dennist041497
asked on
Native routines/SIGALRM on Sun
I have an old legacy database interface system that uses message queues to communicate. We have a Java application that we are trying to write to retrieve information from this database using this legacy system. The native calls to this system were crashing, and I finally found out why. The call to the system call msgrcv() was being interrupted by a signal - SIGALRM (14) to be exact. I theorized that this had something to do with the mouse movements/event-type structure of Java and I think I'm right.
O/S is Sun, Solaris 2.3. We're using JDK1.1.1
To prevent the problem, I tried to ignore the signal (or trap it) just before the msgrcv() call and restore it after the call. Every time I do this, the msgrcv() calls works fine, but the Java VM crashes giving me the single message: "Alarm Clock" on stdout. I have written a very small Java application and native routine (which does nothing but trap and restore this signal when it gets called) and each time the VM crashes with this error. So far, the work-around has been to simply retry the system call for some MAX times until it either works or fails MAX times. While this works so far, I don't know what other system calls may fail. And this seems to be a silly work around.
The question then is, how do I make a native call from Java while ignoring any signals so that the VM doesn't crash?
O/S is Sun, Solaris 2.3. We're using JDK1.1.1
To prevent the problem, I tried to ignore the signal (or trap it) just before the msgrcv() call and restore it after the call. Every time I do this, the msgrcv() calls works fine, but the Java VM crashes giving me the single message: "Alarm Clock" on stdout. I have written a very small Java application and native routine (which does nothing but trap and restore this signal when it gets called) and each time the VM crashes with this error. So far, the work-around has been to simply retry the system call for some MAX times until it either works or fails MAX times. While this works so far, I don't know what other system calls may fail. And this seems to be a silly work around.
The question then is, how do I make a native call from Java while ignoring any signals so that the VM doesn't crash?
ASKER
I'm right about the source of the signal. I can get the signal to happen only if I move the mouse after pressing the button on the Java app that calls the native routine. If the mouse is still, no alarm happens (or happens *very* seldom). Besides, I have a very small C routine (no database calls, just signal handling), that traps the SIGALRM signal and that signal fires when I move the mouse around (only that signal, by the way).
The database is Informix, but that's irrelevant. Many years ago, our company wrote a database interface, in C, so that many processes could concurrently use the database (Ingres at the time) without bringing down the system. The interface was built to be database independent so that if we did change databases (and we did), the applications would not have to change. The msgrcv() call never is interrupted with our C programs that send messages to this interface, presumably because we never generate such a signal. But now, Java is also in the picture. And, with the evidence of the mouse moves... I'd say that's fairly conclusive.
I did get a note from a newsgroup that says I should probably use the jacket routines provided in the green threads package to do system calls. I'll have to research this.
Thanks for responding. Hope this helps clarify my situation.
The database is Informix, but that's irrelevant. Many years ago, our company wrote a database interface, in C, so that many processes could concurrently use the database (Ingres at the time) without bringing down the system. The interface was built to be database independent so that if we did change databases (and we did), the applications would not have to change. The msgrcv() call never is interrupted with our C programs that send messages to this interface, presumably because we never generate such a signal. But now, Java is also in the picture. And, with the evidence of the mouse moves... I'd say that's fairly conclusive.
I did get a note from a newsgroup that says I should probably use the jacket routines provided in the green threads package to do system calls. I'll have to research this.
Thanks for responding. Hope this helps clarify my situation.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
The answer seemed to work, although the code needs to be modified as stated in the answer. All the calls need a pointer to the set, rather than the set itself. Other than that, I believe this work around may help.
The Solaris version of the JVM is only compatible with 1.0.2 of the JDK and won't meet our needs (as we need RMI).
The Solaris version of the JVM is only compatible with 1.0.2 of the JDK and won't meet our needs (as we need RMI).
It's common to set an alarm to interupt a call to msgrcv, since UNIX provides no other mechanism to block the request for a duration between instantaneous and infinite (excusive). I can't say that the Solaris JVM is NOT setting an alarm, but it seems like a less likely scenario, given the symptoms you present.