Re: [Tarantool-patches] [PATCH] fio/coio: handle partial writes

[Vladislav Shpilevoy <v.shpilevoy@tarantool.org>]

Hi! Thanks for the patch!

See 4 comments below.

On 21/04/2020 23:39, Cyrill Gorcunov wrote:
> Writting less bytes than requested is perfectly fine.

1. 'Writting' -> 'Writing'.

> In turn our fio.write/pwrite api simply returns 'true'
> even if only some part of a buffer has been written.
> 
> Thus make coio_write and coio_pwrite to write the
> whole data in a cycle. Note in most situations there
> will be only one pass, partial writes are really the
> rare cases.
> 
> Fixes #4651
> 
> Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
> ---
> issue https://github.com/tarantool/tarantool/issues/4651
> branch gorcunov/gh-4651-partial-write
> 
>  src/lib/core/coio_file.c | 62 ++++++++++++++++++++++++++++------
>  src/lib/core/errinj.h    |  1 +
>  test/app/fio.result      | 73 ++++++++++++++++++++++++++++++++++++++++
>  test/app/fio.test.lua    | 21 ++++++++++++
>  test/box/errinj.result   |  1 +
>  5 files changed, 147 insertions(+), 11 deletions(-)
> 
> diff --git a/src/lib/core/coio_file.c b/src/lib/core/coio_file.c
> index e2345567c..e290214bc 100644
> --- a/src/lib/core/coio_file.c
> +++ b/src/lib/core/coio_file.c
> @@ -164,10 +164,30 @@ coio_file_close(int fd)
>  ssize_t
>  coio_pwrite(int fd, const void *buf, size_t count, off_t offset)
>  {
> -	INIT_COEIO_FILE(eio);
> -	eio_req *req = eio_write(fd, (void *) buf, count, offset,
> -				 0, coio_complete, &eio);
> -	return coio_wait_done(req, &eio);
> +	ssize_t left = count, pos = 0, res, chunk;
> +	eio_req *req;
> +
> +	while (left > 0) {
> +		INIT_COEIO_FILE(eio);
> +		chunk = left;
> +
> +		ERROR_INJECT(ERRINJ_COIO_WRITE_CHUNK, {
> +			chunk = 1;
> +		});
> +
> +		req = eio_write(fd, (void *)buf + pos, chunk,

2. As it was fairly noticed by Timur in one of my
patches for lsregion - lets better avoid 'void' pointer
arithmetic. And use uint8_t * or char *.

> +				offset + pos, EIO_PRI_DEFAULT,
> +				coio_complete, &eio);
> +		res = coio_wait_done(req, &eio);
> +		if (res < 0) {
> +			pos = -1;

3. What if eio_write() returns EAGAIN/EINTR/WOULDBLOCK? Can it
happen at all? Can it happen, if the descriptor is not
blocking? Can it happen if the underlying FS is a network
file system, like sshfs via FUSE?

Coio socket library handles these errors and retries them.

> +			break;
> +		} else {
> +			left -= res;
> +			pos += res;
> +		}
> +	}
> +	return pos;
>  }
> diff --git a/test/app/fio.result b/test/app/fio.result
> index 783fa4fab..73fbd29e5 100644
> --- a/test/app/fio.result
> +++ b/test/app/fio.result
> @@ -865,6 +871,73 @@ fh6:close()
>  ---
>  - true
>  ...
> +-- test partial write/pwrite via one byte transfer

4. Please

- Add the issue reference, like this:

	--
	-- gh-4651: ...
	--

- Start sentences from a capital letter, and finish with dot.