RFR (T): 8211211: vmTestbase/metaspace/stressDictionary/StressDictionary.java timeout
kim.barrett at oracle.com
Tue Nov 27 23:20:44 UTC 2018
> On Nov 27, 2018, at 6:38 AM, Thomas Schatzl <thomas.schatzl at oracle.com> wrote:
> Hi all,
> can I have reviews for this trivial change to the test in the subject
> to avoid timeouts due to some metaspace verification being very slow?
> This tests creates hundreds/thousands of classes to be unloaded. Class
> unloading, and in particular metaspace freeing has some very nasty and
> slow verification code (basically O(n^2)).
> This problem is particular problematic with CMS with precleaning, which
> basically delays remark for 5 seconds (until it times out), during
> which the test continues to add more and more metadata that takes very
> long to clean out. I.e. that n is rather large.
> The other collectors have a much shorter turnaround time, so n is much
> smaller, and the slow metaspace verification code is not problematic.
> This change fixes the test to disable CMS precleaning, avoiding that n
> gets too large.
> I filed JDK-8214356 to look into shortening the actual metaspace
> verification process.
> Running the test in CI with CMS enabled (and disabled), now takes as
> long as with the other collectors
More information about the hotspot-gc-dev