Bug 157141 – inspector/indexeddb/requestDatabaseNames.html is flaky on Mac

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Ryan Haddad]

Flakiness dashboard:
<https://webkit-test-results.webkit.org/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=inspector%2Findexeddb%2FrequestDatabaseNames.html>

Most recent failure:
<https://build.webkit.org/builders/Apple%20Yosemite%20Debug%20WK2%20(Tests)/builds/11744>

--- /Volumes/Data/slave/yosemite-debug-tests-wk2/build/layout-test-results/inspector/indexeddb/requestDatabaseNames-expected.txt
+++ /Volumes/Data/slave/yosemite-debug-tests-wk2/build/layout-test-results/inspector/indexeddb/requestDatabaseNames-actual.txt
@@ -3,17 +3,17 @@
 CONSOLE MESSAGE: line 10: Created Database 'ቍ'
 CONSOLE MESSAGE: line 10: Created Database '𐍆'
 
-PASS: No IndexedDB databases should exist initially
+FAIL: No IndexedDB databases should exist initially
 Created Database 'Database1'
-PASS: A single IndexedDB database should exist
-["Database1"]
+FAIL: A single IndexedDB database should exist
+["Database1","Page 20Cache 20Test"]
 Created Database 'Database2'
-PASS: Two IndexedDB databases should exist
-["Database1","Database2"]
+FAIL: Two IndexedDB databases should exist
+["Database1","Database2","Page 20Cache 20Test"]
 Created Database 'ቍ'
-PASS: Two IndexedDB databases should exist
-["Database1","Database2","ቍ"]
+FAIL: Two IndexedDB databases should exist
+["Database1","Database2","Page 20Cache 20Test","ቍ"]
 Created Database '𐍆'
-PASS: Four IndexedDB databases should exist
-["Database1","Database2","ቍ","𐍆"]
+FAIL: Four IndexedDB databases should exist
+["Database1","Database2","Page 20Cache 20Test","ቍ","𐍆"]

[Brady Eidson]

INTERESTING.

Test test appears to be conflicting with other tests running at the same time using either the same DatabaseProcess or same database directory.

If multiple WKTR are using the same DatabaseProcess, they shouldn't - That's bad.
If multiple WKTRs are pointing their DatabaseProcesses at the same database directory, they shouldn't - That's bad.

I wonder if this is the source of many of the other WK2 flakes?

[Brady Eidson]

I'm going to make a change to this test to try to prove or disprove this.

[Brady Eidson]

Also, just to clarify, DRTs should *also* not share the same databases directory.

I don't know if they are, but they shouldn't.

[Brady Eidson]

Added one tiny additional bit of logging to help support this theory in http://trac.webkit.org/changeset/200202

In a little bit I'm going to turn on debug build logging to see if this is really happening.

[Brady Eidson]

I did a "run-webkit-tests storage/indexeddb" which runs 3 WKTRs
I also did a "run-webkit-tests -1 storage/indexeddb" which runs 3 DRTs

I then monitored the filesystem using fs_usage.

In either the WK1 or WK2 case, each unique DRT or WKTR did, in fact, have an independent directory for its databases.

So that theory seems scratched...

But the results here clearly show tests stomping on each other in some way.

[Brady Eidson]

Aha. It's a page cache test (of course), and causes problems in the same process down the line.

I can reliably reproduce like so:
run-webkit-tests -1 fast/history/page-cache-indexed-opened-db.html inspector/indexeddb --child-processes=1

[Brady Eidson]

This is actually a simple bug in the "get all database names" code that chokes on database names with spaces in them.

[Brady Eidson]

Created attachment 277652 [details]
Patch

[Brady Eidson]

http://trac.webkit.org/changeset/200212