Bug 164951

Summary: Flaky 32-bit JSC test timeouts: stress/op_div-* and ress/op_mod-*
Product: WebKit Reporter: Ryan Haddad <ryanhaddad>
Component: JavaScriptCoreAssignee: Mark Lam <mark.lam>
Status: RESOLVED FIXED    
Severity: Normal CC: ap, fpizlo, ggaren, jfbastien, keith_miller, mark.lam, msaboff, saam, ysuzuki
Priority: P2    
Version: Other   
Hardware: Unspecified   
OS: Unspecified   
See Also: https://bugs.webkit.org/show_bug.cgi?id=155289
Bug Depends on:    
Bug Blocks: 164994    
Attachments:
Description Flags
proposed patch. ysuzuki: review+

Ryan Haddad
Reported 2016-11-18 13:29:15 PST
https://build.webkit.org/builders/Apple%20El%20Capitan%2032-bit%20JSC%20%28BuildAndTest%29/builds/4252 ** The following JSC stress test failures have been introduced: stress/op_div-ConstVar.js.misc-ftl-no-cjit stress/op_div-VarConst.js.misc-ftl-no-cjit stress/op_div-VarVar.js.misc-ftl-no-cjit stress/op_mod-ConstVar.js.misc-ftl-no-cjit stress/op_mod-VarConst.js.misc-ftl-no-cjit stress/op_mod-VarVar.js.misc-ftl-no-cjit
Attachments
proposed patch. (6.27 KB, patch)
2016-11-19 00:13 PST, Mark Lam 		ysuzuki: review+
DetailsFormatted DiffDiff
View All Add attachment proposed patch, testcase, etc.
Ryan Haddad
Comment 1 2016-11-18 13:30:21 PST
Another instance seen here: https://build.webkit.org/builders/Apple%20El%20Capitan%2032-bit%20JSC%20%28BuildAndTest%29/builds/4226 ** The following JSC stress test failures have been introduced: stress/op_div-VarConst.js.misc-ftl-no-cjit stress/op_mod-ConstVar.js.misc-ftl-no-cjit stress/op_mod-VarConst.js.misc-ftl-no-cjit stress/op_mod-VarVar.js.misc-ftl-no-cjit
Mark Lam
Comment 2 2016-11-18 13:36:47 PST
These are all due to time outs. When did these failures started happening? Either we started running on slower bots or potentially, someone introduced a perf regression.
Ryan Haddad
Comment 3 2016-11-18 13:38:10 PST
(In reply to comment #2) > These are all due to time outs. When did these failures started happening? > Either we started running on slower bots or potentially, someone introduced > a perf regression. I'm going through logs now to try to find out.
Ryan Haddad
Comment 4 2016-11-18 13:41:32 PST
Earliest instance I can find on this bot is r208607 / Fri Nov 11 13:56:25 2016 https://build.webkit.org/builders/Apple%20El%20Capitan%2032-bit%20JSC%20%28BuildAndTest%29/builds/4192
Ryan Haddad
Comment 5 2016-11-18 13:42:09 PST
The bot config has not changed.
Mark Lam
Comment 6 2016-11-19 00:13:49 PST
Created attachment 295255 [details] proposed patch. I'm fixing this by simply recognizing that not all tests are equal, and that some just needs more time to run. I've added a --timeoutMultipler=<multiplier> option that can be used on these slower tests to increase their timeout duration. This should fix this issue for good.
Yusuke Suzuki
Comment 7 2016-11-19 00:20:49 PST
Comment on attachment 295255 [details] proposed patch. View in context: https://bugs.webkit.org/attachment.cgi?id=295255&action=review r=me > Source/JavaScriptCore/jsc.cpp:2606 > + auto timeout = std::chrono::microseconds(static_cast<std::chrono::microseconds::rep>(timeoutDuration * 1000000)); Let's use WTF::Seconds.
Mark Lam
Comment 8 2016-11-19 09:41:32 PST
Thanks for the review. I've changed the timeout thread to use WTF::Seconds. Landed in r208925: <http://trac.webkit.org/r208925>.
Mark Lam
Comment 9 2016-11-19 17:58:10 PST
I forgot to also lengthen the timeout for op_mod-* tests. Will do that in https://bugs.webkit.org/show_bug.cgi?id=164994.
Mark Lam
Comment 10 2016-11-23 07:32:09 PST
*** Bug 155289 has been marked as a duplicate of this bug. ***
Note You need to log in before you can comment on or make changes to this bug.