(In reply to Florian Quèze [:florian] from bug 1429796 comment #18)

(In reply to Dana Keeler (she/her) (use needinfo) (:keeler for reviews) from comment #15)

Looks like we need to whitelist the DB file backing cert_storage.

It looks like the <profile>/security_state/data.mdb file is being accessed from the main thread. Is this intentional? If so, what makes the I/O to this file impossible to do off main thread?

This is visible in profiles: http://bit.ly/2V3H6te

In this case, it's a startup profile, and is showing up right after first paint, when we do a nsIOService::SpeculativeConnect call. It's unfortunate that this speculativeConnect perf optimization causes us to fsync on the main thread.

After initialization (which happens on the main thread because we need to access
preferences), cert_storage will first be used on a certificate verification
thread. We can use this to avoid main-thread I/O by lazily opening the DB when
it first gets used rather than at initialization.

Thanks for working on this, Dana! Does this mean that we can remove this xperf whitelist entry?: https://searchfox.org/mozilla-central/rev/3d469329a42644b41e66944d8da22d78d8c0f5ec/testing/talos/talos/xtalos/xperf_whitelist.json#527-532

Ah - thanks for the reminder. Presumably so - I'm trying now with https://treeherder.mozilla.org/#/jobs?repo=try&revision=09bf7f7ea07e9d8e2e8607f490fa321f119e3d36

https://hg.mozilla.org/mozilla-central/rev/3d4f7e72dadb

Seems we can remove that entry: https://treeherder.mozilla.org/#/jobs?repo=try&revision=09bf7f7ea07e9d8e2e8607f490fa321f119e3d36

Setting 67=wontfix because I assume we don't need to uplift this fix to 67 Beta.

https://hg.mozilla.org/mozilla-central/rev/44f7c1b809ee

== Change summary for alert #20155 (as of Thu, 28 Mar 2019 00:26:15 GMT) ==

Improvements:

5% Heap Unclassified osx-10-10-shippable opt stylo 79,391,981.91 -> 75,423,169.70

For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=20155