So I got a pool with media and recordsize of 1M.
while streaming from that zfs I see
that it looks like there are mostly 16k reads ?
I cant find any good docs on the mening of “ind” though .
It appears you are using iostat, but I do not think it applies and correctly reports for ZFS. You should look into zpool iostat or the tool ioztat instead.
The above output is from zpool iostat -w 2
Remember the recordsize is the maximum block size, smaller blocks are still utilized. There are actually a lot of interesting consequences to using different sized blocks through the disk that have come back to bite the project IMO.
zpool iostat’s “ind” is for individual I/O:
zpool-iostat.8 — OpenZFS documentation.
yes ofc its the maximum, but these are large media files and the programs reading the files “ought” to read larger chunks of data.
Im just perplexed that I see 16k reads and not larger reads . very odd.