Code Review Request 7142596: RMI JPRT tests are failing (original) (raw)

Darryl Mocek darryl.mocek at oracle.com
Mon Jul 9 18:14:10 UTC 2012

Stuart,

see inline...

On Fri 06 Jul 2012 11:47:54 AM PDT, Stuart Marks wrote:

Hi, I've reviewed the first half of the files, thru test/java/rmi/registry/reexport/Reexport.java. Basically things look good and make sense, but there are some details to be ironed out and some cleanups to be done. Nothing major, I think, with the exception of SetChildEnv. See discussion below. Second half to follow later. s'marks

------- *** ProblemList.txt The comment for UnbindIdempotent says "Need to be marked othervm, or changed to be samevm safe" but the test isn't marked /othervm. Does the test need to be updated, or is this comment misleading, and the problem really is the port conflict? This was a port conflict, which is why the UnbindIdempotent test was modified. *** ContextWithNullProperties.java (Nitpick: import lines are out of order) The comment says, "Create registry if one is not already running." This is mostly left over from the earlier version which also mentioned port 1099. But, aren't we changing things so that this tests always creates its own registry on a unique port? If so, I'd just remove the comment and remove the catch/ignore of RemoteException (which was suspect in the first place). If for some reason we can't create our own registry, we should just throw the exception out to the caller and let the test error out. If we ignored RemoteException we'd leave registryPort as -1 which would presumably cause some obscure failure later on. Changed. We probably also don't need to declare/initialize registryPort separately anymore, so we can just replace the first 8 lines or so of main() with the following: Registry registry = TestLibrary.createRegistryOnUnusedPort(); int registryPort = TestLibrary.getRegistryPort(registry); This same pattern occurs in these other tests and so a similar fix should be applied to them as well: - UnbindIdempotent - LookupNameWithColon.java Fixed in all places. *** LookupNameWithColon.java This test is missing an @run tag, thus it won't actually get run! Since you've specified an @build tag, you have to specify a separate @run tag to run the test. It's possible to deduce this with a careful reading of the jtreg tag spec: http://openjdk.java.net/jtreg/tag-spec.txt (IMHO this is a terrible usability problem.) I've looked at the other tests in this webrev for similar issues, and I didn't see any, but I might have missed them. Added @run to this test. Regarding the registry, this test is somewhat different from the others in that it was originally coded to use an existing registry if it couldn't create its own. If it were to find a registry running on 1099, it was probably created by some other test, and no assumptions can reliably be made about it. So, I'd just take out this logic and have the test unconditionally create its own registry. Changed. *** StubClassesPermitted.java (Nitpick.) It looks like REGISTRYPORT is a constant, but it's a static field. It should be renamed to have a conventional field name (i.e., mixed case). Renamed. *** UnregisterGroup.java Similar to above, rename PORT. Renamed. *** SetChildEnv.java The testing of the message string from the IOException causes me great concern. This message is defined all the way over in java.io.PipedInputStream, and while it's not localized, it does seem like a pretty fragile dependency. I mean, changing some exception message in java.io might cause an RMI Activation test to fail??!? (Sorry.) Certainly we want to ignore spurious errors, and it sounds from the comment like normal termination of rmid causes these exceptions to be thrown. But I'm wondering if rmid crashes, won't we get the same exception and ignore it, improperly? I don't know what the right thing to do is here. It seems like there ought to be a more definitive way to distinguish between normal termination and pipe closure from an error. I don't see a simple solution right now. I suggest we table this issue and re-visit it after the commit. Another option is to not include the fix for Bug #7161503 with this fix until this issue has been addressed. *** AltSecurityManager.java The registry and rmid fields probably should be final. Maybe all caps too, except that RMID is the name of a class.... Made final, renamed. In run() it should probably bomb if utilityToStart equals neither registry nor rmid. OK. The ensureExit() method has a local variable port, which hides the field named port (well, not really, since ensureExit is static and port is an instance field) but still, this is kind of confusing. Renamed. If the port field is required to be initialized properly, make it a blank final (i.e., declare it final but without an initializer) and error-check it at the point where it's assigned in the constructor. Made blank final. *** MultipleRegistries.java Not really a big deal, but the way the second registry is created seems somewhat roundabout. It's not clear to me why the code can't just do this: Registry registryImpl1 = TestLibrary.createRegistryOnUnusedPort(); Registry registryImpl2 = TestLibrary.createRegistryOnUnusedPort(); int port1 = TestLibrary.getRegistryPort(registryImpl1); int port2 = TestLibrary.getRegistryPort(registryImpl2);

This turned out to be an issue with calling LocateRegitry.createRegistry(0), which occurs in TestLibrary.createRegistryOnUnusedPort(). If LocateRegitry.createRegistry(0) is called within the same VM multiple times, after the first registry is created (and not destroyed), the subsequent calls will fail. See the comment above creating the second registry line:

// Need to get a random port for the second registry.

However, this really isn't the right solution, so I modified TestLibrary.createRegistryOnUnusedPort to catch ExportException (which is thrown if the above issue occurs), get a random port, and attempt to create a registry on that port. See updated TestLibrary.
MultipleRegistries has been changed to what you have above as a result.

I'll post the changes to the comments here and the coming other half after I address the other half as another webrev.

Darryl

-------

On 7/5/12 2:22 PM, Darryl Mocek wrote: Hello core-libs. Please review this webrev to fix Bugs #7142596 and 7161503. Webrev can be found here: http://cr.openjdk.java.net/~dmocek/7142596/webrev.02. This commit fixes concurrency issues with the RMI tests. - Added TestLibrary.createRegistryOnUnusedPort method. This creates an RMIRegistry on an unused port. It will try up to 10 times before giving up. - Added a TestLibrary.getRegistryPort(Registry) method to get the port number of the registry. - Changed almost all tests from using hard port numbers to using random port numbers for running the RMI Registry and RMID. - Removed othervm from those tests which don't need it. - Added parameters for tests which spawn a separate VM to pass RMI Registry and RMID ports in cases where needed. - Added PropertyPermission to security policy files where needed. - Removed java/rmi and sun/rmi from tests which cannot be run concurrently. - Added java/rmi/Naming to list of tests which cannot be run concurrently. Thanks, Darryl

More information about the core-libs-dev mailing list