Re: [PATCH v4 0/8] Introduce the end_syncfs option

On 3/1/26 7:47 PM, Sitsofe Wheeler wrote:
> On Fri, 27 Feb 2026 at 08:13, Damien Le Moal <[email protected]> wrote:
>>
>> This patch series introduces the end_syncfs option and adds support for
>> it to the main IO engines that can handle regular files.
>> This new option improves the precision of file system benchmarks when a
>> large number of files are being handled.
> 
> This looks excellent to me and I especially like that it now groups
> the syncing by the filesystem each file uses. I went through the
> ioengine flags and wondered whether absence of FIO_DISKLESSIO was
> enough to know that an ioengine would support syncfs but there are
> engines (like mmap and sg) that lack FIO_DISKLESSIO but would be
> inappropriate so a new flag makes sense.
> 
> Out of idle curiosity, what happens if you use the syncfs option when
> the file being operated on is a device? Regardless:

In add_file_fs(), I have:

	if (f->filetype != FIO_TYPE_FILE)
		return 0;

So for any file that is not a regular file, we do not try to add a mount
point/FS to the job list and end_syncfs will have no effect for these
non-regular files.


-- 
Damien Le Moal
Western Digital Research