Strange hang maybe due to IP changes

Strange hang maybe due to IP changes

[Daniel Clemente]

Hi;
  since some days my Emacs (compiled from Bazaar) hangs from time to time. It seems to block on I/O so strongly that I can't even debug it with gdb, strace or lsof (they block when trying to attach). A kill doesn't work, and neither does kill -9 (Emacs doesn't react and stays in D status).

  After each hang, I have noticed that my network card's IP had changed via DHCP. Yes, this is absurd and I don't know how it relates to Emacs, but if I run dhclient3 and wait a while (about a minute), Emacs comes to life again and all the C-g I issued take effect.
  I had IP 192.168.5.123 probably until the DHCP lease expired, then it was 192.168.5.190. A „dhclient3“ run put it again at .123
  At the moment of the hang, another Emacs instance where I had Gnus open stopped being able to connect to the IMAP servers, maybe because the new IP couldn't access Internet.


  Versions:
dhcp3-client                               3.1.1-5ubuntu7
Linux CPU107 2.6.28-11-generic #36-Ubuntu SMP Fri Mar 20 19:40:40 UTC 2009 i686 GNU/Linux
GNU Emacs 23.0.92.3 (i686-pc-linux-gnu, GTK+ Version 2.16.0) of 2009-04-20 on CPU107


  Were there changes in network / blocking code that could have triggered these behaviours?


-- Daniel

PS: I had recently another crash involving XftGlyphExtents/XftCharIndex but this seems unrelated. I have reported the bug.

Re: Strange hang maybe due to IP changes

[joakim]

I've been meaning to write a more thourough reply to this, but since I
never do here goes:

- there is a bug in the bugtracker for this, entered by Marcus Triska,
  and also a patch which helps to some extent.

- I get afflicted by this fearsome bug rather frequently because I
  switch IP-adresses on my laptop frequently. I havent been able to fix
  it, but I have a workaround:

  - when you get the horrbile hang, you probably have changed ip
    adresses, and you also probably have a dangling connection to some
    server. Verify this with lsof. take note which interfaces are
    dangling. for me its usualy a gnus connection to gmane.

    - lsof|grep emacs, notice:
emacs      3422    joakim   38u     IPv4     141972       0t0        TCP 10.0.75.184:60871->80.91.229.13:nntp (ESTABLISHED)


  - now, ifconfig local interfaces for the connection:


80.91.229.13 is ger.gmane.org, so its a gnus socket.
10.0.75.184 is no longer valind, eth0 dead!

then i created pseudo interfaces:
ifconfig eth0 80.91.229.13
ifconfig eth0:1 10.0.75.184

This seems to work every time so far, but you need of course adapt to
your own ip adresses.

/Joakim


Daniel Clemente <dcl441-bugs@yahoo.com> writes:

> Hi;
>   since some days my Emacs (compiled from Bazaar) hangs from time to time. It seems to block on I/O so strongly that I can't even debug it with gdb, strace or lsof (they block when trying to attach). A kill doesn't work, and neither does kill -9 (Emacs doesn't react and stays in D status).
>
>   After each hang, I have noticed that my network card's IP had changed via DHCP. Yes, this is absurd and I don't know how it relates to Emacs, but if I run dhclient3 and wait a while (about a minute), Emacs comes to life again and all the C-g I issued take effect.
>   I had IP 192.168.5.123 probably until the DHCP lease expired, then it was 192.168.5.190. A „dhclient3“ run put it again at .123
>   At the moment of the hang, another Emacs instance where I had Gnus open stopped being able to connect to the IMAP servers, maybe because the new IP couldn't access Internet.
>
>
>   Versions:
> dhcp3-client                               3.1.1-5ubuntu7
> Linux CPU107 2.6.28-11-generic #36-Ubuntu SMP Fri Mar 20 19:40:40 UTC 2009 i686 GNU/Linux
> GNU Emacs 23.0.92.3 (i686-pc-linux-gnu, GTK+ Version 2.16.0) of 2009-04-20 on CPU107
>
>
>   Were there changes in network / blocking code that could have triggered these behaviours?
>
>
> -- Daniel
>
> PS: I had recently another crash involving XftGlyphExtents/XftCharIndex but this seems unrelated. I have reported the bug.
>
>
-- 
Joakim Verona