were being tracked. All calls to stored procedures currently appear under the category of 'SQL - Other'. A future agent update is planned which will split out calls to stored procedures separately and report them on the databases tab.
- Added instrumentation for automatically wrapping the WSGI application when using the CherryPy WSGI server bundled with the CherryPy framework.
- Added instrumentation for automatically wrapping the WSGI application when using the WSGI server from the Python standard library wsgiref module. Be aware that due to non conformance with the WSGI specification, it is not recommended that the WSGI server from the wsgiref module be used unless using Python 2.7.4 or higher. Issues with older versions of the WSGI server in the wsgiref module will cause incorrect reporting of data. More details are provided in our status of the Python agent.
- External web calls made using the 'requests' module from a session were not being correctly monitored, resulting in such requests being picked up as being performed by the lower level 'httplib' module and the time taken not being inclusive of the whole request.
- The collection of explain plans by the agent in a background thread when performing a data harvest, was not explicitly performing a rollback. This could result in database proxies such as pgbouncer believing the database connection was now dirty and force closing the connection and ejecting it from the current database connection pool.
- The wrapper used by the agent around wsgi.input passed in the WSGI request environment was preventing non standard WSGI applications directly reaching down to the original socket connection so as to implement extensions such as websockets.
- The object wrapper used in our instrumentation for function traces has been improved so that it will pass type checks performed using builtin Python calls such as isinstance(), isfunction() and isclass(). This addresses issues where the object wrappers could interfere with the correct operation of web frameworks which derived the calling convention to be used when invoking a handler function from the type of the handler. Note that using type() on the wrapper object will still yield its true type, so functions such as isinstance() should always be used where possible.
- The agent could fail on startup when running on a system with a non English locale using a custom date/time format.
This release of the Python agent introduces a major new feature called X-Ray Sessions. This feature allows you to gain deeper insights into a key transaction's performance by showing transaction traces alongside long-running profiler results.
- X-Ray Sessions provide more targeted transaction trace samples and thread profiling for web transactions. For full details see our X-Ray sessions documentation.
- For new agent installations, the agent will now default to using a SSL connection when connecting to our data collector to report any data. If updating from an older agent version, you will however need to manually update your agent configuration file and change the 'ssl' setting to 'true' to always use SSL connections.
- When connecting direct to our data collector over a SSL connection, SSL certificate validation will now be performed to mitigate the possibility of man in the middle attacks.
- Breakdown metrics for transactions are now recorded as unscoped metrics as well as the existing metrics scoped to a web transaction. As custom dashboards can only chart unscoped metrics, what this means is that it is now possible to create custom dashboards that display data on specific breakdown metrics, such as time spent in traced functions. For example, you may wish to chart time spent in creating database connections, or specific middleware and view handler functions.
- When displaying error details, the full path of the exception is now displayed and not just the class name. This makes it easier to correctly identify the full name, with that name then being able to be directly copied into the 'error_collector.ignore_errors' setting of the agent configuration file if it is desired that that exception be ignored.
- Added the 'newrelic.agent.record_custom_metrics()' function to the public API for the agent to allow the recording of many custom metrics in one function call to the agent.
- We have temporarily disabled the generation of CPU burn information for transaction traces. This was displaying inaccurate values after a UI change, but also had an underlying issue whereby it was reporting CPU burn by the whole process and not just the thread handling the request. We are withdrawing the feature until we can improve the accuracy of the reported value.
- We have disabled the ability to get thread profiling information when a WSGI server makes use of coroutine libraries such as gevent or eventlet. This is because the nature of how coroutines are implemented, means we were are not able to generate accurate usable data.
- When queueing Celery tasks from a web application, those web transactions were in some cases being wrongly categorised as background tasks. The name of the transaction was also being labelled with a string with the object representation, resulting in metric grouping issues.
- The older version of the 'requests' package we were bundling and were using for any HTTP calls was not closing socket descriptors immediately when a requests session object was disposed of. This could result in a small number of open socket descriptors being held open until the Python garbage collector kicked in and closed them.
- Login credentials supplied as part of a URL, could in some circumstances appear with the URL parameter associated with a web external node in a transaction trace sample.
- Django 1.5 moved which module the 'Http404' exception was declared in. As a result we were no longer correctly ignoring it when catching exceptions to process as errors.
- A check designed to warn against the use of uWSGI versions prior to 1.2.6, which contain a bug which results in the agent recording incorrect response times, was failing on very old uWSGI versions.
- Updated our gevent instrumentation to accommodate a prototype change in gevent 1.0rc1.
- Updated our bottle instrumentation to handle the non numeric version string used in bottle development versions.
- After five minutes of being unable to connect to our data collector to report data, the agent will now internally reset itself, discarding accumulated metric data and require the agent to reregister itself with our core application before collecting data once again. This is to ensure that we do not unnecessarily keep accruing data and causing an increase in memory usage while we cannot contact our data collector.
- Avoid the potential for a web request to deadlock in a situation where a web framework transaction management system made a copy of the object we use to record details about web transactions or background tasks.