Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8179084

HotSpot VM fails to start when AggressiveHeap is set

    Details

    • Subcomponent:
      gc
    • Introduced In Build:
      b131
    • Introduced In Version:
      9
    • Resolved In Build:
      b168
    • CPU:
      generic
    • OS:
      generic

      Backports

        Description

        FULL PRODUCT VERSION :
        java version "1.8.0_131"
        Java(TM) SE Runtime Environment (build 1.8.0_131-b11)
        Java HotSpot(TM) 64-Bit Server VM (build 25.131-b11, mixed mode)

        FULL OS VERSION :
        Mac OS X Sierra 10.12.4
        Darwin ID-PC14001.local 16.5.0 Darwin Kernel Version 16.5.0: Fri Mar 3 16:52:33 PST 2017; root:xnu-3789.51.2~3/RELEASE_X86_64 x86_64

        A DESCRIPTION OF THE PROBLEM :
        When specifying -XX:+AggressiveHeap as JVM argument the VM now aborts with an error message . This used to work perfectly fine in all previous versions of the JRE/JDK

        THE PROBLEM WAS REPRODUCIBLE WITH -Xint FLAG: Yes

        THE PROBLEM WAS REPRODUCIBLE WITH -server FLAG: Yes

        REGRESSION. Last worked in version 8u121

        STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
        In the simplest form even this command line aborts
        "java -XX:+AggressiveHeap -version" with error message
        "The Parallel GC can not be combined with -XX:ParallelGCThreads=0"

        In previous versions this simple cmd line would print the version. But any other combination doesn't work either.

        Additionally trying to use PrintFlagsFinal to check the actual value still causes the error.
        Many products ship with the setting in some config file out of the box (as do our own when set to production level during install)

        PrintFlagsFInal confirms that it was using 8 threads on my machine prior to u131.

        EXPECTED VERSUS ACTUAL BEHAVIOR :
        That the virtual machine starts up with AggressiveHeap enabled without error
        ERROR MESSAGES/STACK TRACES THAT OCCUR :
        "The Parallel GC can not be combined with -XX:ParallelGCThreads=0"

        REPRODUCIBILITY :
        This bug can be reproduced always.

        ---------- BEGIN SOURCE ----------
        java -XX:+AggressiveHeap -version
        ---------- END SOURCE ----------

        CUSTOMER SUBMITTED WORKAROUND :
        You can work arround this by explicitly specifying a value for ParallelGCthreads only if that argument is earlier in the argument list than AggressiveHeap.

        But having to figure out why certain services/java based apps no longer start and having to edit the config files in the right location and order is annoying at best.

          Issue Links

            Activity

            Hide
            kbarrett Kim Barrett added a comment -
            There don't seem to be any tests at all for -XX:+AggressiveHeap. Also, the documentation for it seems pretty thin.

            I think [~dholmes] suggestion of removing the FLAG_SET_DEFAULT(ParallelGCThreads, ...) form from set_aggressive_heap_flags() is okay for JDK 9 and 8u. In the longer term, we might want to do something a little more principled.

            -XX:+AggressiveHeap is handled in a rather strange way. It is not handled as a normal flag option. (-XX:-AggressiveHeap isn't even recognized.) Instead, while running down the list of CLAs, we call set_aggressive_heap_flags as soon as that option is seen. So the problem at hand can't even be worked around by adding -XX:ParallelGCThreads=<n> to the end of the options; it must preceed the AggressiveHeap option.

            set_aggressive_heap_flags sets a number of options. Some of that looks plausible. But where the value is obtained from something not directly and obviously available at that point, there may be problems. The too early attempt to get the active processor count is one such. I think there might be some similar problems in the heap and new space sizing.
            Show
            kbarrett Kim Barrett added a comment - There don't seem to be any tests at all for -XX:+AggressiveHeap. Also, the documentation for it seems pretty thin. I think [~dholmes] suggestion of removing the FLAG_SET_DEFAULT(ParallelGCThreads, ...) form from set_aggressive_heap_flags() is okay for JDK 9 and 8u. In the longer term, we might want to do something a little more principled. -XX:+AggressiveHeap is handled in a rather strange way. It is not handled as a normal flag option. (-XX:-AggressiveHeap isn't even recognized.) Instead, while running down the list of CLAs, we call set_aggressive_heap_flags as soon as that option is seen. So the problem at hand can't even be worked around by adding -XX:ParallelGCThreads=<n> to the end of the options; it must preceed the AggressiveHeap option. set_aggressive_heap_flags sets a number of options. Some of that looks plausible. But where the value is obtained from something not directly and obviously available at that point, there may be problems. The too early attempt to get the active processor count is one such. I think there might be some similar problems in the heap and new space sizing.
            Hide
            kbarrett Kim Barrett added a comment - - edited
            Fix Request

            Without this fix we have a regression of a failure to start java when using a not uncommon
            command line option.

            The fix is low risk. The removed defaulting of ParallelGCThreads is
            covered by a later defaulting of that option to the same value.

            The proposed change also includes a test to verify simple use of the
            option doesn't crash and has some of the expected effects.

            Webrev of the fix is here:
            http://cr.openjdk.java.net/~kbarrett/8179084/hotspot.00/
            Show
            kbarrett Kim Barrett added a comment - - edited Fix Request Without this fix we have a regression of a failure to start java when using a not uncommon command line option. The fix is low risk. The removed defaulting of ParallelGCThreads is covered by a later defaulting of that option to the same value. The proposed change also includes a test to verify simple use of the option doesn't crash and has some of the expected effects. Webrev of the fix is here: http://cr.openjdk.java.net/~kbarrett/8179084/hotspot.00/
            Hide
            hgupdate HG Updates added a comment -
            URL: http://hg.openjdk.java.net/jdk9/dev/hotspot/rev/f1cca489e9c6
            User: kbarrett
            Date: 2017-04-27 03:32:08 +0000
            Show
            hgupdate HG Updates added a comment - URL: http://hg.openjdk.java.net/jdk9/dev/hotspot/rev/f1cca489e9c6 User: kbarrett Date: 2017-04-27 03:32:08 +0000
            Hide
            wyandi Winston Yandi added a comment -
            It is a critical pre approval to push the fix into the mainline repo.
            Show
            wyandi Winston Yandi added a comment - It is a critical pre approval to push the fix into the mainline repo.
            Hide
            hgupdate HG Updates added a comment -
            URL: http://hg.openjdk.java.net/jdk9/jdk9/hotspot/rev/f1cca489e9c6
            User: lana
            Date: 2017-05-03 19:45:36 +0000
            Show
            hgupdate HG Updates added a comment - URL: http://hg.openjdk.java.net/jdk9/jdk9/hotspot/rev/f1cca489e9c6 User: lana Date: 2017-05-03 19:45:36 +0000

              People

              • Assignee:
                kbarrett Kim Barrett
                Reporter:
                webbuggrp Webbug Group
              • Votes:
                0 Vote for this issue
                Watchers:
                9 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: