one question regarding to crucible on Oxide sled node, according the docs, there is 10 pool created on a physical sled on the 10 SSD, it means there is 1 SSD created a pool , is it true? I’m going to find a way to run PostgreSQL with best performance on crucible, and want to learn how to setup crucible with best performance, any one could help? a big thanks, andyf rmustacc jclulow , could give me any hint, thank you :-)

sommerfeld: socket_vop_getattr() has a better (though still not great) way of choosing inode values FWIW

(it was embarassingly worse -- 2^16 unique values -- before that)

I've asked this before, but it it came up in my mind again

is _anyone_ investigating why 'zfs send/recv' seem to ignore the large blockno feature, and create invalidly big blockids?

when I brought it up initially, people mentioned they knew about it, which I took to mean someone... cared

the symptoms are things like `warning: zfs bug: bn be2 > dn_maxblkid be1`

it's in my use case precisely always 2 blocks over

I'm not sure if that's definite

(you can see the code/warning in loader's libsa/zfs/zfsimpl)

illumos.org/issues/5905 -- I did create that check in loader for this one

→ BUG 5905: wrong dn_maxblkid on dnode received to large_block dataset (New)

I am very unfamiliar with ZFS, so I can't really get further without devoting all my time to learning ZFS-isms

which I'm hoping people who, like, sell things based on ZFS already know.

tsoome: thanks, that's a reference I don't think I had

but I havent really had spent much time digging it:(

I havent seen it doing any harm, but I have only seen it related with boot_archive and thats something we can easily regenerate:)

Yes, I see it on ARM with every new image build

because we build the generic filesystem image, and then send/recv it onto the platform-specific disk

at which point the boot_archive grows two blocks bigger than is possible

but since people were talking about send/recv mysteries on the list, I hoped I could interest someone who already knows things

well, that did not really happen when I did create the issue:D