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

Hi,

On 2023-02-16 16:58:23 +0900, Michael Paquier wrote:
> On Wed, Feb 15, 2023 at 01:00:00PM +0530, Bharath Rupireddy wrote:
> > The v3 patch reduces initialization of iovec array elements which is a
> > clear win when pg_pwrite_zeros is called for sizes less than BLCKSZ
> > many times (I assume this is what is needed for the relation extension
> > lock improvements feature). However, it increases the number of iovec
> > initialization with zerobuf for the cases when pg_pwrite_zeros is
> > called for sizes far greater than BLCKSZ (for instance, WAL file
> > initialization).

In those cases the cost of initializing the IOV doesn't matter, relative to
the other costs. The important point is to not initialize a lot of elements if
they're not even needed. Because we need to overwrite the trailing iov
element, it doesn't seem worth to try to "pre-initialize" iov.

Referencing a static variable is more expensive than accessing an on-stack
variable. Having a first-call check is more expensive than not having it.

Thus making the iov and zbuf_sz static isn't helpful. Particularly the latter
seems like a bad idea, because it's a compiler constant.

> It seems to me that v3 would do extra initializations only if
> pg_pwritev_with_retry() does *not* retry its writes, but that's not
> the case as it retries on a partial write as per its name. The number
> of iov buffers is stricly capped by remaining_size.

I don't really understand this bit?

Greetings,

Andres Freund