...

• JDBC connection pooling - possible enhancements
  • Add option to use a single connection per-thread (e.g. if Pool Max=0)
  • or remove pooling altogether? - does it make sense to share connections between threads, which are supposed to be independent users?
  • use Apache DBCP instead of Excalibur?
• JDBC sampler - use different cache sizes for connection and statement?
  • How do these really work?
  • Should the connection cache be a ThreadLocal item?
  • Need to add debugging to show when cache items are created and destroyed
• XPath Extractor - add user-declared namespaces
• Rework HTTP GUI to make it easier to use and extend
• JDBC IN/OUT/INOUT parameters
• Add HTTP methods dropdown to "SOAP/XML-RPC sampler", bugzilla 42637
• Remove unnecessary code duplication in "SOAP/XML-RPC sampler", rather use code inherited from HTTPSampler2. This is almost done, not committed to svn (Alf Hogemark)
  • Add unit tests for SOAP/XML-RPC sampler
  • Change name exposed to use to "SOAP/XML-RPC/REST sampler"
• Make it possible to send file content for HTTP GET, PUT and DELETE using "HTTP Sampler" and "HTTP Sampler 2", and therefore also "SOAP/XML-RPC sampler"
• Make Axis2 / XFire / CFX Sampler, and possible retire existing "Webservice(SOAP) sampler".
  • A bit unsure about this, not sure what benefit is over "SOAP/XML-RPC Sampler".
  • Will we really be testing Axis / XFire client side peformance here ?
  • Or could we make a nice GUI, so it is easier to test web services ?
• Make command line option available, to easily performance test one url
  • This is meant as a replacement for the "ab" command, and alternative to "faban"
  • I think it would broaden the user of JMeter a lot
  • I will look into making a "ab result listener", which presents the same statistics as "ab" initially (Alf Hogemark)
  • One possible implementation is
    • Add a simple jmeter test plan, with threadgroup, http sampler, and listener, using property values to control behavior to the bin/system_testfiles directory
    • Add shell scripts, for example jmeter_test_http, which just calls the standard jmeter script, with additional "-J" arguments, for example "-Jiterations=10"
    • Need a way of getting the listener output to the console, need some investigation
      • on Windows one can use CON as the file name; /dev/tty for unix - but output is not formatted nicely
      • Summariser can already output to console - that may be enough
• Restructure HTTP Sampler / Settings GUI. bugzilla 41917. A big job, but is becoming necessary, if we are adding more options to the HTTP Sampler
• Add a "View only errors" setting in viewers.
  • Currently, there is a "Log only errors", but I want to have it as a general setting on each listener
  • This will allow a test plan where you can easily look into samples that fails, and get statistics with or without errors (possibly exceptional cases)
    • a "View Results Tree" is set up to only view errors
    • a Summary listener which views all samples
    • a Summary listener only for successes
    • a Summary listener only for errors
  • I find the "Write All Data to File" setting on each listener confusing, since all listeners share the same possible log output, does it not ?
    • For example, the "Graph listener", it does not write graph files to the file does it ? It writes a CSV or XML file ?

    • Wiki Markup
      Could we remove the "Write All Data to File" setting from all listeners, and instead make a "Write All Data listener" ? \[The simple data writer?\]

• Listeners:
  • make distinction between open file for read and write.
    • at present the browse button does both, which is confusing (given the heading) <= this has been changed
    • a file that has just been read will then be used for writing if a test is run - is that sensible, or should the file name be cleared after it has been read?
    • Graph Transactions Per Second (TPS) like the graphing of Response Time.
• Add a "Preview" button to "CSV Data Set Config", which would use the file name and parameter specification, and bring up a dialog or something to show an example of the variables available and their values.
  • Useful for easily checking that you have specified the columns and variables correctly
  • Could this limit the number of emails on the jmeter-user list asking how "CSV Data Set" works
• Improve documentation
  • Currently, there are quite a lot of question duplication on jmeter-user mailing list, would be nice to reduce that number
  • Does people really read the documentation ? Judging by posts on jmeter-user mailing list, I sometime have my doubts.
  • Would be excellent to have documented steps for "view frontpage, log in, view a page, log out" scenario for the following frameworks
    • Struts
    • Spring
    • Spring Webflow
    • .net applications
    • others (please suggest)
    • I think that could reduce number of mails to jmeter-user mailing list
    • the above could start out as Wiki pages
    • it would be useful to provide sample JMX files as well
• Additional sample JMX files for the various scenarios that are FAQed
  • Perhaps a giant example showing lots of different scripts?
• Support for uploading multiple files in HTTPSampler / HTTPSampler2. I think it the one main missing functionality in HTTPSampler that browser provide, and JMeter not.
  • this is bugzilla 19128
  • I think this is dependent on bugzilla 41917, mentioned above
• Improve support for testing web service. This overlaps with some of the points above.
• Make a "HTTP Caching Manager", this is bugzilla 28502.
  • It is useful for performance testing
  • Would also be useful for functional testing, to see if pages are "cacheable".
• Make a "HTTP Proxy Manager", where the user can specify what proxy settings to use
  • Then use would not have to edit jmeter.properties file or command line arguments to use a proxy
  • We could get rid of the "HTTP Proxy" settings in the "Webservice(SOAP) sampler"
  • Could perhaps just extend HTTP Defaults for this?
  • N.B. HTTP (Java) relies on system properties to define proxy items, so it is impossible to have more than one setting.
• Make new, nice?, icons for the GUI elements which currently do not have a unique icon. For example the post and preprocessors.
  • Would make it a lot easier to see what is what in the tree
• Make menu options more context sensitive, so they are only enabled when they make sense
  • There seems to be some context sensitivity currently, but the code seems a bit odd, with explicit methods used by various parts of the code
  • Prevent adding test elements to inappropriate parts of the tree - e.g. controller and sampler must be under a thread group. Should also prevent drag and drop.
• Extend CSV Data Set to read via JDBC
• Assertion Results (or similar) could be used to attach errors to samples - e.g. if a Post-Processor failed, the error could be shown there instead of in the log. Or just add string array for storing such errors?
• Save Graphics should prompt before overwriting files
• Improve Http{{`Mirror}}`Thread class - done, but not submitted (Alf Hogemark)
  • It should allow blocking reads where appropriate, at least when content-length is known
  • Do not use Reader/Writer classes, which uses JVM default character encoding, causing mirrored content to differ
  • Add unit tests
• HTTP Cookie Manager should not clear static cookies each iteration
• Module Controller GUI
  • should it omit leading path name elements as at present?
  • it would be nice if it could check for ambiguous controller names:
    • report error if ambig. name selected? OR
    • flag ambiguous names in the drop-down list?
• user.properties and system.properties should be read from JMeter launch directory if not found in current directory.
• Java Sampler to invoke arbitrary Java method. Would need:
  • classname, method name, params (and types) , classpath
  • this can be done currently with BeanShell, but it would be simpler to have a dedicated GUI.
• Remote Mode selection: might be useful to be able to specify the remote mode for each listener. This would allow the use of immediate mode for a few listeners and statistical mode for most of traffic.
• Hostname is present in the SampleEvent class - can this be saved in the log files?
  • easy for CSV: pass SampleEvent instead of SampleResult.
  • harder for XML, as the default output depends on the class, so using SampleEvent would change the XML element. This would affect any stylesheets etc. However, using SampleEvent as the class would allow the use of a single element instead of both sample and httpSample (and any other future subclasses)
• If Controller - allow functions and variables as alternative to Javascript
  

Non-functional improvements

...

• Use Java 1.5 for building, but only require 1.4 for running. Is this easy to do? Can it be done in Eclipse?
• Look into warnings when compiling code, some deprecated code should be replaced
• Currently, Apache Avalon seems to be used for logging. Avalon seems to have stalled, should we move to log4j ?
  • I looked into adding a GUI / dialog for "Help->View log" to easily view the jmeter log file from the application itself.
    • Seems to be some open source tools available for log4j which does that out of the box.
  • Would we gain functionality by moving to log4j ?
  • Alternatively, move to Commons Logging (as used by HttpClient currently)
  • Or perhaps SL4J - http://www.slf4j.org/
  • What are the risks with continuing to use Avalon, if Avalon is not maintained anymore ?

  • Wiki Markup

    Remark: Http{{`Components project is considering migrating Http}}`Client 4.0 branch off Commons Logging to another logging framework as Commons Logging currently appears unmaintained and the prospects of its further development remain unclear. \[Does it need any development? sebb\] It may be worthwhile to ensure both projects use the same logging framework / compatible logging frameworks. \[Agreed: sebb\]

• Upgrade to httpclient4 ? httpclient4 is still only in Alpha. Should / could we have one sampler class using httpclient3 and one httpclient4 ?
• Remove some code in HTTPSampler, which is there fore workaround for httpclient3.0 bug, which has been fixed in httpclient3.1
• Reorganise documentation
  • component_reference is getting much too big. This will require changes to the help system.
  • sort functions and component ref into more logical order (currently chronological)
  • perhaps use separate XML files for each item, included by main pages ?
  • Could we then add a "Help" button to each GUI element, which would bring up the correct help in the browser ?
  • or extend the existing help menu to load just the individual page.
  • If there are combined and individual help pages, there would probably need to be two copies. Maybe simplest just to split component reference by element type; keep current page as an index into the subsections
  • Help could perhaps be extended to allow loading of linked pages (but one would probably not want to allow external links to be loaded). There is some code in View Tree that might help here.
  Recode SaveService so it does not load classes unnecessarily (code already developed and tested - not committed)
• Re-write ClassFinder:
  • needs general tidyup / javadoc
  • cache results - same classes may be requested multiple times
  • find a way to scan the classes without needing to load them:
    • BCEL, or
    • perhaps use a pre-generated list of class names in a file which is regenerated if any jar files change (MD5)
• TestBean:
  • prepare is probably called too often; can it be done once per test?
  • need some more GUI types - eg. table
  • would be nice to be able to enable/disable fields depending on what else is selected - e.g. JDBC parameters only needed for prepared statements
  • ensure drop-down list size big enough for all entries (within limits!)
• move from Bugzilla to JIRA? More flexible, (but attachments a bit more awkward at present?)
• Change control logic so it does not use Exceptions for normal situations
• Can we add Beanshell jar to the distribution? If so, then some BeanShell elements could be simplified. License is either SPL or LGPL, but versions are not stated; not clear if allowed.
  • According to http://people.apache.org/~cliffs/3party.html, we can include libraries in binary form if licensed under SPL
  • The http://www.beanshell.org/license.html states it is dual licensed under SPL/LGPL
  • We should probably email the author of BeanShell and check if he could be more precise on the licensing on the web page ?
• JMS GUIs should be loadable without needing JMS jars (needs an extra level of indirection, as is done by JMS Publisher) or as below.
• How to handle Gui elements that depend on optional jars:
  • should these be displayable, even though the jars are missing? Convenient for creating and viewing test pans, but not so useful at run-time - should the test plan be allowed to run?
  • or should they be omitted as at present? - this is confusing at build time.
  • or perhaps generate a dummy entry in the list, with a message to say the jar is missing? his would be tricky, as the class is needed to retrieve the name and the menu category. Perhaps the way to do it is to handle it in the GUI by catching the errors, and changing the name or screen comment? May be tedious to do.
• Sort Menu types test tree according to JMeter processing order?
  • Controllers
  • Config
  • Timers
  • Pre-Processors
  • Samplers
  • Assertions
  • Post-processors
  • Listeners
  Sort test tree according to JMeter processing order? This should probably be a separate action, as it would This should probably be a separate action, as it would be confusing for the tree to change as it was editted!
• Re-order HTTP request defaults fields to be like sampler
  • Protocol
  • Host Port
  • Path
  • Encoding
• introduce Generic HTTP Sampler which can do either Java or HttpClient or ...
  • could be new sampler GUI with new underlying test elements
  • but it would be nice if existing test plans were converted into the new sampler. Looked at this a while back, but got stuck with XStream aliasing, which has to be one-to-one at present. However, the code change to SaveService to remove the class loading should make it possible to have multiple aliases for a single class. That might be sufficient, otherwise we'll need a suitable converter.
  • I think the GUI should just contain a dropdown "Client implementation", with values like "Java JRE", "HTTPClient 3.1", "HTTPClient 4.0" etc.
• Improve "org.apache.jmeter.JMeter" class by removing the hack in "ListenToTest", which causes JMeter to always sleep 5 seconds before exiting when run in batch
  • I think this relates to how threads are closed and lingering samples are handled. How is it solved in the "GUI" mode ?
• GUI code refactor
  • there are various table implementations, could they be combined?
  • perhaps the table models could also be combined?
• Add SVN revision number to version? - or at least (already added to Manifests.)

• Wiki Markup
  Consider migration to Maven2 as a build tool for JMeter. This should help simply dependency management and facilitate the use of JMeter for automatic load testing / integration into tools like \[http://maven.apache.org/continuum/ Continuum\]

  • Maven 1 was tried a couple of years ago, and seemed incompatible with the JMeter directory layout and multiple jars; hopefully Maven 2 is more flexible.
    
• If Controller - allow functions and variables as alternative to Javascript
  

Release strategy

Only JMeter committers should change this section, though others may comment.

JMeter 2.3.1

This is about to be released. No further enhancements will be added.

JMeter 2.3.1

This will be the will be the next release.

The following have been implemented:

• File->Revert command, to easily reload existing file.
• File->Open recent files command, bugzilla 31336
• Sort Menu types according to JMeter processing order
  • Controllers
  • Config
  • Timers
  • Pre-Processors
  • Samplers
  • Assertions
  • Post-processors
  • Listeners
• Recode SaveService so it does not load classes unnecessarily
• Add SVN revision number to Manifests.