[cvsnt] CVS HEAD stuck?!?

Mon Jul 2 16:41:46 BST 2007

Thanks for the info.  We don't run AV on our CVS servers.  It is 
pointless since the only way to put anything on the box is via CVS, and 
there any binaries are stored as text so the AV can't find anything in 
the repositories anyway.  It took a little while to convince the 
hardware guys of this, but I've found that installing AV's, even when 
they are completely turned off, can make a 10-20% performance difference 
when reading from disk.  And that's just for the ones where I was able 
to gather data; some of them led to crashes, hangs, or blue screens.

I'm always suspicious of the network. :)

Thanks
chuck

Tony Hoyle wrote:
> Bo Berglund wrote:
> 
>> FWIW:
>> I also have zombies appearing from time to time, but they are orders of
>> magnitude less frequent now than a few years back.
>> But they still happen....
> 
> 
> It's a windows issue that it's not possible to work around 100% - 
> because Windows doesn't support signals it has no way of knowing if one 
> end of a socket dies.
> 
> The circumstances where it happens are fairly rare - it requires either 
> the server or client to die unexpectedly.
> 
> Basically what happens is you have an active connection between server 
> an client.. server enters a blocking read/write and the client dies 
> before it responds wih a packet.  On Unix it detects that, sends a 
> signal to the process and everything works as normal.  Windows has no 
> mechanism for handling it, so the server blocks at kernel level forever 
> waiting for the nonexistant client to do something.
> 
> In normal circumstances you won't see it - clients don't die like that. 
>  I'd be suspicious of a network if it was happening frequently, because 
> unless people are in the habit of starting a client and hitting the 
> power switch the conditions to cause it just won't happen.  It's one of 
> those times I'd remove the Antivirus from the equation to be sure it's 
> not a bug in that causing it.
> 
> Tony