RFR: 8273438: Enable parallelism in vmTestbase/metaspace/stressHierarchy tests
Aleksey Shipilev
shade at openjdk.java.net
Mon Sep 13 08:48:53 UTC 2021
On Thu, 9 Sep 2021 19:54:47 GMT, Mikhailo Seledtsov <mseledtsov at openjdk.org> wrote:
>> Current `vmTestbase/metaspace/stressHierarchy` tests (part of vmTestbase_vm_metaspace suite) contains about 15 tests, each running exclusively. There seem to be no reason to run them exclusively, though: they complete in reasonable time, are single-threaded, and consume the usual amount of memory. There is no evidence in JBS that they ever timed out without a reason, and their history unfortunately predates OpenJDK to see why they were not concurrent from day one.
>>
>> We should consider enabling parallelism for `vmTestbase/metaspace/stressHierarchy` and get improved test performance. Currently it is blocked by `TEST.properties` with `exclusiveAccess.dirs` directives in them.
>>
>> Note there are other exclusive tests in `vmTestbase_vm_metaspace`, but those seem to be the hard stress tests: pushing GC to the limits, or doing many threads, etc.
>>
>> Motivational test time improvements below.
>>
>> Before:
>>
>>
>> $ time CONF=linux-x86_64-server-fastdebug make run-test TEST=vmTestbase_vm_metaspace | ts -s
>> ...
>> 00:24:53 ==============================
>> 00:24:53 Test summary
>> 00:24:53 ==============================
>> 00:24:53 TEST TOTAL PASS FAIL ERROR
>> 00:24:53 jtreg:test/hotspot/jtreg:vmTestbase_vm_metaspace 25 25 0 0
>> 00:24:53 ==============================
>> 00:24:53 TEST SUCCESS
>> 00:24:53
>> 00:24:53 Finished building target 'run-test' in configuration 'linux-x86_64-server-fastdebug'
>>
>> real 24m53.389s
>> user 53m2.029s
>> sys 1m1.849s
>>
>>
>> After:
>>
>>
>> $ time CONF=linux-x86_64-server-fastdebug make run-test TEST=vmTestbase_vm_metaspace | ts -s
>> ...
>> 00:04:04 ==============================
>> 00:04:04 Test summary
>> 00:04:04 ==============================
>> 00:04:04 TEST TOTAL PASS FAIL ERROR
>> 00:04:04 jtreg:test/hotspot/jtreg:vmTestbase_vm_metaspace 25 25 0 0
>> 00:04:04 ==============================
>> 00:04:04 TEST SUCCESS
>> 00:04:04
>> 00:04:04 Finished building target 'run-test' in configuration 'linux-x86_64-server-fastdebug'
>>
>> real 4m4.574s
>> user 56m10.582s
>> sys 1m4.725s
>
> This looks like a good change to me. Please allow me some time to run multiple stress testing of these tests with exclusiveAccess removed. I should have the results tonight PST, or tomorrow.
Thank you @mseledts. I guess I need a second reviewer for this. @iignatev, @dholmes-ora?
-------------
PR: https://git.openjdk.java.net/jdk/pull/5391
More information about the hotspot-dev
mailing list