NPE on Strand.park, no missing instrumentation highlighted with verifyInstrumentation turned on #182
Comments
|
Hi, I'm not too familiar with OSGI but how does the additional Quasar instrumentor work? There's no message at all from instrumentation verification? If you could share a small program that reproduces the issue that would be optimal. |
|
Hi, OSGi offers a hook which allows the registration of Pre-Classloading functionality, such as bytecode instrumentation: http://www.eclipse.org/equinox/weaving/ I did originally get instrumentation errors for classes I forgot to add to my suspendables and suspendables-super files. So fundamentally, the instrumentation verification is working. However, now that I have added the classes and interfaces that were missing instrumentation, I am getting this exception. I'll try to see if I can come up with a way to share a minimal way with which the issue is reproducible - that setup would likely however still include Equinox as its execution environment. Is there maybe a way with which I can evaluate whether all Quasar framework code was actually instrumented? One guess I had was that the Semaphores or methods called within the Semaphores might not have been instrumented. |
|
Quasar code is AOT-instrumented. You'll find the The issue is that of a strange exception (w.r.t. the source line it points to), so probably occurring in the added bytecode and probably because the fiber stack cannot be found, which means there's something uninstrumented (likely only a call site) towards the top of the stack frame: that's where I suggest to start looking for uninstrumented methods and/or uninstrumented call sites (they also can be due to unmarked super/interface/abstract methods) with the help of instrumentation debug as above. What is puzzling is that instrumentation verification doesn't point out the issue and (modulo remaining bugs that haven't been found for several months now, after it has been substantially improved) it really should. I'll have a closer look at the OSGI-attached instrumentor as well. |
|
This seems to be a scheduler issue (the fiber appears twice in the trace, and in a not in a FJPool worker thread). What scheduler are you using for the fiber? |
|
Sorry, it's right there: |
|
Thank you for your quick response. I am using the SameThreadExecutor as it is used by a simulation framework (DESMO-J) which we are using: https://sourceforge.net/p/desmoj/code/HEAD/tree/src.core/desmoj/core/simulator/FiberSimStrandFactory.java It also was in line with my mental model of Fiber execution which I had put together until then. We do not want to use Fibers to parallelize workload. Rather, we want to use them as a replacement of Thread-based execution of small simulation 'transactions' which one could probably also describe as co-routines. Of the execution threads, only one can actually do meaningful work at any point in time, which is why the SameThreadExecutor seemed like a reasonable fit. After reading your response and looking at our 'working' small-scale example, it (seemingly) is working as long as you don't start spawning fibers within fibers. I would be very much interested in your explanation. Many thanks, |
|
The issue still occurs after switching out the executor. The problem seems to be that the constructors are not instrumented ( in the above logs). When I explicitly mark them as Suspendable (e.g. by throwing SuspendExecution), I now get an UnableToInstrumentException: [quasar] ERROR: while transforming de/uka/ipd/sdq/simulation/abstractsimengine/desmoj/DesmoJSimProcess: Unable to instrument de/uka/ipd/sdq/simulation/abstractsimengine/desmoj/DesmoJSimProcess#(Lde/uka/ipd/sdq/simulation/abstractsimengine/processes/ISimProcessStrategy;Lde/uka/ipd/sdq/simulation/abstractsimengine/AbstractSimProcessDelegator;Lde/uka/ipd/sdq/simulation/abstractsimengine/desmoj/DesmoJModel;Ljava/lang/String;)V because of special method Is there a way I can get Quasar to instrument constructors? |
Fibers are very much intentionally not continuations. They are (abstract) threads and, like all threads, are continuations plus a scheduler. This is important precisely so that fibers could use constructs like channels etc., to interact with other fibers. Also, in 99% of the cases where people say they want continuations, they really want threads with better control over scheduling, which is what fibers provide, and I believe your case is no different. In Java 9 we will have a proper API for continuations, but even then we expect fibers to be a better fit for 99% of use cases. It is perfectly OK to use fibers without parallelization, but in order for fibers to wake-up other fibers, the executor needs to have some sort of an execution queue. Try giving your fibers a single-threaded thread pool (
Yes, constructors are not allowed to block. Blocking constructors are strongly discouraged and considered a code-smell even when plain Java threads are concerned. It's best not to leave a thread (or a fiber) hanging while an object is in an undetermined state (partly constructed). There's always a way around that, which is probably better anyway. |
|
Hi, having ruled out calls from the constructors and changing the used scheduler, I am still running into the same issue. I also made the switch to a FiberExecutorScheduler based on the SingleThreadExecutor. Any pointers are greatly appreciated. [pool-4-thread-1] DEBUG: Initialising random number generator! [Stream ID = 0] |
|
This stack looks regular to me so I think the previous considerations and suggestions about instrumentation checking now apply. Have you tried checking manually for instrumentation, with the help of instrumentation debug, the methods and call sites that appear in the trace? Are these classes instrumented by the agent or by the OSGI-based instrumentor? |
|
I'll evaluate whether the OSGi-based instrumentor actually verifies the instrumentation and then report back on this, thank you. |
|
Is this related? #255 |
Hi,
I'm currently getting the following error when using Fibers. I have turned on verifyInstrumentation. I am running the latest 0.7.5 JDK8 build and my agent is running. On top of the agent instrumentation, I am also running a separate QuasarInstrumentor that handles instrumentation of classes that are loaded via the platform's (Equinox OSGI) dependency injection mechanism. I have a minimal test setup that is working in conjunction with this setup.
However, when I use Semaphores I run into the following error.
I would greatly appreciate any help on this matter.
Best regards,
Christian
java.lang.NullPointerException
at co.paralleluniverse.strands.Strand.park(Strand.java:493)
at co.paralleluniverse.strands.concurrent.AbstractQueuedSynchronizer.parkAndCheckInterrupt(AbstractQueuedSynchronizer.java:838)
at co.paralleluniverse.strands.concurrent.AbstractQueuedSynchronizer.doAcquireShared(AbstractQueuedSynchronizer.java:969)
at co.paralleluniverse.strands.concurrent.AbstractQueuedSynchronizer.acquireShared(AbstractQueuedSynchronizer.java:1302)
at co.paralleluniverse.strands.concurrent.Semaphore.acquireUninterruptibly(Semaphore.java:333)
at de.uka.ipd.sdq.simulation.abstractsimengine.desmoj.processes.SimProcessFiberStrategy.suspendProcess(SimProcessFiberStrategy.java:78)
at de.uka.ipd.sdq.simulation.abstractsimengine.processes.SimulatedProcess.suspend(SimulatedProcess.java:93)
at de.uka.ipd.sdq.simulation.abstractsimengine.processes.SimulatedProcess.passivate(SimulatedProcess.java:73)
at de.uka.ipd.sdq.simulation.abstractsimengine.AbstractSimProcessDelegator.passivate(AbstractSimProcessDelegator.java:62)
at de.uka.ipd.sdq.scheduler.resources.active.SimDelayResource.doProcessing(SimDelayResource.java:59)
at de.uka.ipd.sdq.scheduler.resources.active.AbstractActiveResource.process(AbstractActiveResource.java:44)
at de.uka.ipd.sdq.simucomframework.SimuComSimProcess.hold(SimuComSimProcess.java:158)
at de.uka.ipd.sdq.simucomframework.usage.OpenWorkload.waitForNextUser(OpenWorkload.java:99)
at de.uka.ipd.sdq.simucomframework.usage.OpenWorkload.internalLifeCycle(OpenWorkload.java:68)
at de.uka.ipd.sdq.simucomframework.SimuComSimProcess.lifeCycle(SimuComSimProcess.java:177)
at de.uka.ipd.sdq.simulation.abstractsimengine.processes.SimulatedProcess.actions(SimulatedProcess.java:58)
at de.uka.ipd.sdq.simulation.abstractsimengine.processes.ProcessRunnable.run(ProcessRunnable.java:28)
at co.paralleluniverse.strands.SuspendableUtils$VoidSuspendableCallable.run(SuspendableUtils.java:44)
at co.paralleluniverse.strands.SuspendableUtils$VoidSuspendableCallable.run(SuspendableUtils.java:32)
at co.paralleluniverse.fibers.Fiber.run(Fiber.java:1027)
at co.paralleluniverse.fibers.Fiber.run1(Fiber.java:1022)
at co.paralleluniverse.fibers.Fiber.exec(Fiber.java:732)
at co.paralleluniverse.fibers.RunnableFiberTask.doExec(RunnableFiberTask.java:95)
at co.paralleluniverse.fibers.RunnableFiberTask.run(RunnableFiberTask.java:86)
at co.paralleluniverse.common.util.SameThreadExecutor.execute(SameThreadExecutor.java:30)
at co.paralleluniverse.fibers.FiberExecutorScheduler.execute(FiberExecutorScheduler.java:77)
at co.paralleluniverse.fibers.RunnableFiberTask.submit(RunnableFiberTask.java:264)
at co.paralleluniverse.fibers.RunnableFiberTask.unpark(RunnableFiberTask.java:244)
at co.paralleluniverse.fibers.RunnableFiberTask.unpark(RunnableFiberTask.java:204)
at co.paralleluniverse.fibers.Fiber.unpark(Fiber.java:1275)
at co.paralleluniverse.strands.Strand.unpark(Strand.java:685)
at co.paralleluniverse.strands.concurrent.AbstractQueuedSynchronizer.unparkSuccessor(AbstractQueuedSynchronizer.java:666)
at co.paralleluniverse.strands.concurrent.AbstractQueuedSynchronizer.doReleaseShared(AbstractQueuedSynchronizer.java:693)
at co.paralleluniverse.strands.concurrent.AbstractQueuedSynchronizer.releaseShared(AbstractQueuedSynchronizer.java:1373)
at co.paralleluniverse.strands.concurrent.Semaphore.release(Semaphore.java:424)
at de.uka.ipd.sdq.simulation.abstractsimengine.desmoj.processes.SimProcessFiberStrategy.resumeProcess(SimProcessFiberStrategy.java:66)
at de.uka.ipd.sdq.simulation.abstractsimengine.processes.SimulatedProcess.resume(SimulatedProcess.java:108)
at de.uka.ipd.sdq.simulation.abstractsimengine.desmoj.DesmoJSimProcess.access$0(DesmoJSimProcess.java:1)
at de.uka.ipd.sdq.simulation.abstractsimengine.desmoj.DesmoJSimProcess$1.eventRoutine(DesmoJSimProcess.java:62)
at desmoj.core.simulator.Scheduler.processNextEventNote(Scheduler.java:608)
at desmoj.core.simulator.Experiment.doProceed(Experiment.java:1477)
at desmoj.core.simulator.Experiment.access$0(Experiment.java:1469)
at desmoj.core.simulator.Experiment$2.run(Experiment.java:1441)
at co.paralleluniverse.strands.SuspendableUtils$VoidSuspendableCallable.run(SuspendableUtils.java:44)
at co.paralleluniverse.strands.SuspendableUtils$VoidSuspendableCallable.run(SuspendableUtils.java:32)
at co.paralleluniverse.fibers.Fiber.run(Fiber.java:1027)
at co.paralleluniverse.fibers.Fiber.run1(Fiber.java:1022)
at co.paralleluniverse.fibers.Fiber.exec(Fiber.java:732)
at co.paralleluniverse.fibers.RunnableFiberTask.doExec(RunnableFiberTask.java:95)
at co.paralleluniverse.fibers.RunnableFiberTask.run(RunnableFiberTask.java:86)
at co.paralleluniverse.common.util.SameThreadExecutor.execute(SameThreadExecutor.java:30)
at co.paralleluniverse.fibers.FiberExecutorScheduler.execute(FiberExecutorScheduler.java:77)
at co.paralleluniverse.fibers.RunnableFiberTask.submit(RunnableFiberTask.java:264)
at co.paralleluniverse.fibers.Fiber.start(Fiber.java:1058)
at co.paralleluniverse.fibers.Fiber.start(Fiber.java:79)
at desmoj.core.simulator.Experiment.proceed(Experiment.java:1444)
at desmoj.core.simulator.Experiment.start(Experiment.java:2324)
at desmoj.core.simulator.Experiment.start(Experiment.java:2259)
at de.uka.ipd.sdq.simulation.abstractsimengine.desmoj.DesmoJExperiment.startSimulator(DesmoJExperiment.java:49)
at de.uka.ipd.sdq.simulation.abstractsimengine.AbstractExperiment.start(AbstractExperiment.java:61)
at de.uka.ipd.sdq.simucomframework.ExperimentRunner.run(ExperimentRunner.java:46)
at de.uka.ipd.sdq.simucomframework.ExperimentRunner.run(ExperimentRunner.java:61)
at org.palladiosimulator.simulizar.runtimestate.AbstractSimuLizarRuntimeState.runSimulation(AbstractSimuLizarRuntimeState.java:153)
at org.palladiosimulator.simulizar.launcher.jobs.PCMStartInterpretationJob.execute(PCMStartInterpretationJob.java:77)
at de.uka.ipd.sdq.workflow.jobs.SequentialJob.executeWithDelayedCleanUp(SequentialJob.java:111)
at de.uka.ipd.sdq.workflow.jobs.SequentialJob.execute(SequentialJob.java:79)
at de.uka.ipd.sdq.workflow.jobs.SequentialBlackboardInteractingJob.execute(SequentialBlackboardInteractingJob.java:86)
at de.uka.ipd.sdq.workflow.jobs.SequentialJob.executeWithImmediateCleanUp(SequentialJob.java:150)
at de.uka.ipd.sdq.workflow.jobs.SequentialJob.execute(SequentialJob.java:77)
at de.uka.ipd.sdq.workflow.BlackboardBasedWorkflow.execute(BlackboardBasedWorkflow.java:82)
at de.uka.ipd.sdq.workflow.Workflow.run(Workflow.java:90)
at de.uka.ipd.sdq.workflow.launchconfig.AbstractWorkflowBasedLaunchConfigurationDelegate.createAndRunWorkflow(AbstractWorkflowBasedLaunchConfigurationDelegate.java:244)
at de.uka.ipd.sdq.workflow.launchconfig.AbstractWorkflowBasedLaunchConfigurationDelegate.launch(AbstractWorkflowBasedLaunchConfigurationDelegate.java:91)
at org.eclipse.debug.internal.core.LaunchConfiguration.launch(LaunchConfiguration.java:885)
at org.eclipse.debug.internal.core.LaunchConfiguration.launch(LaunchConfiguration.java:739)
at org.eclipse.debug.internal.ui.DebugUIPlugin.buildAndLaunch(DebugUIPlugin.java:1039)
at org.eclipse.debug.internal.ui.DebugUIPlugin$8.run(DebugUIPlugin.java:1256)
at org.eclipse.core.internal.jobs.Worker.run(Worker.java:55)
The text was updated successfully, but these errors were encountered: