[Users] System hangs

[Kirill Korotaev]

Arnold,

> Okay. I thought it might be something interesting, because the cron was
> in 'simfs:sim_systemcall', and afaik that is openvz specific?
simfs is openvz specific. However, it was sleeping in another place  at pipe_dowait().
calltraces have all the addresses which look like the kernel addresses,
so it contains some garbarge usually, which is the case.
it is possible to check via assembler dump of vmlinux file.

> I can, but it's about 400KB (18KB compressed, only the sysrq parts) so
> I'd rather not post it to a mailinglist. But if you wish, I can put it
> online or email it separately?
email privately to me plz. I'll take a look when have some time.

> The clients defaulted to TCP, hard mounts. I've switched over to UDP,
> soft mounts, 8KB block sizes, and enabled jumbo frames as specified by
> the NFS manpages. The systems survived over the weekend.  I'll switch
> one back to the original TCP setup, and leave one at UDP, to verify that
> its indeed the cause.
aha, so you mean your mount changes cured the problem?

Thanks,
Kirill