On Fri, Jan 28, 2022 at 7:28 AM Andres Freund <andres@anarazel.de> wrote:
> On 2022-01-27 19:55:45 +0200, Heikki Linnakangas wrote:
> > I was not able to reproduce this without the tablespace on a different
> > virtual disk, I presume because ext4 orders the writes so that the
> > checkpoint implicitly always flushes the creation of the file to disk.
>
> It's likely that the control file sync at the end of a checkpoint has the side
> effect of also forcing the file creation to be durable if on the same
> tablespace (it'd not make the file contents durable, but they don't exist
> here, so ...).
It might be possible to avoid that on xfs or pretty much any other
file system. I wasn't following this closely, but even with ext4's
recent fast commit changes, its fsync implementation still
deliberately synchronises data for other file descriptors as a side
effect as summarised in [1], unlike xfs and other systems. So they've
caught up with xfs's concurrent writes (and gone further than xfs by
doing it also for buffered I/O giving up even page-level atomicity, as
discussed in a couple of other threads), but not yet decided to pull
the trigger on just-fsync-what-I-asked-for.
[1] https://lwn.net/Articles/842385/