RESOLVED FIXED 78004
IndexedDB: Layout Test storage/indexeddb/open-close-version.html is flaky
https://bugs.webkit.org/show_bug.cgi?id=78004
Summary IndexedDB: Layout Test storage/indexeddb/open-close-version.html is flaky
Joshua Bell
Reported 2012-02-07 11:04:50 PST
storage/indexeddb/open-close-version.html is flaky on all platforms This test uses setTimeout() calls to try and keep transactions alive, which is not supported in the spec and probably entering a race condition with timers. The test should be reworked to use IDB correctly.
Attachments
Joshua Bell
Comment 2 2012-02-07 13:56:46 PST
Looking at the flake output, it doesn't appear that the setTimeout is at fault. (And the usage appears correct, contrary to my recollection). Here are three examples: =========================================== Example 1: =========================================== ... TEST: open and setVersion blocked if a VERSION_CHANGE transaction is running - close when blocked window.dbname = 'test3'; window.ver = 1; window.steps = [] 'h1.open' 'h1.open.onsuccess' 'h2.open' 'h2.open.onsuccess' 'h1.setVersion' 'h2.setVersion' <TIMEOUT> =========================================== =========================================== Example 2: =========================================== ... TEST: two setVersions from the same connection window.dbname = 'test6'; window.ver = 1; window.steps = [] 'h1.open' 'h1.open.onsuccess' 'h1.setVersion' 'h1.setVersion' 'h1.setVersion.onsuccess' 'h1.setVersion.transaction-complete' half done <TIMEOUT> =========================================== =========================================== Example 3: =========================================== ... TEST: two setVersions from the same connection window.dbname = 'test6'; window.ver = 1; window.steps = [] 'h1.open' 'h1.open.onsuccess' 'h1.setVersion' 'h1.setVersion'<TIMEOUT> =========================================== Analysis: All three of these point to setVersion calls getting wedged, so that either no events go out (so one handle can close) or the call is never processed. This looks very similar to http://crbug.com/80111 but without the tab closure issue - all of these are occurring within the same renderer process. =========================================== Example : =========================================== ... <TIMEOUT> =========================================== Analysis: ===========================================
Joshua Bell
Comment 3 2012-02-13 17:16:59 PST
This has been happy since http://webkit.org/b/78167 landed in http://trac.webkit.org/changeset/107278 so declaring victory
Note You need to log in before you can comment on or make changes to this bug.