RFR: 8141278: New tests for PLAB testing

Michail Chernov michail.chernov at oracle.com
Tue Jan 19 17:11:31 UTC 2016


Hi Thomas,

Thanks a lot for your comment!

Updated webrev
http://cr.openjdk.java.net/~mchernov/8141278/webrev.01/
http://cr.openjdk.java.net/~mchernov/8141278/webrev.01_to_00/

See comments below:

On 14.01.2016 17:04, Thomas Schatzl wrote:
>
> General comments:
>
> - would it help if the PLAB logging would print out what generation the
> particular output is for? That would probably make the code easier to
> understand than the current way of filtering out every second printout
> (and potentially decrease the size of LogParser a lot).
Yes, it would be good. Appropriate CR was created couple months ago: 
https://bugs.openjdk.java.net/browse/JDK-8141141
>
> Also, unified logging should help you getting only the messages you are
> interested in.
New logging options are in PLABUtils.java - -Xlog:gc=debug,gc+plab=debug
> TestPLABNoResize.java
>
> - the @summary description should be more concise, e.g. change
> @summary: "Test for PLAB allocation in Survivor and Old" to "Test that
> PLABs are never resized if PLAB resizing is disabled."
Summary was updated. Also TestPLABNoResize was renamed to 
TestPLABPromotion. I've made it because this test is also applicable for 
resizing PLAB. This test doesn't check fixed size of PLAB, but common 
PLAB's behavior. So, added new test cases with -XX:+ResizePLAB 
(TestPLABPromotion.java line 100+)
>
> - line 65: the comment does not tell what the threshold is used for.
> E.g. instead of "some threshold" use "Threshold to determine whether
> the correct amount of objects were promoted. This is only an
> approximate threshold for these checks."
Comment was updated. Threshold is so big because on some cases on some 
platforms I got next log:

Test case details:
   Young gen size : 64M
   Predefined PLAB size : 1024
   Parallel GC buffer waste pct : 30
   Chunk size : 1000
   Parallel GC threads : 3
   Objects are created : reachable
   PLAB size is fixed: yes
Test expectations:
   PLAB allocation : unexpected
   Direct allocation : expected
Command line: [/export/home/aurora/CommonData/TEST_JAVA_HOME/bin/java -d64 -cp /export/home/aurora/CommonData/jtreg_dir/lib/javatest.jar:/export/home/aurora/CommonData/jtreg_dir/lib/jtreg.jar:/export/home/aurora/sandbox/results/workDir/classes/1/gc/g1/plab:/export/home/aurora/CommonData/j2se_jdk/hotspot/test/gc/g1/plab:/export/home/aurora/sandbox/results/workDir/classes/1/testlibrary:/export/home/aurora/sandbox/results/workDir/test/lib:/export/home/aurora/sandbox/results/workDir/classes/1 -Xmixed -server -Xmixed -server -Xbootclasspath/a:. -XX:+UnlockDiagnosticVMOptions -XX:+WhiteBoxAPI -Xlog:gc=debug,gc+plab=debug -XX:+UseG1GC -XX:G1HeapRegionSize=1m -XX:OldSize=64m -XX:-UseAdaptiveSizePolicy -XX:-UseTLAB -XX:SurvivorRatio=1 -XX:ParallelGCThreads=3 -XX:ParallelGCBufferWastePct=30 -XX:OldPLABSize=1024 -XX:YoungPLABSize=1024 -XX:-ResizePLAB -Dchunk.size=1000 -Dreachable=true -XX:NewSize=64m -XX:MaxNewSize=64m -Dmem.to.fill=33554432 gc.g1.plab.lib.AppPLABPromotion ]
Survivor PLAB allocated:             4224 Direct allocated:          33551368 Mem consumed:         33554432
Old      PLAB allocated:             4224 Direct allocated:          33822640 Mem consumed:         33554432

[0.077s][info   ][gc] Using G1
[5.564s][info   ][gc] GC(0) Pause Full (Last ditch collection) 1M->0M(128M) (5.497s, 5.564s) 66.905ms
[6.130s][debug  ][gc,plab] GC(1)  (allocated = 682 wasted = 0 unused = 154 used = 528 undo_waste = 0 region_end_waste = 336 regions filled = 0 direct_allocated = 4193921 failure_used = 0 failure_waste = 0)
[6.130s][debug  ][gc,plab] GC(1)  (allocated = 0 wasted = 0 unused = 0 used = 0 undo_waste = 0 region_end_waste = 0 regions filled = 0 direct_allocated = 0 failure_used = 0 failure_waste = 0)
[6.143s][info   ][gc     ] GC(1) Pause Young (WhiteBox Initiated Young GC) 32M->32M(128M) (5.967s, 6.143s) 175.963ms
[6.558s][debug  ][gc,plab] GC(2)  (allocated = 0 wasted = 0 unused = 0 used = 0 undo_waste = 0 region_end_waste = 0 regions filled = 0 direct_allocated = 0 failure_used = 0 failure_waste = 0)
[6.558s][debug  ][gc,plab] GC(2)  (allocated = 682 wasted = 0 unused = 154 used = 528 undo_waste = 33909 region_end_waste = 336 regions filled = 0 direct_allocated = 4227830 failure_used = 0 failure_waste = 0)
[6.570s][info   ][gc     ] GC(2) Pause Young (WhiteBox Initiated Young GC) 32M->33M(128M) (6.144s, 6.570s) 426.536ms

Difference for Survivor is 3064, for Old is 268208. Does it mean that 
PLAB works incorrectly? No, it means that test should handle this 
'special case'. This is the reason, why threshold is so big.
>
> Actually I am not sure whether it is useful to all these checks with
> the threshold, because it seems to be very large.
This test doesn't check that VM promotes exactly the same amount of 
memory from on gen to other, so at my point of view using of threshold 
here is good way to solve some potential problems which this test can 
cause in nightly testing.
>
> - line 144: I think "No dead object should be promoted." reads better
Fixed.
> - line 145+: if we really need to check the correct amount of promoted
> bytes (not sure if this is useful here), also check that the sums of
> plabAllocated and directAllocated are below that threshold.
Added, see TestPLABPromotion 188+.
> - maybe the TestCase class could (but possibly should not?) be merged
> in some way that the TestCase for PLABResize seems to be a part of the
> TestCase for PLABNoResize
Don't think that there are two similar cases. It can be one class, but 
it will have a lot of extra unnecessary code for both cases.
> The Storage class ...
> This class will not store than a given number of objects...
> If its capacity is exceeded, old objects will be overwritten.
> E.g. if capacity is one, only the last object added will be saved.
Storage was renamed to MemoryConsumer. Also comments were  fixed. Thanks 
Michail
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.java.net/pipermail/hotspot-gc-dev/attachments/20160119/5e4fda5b/attachment.htm>


More information about the hotspot-gc-dev mailing list