933074 - Allow a pldhash to be reasonably overloaded if it cannot be grown further.

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Nicholas Nethercote [inactive]]

It used to be that case that pldhash would allow tables to be overloaded to
100% full if they could not be grown further.  In bug 927705 I changed that so
that 75% full became a hard limit.

Then bug 932781 hit, where pldhash growth failures were causing CC to not
complete and shutdown leaks to occur on Win7 Mochi-2 tests.  The underlying
caused turned out to be something else (badly written tests, primarily) but it
got me thinking... it is reasonable to allow overloading on failure-to-grow, so
long as you don't go all the way to 100%, because then performance truly
plummets.

Comment 1

[Nicholas Nethercote [inactive]]

Attachment: Allow a pldhash to be reasonably overloaded if it cannot be grown further.

This patch allows a pldhash to be overloaded up to 96.875% full if it cannot be
grown.  It makes the fill-to-failure test in TestPLDHash.cpp take about 1.4x
longer, which isn't too bad given that it allows 1.3x more elements.

Comment 2

[Andrew McCreight [:mccr8]]

Marking as MemShrink because it will make the CC fall over less often.

Comment 3

[Nicholas Nethercote [inactive]]

BTW, I tried a bunch of different N values in the |size - (size >> N)| value and N=5 gave good results -- much beyond that didn't seem worthwhile.

Comment 4

[Jason Orendorff [:jorendorff]]

Comment on attachment 825031 [details] [diff] [review]
Allow a pldhash to be reasonably overloaded if it cannot be grown further.

Review of attachment 825031 [details] [diff] [review]:
-----------------------------------------------------------------

Nice.

Comment 5

[Nicholas Nethercote [inactive]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/542b953e3bff

Comment 6

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/542b953e3bff