Re: Solaris 10 11/06 x86_64: almost impossible to shutdown the system

Andrew Gabriel wrote:

In article <en0njc$d68$1@xxxxxxxxxxxxxxxx>,
Miroslav Zubcic <news@xxxxxxxxxxxxx> writes:

Meanwhile, I found that reboot(1M) _does_ umount file systems on reboot,
but poweroff doesn't (no RAID rebuild on next boot).

Struggling to understand what you mean here.

Sorry for my bad english ... What I want to say is: when shutdown
procedure finishes (hangs), I can reboot system without getting
inconsistent SVM raid or filesystem check delay on next system boot.
(From still-alive ssh connection, or scheduled reboot in rc5 script).

reboot and poweroff do virtually the same thing from the OS
perspective (actually they're the same command).

Hm ... yes, they are the same on diff. Hard links probably. This means
that I can even use poweroff in /etc/rc5. Problem remains when going to
single user mode or runlevel 6 or 5 - console hangs:

/var/svc/log/svc.startd.log:
pro 28 02:56:26/6: couldn't get psinfo data for svc:/system/sac:default
(No such file or directory); assuming failed

And then poweroff (and probably reboot) doesn't sync and umount
filesystems. Only situation when I do not have long delay on next boot
when it prints "checking ufs filesystems" and/or RAID autorebuild, is
when system is more/less shut down, and then reboot/poweroff occurs
(from rc5 script where I put sleep 48 && reboot).

BTW:

/var/adm/messages:
Dec 29 01:36:05 hegel genunix: [ID 672855 kern.notice] syncing file
systems...
Dec 29 01:36:05 hegel genunix: [ID 904073 kern.notice] done

This is displayed 1 second before poweroff/reboot! If filesystems are
umounted (and syslog is still alive as we see) how can syslog still
write to /var/adm/messages on /var partition? Looks like partitions are
never unmounted before reboot.

There have been problems in the past
with IDE drives with write caches enabled failing to get the write
cache flushed before the power was cut, which isn't going to be
the case with a reboot as it remains powered after the OS has
vanished.

I have 2 SATA II disks with 5 partitions in SVM RAID1 configuration. It
worked on Solaris 10 update 2. Problem escalated after I upgraded to
update 3, that is, from 118855-19 to 118855-33 kernel patch revision.

sync; sync; sync

Well, that shouldn't be necessary nowadays, but if you must,
then you might as well stick a "lockfs -fa;" between each
of those syncs too to clear out the UFS logs.

lockfs ? ... Good idea for workaround. Thanks. Will try, but I will
really like to solve this strange problem, and shutdown/reboot normally
with init 5, init 6, shutdown, telinit ...


--
Man is something that shall be overcome.
-- Friedrich Nietzsche
.