Re: [Wrapper-user] clustering and java service wrapper

[Christian M. <chr...@ta...>]

Hello Jeroen,

I don't think you are able to start the service by issuing "wrapper.exe -s
..\conf\wrapper.conf" from console or any user session.

When running the Wrapper with -t, it tells the Service Control Manager, to
start the service. Using the API call StartService:
http://msdn.microsoft.com/en-us/library/windows/desktop/ms686321%28v=vs.85%29.aspx

For instance, if you call, net start {service}, it will also call
StartService quite like the Wrapper when run as wrapper -t

With the StartService call, the service manager will start then the service
command (wrapper -s) and waits for the process to start the control
dispatcher and connect to the service control manager. This is being done
in the service process by calling StartServiceCtrlDispatcher.
http://msdn.microsoft.com/en-us/library/windows/desktop/ms686324%28v=vs.85%29.aspx
Therefore, running bin\wrapper -s, from console without the service manager
waiting for a connection, will ultimately cause a failure.

That mechanism is a bit similar for instance to the TCP handshake.

Hope this helps you out.

Cheers,
Christian


On Wed, Apr 4, 2012 at 6:57 PM, jeroen vranckx <jvr...@gm...>wrote:

> hey christian,
>
> If I look at the created windows service that I made by using the
> wrapper.exe -i wrapper.conf option, when i look at the path to the
> executable it says X:\Foo\Bar\app\bin\wrapper.exe -s
> X:\Foo\Bar\app\conf\wrapper.conf.
>
> Now, I have the possibility when i create a generic application(for
> clusterunaware services), to just specify the service that i want to make
> generic. I have the possibility to give the generic service a specific
> startup command, so I could definatly use the -t method if nesecairy.
>  If I run it, it works perfectly fine without the us of the -t
> parameter(it will probably use the -s parameter by default). What does the
> -t parameter do actually?
> What's the difference if i run it with -s or --t?
>
> Indeed it says on the website(just as you say), that you need to call the
> -t parameter for a service, but the -s parameter(that is default when you
> install by the -i parameter) kinda confused me.
>
> Looking forward to your replye, and tnx in advance
>
> greetings,
> Jeroen Vranckx
>
>
> On 4 April 2012 07:10, Christian Mueller <
> chr...@ta...> wrote:
>
>> Hello,
>>
>> the parameter '-s' should not be called by the user but the service
>> manager in order to control the wrapper as service.
>>
>> when you want to start the installed service from console or script,
>> please run:
>> X:\Foo\Bar\app\bin\wrapper.exe -t X:\Foo\Bar\app\conf\wrapper.conf
>>
>> for all available parameters, please run bin\wrapper -?
>>
>> The wrapper.internal.namedpipe property is for internal use only and
>> shouldn't be set.
>> Actually it is only being used during installation and actually I think
>> it's redundant to have it on the servicemanager command.
>> When you interact with a service, administrator privileges are required.
>> However starting with the introduction of UAC in Windows Vista, an
>> Administrator account also needs to elevate a process to run under high
>> security context.  The Wrapper will check if it runs already elevated and
>> if not, it will pop-up the elevation dialog, requesting the elevation. This
>> will actually spawn an elevated process, which will do the required task.
>> The IPC of the elevated process and the original process is done via a
>> named pipe the original process allocates. The name of the named pipe is
>> random, so that's why you see two different names. Using random names makes
>> it possible to interact with multiple services simultaneously without
>> clashing. To tell the elevated process at which pipe the original wrapper
>> process is waiting for, the wrapper.internal.namedpipe is being defined.
>> The elevated process then redirects stdin/stdout/stderr to the pipes and
>> because by default stdin/stdout/stderr is buffered to save some
>> performance, but in this case it will cause some timing issues and since
>> the install/remove/start/stop actions are only running for a rather short
>> time, performance doesn't matter that much.
>>
>> Hope this answers all your questions.
>>
>> Please let me know if you need any further information.
>>
>> Thank you,
>>
>> Christian
>>
>>  On Tue, Apr 3, 2012 at 5:43 PM, jeroen vranckx <jvr...@gm...>wrote:
>>
>>>  Just figured out something ^^
>>> Seems like the wrapper.console.flush param and namedpipe param are a
>>> indirect consequence of installing the service when you don't use
>>> administrator params in cmd.
>>> So if you're doing something like clustering, this will definatly mess
>>> things up.
>>>
>>> Fixed this issue, and not having issues anymore.
>>> Scripted the wrapper.exe -i option in windows powershell, and managed to
>>> get the service running now. I'll be integrating it with the automatic
>>> generation of the windows service as a generic service in the cluster.
>>>
>>> I still would like to know what those param do, as they do seem to be
>>> wrapperrelated
>>>
>>>
>>> On 3 April 2012 10:03, jeroen vranckx <jvr...@gm...> wrote:
>>>
>>>> Well, to keep you guys updated on my progress.
>>>>
>>>> I managed to get the fix the logging off problem by using the
>>>> ignore_user_logoffs=TRUE option. Seems to do the trick, but I'm checking
>>>> out the other method to where I use the application as Windows Service too.
>>>> I managed to get that working too. Now when i go check out the service
>>>> that is installed on both nodes(well on the windows server cluster, you
>>>> need to install the service on both nodes locally), i can see that the path
>>>> to the executable is something like this:
>>>>
>>>> X:\Foo\Bar\app\bin\wrapper.exe -s X:\Foo\Bar\app\conf\wrapper.conf
>>>> wrapper.console.flush=true wrapper.internal.namedpipe=0643816939
>>>> The named pipe is different on both servers
>>>>
>>>> What does the wrapper.console.flush parameter do? And the namedpipe is
>>>> dependant on the wrapper?
>>>> Reason why I am asking this, is because I'm probably gonna need to
>>>> script some managementtools. Already tried to script something in
>>>> powershell, where i kept everything like that, but took a random number for
>>>> the namedpip(kept the number of digits though).
>>>>
>>>> The service did get created, but it didn't start up. So Im' woundering
>>>> if i need to consider some stuff, when creating the windows service myself
>>>> in powershell instead of using the wrapper.exe -i method in cmd.
>>>>
>>>> thanks in advance
>>>>
>>>> jeroen Vranckx
>>>>
>>>>
>>>>
>>>> On 28 March 2012 09:10, Christian Mueller <
>>>> chr...@ta...> wrote:
>>>>
>>>>> Hello Vranckx,
>>>>>
>>>>> about the restart on log off, you are probably right about that the
>>>>> application gets restarted due to the cluster setting.
>>>>> If that's the case, you should actually see that in the Wrapper log
>>>>> file as well.
>>>>>
>>>>> The 2 options are actually the most easy ways of telling the
>>>>> Wrapper/JVM to ignore the logoffs.
>>>>>
>>>>> However, there is also another way.
>>>>> If you are willing to do some coding, you can take a look at
>>>>> integration method 3:
>>>>> http://wrapper.tanukisoftware.com/doc/english/integrate-listener.html
>>>>>
>>>>> The controlEvent(..) method receives signals/events raised by the
>>>>> System and allows you to implement your own logic on the various signals.
>>>>>
>>>>> The advantage of this option is that you wouldn't need to change the
>>>>> Wrapper version and also can keep running the applications the same way
>>>>> they are running right now.
>>>>> The API for the Wrapper can be found here:
>>>>> http://wrapper.tanukisoftware.com/jdoc/index.html
>>>>>
>>>>> If you have any questions in the implementation, please let me know.
>>>>>
>>>>> Best Regards,
>>>>> Christian
>>>>>
>>>>>
>>>>> On Tue, Mar 27, 2012 at 10:36 PM, jeroen vranckx <
>>>>> jvr...@gm...> wrote:
>>>>>
>>>>>> Thanks for the outstanding service Mr. Mueller,
>>>>>>
>>>>>> Your explanation makes perfect sense. Seems like I was thinking in
>>>>>> the right direction.
>>>>>> I went to check on the tips that you gave me, and was surprised that
>>>>>> i couldn't find the wrapper.ignore_user_logoffs setting.
>>>>>> Did a little resource on what version they are working with, and
>>>>>> seems like they are working with a  3.2.1version. (So that explains the
>>>>>> missing setting ^^).
>>>>>>
>>>>>> Went to check on the programmers also, and it seems that all
>>>>>> applications are running as console applications, and not as services. This
>>>>>> explains pretty much everything.
>>>>>> As they weren't actually running as services but as console
>>>>>> applications, and because they were using an outdated version that didn't
>>>>>> have the wrapper.ignore_user_logoffs setting,
>>>>>> it makes perfect sense that the JVM would shutdown.
>>>>>>
>>>>>> i need to clear a little misunderstanding though. The "so called"
>>>>>> services restarted after a log-off.
>>>>>> This is probably caused by the JVM shutting down like you said,
>>>>>> because of the fact that they are running as console applications.
>>>>>> The cluster sees that his resources went down, so he tries to bring
>>>>>> them back online. I think he probably initiates new JVM's then.
>>>>>>
>>>>>> As it is a production environment, I can't recreate the scenario.
>>>>>> It's a shame, but I think we have it pretty much figured out now, why it
>>>>>> was going wrong ^^.
>>>>>> Don't think there is a way to fix the issue in the currently running
>>>>>> environment.(without having to convert everything to services, or upgrading
>>>>>> to the new version of the wrapper).
>>>>>>
>>>>>> So I'm just gonna concentrate on the new environment.
>>>>>>  Now, for the new environment I'll need to make some decissions and
>>>>>> if i understand correctly i have 2 options to avoid the log off issue:
>>>>>>
>>>>>> 1: running the application as a real service (that will ignore the
>>>>>> logoff by default)
>>>>>>
>>>>>> 2: keep using the console application but use the option
>>>>>> ignore_user_logoffs=TRUE (when using at least the 3.3.1 version(i'll be
>>>>>> using the latest version of course ^^))
>>>>>>
>>>>>> Is this correct, or are there other options? Maybe keep in mind, that
>>>>>> we are using it in a clusterevironment.
>>>>>>
>>>>>> Thanks for your help in advance!
>>>>>> Vranckx Jeroen
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 27 March 2012 13:51, Christian Mueller <
>>>>>> chr...@ta...> wrote:
>>>>>>
>>>>>>> Hello Vranckx,
>>>>>>>
>>>>>>> thank you for your mail.
>>>>>>>
>>>>>>> The native library of the Wrapper (wrapper.dll) is installing a
>>>>>>> signal handler for the JVM, making it possible to intercept signals/events,
>>>>>>> such as the logoff event, the system sends to all running processes,
>>>>>>> whenever a user logs off from a session.
>>>>>>> As you already have mentioned, the JVM is actually a process, which
>>>>>>> is by design not thought of running as a Service by itself. So whenever the
>>>>>>> JVM receives a logoff event from the system, it's default behavior is to
>>>>>>> shut itself down.
>>>>>>>
>>>>>>> When running as service, by default, the Wrapper catches this signal
>>>>>>> allowing the JVM to keep running even after the logoff signal (among some
>>>>>>> other signals) have been received.
>>>>>>>
>>>>>>> When running as console application, which is been done when running
>>>>>>> wrapper.exe -c, the default action is actually to forward the event to the
>>>>>>> JVM, which then will shut itself down.
>>>>>>> This behavior can be changed easily by setting the following
>>>>>>> property into your conf file:
>>>>>>> wrapper.ignore_user_logoffs=TRUE
>>>>>>>
>>>>>>> http://wrapper.tanukisoftware.com/doc/english/prop-ignore-user-logoffs.html
>>>>>>>
>>>>>>> I don't think that's the issue, but just in case, if the Wrapper
>>>>>>> wasn't able to load the native library, it will print out a warning,
>>>>>>> indicating the reason of the failure.
>>>>>>> Please make sure that the wrapper.java.library.path.<n> property is
>>>>>>> being set correctly to the path where the native library is located.
>>>>>>> But since you said, that the services actually keep running after a
>>>>>>> logoff, I assume that the JVM was already loading the native library
>>>>>>> successfully.
>>>>>>>
>>>>>>>
>>>>>>> Please let me know if you have any further questions.
>>>>>>>
>>>>>>> Best Regards and Good luck with your thesis!
>>>>>>>
>>>>>>> Christian Mueller
>>>>>>> Tanuki Software, Ltd.
>>>>>>>
>>>>>>>
>>>>>>>  On Tue, Mar 27, 2012 at 7:00 PM, jeroen vranckx <
>>>>>>> jvr...@gm...> wrote:
>>>>>>>
>>>>>>>>  Hi everybody,
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> This is my first time for posting a question here, but I thought it
>>>>>>>> would be best to ask some help with people who are familiar with the java
>>>>>>>> service wrapper already.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> So maybe it’s best that I give you some information about why I’m
>>>>>>>> asking this question and why I need your help.
>>>>>>>>
>>>>>>>> So I’m a student from Belgium, and I’m currently doing my
>>>>>>>> thesis(graduation paper?).
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> The company were I’m doing this asked me to revise and migrate an
>>>>>>>> existing servercluster infrastructure. I’ve been working the past 2 weeks
>>>>>>>> on checking out the possibilities to upgrade and improve the existing
>>>>>>>> structure,
>>>>>>>>
>>>>>>>> And I think I have given them some options on that side.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Now I have come to the actual clustering. The servercluster is only
>>>>>>>> used to cluster java-applications and services.
>>>>>>>>
>>>>>>>> There are a number of generic applications that run on the cluster,
>>>>>>>> which are using the java service wrapper.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Now; one of the problems that they are having, is that when an
>>>>>>>> administrator logs on to one of the servernodes, and logs off again, all
>>>>>>>> the generic applications restart.
>>>>>>>>
>>>>>>>> They were thinking that this problem was probably caused by the
>>>>>>>> wrapper.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> After reading the documentation, I seem to have found some
>>>>>>>> indications that indeed point in this direction.
>>>>>>>>
>>>>>>>> Not saying that the wrapper itself is doing anything wrong, but I
>>>>>>>> think that it’s probably used in a wrong manner.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> *“The problem is that Java on its own can not be run as a service.
>>>>>>>> Many simple tools like the Windows "sc" command can be used to run Java as
>>>>>>>> a service, but the user doing something as simple as logging off of the
>>>>>>>> machine will cause Java to shutdown.“*
>>>>>>>>
>>>>>>>> * *
>>>>>>>>
>>>>>>>> *“Most Java applications die rather abruptly if the user presses
>>>>>>>> CTRL-C, logs out of Windows, etc. You can work around some of
>>>>>>>> these issues with a Shutdown Hook, but the Wrapper implements this by using
>>>>>>>> a **native library<http://wrapper.tanukisoftware.com/doc/english/prop-native-library.html>
>>>>>>>> ** to directly capture the system signals. This makes it possible
>>>>>>>> to have a Java application installed as a Windows Service without it being
>>>>>>>> stopped when a user logs out. “*
>>>>>>>>
>>>>>>>> * *
>>>>>>>>
>>>>>>>> * *
>>>>>>>>
>>>>>>>> Now, what I think is what happens: the generic applications aren’t
>>>>>>>> correctly set-up to run as a windows service(probably combination of
>>>>>>>> clustering resource and java service wrapper misconfiguration),
>>>>>>>>
>>>>>>>> which causes the JVM’s to stop when logging off. The cluster
>>>>>>>> notices that there is something wrong with the clustering resource(generic
>>>>>>>> application), and starts it back up.
>>>>>>>>
>>>>>>>> This caused them to think at the company that the resources
>>>>>>>> restarted, but I think this isn’t the case. It’s the JVM that stops, and
>>>>>>>> the clustering that steps in.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Now, I’m wondering where I need to start looking to fix this
>>>>>>>> problem. I think the problem lies with the fact that the java-application
>>>>>>>> isn’t running as a windows service.
>>>>>>>>
>>>>>>>> I did already notice something with the command line parameters of
>>>>>>>> the generic resources. They are using the following commands to run the
>>>>>>>> java applications:
>>>>>>>>
>>>>>>>> X:\Foo\Bar\FooBAR\bin\wrapper.exe -c  X:\ Foo \ Bar \ FooBAR
>>>>>>>> \conf\wrapper.conf
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> If my first quick read of the documentation was correctly, that
>>>>>>>> will let it run as a console-application. Don’t know if this has anything
>>>>>>>> to do with it, just guessing.
>>>>>>>>
>>>>>>>> I’m probably going to re-read the whole documentation again, to get
>>>>>>>> myself more familiar with the java service wrapper.
>>>>>>>>
>>>>>>>> I’m not just looking for the answer, I really want to look into the
>>>>>>>> problem here. That’s why I’m looking for some input here, to get a point to
>>>>>>>> work from.
>>>>>>>>
>>>>>>>> My time is limited, so I need to work efficient(need to do
>>>>>>>> management and probably a whole lot of testing), so I thought it would be
>>>>>>>> best to ask for some help here.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> So is there someone who has some experience with clustering and
>>>>>>>> java service wrapper?  Or somebody that has a hunch what the problem is?
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Your help will certainly be appreciated.
>>>>>>>>
>>>>>>>> I apologize for my bad English, and hoping to hear your ideas on
>>>>>>>> this one.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Vranckx Jeroen
>>>>>>>>
>>>>>>>>
>>>>>>>> ------------------------------------------------------------------------------
>>>>>>>> This SF email is sponsosred by:
>>>>>>>> Try Windows Azure free for 90 days Click Here
>>>>>>>> http://p.sf.net/sfu/sfd2d-msazure
>>>>>>>> _______________________________________________
>>>>>>>> Wrapper-user mailing list
>>>>>>>> Wra...@li...
>>>>>>>> https://lists.sourceforge.net/lists/listinfo/wrapper-user
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ------------------------------------------------------------------------------
>>>>>>> This SF email is sponsosred by:
>>>>>>> Try Windows Azure free for 90 days Click Here
>>>>>>> http://p.sf.net/sfu/sfd2d-msazure
>>>>>>> _______________________________________________
>>>>>>> Wrapper-user mailing list
>>>>>>> Wra...@li...
>>>>>>> https://lists.sourceforge.net/lists/listinfo/wrapper-user
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>> ------------------------------------------------------------------------------
>>>>>> This SF email is sponsosred by:
>>>>>> Try Windows Azure free for 90 days Click Here
>>>>>> http://p.sf.net/sfu/sfd2d-msazure
>>>>>> _______________________________________________
>>>>>> Wrapper-user mailing list
>>>>>> Wra...@li...
>>>>>> https://lists.sourceforge.net/lists/listinfo/wrapper-user
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>> ------------------------------------------------------------------------------
>>>>> This SF email is sponsosred by:
>>>>> Try Windows Azure free for 90 days Click Here
>>>>> http://p.sf.net/sfu/sfd2d-msazure
>>>>> _______________________________________________
>>>>> Wrapper-user mailing list
>>>>> Wra...@li...
>>>>> https://lists.sourceforge.net/lists/listinfo/wrapper-user
>>>>>
>>>>>
>>>>
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Better than sec? Nothing is better than sec when it comes to
>>> monitoring Big Data applications. Try Boundary one-second
>>> resolution app monitoring today. Free.
>>> http://p.sf.net/sfu/Boundary-dev2dev
>>>
>>> _______________________________________________
>>> Wrapper-user mailing list
>>> Wra...@li...
>>> https://lists.sourceforge.net/lists/listinfo/wrapper-user
>>>
>>>
>>
>>
>> ------------------------------------------------------------------------------
>> Better than sec? Nothing is better than sec when it comes to
>> monitoring Big Data applications. Try Boundary one-second
>> resolution app monitoring today. Free.
>> http://p.sf.net/sfu/Boundary-dev2dev
>> _______________________________________________
>> Wrapper-user mailing list
>> Wra...@li...
>> https://lists.sourceforge.net/lists/listinfo/wrapper-user
>>
>>
>
>
> ------------------------------------------------------------------------------
> Better than sec? Nothing is better than sec when it comes to
> monitoring Big Data applications. Try Boundary one-second
> resolution app monitoring today. Free.
> http://p.sf.net/sfu/Boundary-dev2dev
> _______________________________________________
> Wrapper-user mailing list
> Wra...@li...
> https://lists.sourceforge.net/lists/listinfo/wrapper-user
>
>