>
> Wojciech Puchar wrote:
> > it happens (mostly observed) when squid -k rotate is started and 150MB
> > squid process travels it's all VM space to write cache directory to
> > /var/spool/squid/swap.state.
> >
> >
> > it's not complete crash, i still can ping machine, routing/NAT still
> > works, but anything else.
>
> OK, so it's not crash as in panic, good. If you wait a while,
> does it recover? The machine might be just thrashing. Turning

unfortunately not. no disk I/O and stall. (3 hours now after crashed)

> softdeps on for the squid cache filesystem might help ...


one filesystem, softdep and noatime on.


in an hour i will send all kernel config, and average state as top and
vmstat reports.

and sysctl.conf but changing vm.filemin/max/bufcache/anonmax doesn't
change things at all.

it's that kernel

NetBSD chylonia.3miasto.net 1.6ZK NetBSD 1.6ZK

but older doesn't work too anyway...same effects