Alexander Motin

I don't know why one of the datasets is reported differently. While about "refer are slightly different", it seems you are requesting "lrefer" ("logicalreferenced"), while default output shows "referenced" property....

Some pool features are activated on per-dataset basis. Though I don't remember what's special about those specific properties. In the man page I see this: ``` usedby* The usedby* properties...

I am not sure about this change motivation, and respectively adequate solution choice. If some device got lost, but it was redundant and did not cause pool suspension, then ZFS...

@Nikratio It should not "just hang", if hardware is behaving (disks are not hanging indefinitely on I/Os). But depending on versions, workload and other factors sometimes it may take time....

@Nikratio Import process logs what it is doing into ZFS debug messages (dbgmsg), which you might get via procfs on Linux or sysctl on FreeBSD, as you may see @quadniko...

It seems doing something with meta objset in a sync context, but I don't see what exactly. Considering your use of HDDs and RAIDZ3, your vdevs may be pretty wide...

There is such thing as deferred delete, but I don't remember much details out of my head.

KM_NOSLEEP makes sense only if caller is ready to handle allocation errors, as a packet loss in network stack. ZFS can not just return ENOMEM errors randomly, so it would...

Debian's kernel config does not have this option. Wonder if it was added later, or they don't have it for a reason. Also on most of server systems things like...

I am not about SPD. IIRC the bus where the sensor reside did not have proper arbitration to prevent simultaneous access by OS and CPU microcode.