Hey guys -

 

I built an unRAID server for a friend of mine which has 10x 3TB and 10 2TB drives in it. He is asking to replace all the 2TB's with 3TB's and I'm scratching head here trying to come up with the best way to approach this upgrade... Any thoughts or suggestions? Obviously, replacing and upgrading one-by-one would take FOREVER...

 

 

Thank you,

Better forever to do the replacements than to lose data. 

 

If I were faced with this problem.  I would actually want to take some time.  (Remember a failure of more than one disk in the array will cause a data loss on all the disks that failed.)  It is important to realize that a hard disk is most likely to fail early in its life or it will last until it reaches its end-of-life.  (Google Bathtub curve for more info on this.)

 

First, I would preclear every disk with, at least, two (preferably three)  preclear cycles.  This will catch many failures due to infant mortality.  (BTW, this will probably take three to four days to complete.)

 

Second, I would then replace one disk. Rebuilt the array.  After a day or two run a non-correcting parity check.  You should have no errors.  If you do, find and fix the problem.

 

Third, wait another week.  Again run the non-correcting parity check.  If you have successfully run two checks without error.  You are ready to replace the next disk.

If I was doing this I would build a second system and rsync the data after preclearing all the disks a couple of times.

 

I have the luxury of having lots of hardware lying around though.

 

Even then, the process would take a long time.  Probably quicker than doing 10 rebuilds tho....

 

 

Replacing 10 drives at at once - it it were me I'd probably run at least 6 preclear cycles on the new drives.  It will take a while but I'd sure rather find a bad drive with the preclear than in production.

Slow and steady is the way you are going to want to go on this one.

 

I would 2-3 pass preclear the new drives and then just start swapping them out and letting unRAID rebuild the data onto the new drives.

Hey guys -

 

I built an unRAID server for a friend of mine which has 10x 3TB and 10 2TB drives in it. He is asking to replace all the 2TB's with 3TB's and I'm scratching head here trying to come up with the best way to approach this upgrade... Any thoughts or suggestions? Obviously, replacing and upgrading one-by-one would take FOREVER...

 

 

Thank you,

unless your friend has an immediate need for the additional 10TB of space, I'd only upgrade one or two drives at this time.  The money spent for a 3TB drive today might be the cost of a 4 or 5 TB drive in a year when the extra space is actually needed.

 

Joe L.

I know I always get annoyed when someone tells me an answer other than what I'm directly asking, but sometimes they're good for seeing the whole picture, in that vein I would like to mention:

 

Since he's at 10 drives he seems to already be at "Pro", so is there a reason he can't just add a few more 3TB drives rather than replacing all the 2TB drives to get his 33% increase in space?  That way you'd just have to replace the Parity drive, put the 2TB Parity drive in the array, then add 3x 3TB drives and end up with 1 more TB than you originally bargained for.  And, other than pre-clear time, the only time it takes is the parity build of the new Parity drive, down time is fairly minimal.

 

Of course, hardware limitations might be an issue, but for the cost saved in drives (unless he already bought them, of course) you might be easily able to engineer around those limitations (SATA/SAS expanders, another SATA card, External SATA enclosure, etc.)

All good advice, here. Slow and steady. Be sure to run at least 1 preclear on every drive.

 

I would do a 4 pass badblocks test, then do 1 preclear,

Then do a non correcting parity check. If all looks good.

Upgrade parity.

 

Wait a week or so

Then for each data drive do the same.

4 pass badblocks test, 1 preclear

Swap data drive.

 

If you really needed to, you could do the badblocks test and preclear on 2 3tb drives.

Upgrade parity after that has completed, upgrade 1 data drive, wait a week before upgrading others.

 

The 4 pass badblock write test and/or multipass preclear are going to take days anyway.

 

Although some may not agree with the badblocks test. I've weeded out bad drives with it that were marginal. It also forced reallocation of sectors that were weak.

However, some drives can pass the badblocks test with the kernel retrying bad blocks.

The preclear as the last step is paramount since it will compare the smart information and warn of pending sectors.

 

You can run multiple preclears, however I prefer a badblocks in 4 pass write mode pass since it becomes obvious if you are going to have trouble with the drive after reviewing the syslog and smart reports.

There's something to be said for exercising the drive with shifting bit patterns.

 

I would take this slow and steady and buy drives as they are needed. Buying in large quantities may save you money, but it may also hinder recovery if you happen to have a bad batch of drives.

I've seen where drives purchased in batches start to fail in very close proximity to one another over time.

Will he want to consolidate or move data among the drives?

 

I don't like being subjected to long periods of a second-disk failure risk.  For that reason, I do not swap drives and let them rebuild (except parity).

 

First, get very familiar with the "trust my parity" procedure.

 

I pre-clear and add the new, larger drive(s). 

 

Then copy files from the old drives (sometimes consolidating from 2 or more old drives).

 

Create a root directory folder on old drives that will be removed as "__DELETE_ME__" and move all contents of that drive into that folder.  (keeps me from putting stuff on an old drive my mistake)

 

Wait a couple of weeks to confirm stability.

 

Zero out old drives, shutdown and remove them, and trust parity.

 

Run a parity check.