-
Bug
-
Resolution: Unresolved
-
P4
-
8, 9
RedefineBigClass.sh and RetransformBigClass.sh are targeted to detect a memory leak in Redefine/Retransform operations.
These are memory stress tests that have been failing intermittently several times, most often due to stressed running context.
We should move them to some stress suite (may be Kitchensink ?) to ensure the running context is fully controlled, and avoid/control any intermittent/unexpected failure.
Some refactoring is also needed to ensure a proper (safe) memory leak detection when it fails (many false alarms in the past):
- memory leak detection should ONLY be based on raising OutOfMemory exception (as was the case in their initial design).
- as any such kind of test, they should incorporate three running phases: warning-up (current is somewhat too short), steady state(stress phase), warming-down (let VM do any possible cleanup)
- all collected figures (if still needed after OOM refactoring) should use Native Memory Tracking (as is the case today).
- vm memory flags (-Xm*) must be chosen carefully to effectively stress memory and force GCs operations, possibly depending on the running arch.
-----------------------
I've closed this Enhancement and re-opened as a bug to cover the test failures in nightly.
We do not have resources at this point to work on the test refactoring.
These are memory stress tests that have been failing intermittently several times, most often due to stressed running context.
We should move them to some stress suite (may be Kitchensink ?) to ensure the running context is fully controlled, and avoid/control any intermittent/unexpected failure.
Some refactoring is also needed to ensure a proper (safe) memory leak detection when it fails (many false alarms in the past):
- memory leak detection should ONLY be based on raising OutOfMemory exception (as was the case in their initial design).
- as any such kind of test, they should incorporate three running phases: warning-up (current is somewhat too short), steady state(stress phase), warming-down (let VM do any possible cleanup)
- all collected figures (if still needed after OOM refactoring) should use Native Memory Tracking (as is the case today).
- vm memory flags (-Xm*) must be chosen carefully to effectively stress memory and force GCs operations, possibly depending on the running arch.
-----------------------
I've closed this Enhancement and re-opened as a bug to cover the test failures in nightly.
We do not have resources at this point to work on the test refactoring.
- relates to
-
JDK-8061177 java/lang/instrument/RetransformBigClass.sh failed with ExitCode: 134 (corrupted control word detected)
- Closed
-
JDK-8178700 ProblemList.txt update for RedefineBigClass.sh and RetransformBigClass.sh
- Closed
-
JDK-8067353 java/lang/instrument/RedefineBigClass.sh timeout
- Closed