Re: Use pg_pwritev_with_retry() instead of write() in dir_open_for_write() to avoid partial writes?

Hi,

On 2023-02-14 18:00:00 +0530, Bharath Rupireddy wrote:
> On Mon, Feb 13, 2023 at 11:09 PM Andres Freund <andres(at)anarazel(dot)de> wrote:
> >
> > > Later code assigns iov[0].iov_len thus we need to provide a separate
> > > iov non-const variable, or can we use pwrite instead there? (I didn't
> > > find pg_pwrite_with_retry(), though)
> >
> > Given that we need to do that, and given that we already need to loop to
> > handle writes that are longer than PG_IOV_MAX * BLCKSZ, it's probably not
> > worth avoiding iov initialization.
> >
> > But I think it's worth limiting the initialization to blocks.
>
> We can still optimize away the for loop by using a single iovec for
> remaining size, like the attached v2 patch.
>
> > I'd also try to combine the first pg_writev_* with the second one.
>
> Done, PSA v2 patch.

This feels way too complicated to me. How about something more like the
attached?

> 2) A small test module passing in a file with the size to write isn't
> multiple of block size, meaning, the code we have in the function to
> write last remaining bytes (less than BLCKSZ) gets covered which isn't
> covered right now -

FWIW, I tested this locally by just specifying a smaller size than BLCKSZ for
the write size.

Greetings,

Andres Freund