Re: [PATCH 2/4] fsstress: reduce the number of events when io_setup

[Zorro Lang <zlang@redhat.com>]

On Wed, Jul 29, 2020 at 03:43:39PM -0400, Jeff Moyer wrote:
> Zorro Lang <zlang@redhat.com> writes:
> 
> > The original number(128) of aio events for io_setup is a little big.
> > When try to run lots of fsstress processes(e.g. -p 1000) always hit
> > io_setup EAGAIN error, due to the nr_events exceeds the limit of
> > available events. So reduce it from 128 to 64, to make more fsstress
> > processes can do AIO test.
> 
> It looks to me as though there's only ever one request in flight.  I'd
> just set it to 1.
> 
> Also, you've included another change not mentioned in your changelog.
> Please make sure the changelog matches what's done in the patch.

Thanks Jeff, I'll rewrite this patch:) Do you have any review points about
those two IO_URING related patches (1/4 and 4/4), or it looks good to you?

Thanks,
Zorro

> 
> -Jeff
> 
> >
> > Signed-off-by: Zorro Lang <zlang@redhat.com>
> > ---
> >  ltp/fsstress.c | 7 ++++---
> >  1 file changed, 4 insertions(+), 3 deletions(-)
> >
> > diff --git a/ltp/fsstress.c b/ltp/fsstress.c
> > index 388ace50..a11206d4 100644
> > --- a/ltp/fsstress.c
> > +++ b/ltp/fsstress.c
> > @@ -28,6 +28,7 @@
> >  #endif
> >  #ifdef AIO
> >  #include <libaio.h>
> > +#define AIO_ENTRIES	64
> >  io_context_t	io_ctx;
> >  #endif
> >  #ifdef URING
> > @@ -699,8 +700,8 @@ int main(int argc, char **argv)
> >  			}
> >  			procid = i;
> >  #ifdef AIO
> > -			if (io_setup(128, &io_ctx) != 0) {
> > -				fprintf(stderr, "io_setup failed");
> > +			if (io_setup(AIO_ENTRIES, &io_ctx) != 0) {
> > +				fprintf(stderr, "io_setup failed\n");
> >  				exit(1);
> >  			}
> >  #endif
> > @@ -714,7 +715,7 @@ int main(int argc, char **argv)
> >  				doproc();
> >  #ifdef AIO
> >  			if(io_destroy(io_ctx) != 0) {
> > -				fprintf(stderr, "io_destroy failed");
> > +				fprintf(stderr, "io_destroy failed\n");
> >  				return 1;
> >  			}
> >  #endif
>