Red Hat Bugzilla – Bug 445573
firefox disk write IO storms
Last modified: 2008-07-23 13:04:58 EDT
firefox-3.0-0.60.beta5.fc9.i386 (J5's machine) firefox-3.0-0.60.beta5.fc9.x86_64 (warren's machine) Two days ago I began noticing firefox using tons of disk IO sporadically while the browser is supposed to be idle. iotop shows constant write activity but no read. J5 began noticing this problem yesterday on his i386 machine. Write activity happens while sitting on an idle web page (even as simple as Fedora Wiki) and it goes constantly for a few minutes, then stops. During the disk I/O firefox is still usable but extremely unresponsive. Even closing Firefox takes as much as a minute or two as it continues writing. http://forums.fedoraforum.org/showthread.php?t=187332 mcepl pointed out this thread that hints at urlclassifier*.sqlite files being the cause. https://bugzilla.mozilla.org/show_bug.cgi?id=402469 mcepl pointed out complaints about urlclassifier3.sqlite here, but I don't see any complaints about disk IO.
Cannot reproduce here, but I have heavily working Firefox with insane amount of tabs, so maybe it is just hidden in other activity.
*** Bug 445568 has been marked as a duplicate of this bug. ***
I'm also seeing this problem with: xulrunner-1.9-0.60.beta5.fc9.x86_64 firefox-3.0-0.60.beta5.fc9.x86_64 Firefox starts thrashing the disk almost every time a few minutes into the session no matter what website. Vmstat tells me the CPU is 50%-90% waiting on IO fairly consistently for more than 5 minutes at a time.
I have the same problem here. iotop's (yum install iotop) most actives (examples while Firefox is running): PID USER DISK READ DISK WRITE SWAPIN IO> COMMAND 477 root 0 B/s 22.64 K/s 0.00 % 99.99 % [kjournald] 3293 ferrogoo 116.37 K/s 72.73 K/s 0.00 % 99.99 % firefox -UILocale de ... PID USER DISK READ DISK WRITE SWAPIN IO> COMMAND 214 root 0 B/s 0 B/s 0.00 % 99.99 % [pdflush] 477 root 0 B/s 0 B/s 0.00 % 99.99 % [kjournald] ... PID USER DISK READ DISK WRITE SWAPIN IO> COMMAND 3222 ferrogoo 290.78 K/s 207.18 K/s 0.00 % 79.13 % firefox -UILocale de 477 root 0 B/s 0 B/s 0.00 % 61.14 % [kjournald] 214 root 0 B/s 0 B/s 0.00 % 19.37 % [pdflush] It takes about three minutes to complete. Part of the time Firefox does not respond to any user interaction.
Changing version to '9' as part of upcoming Fedora 9 GA. More information and reason for this action is here: http://fedoraproject.org/wiki/BugZappers/HouseKeeping
Per http://jasondclinton.livejournal.com/66300.html , looks like this may have something to do with upstream bug https://bugzilla.mozilla.org/show_bug.cgi?id=421482 (and peripherally http://bugzilla.kernel.org/show_bug.cgi?id=9546 )
I see this, too. The IO bursts seem to be periodic, and strace'ing firefox during one of those peroids shows it happily working it's way through urlclassifier*.sqlite. Streaming reads from the same filesystem (movie playback) completely stalls during those reads.
*** This bug has been marked as a duplicate of 439908 ***
*** Bug 452718 has been marked as a duplicate of this bug. ***