System crashing when swap hits 25%

[Kevin J. Cummings]

Todd Denniston wrote:
> Simon Slater wrote, On 03/10/2008 06:37 PM:
>>     G'day all,
>>         This is an inconvenience that I would like to get to the 
>> bottom of.  I
>> have a PII 233MHz 512 MbRAM running FC6. All is fine until swap reaches
>> about 25% used then the system hangs.  It will run for days with the
>> swap at 10-15%.  My first thought was to run badblocks to see if that
>> was the problem, since it is an old drive, but couldn't find the swap
>> device with mount.
>>
>
> Been there, done that.
>
> WARNING: before using badblocks in full rw mode on swap space, 
> remember two things:
> 1) some how you would have to convince mkswap to retake what ever 
> label is being used in /etc/fstab, or you would need to use a new 
> label/device name in /etc/fstab.
> 2) don't be using it for swap while checking it. :)
>
> /sbin/swapon -s
> will display which partitions/files are being used in the system for 
> swap.
> /sbin/swapoff -a
> will turn swap off. probably should shut down any programs down that 
> you don't need so that you can get under 512MB, then badblocks (with 
> full read/write test and re-`mkswap -c` on it, or Use non-destructive 
> read-write mode) sounds like a reasonable thing.
>
> If you find a single bad block, you should really be looking for new 
> hardware.
> I have done the thing where you partition around the bad blocks, but 
> it only buys you a little time.
> If I understand the man-page for mkswap the -c will print the bad 
> block locations, it says nothing of setting up swap to avoid them.
Here's a thought, create a swap *file*, at least as big as your swap 
partition if you can, but at least as big as necessary to hold what is 
in swap now.  Enable it with "swapon pathname".
*then* turn off your old swap partition.  *then* check it for bad blocks 
by re-running mkswap on it with the -c option.  Now you will be 
protected in case in case you *can't* swap everything back in.
>>     My questions are: what would cause this system to hang when the swap
>> file is used?; 
>
> [from experience I know] Bad blocks in swap cause errors that look 
> _just_ like failing ram.  That is, programs just die, system locks up 
> in weird ways.
>
>> where is the swap file?; 
>
> Most times it is a swap partition, `/sbin/swapon -s` or `cat 
> /proc/swaps` will help here.
>
>> if in the VG, how do I umount it
>> to run badblocks?  
>
> /sbin/swapoff
>
>> Here is some output that may be useful.
>>
>>     Thanks.
>
> Good luck.
>