Am 19.08.2010 18:05, schrieb Dustin J. Mitchell: > On Thu, Aug 19, 2010 at 10:32 AM, Stefan G. Weichinger <[email protected]> > wrote: >> And why does it work then at the second time, when I start amdump >> manually? The amount of input should be the same at the 2nd time ... ? > > That's why I suspect it's a race condition. Recall we just found out > about the GNU tar race condition that affects amcheckdump recently, > even though it's been in tar since 1.21. I don't *think* that same > problem would be surfacing here, but it could be.
Interesting ... > The index write failed 1s or less after starting the dump. Does the > dump of this filesystem usually take longer than 1s? ;-) For sure, yes. And to repeat: that error was thrown for more than 20 DLEs at that run. So there were around 20 DLEs failing with the same error, which to me means that there were also 20 separate calls to the wrapper failing. And after the whole amdump failed and sent the mail-report, my manual call of amdump succeeded. - Another thing that might be the problem: I create several LVM-snapshots before I start amdump (everything via cron, sure). The snapshots are created 5 mins before I start amdump. The snapshotting normally takes less than a minute, but I will change that to 10 minutes to get that possibility out of the way. (It's unlikely that this is the reason, the errors look different when I don't have the snapshots mounted, I saw that sometimes when I tested stuff and having forgotten to create the LVM-snapshots) - What could I do to spot the problem? Show the wrapper? Browse for other logs? Thank you, Dustin ... Stefan
