RFR (XXS): 8205633: TestOptionsWithRanges.java of '-XX:TLABSize=2147483648' fails intermittently
Thomas Schatzl
thomas.schatzl at oracle.com
Thu Aug 9 10:47:41 UTC 2018
Hi all,
can I have reviews for this tiny change that excludes maximum range
testing for the TLABSize option from TestOptionsWithRanges?
Manually setting TLABSize forces the VM in serial, parallel and cms to
use that TLAB size. This means that for ergonomically determined heap
sizes, there is a high chance that a single java thread allocating a
TLAB will exhaust eden. Another Java thread allocating objects would in
the normal case cause a GC, but when we are starting up, we can't do a
GC yet and error out as described in the bug.
The suggested fix is to disable doing the maximum range check for
TLABSize in the test because
- there are setups where this TLAB size is valid (i.e. eden large
enough)
- otherwise limiting maximum TLAB size in these collectors could
decrease performance; particularly at startup we can not determine a
good value yet because eden has not been fully sized yet.
CR:
https://bugs.openjdk.java.net/browse/JDK-8205633
Webrev:
http://cr.openjdk.java.net/~tschatzl/8205633/webrev/
Testing:
Failing test case with serial/parallel/CMS 400 times without failure
Thanks,
Thomas
More information about the hotspot-gc-dev
mailing list