455017 - entire user interface freezes while loading completions for first URL typed on cold start

Status: RESOLVED INACTIVE

(Firefox :: Address Bar, defect)

Description

[Zack Weinberg (:zwol)]

On cold start - that is, with a just-started browser process and none of the profile data in the OS's memory cache - the entire UI freezes up for (what feels like) entire seconds immediately after I type one character in the URL bar, and then again shortly before the first page load completes.  Similar but shorter hangups happen on typing into the URL bar in new tabs, until (I am guessing) the entire history database has been read into memory, at which point they stop.

I am using an XFS partition for everything, so I doubt this is due to the infamous fsync problem.

I attach the output of strace -ff -tt -p <browser-pid>, starting after the initial display was completely loaded (this is about:blank for me) and ending, um, when nspluginwrapper crashed and took the browser with it.  In that window I opened three tabs and requested the load of two web pages: https://mail.google.com/mail/#inbox and http://zwol.livejournal.com/friends/ (in both cases, by typing the first two characters of the domain name and then selecting the first result from the completions list).  I would have loaded another web page in the third tab if the browser hadn't crashed.

Probably a lot of the traces in the tarball are uninteresting, but I don't know which thread is which.  I did find one thread that seemed to switch back and forth between lots of read/write/select on one fd and lots of lseek/read on another fd -- if one of those is communication with the X server and the other is a disk database, that would explain the UI hickups.

Comment 1

[Zack Weinberg (:zwol)]

Attachment: tarball of system call traces with times

Comment 2

[mmc]

I've definitely experienced this on Windows as well.

Comment 3

[Zack Weinberg (:zwol)]

seems plausible that this would bite anyone anywhere, if the problem really is that the UI thread is tied up with the database I/O.

Comment 4

[Ria Klaassen (not reading all bugmail)]

This was already reported a few times so I guess this is a duplicate.

Comment 5

[Marco Bonardo [:mak]]

can someone still reproduce this?

Comment 6

[Zack Weinberg (:zwol)]

It doesn't happen for *me* anymore with current builds on similar hardware and OS.

Comment 7

[BMO Automation]

Per policy at https://wiki.mozilla.org/Bug_Triage/Projects/Bug_Handling/Bug_Husbandry#Inactive_Bugs. If this bug is not an enhancement request or a bug not present in a supported release of Firefox, then it may be reopened.