Connection Reset by peer

[David BORDAS]

----- Original Message -----
From: "Martin Pool" <mbp at samba.org>
To: "David BORDAS" <bordas at jeuxvideo.com>
Cc: <rsync at samba.org>
Sent: Wednesday, April 10, 2002 11:35 AM
Subject: Re: Connection Reset by peer


> On 10 Apr 2002, David BORDAS <bordas at jeuxvideo.com> wrote:
>
> > All server are running rsync 4.6.xx installed from a red-hat rpm.
> > All OS are Linux red hat on intel platform.
> >
> > On client i tried the same rpm, or several tar.gz version from 4.6.2 to
> > 4.5.5.
>
> I suppose you mean 2.4.6 and 2.5.5?
>

Yes, sorry for this ...

> > I've also tried the "Patch to avoid 'Connection reset by peer' error for
> > rsync on cygwin" found in the list.
> > I know that i'm not using cygwin but i won't to give it a try withou any
> > luck ...
> >
> > Client is like this : intel celeron 600 + 256 Mo of SDRAM and 2 * 30 Go
> > 7200tr UDMA drives.
> > All drive have at least 15 Go free space.
> > Internet connection : 1.2Mb ADSL permanent.
> >
> > The most common "Connection reset by peer" appear when backuping big
> > text files like SQL dump.  Each dump is around 600MB.
> >
> > All server are running rsync as a daemon and client is connecting to
rsync
> > like this :
> > /usr/local/bin/rsync -avrpogtzH --timeout=3600 XXX.XXX.XXX.XXX::backup
> > /unit2/backup/
> >
> > I also have a look to rsync daemon error log on server and i can see
> > something like this :
> > << 2002/04/10 04:50:47 [28350] transfer interrupted (code 11) at
main.c(295)
> > 2002/04/10 04:56:59 [26821] transfer interrupted (code 30) at
> > io.c(65) >>
>
> Code 11 is "IO error", code 30 is "timeout".  (2.5 gives better error
> messages including this text.)
>
> So it seems the server suspects a timeout or disk/network io error,
> and is closing the connection.  Unfortunately it is a bit hard to tell
> from here what exactly is causing those errors to occur.  (Better
> error reporting is a priority for me in the near future.)
>
> You could try installing 2.5.5 on the server, perhaps running on a
> different tcp port using the --port option.
>

Web servers are in production and replication works well, so i can't change
rsync version on it.
Can i launch 2 rsync daemon on different port ?
Is this ok, so on one server i can install a 2.5.5 one with no make install
and try ...

> You could also use the --partial option.  Repeated partial transfers
> might get the whole file across even if there are network problems.
>
> If you want to do get your hands dirty:
>
> You can use a debugging tool on the server to find more information
> about what is going wrong, which will help us fix rsync and resolve
> your problem.  After rsync has started on the server, please find it's
> pid from "ps", and then do
>
>   strace -o /tmp/rsync.trace -f -p PIDOFRSYNC
>

Rsync daemon pid ?

> Try to find the relevant failure in that file (probably near the end)
> and send it a short section through to the list.
>
> Alternatively, you could attach to rsync with gdb and set a breakpoint
> on _exit_cleanup.  The "where" command will then show some useful
> information.
>
>   $ gdb
>   (blah blah)
>   gdb> attach PIDOFRSYNC
>   gdb> break _exit_cleanup
>   gdb> continue
>   (wait)
>   gdb> where
>
> --
> Martin
>

I thought a client problem in fact.
I've set in crontab a little shell script like this :
#!/bin/sh
echo "############################"
date
free -o
echo "############################"

This script will be launch each minute during backuping.
So, tomorow i'll know if i've got a memory problem or not ...

David