Open matdibu opened 1 week ago
That would be a data-loss regression for anyone relying on the current behaviour, so I'm heavily against that. The idea is to allow performance improvements at the cost of time reversal if the system crashes; not to allow it at the cost of data loss on every single reboot.
I'm not sure if zpool sync
will do the job, but that's testable. If it does not, how about making the forced sync only happen if sync happens to be disabled on some filesystem?
I understand the risks, but so should anyone who goes out of their way to set sync=disabled
, it's not a default setting and it comes with its own documented caveats.
As it stands, we override what the user set in their zfs config, just because we know better than them.
In my mind, this is in the same spirit as shipping a patch to openzfs that outright disables the feature, but still accepts the flag and doesn't notify the user of the non-upstream behaviour.
Issue description
Half of my
zpool history
is thezfs set nixos:shutdown-time=...
message set in #16903 by @BaughnAnd that PR explicitly modifies the behaviour of ZFS'
sync=disabled
; I don't think we should go against the users that decide to use that feature.And for the large majority who left sync enabled, it's not doing much aside from setting a timestamp
Overall, I suggest removing this overriding behaviour.
Steps to reproduce
zpool history