[Pan-users] Re: Pan 0.120 only queuing tasks

[SciFi]

On Mon, 30 Apr 2007 21:58:40 +0000, Frederik Himpe wrote:
> On Mon, 30 Apr 2007 10:15:55 +0100, Steve Davies wrote:
>> Perhaps someone could install "Wireshark" and capture a trace with
>> "tcp.port == 119" when this problem occurs. That way if it is a network
>> level problem as suggested earlier in the thread, it could be
>> diagnosed.
> 
> It seems this was a great suggestion, as I know what happens now after
> looking at Wireshark's output.
> 
> I have different news servers configured. One of the servers, is not
> valid anymore now (netcat to port 119 of this address gives "Connection
> refused").
> 
> It seems like Pan correctly gets the new headers van GMane. At this
> moment, the status bar changes from "Tasks: 1/1 to 0/1". But then Pan
> still tries to contact the server which refuses connections, with 5
> seconds intervals. Instead, I think Pan should silently give up trying
> to connect to that server if it does not work the first time.
> 
> If interested, I can send the wireshark log (160KB), but I don't think
> it adds more information to this summary.

I wonder if this info might be useful, it comes directly from the
developers of Unison for MacOSX at www.panic.com (yes that's
correct ;) ) via private e-mail to me.

I wrote to the Unison developers:
> (1)  Point to the gMane news-servers at news.gmane.org, get
> the list of groups, then get the headers for either of these
> two groups:
> gmane.comp.video.ffmpeg.devel
> gmane.comp.video.mplayer.user
> 
> (gMane is free & open to the public, no id/pswd required, but
> if you want to post, you need to enter a valid e-mail addy in
> your From: field and answer the resulting msg before it'll be
> posted.  I picked gMane for you-all to test with because we
> all can then see exactly the same articles on the same
> servers. ;) )
> 
> When it comes time to thread the articles in either of the two
> groups mentioned above, you'll see the spinning beachball and
> CPU go to near 100% for a _very_ long time.  My Dual G5 2.7GHz
> usually pins the 2nd CPU at 100%, while the 1st CPU is quite
> busy trying to handle the leftovers.
> 
> There's not that many headers in these groups, I mean as
> compared to the hdtv group let's say (which can be explained
> why hdtv takes so long to thread ;) ).
> 
> The spinning beachball happens _every_ time fetching the
> latest headers is done for these groups, too, even if only one
> header is "new".

Unison developers replied:
> For what it's worth, it appears that gMane is doing something weird.   
> When I do an XOVER in gmane.comp.video.ffmpeg.devel, it repeats a 
> section over and over.

I have filed Pan2 bug #430628 for the long pause problem except it
occurs on all groups and no matter the servers being used, and is
*extremely* irritating esp. on the large groups. 
(Makes me wonder if Unison has stolen code from Pan2? nah say it's
not so ;) )
Pan2 is actually saving/merging/sorting the headers during this
long pause ... OTOH I have no idea what Unison is doing during its
long pause.

To try getting around this for gmane only, I've restricted Pan2 to
use only one session.  I also have some highly tweaked sysctl
settings that help keep the network pipe being filled while Pan2
or Unison takes its time finishing the single-threaded task, then
I see our network monitor sends a huge wad of ACKs when
Pan2/Unison gets that task freed up.  ;)

...anyway, fwiw...