clone and pull "premature EOF"

[Matt Mackall]

On Fri, 2010-05-07 at 13:25 -0700, Paul Copeland wrote:
> Following up on prior topic "hg pull stalls" -
> 
> I switched to ssh from https and was hopeful that would take care of 
> it.  No luck - Now I it quits with an error with ssh - so that's better 
> than stalling.  When I use --debug it again stops on the kind of file 
> where it stalled with https.  hg clone also fails.
> 
> This is a large server with many projects.  They just moved to a new 
> datacenter.  They have lots of users and they claim that only me and a 
> few others on my project are having this problem.
> 
> Does the error from hg clone below give any hints about the problem?  
> This was all working until a few days ago.  Some of the users on my 
> project are not having this problem and some are.

Is there any pattern to the operating systems involved? Are any Linux
clients failing?

> Appreciate any suggestions.  I'm really stuck on this.
> 
> > hg clone ssh://myname@MYHOST/jfxphoneapps~javafxphoneapps
> > Enter passphrase for key '/cygdrive/c/Users/jot/.ssh/id_rsa':
> > destination directory: jfxphoneapps~javafxphoneapps
> > requesting all changes
> > adding changesets
> > adding manifests
> > adding file changes
> > transaction abort!
> > rollback completed
> > abort: premature EOF reading chunk (got 57157 bytes, expected 978254)
> > remote: Read from remote host MYHOST: Software caused connection abort

Almost all problems with push/pull are on the server side, and there's
no sign that this is an exception. That it's sporadic and affects only
some clients reinforces that. It's very likely hitting a resource limit
or timeout or similar. How long does it take to fail?

If the server folks still aren't buying it, feel free to direct them to
me.

In the mean time, you might be able to make some forward progress using
partial pulls (see pull -r)

-- 
Mathematics is the supreme nostalgia of our time.