File System ???

[Larry Grover]

Stephen R Laniel wrote:
> On Thu, Jul 14, 2005 at 03:59:36AM +0300, ZIYAD A. M. AL-BATLY wrote:
> 
>>On Tue, 2005-07-12 at 22:35 -0400, Larry Grover wrote:
>>
>>>Are you exporting with the "async" option?  According to the exports
> 
> [snip]
> 
>>This is no longer true with journal file-systems (like in this case:
>>EXT3)!  One of the most important features of a journal file-system is
>>to prevent such corruption in all cases (at the expense of some
>>performance lose).
> 
> 
> I think you're confusing layers, Ziyad. If the original
> poster is using NFS to send data over to another hard drive,
> he can lose data *even if* that other hard drive is using a
> journaling FS. That's the trouble with the 'async' option:
> it doesn't wait for the first block that went over the wire
> to be committed to disk on the other end before it sends the
> next block.

Thanks for lifting the confusion with your clear explanation.


>>Also, "async" is *on* by *default* unless you turn it off (by passing
>>the "sync" option after all other options related to the subject)!
> 
> 
> Um ... right. Which backs up Larry's point: if async is on
> (as it is by default), there's some danger of data loss.
> 
> Personally, I wouldn't suspect that this is the main
> culprit for the original poster's problem, but it's
> something worth checking out.

Hugh (the original poster) indicated (just above, in this thread) that 
he is not using "async", which rules out my suggestion.

Since Hugh's problem is not attributable to NFS, and ext3 is 
ordinarily very reliable, I'd have to guess that the cause is hardware 
related.

I've seen an NFS exported drive get seriously corrupted after an 
unclean shutdown (and I'm pretty sure it was exported "sync,ro").  In 
this case, the shutdown was caused by a failing power supply -- which 
was almost certainly the direct cause of the disk corruption, too.

I'm not suggesting that Hugh's problem is caused by a bad power 
supply, but if it was my system, I'd look to the hardware first 
(starting with the hard drive).

Regards,
Larry