There seems to be a rash of posts lately where people are resetting or
rebooting without getting any data, so I thought I''d post a quick
overview on collecting crash dumps. If you think you''ve got a hang
problem with ZFS and you want to gather data for someone to look at,
then here are a few steps you should take.
If you already know all about gathering crash dumps on Solaris, feel
free to delete now.
1) Make sure crash dumps are enabled
Enable saving of crash dumps by executing as root or with pfexec
''dumpadm -y''.
The most reasonable trade-off of information vs size in the crash dump
is probably ''dumpadm -c curproc''
If you''re running Opensolaris you likely already have a dedicated zvol
as a dump device. If you''re running SXCE you may need to dedicate a
disk or slice for dump purposes. Change or dedicate the dump device
with ''dumpadm -d <device>''
See dumpadm(1M) for more info.
2) There are at least a couple of ways to capture crash dumps.
As root or with pfexec run ''savecore -L''. This is a
point-in-time
capture of what''s happening on the live system. The system continues
to run during the capture so the results can be slightly inconsistent,
but the machine won''t reboot. Good if you think whatever is hung is
still making progress.
If you really don''t mind rebooting, then ''reboot -nd''
will most often
get a dump without the dump also hanging up and forcing you to hard
reset anyway.
---
Number 1 is best done now, before you have a hang. It won''t hurt
anything to have crash dumps enabled - and if you ever get a panic
you''ll have the data needed for someone to analyze the issue.
If the crash dump saving works, the default location for the dumps to be
stored is the directory /var/crash/<hostname>.
-tim
