When unRAID is booted after a "dirty" shutdown (i.e., after a power failure) it proceeds to immediately perform a parity check.  It begins BEFORE the drives have had a chance to process their journal logs and apply transactions to the disks.  This creates race conditions and delays writing of the journal transactions until AFTER the parity has already been checked on the early sectors of the disk.  This can require a SECOND parity check to correct sync errors.

 

I request that unRAID wait until all pending journal transactions are applied before starting to performing parity checks.

I don't think this is an issue.  If a parity check detects bad parity it will only write the correct parity to the parity disk - it will not write any data disk.  Any writes that take place during the parity check will update parity in the usual manner even when parity check is running.

Tom, I believe I was the source of the concern here, and I'm perfectly willing to trust you.  However, I wondered if you had seen the related thread below, especially his first syslog at the end, which made me very nervous about what might be happening.  A few posts later, he mentions that a second Parity Check immediately following had one more parity error in the same region, where the first parity check and transaction replaying *may* have coincided.  I commented that it seemed safer to me to wait on drive being completely ready before beginning the Parity Check, but if you can confirm it's safe, then I'm happy.

 

http://lime-technology.com/forum/index.php?topic=1562.0

 

This is off-topic to this thread, but we had a second concern as to why data did not seem to have been flushed to disk, even though it had idled long enough to spin down.  It doesn't seem as if there should be any transactions left, after a few seconds of inactivity.

 

This is off-topic to this thread, but we had a second concern as to why data did not seem to have been flushed to disk, even though it had idled long enough to spin down.  It doesn't seem as if there should be any transactions left, after a few seconds of inactivity.

The journal entries seem to have been flushed/written, but the subsequent "commit" entry did not seem to be, even though the disk was idle long enough to spin down before the power failure occurred.  That was puzzling.

 

As you said, it probably warrants a separate thread.

 

Joe L.