severe cache coherency problem

[Love]

"Mattias Engdegård" <f91-men at nada.kth.se> writes:

> I'm creating 2 files, a and b, on host X (Solaris, transarc AFS client).
> They are visible and readable from host Y (Linux 2.2.6, arla 0.23, libc5.4.46).
> When X removes file a, arlad on Y says:
> 
>   Thu Apr 22 00:45:31 1999: arlad: callback (130.237.42.231)
>   Thu Apr 22 00:45:31 1999: arlad: -1: (536880467, 23, 517742)
>   Thu Apr 22 00:45:31 1999: arlad: callback for non-existing file (-1, 536880467, 23, 517742)
> 
> but ls on Y can see both. When X removes file b, arlad reports nothing
> (running with debug=almost-all).

As the -1 hints arlad have seams to miss the address in the connection
cache, and thus fails to find its cell. Could you start arlad with -n
and when this happens do `fs venuslog', this should output a dump of the
state of all parts of arla. Please verify that arla doesn't have
130.237.42.231, in the connection cache.

There should probably be some refcount on the connection from the
FCacheEntry too.

> Trying to open the files produces
> 
>   Thu Apr 22 00:46:46 1999: arlad: worker 0: processing
>   Thu Apr 22 00:46:46 1999: arlad: Rec message: opcode = 12 (open), size = 40
>   Thu Apr 22 00:46:46 1999: arlad: read_data
>   Thu Apr 22 00:46:46 1999: arlad: Error reading length: Network dropped connection because of reset
>   Thu Apr 22 00:46:46 1999: arlad: multi-sending wakeup: seq = 28, error = 102
>   Thu Apr 22 00:46:46 1999: arlad: worker 0: done
>   Thu Apr 22 00:46:46 1999: arlad: worker 0 waiting
>   Thu Apr 22 00:46:46 1999: arlad: worker 0: processing
>   Thu Apr 22 00:46:46 1999: arlad: Rec message: opcode = 10 (inactivenode), size = 32
>   Thu Apr 22 00:46:46 1999: arlad: worker 0: done
>   Thu Apr 22 00:46:46 1999: arlad: worker 0 waiting
>   cat: b: Network dropped connection because of reset

This error should be handled better, but will never occur if the
callback-code works.

Love