
Hosted by:
|
|  |
Java Service Wrapper Release Notes
 |
-
Added support for SGI Irix. Big thanks to Andreas Wendt for supplying the
patch.
-
Due to a bug in the build, the native library was not included in the 3.0.3
or 3.0.4 binary releases for OSX, building from source was working correctly.
This has been fixed and the build greatly simplified to avoid such problems
in the future.
Bug #791755.
-
Changed the default location of the pid file generated by the sh script to
exist in the same directory as the sh script rather than in the /var/run.
This can be changed by setting the PIDDIR variable in the sh script used to
launch the Wrapper.
-
Added support for the wrapper.pidfile
property on the Windows platform.
WARNING
|  |
(WINDOWS USERS) If you are using a wrapper.conf file that was created prior
to version 3.0.0 of the Wrapper, then you may have this property defined in
your configuration file. You will get an error on startup if the specified
path does not exist.
|
-
Added the wrapper.java.pidfile
property which will cause the pid of the java process to be written to a
specified file.
-
Stop clearing the file creation mask when the Unix version of the Wrapper is
run as a daemon process. The file creation mask will not be inherited from
the process which launches the Wrapper.
Bug #788849.
-
Modify the sh script so it works on Linux, then deprecate the bash script.
This means that all Unix platforms can now use the same script to control
the Wrapper. Thanks to Mike Castle for the patch. The bash script can still
be found in the release, but it is deprecated and will be removed in a
future version.
-
Modified the sh script so it is now possible to set the nice priority in the
script configuration block.
-
Remove output to System.out in the WrapperManager.requestThreadDump() method.
If some JVM threads were hung while accessing the System.out object,
attempting to do a thread a dump would cause the calling thread to hang as
well. Thanks to Thomas Hart for the patch.
-
Make it obvious in the log whether or not the Wrapper was started as a
daemon process on UNIX systems.
-
Modify the way restarts requested from the JVM, or caused by a filter are
handled. The Wrapper will no longer reset the restart count in either of
these cases. If an application runs for longer than the
wrapper.successful_invocation_time
timeout then the count will still be reset back to 0.
-
Added a new wrapper.ignore_signals
property which makes it possible to configure the Wrapper so it will ignore
CTRL-C, HALT and INT signals.
-
Modify the WrapperManager.isLaunchedAsService() method on UNIX systems so it
now returns true if the Wrapper was launched with the
wrapper.daemonize flag set.
-
Added a pair of MBean interfaces which allow the Wrapper to be controlled
using JMX. See the new JMX section in the
documentation for details. Thanks to Sal Ingrilli for help with testing.
-
Modify the Windows build so the Wrapper.exe and Wrapper.dll files can now
be built from Ant if MSVC is
installed.
-
Added a new wrapper.ping.interval
property which lets users control the frequency that the Wrapper pings the JVM.
Feature Request #607768.
-
When a JVM refuses to shutdown, the Wrapper can be configured to request a
thread dump using the wrapper.request_thread_dump_on_failed_jvm_exit
property. The Wrapper was then waiting 1 second before the process was
killed. This was not always long enough, resulting in a truncated thread
dump. Increased the pause to 3 seconds.
Feature Request #633761.
-
Fix a bug where wrapper.app.parameter.<n>
and wrapper.java.additional.<n>
properties declared from the Windows
command line were not correctly
handling spaces in their values.
Support Request #802139.
|
 |
-
Fix a problem on UNIX systems where requesting a second thread dump any time
during the life of a single Wrapper process would cause the Wrapper and JVM
to shutdown rather than perform the thread dump.
-
Fix a problem where a, user without permission, attempting to stop an
application was able to delete the pid file even though they were unable
to stop the application itself. This would make the scripts think that
the application was stopped when was actually still running.
-
Fix a problem where an application was being killed prematurely if it took
longer than 6 seconds to exit on its own. The scripts now make sure that
an application always has enough time to shutdown cleanly.
-
Improve the debug output so that packet codes are now shown using a name
rather than a raw number.
-
Reduce the frequency of "Waiting to stop..." messages displayed when removing
an NT service that is currently running. Decreased frequency from once per
second to once every five seconds.
-
Fix a minor problem where the hour in the date returned by
WrapperInfo.getBuildTime() was not base 24.
-
Added -t and -p command line options to the Windows version of the Wrapper
to sTart and stoP the Wrapper as an NT service. This can be used in place
of "net start" and "net stop", which do not always work correctly when a
service takes a long time to start up or shutdown. See the
Launching Your Application (Win32)
section for more details.
-
Add a new method WrapperManager.stopImmediate which will cause the JVM to
exit immediately without calling any stop methods or shutdown hooks.
-
Add a new class, WrapperActionServer, which makes it easy to remotely control
the Wrapper remotely by opening a socket and sending commands. See the
javadocs of the class for more
details.
-
Fix bug #744801. A Java GUI was not being displayed when the application was
run in either console mode or as a service with wrapper.ntservice.interactive
enabled. This problem was introduced in Version 3.0.0 when using 1.2.x or
1.3.x versions of Java. To use interactive services with 1.2.x or 1.3.x
versions of java, please review the documentation for the
wrapper.ntservice.interactive property.
-
Fix a problem where the JVM was not receiving CTRL-C and CTRL-CLOSE events
when running under the Wrapper on Windows. This was not a problem in most
cases as the Wrapper was taking care of the processing of the events. But
the WrapperListener.controlEvent() method was not being called as documented.
-
Changed the way the WrapperSimpleApp and WrapperStartStopApp respond to
control events so that the JVM will respond and call WrapperManager.stop()
even when being controlled by the Wrapper.
-
Modified the suggested behavior of the WrapperListener.controlEvent() method.
Users who have implemented the WrapperListener interface themselves should
review the javadocs. The changes are not
required and applications will continue to function as they did before.
-
Added support for DEC OSF1 (Alpha). Big thanks to Andreas Wendt for
supplying the patch.
-
Fix a problem where the sh and bash scripts were failing if the path to the
script contained spaces.
-
Fix a problem where the JVM would sometimes hang when trying to shutdown if
the wrapper.key parameter was passed to the JVM while not being controlled
by the Wrapper. This would happen if a user copied the command from the
Wrapper's debug output and attempted to run it as is without first removing
the wrapper.key parameter.
-
Implement the ability to specify an NT service's load order group in response
to feature request #764143. See the javadocs for the new
wrapper.ntservice.load_order_group
property for details.
-
Improve the error message displayed when the NT EventLog is full in response
to feature request #643617. The EventLog output will now be disabled if any
errors are encountered while logging events. This prevents the error from
repeating.
-
Improve the error message displayed on Windows when the configured Java
command can not be executed or does not exist.
-
Fix a problem where the Wrapper was leaving a pipe unclosed each time the JVM
was restarted on all UNIX platforms. This was causing the Wrapper to run out
of file handles. Bug #767267, discovered and patched by David Wong.
-
Fix a problem where the '#' character, which signifies a comment, could not
be included in property values. A double hash, '##' will now resolve into a
'#' within the property value. Bug #777303.
-
Added support for FreeBSD. Big thanks to Alphonse Bendt for supplying the
patch.
-
Make the wrapper.port property optional.
-
Changed the way environment variables are loaded from the registry on Windows
platforms so users will no longer get warning messages about not being able
to handle very large environment variables. Prior versions could only handle
environment variables whose expanded value was less than 2048 characters in
length.
-
Fix a problem on UNIX platforms where a shell used to start the Wrapper
running as a detached process would hang when the user attempted to exit
the shell. Thanks to Mike Castle for this patch.
|
 |
-
Added support for Mac OS X. Big thanks to Andy Barnett for supplying the
patch.
-
Fix a segmentation fault on UNIX systems when the first console output
from the JVM was an empty line. Thanks to Mike Castle for finding this.
-
Fix a problem where a 0 length malloc was being called if there were no
configured filters. This was fine on most platforms but caused a crash
on MAC OS X.
-
Rework the initialization of the bash and sh scripts so that they will
work correctly when referenced as symbolic links. Thanks go out to Richard
Emberson for the code to resolve symbolic links.
-
Deprecated the realpath binary in the *NIX distributions as it is no longer
used by the bash or sh scripts. It is being left in for now so as not to
break the build scripts of other projects, but it will be removed after a
couple more releases.
-
Added a test to make sure that
wrapper.ntservice.interactive
is not set to TRUE when an account is specified using
wrapper.ntservice.account.
|
 |
-
Modified the sh and bash scripts so that console log output is disabled by
default when the scripts are launched with the 'start' action. Running with
the 'console' action will still send output to the console. Logging to the
file is still enabled.
-
Modified the wrapper.ping.timeout property so it also controls the ping
timeout within the JVM. Before the timeout on responses to the Wrapper
could be controlled, but the ping timeout within the JVM was hardcoded to
30 seconds.
-
In the last release, some work was done to avoid false timeouts caused by
large quantities of output. On some heavily loaded systems, timeouts were
still being encountered. Rather than reading up to 50 lines of input, the
code will now read for a maximum of 250ms before returning to give the main
event loop more cycles.
-
Fix a problem where the values of environment variables set in the
configuration file were not correct when those values included references
to other environment variables.
-
Fix a potential buffer overflow problem if configuration properties
referenced extremely large environment variables.
-
Fix a potential problem where the inability to expand very large environment
variables would have led to an access violation when run as an NT service.
-
Add some extra checks in the event where the native library can not be loaded
so that the WrapperManager can differentiate between the library missing and
not being readable due to permission problems.
-
Remove the wrapper.ntservice.process_priority from the default wrapper.conf
because its use can produce unexpected results if used improperly. Please
see the property documentation for details.
-
Fix a problem where environment variables in the registry which had no value
were causing the Wrapper to crash with an access violation. This was
introduced in version 3.0.0 with the feature to load environment variables
from the registry. The offending registry entry was WV_GATEWAY_CFG which
appears to be related to Oracle.
|
 |
-
Fix a problem with the wrapper.disable_shutdown_hook. Due to a typo in the
source, the property was being ignored. This was broken in the 3.0.0
release.
-
Fix a problem with the HP-UX release build reported by Ashish Gawarikar.
-
Add the ability to set environment variables from within the configuration
file or from the command line. See the
Environment Variables section for
details.
-
Fix a problem on HP-UX and AIX machines where the stop() function in the
shell scripts was causing a syntax error due to a conflict with a like named
shell command on those platforms. This appears to be an issue with the
Korn shell on all platforms.
-
Fix a problem where very heavy output from the JVM can cause the Wrapper to
give a false timeout. The Wrapper now only reads 50 lines of input at a time
to guarantee that the Wrapper's event loop always gets cycles.
-
Fix a problem on UNIX versions where extra line breaks would sometimes be
added to the logged output when there was large amounts of output being
sent from the JVM.
-
Fix a problem where a large number of calls to WrapperManager.log()
immediately before the JVM exits could lead to the Wrapper incorrectly
reporting that the JVM exited unexpectedly.
|
 |
-
Deprecated the com.silveregg.wrapper package in favor of
org.tanukisoftware.wrapper. The classes and interfaces in the silveregg
package will continue to function, but migration to the new package should
be done when possible. See the project
history for details.
-
On Windows systems change any forward slashes in the wrapper.java.command
property to back slashes. Some users had reported having problems on
Windows XP.
-
Implemented feature request #633178. Added WrapperManager.requestThreadDump()
to force the current JVM to immediately perform a thread dump.
-
Fixed bug where wrapper.logfile.maxsize was being set to 0 if the 'k' or 'm'
unit was omitted.
-
Add the ability to specify an account name and password when installing an
NT service.
-
Add a property, wrapper.ntservice.interactive, which makes it possible to
control whether or not the Java process can gain access to the desktop while
it is running as an NT service.
-
Add limited support for 1.2.x versions of Java. Shutdown hooks are
supported until Java 1.3 so those functions will be disabled. If the
application displays a GUI then Java 1.3 should be used as the GUI can not
currently be displayed when using Java 1.2.x.
-
Made it possible to use the wrapper.pidfile property on all *nix platforms.
Please notice that the property has been removed from the default
wrapper.conf file. The property is not needed when the wrapper is launched
with the bash shell script. The sh shell script will set the wrapper.pidfile
when the wrapper is launched. If either of the scripts provided with the
Wrapper distribution are used then the wrapper.pidfile should always be
removed from your wrapper.conf file.
-
Added a new wrapper.daemonize property which, when set, will form the wrapper
process to be a detached non-session group leader. This makes it possible to
launch the wrapper in such a way that it will not be terminated when the user
launching the process logs out. This had been a problem on Solaris systems
when using the sh shell. The default sh and bash scripts both make use of
this in the default. Please update your scripts for use with this version.
Thanks to Rajiv Subrahmanyam for the patch.
-
Fix a problem where the Wrapper was incorrectly counting the number of
non-daemon threads in BEA's JRockit Virtual Machine. This was causing the
application to shutdown when the non-daemon thread count dropped to 1.
-
Added support for building the wrapper on AIX and HP-UX systems. Thanks for
the patches involved go out to Ashish Gawarikar and William Lee.
-
Implement feature request #653131 to force the JVM to immediately exit when
the user presses CTRL-C multiple times.
-
Added a 'console' action to the bash and sh scripts to make it possible to
launch the Wrapper in the current shell process. The 'start' task will launch
the Wrapper as a spawned daemon process.
-
Fixed a problem where missing environment variables specified in classpath
or library path properties were not being handled correctly.
-
Implemented feature request #676599 to enable the filtering of JVM output to
trigger JVM restarts or Wrapper shutdowns. See the new
wrapper.filter.trigger.n and wrapper.filter.action.n properties.
-
Modify the Win32 version of the Wrapper so that Environment Variables are
always read from the system registry when the Wrapper is run as a service.
By doing this, it makes it possible to change or add the system environment
variables and have them take effect without having to first reboot the
machine.
-
Implemented cascading configuration files. See the
Cascading Configuration Files
section for details.
-
Changed the default value for the wrapper.java.initmemory property to be 3Mb.
The default on Windows and Linux JVMs is 2Mb, but the Solaris JVM requires
a minimum of 3Mb. The minimum value accepted by the Wrapper was changed
from 8Mb to 1Mb to make it possible to reduce the footprint of applications
to what is possible without using the wrapper.
-
Improve the parsing of configuration files so that leading and trailing white
space is now correctly trimmed. It is also now possible to have comments at
the end of a line containing a property.
-
Modify the way exceptions thrown by an application's main method are
presented to the user by the WrapperSimpleApp and WrapperStartStopApp so
they no longer look like a problem with Wrapper configuration.
|
 |
-
Added a new property, wrapper.restart.delay, which allows the user to control
the amount of time to pause between a JVM exiting and a new JVM being
launched.
-
Fixed bug #611024. The Wrapper would sometimes fail to start if
wrapper.max_failed_invocations is set to 1.
-
Fix a problem where the number of non-daemon threads was not being calculated
in some cases.
-
Implemented feature request #491443. Environment variables referenced in the
wrapper.conf file will now be evaluated as the file is loaded. The windows
syntax for environment variables is used on all platforms to make them
platform independent.
-
Fixed a problem where the wrapper.conf was being open with both read and
write locks when a read lock is all that is needed. Made the wrapper fail
on startup if another application held a read lock on the conf file.
-
Modified the message displayed when the native library could not be found,
so that it is much more descriptive. Hopefully it will cut down on questions
caused by configuration problems.
-
Implemented feature request #613539. Modified the wrapper.java.library.path
to function like the wrapper.java.classpath.n properties so that multiple
directories can be specified in the library path in a platform independent
way. The old property is still supported, but deprecated.
-
Fix Bug #632215. The WrapperManager.isLaunchedAsService() method was always
returning false, even when run as a service under Windows. On Linux, the
Wrapper is always run as a console app, so this method will always return
false.
-
Improve the message thrown when user code attempts to access System.in from
within a JVM being controlled by the Wrapper. System.in will not work
because the JVM is a spawned process.
|
 |
-
Fixed a compiler problem on Solaris some systems.
-
Added a new property, wrapper.cpu.timeout, which allows the user to control
how much time without receiving any CPU the Wrapper will tolerate before
displaying a warning message. The CPU timeout feature was added in 2.2.7
but the default timeout of 10 seconds was not configurable.
-
The Wrapper was only allowing 5 seconds between the JVM informing the
Wrapper that it was going to exit and the JVM process actually exiting.
This would cause the Wrapper to terminate the process prematurely in
cases where an application shutdown thread took longer than 5 seconds to
complete. The Wrapper not allows wrapper.jvm_exit.timeout seconds for
the JVM process to exit on its own before being forcibly terminated.
-
When there is a configuration problem or a resource is unavailable, a JVM
will sometimes exit abnormally very shortly after being launched. This
can lead the the JVM being infinitely restarted due to a simple class
path misconfiguration. To work around this, the Wrapper has always had
a hard limit of 5 restarts within a short period of time. If the JVM
has been running for more than a few minutes, then the count was reset.
In this version, a new property. wrapper.max_failed_invocations was added
to allow the max value to be set. The time period which the JVM must
now be running for the JVM launch to have been considered a success for
restart purposes is set using the new wrapper.successful_invocation_time
property.
-
The number of advanced properties which most users do not need has been
increasing as the Wrapper has been made more and more flexible. This
has been causing confusion in their usage by people who play with them
without first reading the documentation. To solve this, the advanced
properties were removed from the default configuration file. They still
function. But users must now read to the advanced configuration
documentation to learn about their existence. Added quite about to the
descriptions of these properties to hopefully clear up any confusion
about their usage.
-
When the JVM exits abnormally, the Wrapper will pause for a few seconds
before starting another JVM. If the user pressed CTRL-C during this
pause, a new JVM would still be launched. The new JVM was exiting
immediately but it was a waste of time. The Wrapper now recognizes the
event and aborts launching the new JVM.
-
Added a page to the documentation which shows inline javadocs. This
will hopefully make it easier to navigate them as part of the full
documentation set.
-
Added a new method to the WrapperManager which enables user code to
log at any log level.
-
Added a new Helper class WrapperStartStopApp which allows users to easily
integrate applications like Tomcat which use a separate class to stop
the application.
-
Added a samples section to the documentation. Just includes Tomcat 4
for now.
|
 |
-
Fix a problem where the JVM was trying to reconnect the Wrapper as it was
being shutdown. This was causing problems if the JVM was being restarted.
-
Added support for the system being suspended to RAM or disk. Also improved
wrapper performance when a system is under 100% load. See the new example
output in the example section.
-
Fix a problem where the log output was not being directed to a file called
wrapper.log in the same directory as the Wrapper binary in the event that the
configured wrapper log file could not be accessed.
-
Fix a problem where the Wrapper was not shutting down the JVM correctly when
all non daemon threads completed. Normally a JVM will exit when all of its
non daemon threads have completed so this was causing some problems.
(Thanks to Jung Tamas)
-
Added the ability to set the priority of the Wrapper and its JVM when run
as an NT service or console application. The same thing can be better
achieved on Unix systems by using "nice" in the shell script used to launch
the Wrapper. See the documentation for for details.
-
JVM information was not being displayed correctly when the Wrapper native
library could not be loaded.
-
Added a new property to cause the wrapper to attempt to request a thread dump
when the JVM does not exit on request.
-
Improved the documentation of the WrapperSimpleApp and WrapperListener classes.
-
Adding a new property wrapper.shutdown.timeout to allow the user to extend the
length of time that an application is allowed to take shutting down.
-
Rework the way the shutdown process works so that System.exit will never be
called before the stop method in WrapperListener has had a chance to complete.
-
Add a Restart button to the TestWrapper application.
-
Fix a problem on Unix versions where '%' characters in Java output would
sometimes cause the wrapper to crash. Somehow missed getting this into the
last release.
-
Added a test to make sure that WrapperManager.stop is not called recursively.
-
Added support for building under Windows XP. Prebuilt installations had
already been working.
|
 |
-
Fix a problem where '%' characters in Java output would sometimes cause the
wrapper to crash. (Thanks to Frode Moe)
-
Added support for requesting a Java thread dump without shutting down the
Java process.
-
Fixed a problem on windows where the java command was looking in the windows
system and system32 directories for the java executable before checking the
path when the full path to the java executable was not specified in the
configuration file. This could lead to different JVM being run from the
Wrapper than was run if java -version was run from the command line. The
Wrapper will now attempt to resolve the full java path using the PATH
environment variable.
-
Added debug output showing Java version information when the JVM first starts.
-
Modified c source to use /* */ style comments rather than // style comments.
Some people were having problems with some compilers.
|
 |
-
Fixed bug issue when reading configuration files from Windows on Unix.
-
Deprecated the wrapper.debug property in favor of loglevels.
-
Added support for service descriptions for Win2k and XP.
-
Added new logger functionality includes the following features: Loglevels like
Log4j, NT Eventlog support, UNIX syslog support and rolling log files.
-
Added wildcard support for classpath entries in wrapper.conf.
-
Added the ability to specify configuration properties from the command line.
-
Changed the way NT services are installed so that a patched version of the
Wrapper.exe file no longer needs to be created to reference the wrapper.conf
file.
|
 |
-
The value of APP_NAME in the bash or sh scripts no longer needs to be the
same as the script.
-
Added the ability to format and/or disable file logging and output to the
console.
-
Set mode of executables in binary release tar files so that they can be run
without modification after being extracted.
-
Fixed line feeds in release so that bat files are always CRLF, unix scripts
are always LF. Other source files are always CRLF in ZIP archives and LF in
tar.gz archives.
-
Make the build fail if Wrapper.exe or Wrapper.dll do not exist for Windows
builds.
-
Added an empty wrapper.log to the releases so that the TestWrapper example
program runs out of the box.
|
 |
-
Added template scripts and conf files for ease of integration with other
applications.
-
Cleaned up the build.
-
The WrapperSimpleApp method of launching applications was not working correctly
with applications whose main method did not return.
-
Add sample scripts and wrapper.conf files in the bin and lib directories.
These scripts are used to start a sample application which runs out of the box.
See the new example.html page in the
documentation for more details.
-
Get rid of the platform specific directories in the bin and lib directories.
-
Enable relative paths for Windows version. In previous versions of Wrapper,
it was necessary to always use absolute paths because the working directory of
the wrapper would be then NT System32 directory when run as a service.
-
On the windows version, the wrapper always sets the current working directory
to the location of the wrapper executable immediately after startup.
-
Improvements to the documentation / web page.
|
|
by Leif Mortenson
|
|