[Netatalk-admins] Re: [Netatalk-devel] "dsi_stream_write: Broken pipe" errors

[didier <dga...@ma...>]

Andrew wrote:
>     Hello
>=20
>     The users aren=92t experiencing any actual connection problems at a=
ll.
>     I did have that with 1.5.3.1 and found that where I couldn=92t conn=
ect
>     to the Server with a login that this would be rectified, as you=92v=
e
>     mentioned, by removing the top level .AppleDB from that share. This
>     worked fine as required and I have never had this problem (even
>     once) with 1.5.5 or 1.6. In fact if the load average wasn=92t
>     increasing they would have no idea that there was a problem.
>=20
>     These errors seem (could be wrong !) to correspond to a workstation
>     that unexpectedly crashes while connected to the Server. If they
>     reboot they can connect in completely fine but it leaves a running
>     process that corresponds ID-wise to the broken pipe errors. On one
>     site there may be 10-12 lines such as =9327482 ?  R    464:06
>     /usr/local/sbin/afpd -g nobody -c 20 -n FileServer=94 while there a=
re
>     physically only 3 machines connecting to it. I can only explain thi=
s
>     with the workstation being force rebooted while connected to the
>     Server although like I say they can connect in fine when the machin=
e
>     boots up again (I would suggest that this means that the CNID
>     database is ok) . Because of the unexpected quit the process remain=
s
>     running at the Server, and correspondingly the load average increas=
es
Can you attach gdb to one of those run away afpd process? I'll commit a=20
fix to the broken pipe loop but it will only mask the problem.

Didier