ntservice: code in Win32Service.onStart() callback fails to look up classes about jna HOT 8 CLOSED

Well, it seems like no workaround I try is really working either. I tried doing the startup of our service separately but then there is a risk of the service initialisation starting up at the same time, which appears to cause JNA to deadlock. If the service control dispatcher is running, another thread attempting to make a JNA call stalls and never completes. Since we use JNA to fetch information about system memory for outputting a diagnostics line, this kills us before the first line of text is written to the log.

I tried separating it such that we delayed our own startup (using locks) until the service onStart() callback has been called, but that is apparently no use either - if we do it this way around, onStart() is never called.

The frustrating thing is that I can't see what's wrong in my code which makes it fail whereas YAJSW supposedly works. The next step of course is to see if YAJSW actually works.

from jna.

Check your classpath. I believe the NTService stuff samples the classpath when you install the service (via Java) in order to set up the service callback parameters. This is potentially something that YAJSW might handle differently.

from jna.

My workaround is to do the startup code outside the callback, and this works, so clearly the classes I'm calling are on the classpath. Unless somehow the service wrapper is changing the classpath before calling onStart(), which would be really bad.

Our service itself launches via an .exe which sets up the classpath properly. Also, the executable we run as a command-line app is the same executable we use to run the service - we just fall back to starting up normally if we get told we don't have permission to start up as a service.

I have just checked the value of the java.class.path system property from inside onStart() anyway and it's the same value I get if I look it up immediately after startup.

from jna.

How does the ntservice code compare to that within the platform library? Any improvements in the former should be incorporated into the latter and the ntservice version dropped.

from jna.

The platform version was a port of the example, with fixes, cleanup and tests done by Eugene Levitov. There's a bit of work remaining there to add an actual service and extend tests per the TODOs in W32ServiceTests.java.

The contrib version should be using this and serve as an example. But that was never finished, PRs welcome.

from jna.

Evidence on stackoverflow that this might be a common problem related to JNI:

http://stackoverflow.com/questions/10514495/service-loader-does-not-locate-service-provider-class-even-though-class-is-in-s

But what doesn't make sense is, our main application launcher uses JNI to launch the JVM in the first place. So why does that work, when this doesn't?

from jna.

Thread.currentThread().getContextClassLoader() returns an instance of AppClassLoader on the main thread but in the callback, it returns null.

Putting in a hack to set it properly after the callback is called seems to fix the issue, so maybe all that JNA has to do is set the context class loader when calling any callbacks?

I still don't understand this though. How did ServiceLoader manage to load the services file, if it didn't have the right ClassLoader?

from jna.

Closing since I don't use the class anymore and it seems like nobody else has joined.

from jna.