#21 — zpool status CKSUM update broken?

 Hi list

I tried to test the scrubbing mechanism of a two member raid today and here's
sth odd I don't understand:

I created a zfs-fuse test raid, filled it with 70GB of video material, scrubbed
the array and everything was fine as expected.

With one drive unplugged I booted into a live system. I used dd_rescue to
spread 20 errors all over the zfs partition. In fact I overwrote some 512 Byte
blocks with /dev/zero. So this raid member became the defective one.

I rebooted the zfs file system with both drives attached and run a binary
compare of all video files in the zfs pool. The idea behind this was to make
sure as many data as possible gets read including data from the defective
drive.

Meanwhile I checked zpool status to see, if the reading of both drives (one
damaged, one intact) would cause the CKSUM value to rise. Nothing happened -
the binary compare finished with no CKSUM errors.

So I forced the scrubbing of both drives with zpool scrub <pool> - now 2 CKSUM
errors were noticed right at the begining of scrub (less than 10% was scrubbed
at that time) with 384kb to re-silver. For the rest of the 90% no more errors
got detected.

Two things I don't understand here, I hope someone can explain if it's a bug
what I should expect to happen with zfs in this case:

a) Why doesn't zpool status show checksum errors immediately when it reads
from the faulty drive? I expected zpool status to update every checksum error,
reading error and writing error immediately.

b) Status after scrubbing: why are there only 2 CKSUM errors mentioned right
at the begining of scrub, when there are about 20 error blocks of 512byte size
spread in the rest of the unchecked 90%?

thanks for some comment on that