I've read this parity swap procedure on the wiki: https://lime-technology.com/wiki/The_parity_swap_procedure

 

Correct me if I'm wrong, but doesn't this mean that during the procedure the array is entirely without protection? Meaning if any of the drive fails, I lose all the contents of parity, the failed drive and the drive I was replacing. Wouldn't it be safer to first fully replace the parity (having the old parity as backup to put back in, in case something goes wrong), and only after it was replaced successfully I would replace the smaller data drive with the old parity drive?

Parity swap is for when there's a failed drive, so the array is already at risk, if there's no failed drive you don't need to do it, though if you keep the old parity intact the risk is similar since it's a 2 step procedure, first parity is copied from the old to the new parity disk, then the disabled disk is rebuilt.

Is there a speed difference when copying the parity data as opposed to calculating it from scratch?

15 minutes ago, Krzaku said:

Is there a speed difference when copying the parity data as opposed to calculating it from scratch?

Copying parity data can be done by reading from one drive and writing to another.

 

Building parity requires that every data drive is read to get the input data for computing parity that can then be written to the parity disk.

 

So there is less work to copy parity - but actual speed difference depends on number of data disks and how much bandwidth the machine is able to get and process.

Thanks. I hope all goes well when I'll be replacing my drives.

There is an alternative. You can insert a new disk as a UD, copy the contents of the failed disk to the new drive. Then do a new config and build parity. 

 

I prefer this method. It starts getting the data off of the emulated disk immediately. It copies a file at a time, so each file copied is complete. (If a problem occurs part way of rebuilding a disk sector by sector,  you have no idea what is complete and what is not.) You can copy data in priority order, getting most important data off first. It also results in a defragmented disk rather than a mirror image of the original disk that might have been heavily fragmented.