Do you have experimental flags enabled on your device? They can cause Chrome to restart following login from cold start, causing a delay like this one.

Was fine (~4 seconds) on another Pixel device running 55.0.2883.103 using the same account.

Upgrading the first device to 56.0.2924.28 didn't help.

So presumably something is strange with the profile on the slow device and wiping it will fix the issue.  IMHO the urgent thing to do here is to verify whether this anomaly is being reported correctly via UMA, and once we're sure it is worry only about the 99th percentile from UMA (if my case is rarer than 1% it's perhaps not worth digging into).

Since this shows up only on cold boot, perhaps there should be a separate UMA metric for cold and warm login?  Even 1% of cold logins having this problem would be pretty bad.

Let me know if you'd like my help investigating from this device (not in dev mode, so debugging options are limited - probably to the feedback report I've sent).  Otherwise in a couple weeks I'll reimage this device to make it not so terribly annoying ;-)

chrome://flags don't sync across devices, so it's still possible that that's the source of the problem - could you try resetting all flags to default and see if it improves?

If not, agree that we definitely need to be tracking this in UMA.

(sorry when I wrote c#2 I hadn't seen c#1).

Yep that's it - resetting flags brought it down to 4s.  Is there really no bug there though?  Starting chrome in the first place when booting didn't take anywhere near 22 seconds, so why should restarting it take that long?  I often need the "experimental web platform features" flag enabled to do my job (but that's rare I guess so perhaps not worth worrying too much about).

Also that explains why I wasn't seeing BootTime.Login show up in chrome://histograms (I do after disabling flags).  That must mean the login metric isn't reported at all (because the browser was restarted between the login starting and when it finished), right?  Should we perhaps be writing a login start timestamp out to disk, and then reporting the metric once the browser restart is complete so that BootTime.Login represents an accurate picture (or perhaps better, using a new BootTime.LoginRestart histogram so as not to conflate the existing tone)?

I can't speak for why the restarting process takes so much longer (I am but a humble PM) but perhaps worth investigating once Albert is back.

Seeing how often this affects users with a new histogram would be interesting.

One thing we'd like to do in an upcoming release is surface a notification to users after flags cause Chrome to restart so that they understand what has happened. Wouldn't solve your case, but would help those who flip a flag once and then forget about it.

Thanks.  Yeah not urgent - should totally wait for the new year.  Hopefully adding a new histogram would be pretty trivial, then we can check the 95th percentile and compare the absolute count to the normal case and see if it's worth worrying about.  Perhaps my time is still unusually slow or two few people set flags to make it worthwhile.

But if my experience is representative and setting flags isn't _that_ rare, then IMHO it's worth a deeper investigation.  It makes ChromeOS feel a lot more like Windows IMHO - staring at a black screen wondering what the hell happened to "Chrome speed" ;-)

@zalcorn can you follow up on this?

steel@, would creating this UMA histogram be a good thing for wzang@ to tackle?

Not having the total time login took (whether due to flags restart or not) is a considerable issue.

Please file an issue for that and assign it to me, I'll find someone to work on it (probably wzang@).

The metric not being reported at all is a serious problem. Spoke with Alexander, he will look into this.

Glad to hear it (I thought so too), thank you!

I vaguely recall there was some display panel re-initialization that happen on Chrome restart, perhaps specific to Pixel, that made it take forever. marcheu might know more.

Sonny, this sounds much like issue 681278.
WDYT?

Assigning to Sonny as this sounds very much like 681278.
I do not mark this as duplicate because this one is about regression at boot.