Menu

#25 CPU hog bug

open-works-for-me
nobody
None
5
2001-10-22
2001-10-19
No

When involuntarily disconnected from a server
(either kick or entire server shut down), an action
causes the CPU usage from < 0.1% to 95% or
more.

Discussion

  • Misha Nasledov

    Misha Nasledov - 2001-10-22

    Logged In: YES
    user_id=6597

    I can't reproduce this ... Can you provide me an exact
    method for reproducing this?

     
  • Misha Nasledov

    Misha Nasledov - 2001-10-22
    • status: open --> open-works-for-me
     
  • Travis Reitter

    Travis Reitter - 2001-10-22

    Logged In: YES
    user_id=346641

    I'll try to produce it when I get home today.

    I think that it could have to do with a long-term (several
    days to a week or more) connections, if you weren't
    able to reproduce it (assuming you logged into a server
    and tried having someone kick you).
    The server I've had it happen on has a lot of chat and a
    very large News, and is running server 1.2.3, if that
    helps. Maybe a loop isn't terminated on kick under
    certain circumstances.. I'll try to have a look at the code
    when I can.

     
  • Travis Reitter

    Travis Reitter - 2001-10-23

    Logged In: YES
    user_id=346641

    OK: Did a little scientific experimentation:

    Joining the server I mentioned in the last post (large news,
    constant chat [though I don't think chat is a factor], 1.2.3
    server), I kicked a copy of myself. Instantly, the CPU usage
    surged.

    Then, I connected to my own server (running HXD) twice. I
    kicked myself, and there was an instant spike of CPU, but
    then it dwindled down to nothing just as quickly, though
    really short spikes (maybe < 3%) continually showed up until
    the application was quit.

    -------

    OK.. a little more testing. I connected remotely to myself
    (rather than looped through address 0 directly), loaded the
    news to 10 MB without any added stress. I think HXD only
    loads so much news, regardless of the news size (since the
    other server loads a lot more news at once, and most of the
    news file was cut off). So, it still may be an issue with
    the news on some servers being to large for GtkHx to handle
    when kicked.

    Sidenote: kicking directly through 0 (127.0.0.1) caused the
    same amount of stress as remotely (through my domain name).

     
  • Misha Nasledov

    Misha Nasledov - 2001-10-23

    Logged In: YES
    user_id=6597

    Hmm I'm still confused.

    I tried kicking myself off remote servers, off local
    servers, killing local servers abruptly, suspending them and
    then disconnecting myself or waiting to time out...I
    honestly find no bug. I don't notice any increase in GtkHx's
    CPU usage, GtkHx does not lock up and I don't have to quit
    it and restart it or anything. It seems to handle
    connections perfectly fine for me! It recognizes when it's
    been kicked and everything! News/chat are not factors. There
    aren't any busy loops I created. I am simply using GTK+'s
    main loop with its gdk_input code that does select() on my
    file descriptors and its own events.

     

Log in to post a comment.

MongoDB Logo MongoDB