By clive on Sep 30, 2009
The Sun Corp User Group meeting yesterday went well. We had a fair turn out, some good questions and I have been asked to do a couple of company specific sessions as follow on.
Sometimes you don't always give the best answer when answering cold. Once you have had a little time to mull it over you think of a possible better answer or an equally valid alternative approach(or you were just wrong the 1st time, it happens sometimes).
To the gentleman with the umount problem which fails every few mounts prior to a backup, I would also make sure the assumption about why umount is failing with a busy file. If you are getting an error message in the log which points that way, then fine, if not, it would be a good idea to get the error code of the umount2 system call when it does fail. We could use DTrace, but we could also wrap the umount command inside truss in the script
truss -t umount2 -v umount2 umount /wibblemake sure it is logged. Still think the live dump is the right way to chase this if you can script it right and don't want to sit by the machine for months on end at 2am when the backup kicks off. This is still the wrong answer as ZFS snapshots would allow you to avoid the umount altogether.
To the people concerned about Oracle performance on ZFS, the offer of 30 minutes or so of SharedShell (http://www.sun.com/123) still stands. One question I should of also asked is if the ZFS intent log is split out and on separate fast storage.
On the way home I went for a run near New Rador. Very nice area indeed and great running. I did come across a flag pole with a red flag and some signage which discussed explosives and this being a restricted area. From later discussions with a local this was a test range and people with metal detectors turn up, hoist the flag and dispose of the various bits of scattered ordnance and you really know when they are there.
I am staying to the paths next time. The thought does not appeal of either arriving on a world war 2 morter shell at the pearly gates or being an minor character in 1 episode of Old Harry's Game who spent eternity debugging Windows and Linux performance issues with the promise of DTrace port always a few days away.
View New Radnor in a larger map
In spite of the possible objective danger, some great running but I am keeping to the paths until the clearance work is complete in 2011.