According to Telemetry, the median shutdown time in Nightly 32 has more than doubled, starting with the May 30th Nightly. http://telemetry.mozilla.org/#filter=nightly%2F32%2FSIMPLE_MEASURES_SHUTDOWNDURATION!nightly%2F31%2FSIMPLE_MEASURES_SHUTDOWNDURATION!nightly%2F30%2FSIMPLE_MEASURES_SHUTDOWNDURATION!nightly%2F29%2FSIMPLE_MEASURES_SHUTDOWNDURATION!nightly%2F28%2FSIMPLE_MEASURES_SHUTDOWNDURATION!nightly%2F27%2FSIMPLE_MEASURES_SHUTDOWNDURATION&aggregates=median&evoOver=Builds&locked=true&sanitize=true&renderhistogram=Table

In some of the newer slow shutdown profiles I had seen that we were doing more expensive more NukeCrossCompartement. Last year killing those was one of our biggest win with shutdown performance. I'd suggestion first investigating that this could did not get brought back on shutdown.

Do we have a regression range for May 30th?

(In reply to David Rajchenbach Teller [:Yoric] from comment #2) > Do we have a regression range for May 30th? No, we don't. The Talos shutdown test is not useful (bug 1022824) :(

Fwiw, May 30th appears to match to https://tbpl.mozilla.org/?fromchange=1e712b724d17&tochange=e6f113c83095

On a hunch, possible suspects: - Asaf Romano – Bug 1016831 - make sure the shared, readonly-db-connection is closed on shutdown. r=mak, brain=yorik - Kyle Huey – Bug 965309: Shut down the sync loop, if any, when we shut down the XHR. r=bent - Shian-Yow Wu – Bug 1010784: Null out sync loop target to avoid assertions when reusing sync XHR. r=khuey - Dan Gohman – Bug 1005113 - IonMonkey: Mark more LUses as AtStart on x86/x64. r=bhackett - David Rajchenbach-Teller – Bug 917883 - Use AsyncShutdown instead of spinning the event loop in healthreporter.jsm. r=gps, r=rnewman