Re: 100% CPU Utilization, looping with gettimeofday calls
Ruari Ødegaard <[email protected]> Wed, 23 Sep 2009 12:47:30 +0200
| Newsgroups | gmane.comp.web.opera.linux |
|---|---|
| Organization | Opera |
| Message-ID | <op.u0phlg1m2xhcog@localhost> |
On Mon, 21 Sep 2009 15:44:12 +0200, Witold Baryluk <[email protected]> wrote: > Dnia 2009-09-21, pon o godzinie 09:46 +0200, Ruari Ødegaard pisze: >> On Sun, 20 Sep 2009 20:06:47 +0200, Witold Baryluk >> <[email protected]> wrote: >> >> > Dnia 2009-08-24, pon o godzinie 14:42 +0200, Ruari Ødegaard pisze: >> >> On Thu, 20 Aug 2009 14:01:43 +0200, Witold Baryluk >> >> <[email protected]> wrote: >> >> >> >> > Dnia 2009-08-19, śro o godzinie 13:55 +0200, Ruari Ødegaard pisze: >> >> >> On Wed, 19 Aug 2009 12:34:29 +0200, Witold Baryluk >> >> >> <[email protected]> wrote: >> >> >> >> >> >> Hi Witold, >> >> >> >> >> >> > Hi guys. >> >> >> > >> >> >> > I was reporting this bug many times. But still probably >> >> >> > Opera developers are ignoring this. >> >> >> > >> >> >> > Opera is running tight loop with millions of gettimeofday calls. >> >> >> > >> >> >> > >> >> http://smp.if.uj.edu.pl/~baryluk/opera10_few_seconds_of_strace.txt.gz >> >> >> > >> >> >> > This is present sine 9.6x series, and is present in all >> snapshots, >> >> >> > including Beta3 and 4546. >> >> >> >> >> >> Below is some information from Arjan van Leeuwen, >> >> >> >> >> >> "Unfortunately, this information is not useful to developers. >> >> Although >> >> >> the >> >> >> gettimeofday() calls are the only thing you can see with strace >> >> (since >> >> >> there are no other system calls), they are not what is causing the >> >> 100% >> >> >> CPU usage. And since no one at Opera has ever reproduced this bug, >> >> it's >> >> >> pretty difficult to see what's going on on your system. >> >> >> >> >> >> "What might help is to create a crash log for Opera once this is >> >> >> happening. >> >> >> To do this, wait until you are in this situation, and then send >> >> Opera a >> >> >> SIGABRT signal using kill. This should trigger the crash report >> >> dialog. >> >> >> Make sure to describe the situation you are in within the crash >> >> report, >> >> >> tell >> >> >> us it's a freeze and not a crash, and provide your email address >> so >> >> that >> >> >> we can contact you for further details. You can send several >> reports >> >> >> like >> >> >> this - they are likely to be different, and one of them might >> provide >> >> >> the >> >> >> hint that is needed." >> >> >> >> >> > >> >> > Hi, i prepared simple script, which: >> >> > 1. restores ~/.opera (rsync with another 700MB directory) >> >> > 2. starts opera with no tabs >> >> > 3. measure CPU utilisation of opera >> >> > 4. sends ABRT, wait, and send it again >> >> > 5. copy crash report >> >> > 6. kill all operas >> >> > 7. goto 1 >> >> > >> >> > >> >> > >> >> > Now i have about 200 crash reports, and about 100 of them >> >> > have this problem with high CPU utilisation. (in top it is 85%, >> >> > in ps it is about 50-60^). >> >> > >> >> > I have posted another post but it is awaiting moderation (it have >> >> 26KB) >> >> > >> >> > Where to send this crash report files crash*.txt ? >> >> > I can provide them to trusted Opera developer using my HTTP server. >> >> > It is 90MB uncompressed now. >> >> >> >> Considering your method I doubt there is that much variation. Send me >> >> the >> >> first 5 and I'll have a look to see. >> >> >> > >> > I send you this data about month ago, any progress. AFAIK nobody from >> > Opera even downloaded this data from my server. >> >> I have been on holiday most of that time and this is not the only issue >> I >> am looking at. Also I was expecting you to send me 5 hand picked samples >> directly as I suggested above, rather than a URL you sent with a link >> 50+. >> I guess we misunderstood what the other expected. >> > I completly uderstand this. I just was hoping for some acknowledge. > Thanks you for information. > >> > Of course newest versions of opera is still having this problem. >> >> We still do not have a way to reliably reproduce, nor a full >> understanding >> of what is going on fromthe bug reports. Hence we can't fix this until >> we >> understand what is broken. In addition, I should remind you that the >> vast >> majority of users don't have this issue, hence whilst I completely >> understand that this is very high priority for you it is not the number >> one issue we are looking at right now and hence does not trump all other >> work we are doing. > > I also understand this. I want to provide useful information when > possible. > >> >> I have quickly scanned your data now. It doesn't look like your method >> to >> automate collection was as useful as would be hoped given that none of >> these display 100% CPU utilization (like your original description), >> albeit they are often very high (e.g. several are at 70%+). > > ps and top utilities computes cpu utilization quite differently. when > top shows 100% for opera, in the exactly the same time ps shows about > 70%. > >> >> That said, we >> are not done looking through what you have provided. I may need you to >> run >> a couple more tests or tell me more about your setup but I'll come back >> to >> you on that. Thanks for all your assistance so far, it is appreciated! >> >> I'll send you an email directly as well. The developers did indeed glean some useful information from the logs you sent us and have a couple of ideas. However, we still need to do more testing to work out _exactly_ what is going wrong here. I see that you are restoring the same profile each time. Is this your main ~/.opera profile? Would you consider sharing it with us purely for the purpose of us better understanding what is going on. It will obviously contain personal and hence sensitive information and I would understand completely if you would prefer not do this. Please be assured however that we would _only_ use this for the purposes of understanding this bug and delete it once finished. If you are still not happy to share the profile perhaps you could try creating a stripped down profile that still demonstrates the the issue by removing parts of the profile that you would prefer not share. Alternatively you might be able to generate a new faulty profile by backing up your regular one, starting clean and gradually recreating your regular usage behaviour until you hit this bug. If you have a profile you are willing to share, send me a message and I'll find you a place to upload it. If none of these are viable options we will be continuing to investigate anyway. We just felt this might help speed up the process. P.S. The bug number for this issue is DSK-253633 (I have linked a couple of bug numbers you raised to this master bug number). -- Using Opera's revolutionary e-mail client: http://www.opera.com/mail/ -- Opera-Linux: https://list.opera.com/mailman/listinfo/opera-linux More lists: https://list.opera.com/mailman/listinfo/ Unsubscribe: mailto:[email protected]?subject=unsubscribe