lkml.org 
[lkml]   [2009]   [Dec]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    Date
    SubjectRe: git pull on linux-next makes my system crawl to its knees and beg for mercy
    On Wed, Dec 23, 2009 at 7:27 AM, Denys Vlasenko
    <vda.linux@googlemail.com> wrote:
    > On Fri, Dec 18, 2009 at 11:03 PM, Luis R. Rodriguez <mcgrof@gmail.com> wrote:
    >> On Fri, Dec 18, 2009 at 1:56 PM, Stephen Rothwell <sfr@canb.auug.org.au> wrote:
    >>> Hi Luis,
    >>>
    >>> On Fri, 18 Dec 2009 09:26:29 -0800 "Luis R. Rodriguez" <mcgrof@gmail.com> wrote:
    >>>>
    >>>> I tend to always be on a 2.6.32 kernel + John's queued up patches for
    >>>> wireless for the next kernel release (I use wireless-testing). My
    >>>> system is a Thinkpad T61, userspace is Ubuntu 9.10 based (ships with
    >>>> git 1.6.3.3) and I kept an ext3 filesystem to be able to go back in
    >>>> time to 2.6.27 at will without issues.  I git clone'd linux-next a few
    >>>> weeks ago. After a few days I then tried to git pull and my system
    >>>> became completely unusable, It took *ages* to open up a terminal and
    >>>
    >>> The start of the daily linux-next boilerplate says:
    >>>
    >>>> If you are tracking the linux-next tree using git, you should not use
    >>>> "git pull" to do so as that will try to merge the new linux-next release
    >>>> with the old one.  You should use "git fetch" as mentioned in the FAQ on
    >>>> the wiki (see below).
    >>>
    >>> (Unfortunately, the wiki seems to be unavailable at the moment)
    >>>
    >>> I am guessing that the merge that git is attempting is killing your
    >>> laptop (though besides the number of common commits I am not sure why).
    >>> Please try using "get fetch" instead.
    >>
    >> Indeed, I learned my lesson now. Thanks for the details.
    >>
    >> Now granted, even if 'git merge' is killing my laptop due to the
    >> conflicts of the insane merge I was trying to do it *still* should not
    >> make my box completely unresponsive for so long. And given that I'm
    >> using mostly distribution specific kernel config options and my have
    >> ruled out my hard drive it seems a general serious kernel issue even
    >> down to 2.6.27. Whatever git is doing I'm sure other userspace
    >> software can also end up generating and would make any user go
    >> completely bananas. I was about to rip my hair out.
    >
    > Git gurus would know it by heart, but I am not one. So if I were you,
    > I would just do a generic diagnostic run.

    Right its the first thing I did, but its to the extent that even doing
    that is not possible unless you're willing to wait 5-10 minutes for
    some output. I'm not kidding.

    > What is it git is doing
    > so that machine slows down that much? Is it spawning a lot
    > of running processes?

    Doesn't seem like it, the only visible git process is get-merge, I
    forgot to grep for all git processes though, but I think that was the
    only one.

    > Is it allocating/using so much memory
    > that your box goes into a severe swap storm?

    Could be, 979M virtual, 298M resident size (non swapped), 58665 shared.

    Unfortunately when this happens I cannot log into my box and run good
    diagnostics, that's how much of a pain in the bolas this is. Some
    morning I had enough patience I did leave vmstat and iostat running
    and didn't see much out of the ordinary except CPU wait time was
    pretty high. I did manage to get at least htop running once and took a
    screenshot (and this took me about 10 minutes to generate):

    http://bombadil.infradead.org/~mcgrof/images/2009/12/git-merge.jpg

    So if anything it could be the later, that of a swap storm.

    What I should have running is sar, that way I can treck back in time
    when I want to.

    But even when compiling the kernel my machine becomes unusable for a
    few seconds when the linking for vmlinux.o starts and in that case my
    swap usage is about 45 - 125 M. A silly example, pandora reliably poos
    out on firefox requiring a pkill on firefox to get it back while
    vmlinux.o is linking.

    > I guess it is the latter.

    Only it seems to happen with some other things like compiling the kernel.

    I'll see if I can upgrade the memory on this thing.

    > If it is, then it's not a kernel problem -
    > kernel can't magically make your system adequately handle a workload
    > which needs 3 GB for working set when the box only has 2 GB of RAM.
    > It _will_ be very slow.

    Sure, I'll try to keep my eye out on swap overuse, I suppose it could be that.

    I started to be suspicious about the CPU freq governor but I'll note
    on both systems even if I set the freq static to the highest I still
    had issues. I'll also note on both 2.6.27 and 2.6.32 I used:

    CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y

    I started testing CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE=y

    but don't really notice an improvement.

    Luis
    --
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2009-12-23 17:23    [W:0.027 / U:268.124 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site