When and how often should a parity check be performed?

 

When should it be a corrective parity check, and when not?

There is a unmenu script that will run monthly no-correct parity checks.

When and how often should a parity check be performed?

 

When should it be a corrective parity check, and when not?

A non correcting parity check is a good exerciser to make sure you don't have a drive fail on you and you not realize it. During normal operation, the only spots on the drives that are read and written are where the actual files currently reside. If a drive fails a read during normal operation, the array will spin up all the drives, compute what data is supposed to be there, and attempt to write the data back to the spot that failed the read. If that succeeds, you just get an error count in the gui, and the drive remains online. If it fails, the drive is immediately red balled, and all further reads are computed from all the rest of the disks. The problem is, if you have a spot on a drive that is never normally accessed go bad silently, then the first time there is a read failure on any other drive in that relative location, that read will fail totally and you will lose data.

 

A non correcting parity check reads ALL the spots on ALL the drives and does the math to check if the parity drive is in sync. If you see errors on a non correcting check, STOP and figure out which drive is bad from smart reports and the syslog. If you can't figure out which drive is at fault, post and ask for help. Occasionally there will be a parity error that just needs a correcting run to straighten out and that's the end of it, but it's never a good idea to just go on and blindly assume things are ok.

 

In my opinion the only time a correcting check should be run is if the array was shut down uncleanly, or a drive was otherwise altered when parity wasn't active, or you are rebuilding parity from known good data drives.

 

Non-correcting checks should be run on a regular basis, once a month at least, more often if you suspect an error.

 

If you can't complete a non- correcting parity check without error, you can't rebuild a bad drive.