On 29.10.2021 1:48, Andres Freund wrote:
> That wouldn't protect against e.g. a logic bug in ZFS. Given its copy-on-write
> nature corruption could very well manifest as seeing an older version of the
> data when re-reading data from disk. Which could very well lead to the type of
> corruption we're seeing here.
> Didn't 14 change the logic when index vacuums are done? That could cause
> previously existing issues to manifest with a higher likelihood.
Note that I ran 13.2 (that came with open_datasync()) and 13.3 (which
switched back to fdatasync()) on openzfs for almost a year since
upgrading to 13.0-RELEASE, with no discernible issues. The 13.3 data
directory has been kept intact since the upgrade to 14.
--
K. R.