Jemalloc 4.0 Released

github.com

75 points·zx2c4··6 comments

Apparently a bug we saw, very rarely, with Redis, where there were strange crashes on OOM, is fixed. This release allows us to just pass parameters with configure in order to obtain a "24 bytes" tiny class, which was introduced recently with Redis unstable. It's cool that jemalloc is actively developed with care.

Do you have a reference to the jemalloc issue/tracking for this bug? Or maybe the Redis one at least? Thanks!

EDIT: NM, it's clear from the release notes:

> "Fix an OOM-related regression in arena_tcache_fill_small(), which could cause cache corruption on OOM. This regression was present in all releases from 2.2.0 through 3.6.0."

https://github.com/jemalloc/jemalloc/issues/98

Are there plans afoot to upgrade to this in Firefox? It'd be interesting to see how much effect these improvements have on Firefox [1], which has been gaining weight for the last 3 years...

[1] https://areweslimyet.com/

According to a comment[1] I read, it's being worked on but jemalloc 3 still isn't enabled by default. I guess it could improve performance for some things, but I think less fragmentation would be the only major win for overall footprint.

[1] http://www.phoronix.com/forums/forum/software/programming-co...

Here's the bug for jemalloc3 by default: https://bugzilla.mozilla.org/show_bug.cgi?id=762449

As of tomorrow's nightly, Firefox should be using jemalloc 4.0+ (actually current dev branch tip + 2 extra patches). It's not set to ride the trains yet, meaning it won't be in Developer Edition 43.0a2 in a few weeks unless things change.

Reply on news.ycombinator.com