[PATCH 1/3] lib/core/fiber: Increase default stack size
Vladimir Davydov
vdavydov.dev at gmail.com
Wed Mar 13 17:13:05 MSK 2019
On Wed, Mar 13, 2019 at 04:53:07PM +0300, Cyrill Gorcunov wrote:
> On Wed, Mar 13, 2019 at 04:42:53PM +0300, Vladimir Davydov wrote:
> > On Wed, Mar 13, 2019 at 01:47:19AM +0300, Cyrill Gorcunov wrote:
> > > The default 64K stack size used for years become too small
> > > for modern distors (Fedora 29 and etc) where third party libraries
> > > (such as ncurses) started to use 64K for own buffers and we get
> > > SIGSGV early without reaching interactive console phase.
> > >
> > > Thus we increase default size up to 512K which should fit
> >
> > May be, better use 1 MB, just to be sure?
>
> Well, you know iirc 512 should fit even on current MacOS.
OK I see. Let's leave it as is then.
> Also 1M is not guarantee us anything we have to make
> this value configurable on early init. And I'm gonna
> make it so.
>
> > > for common case. Later we will make this value configurable
> > > to address arbitrary stack sizes without a need to rebuild
> > > the whole code.
> > >
> > > Note the values are switched to 4K page granularity for sake
> > > of future modifications -- we gonna manipulate pages to
> > > relax rss usage if OS allows.
> > >
> > > Closes #3418
> >
> > Should be "Part of #3418". We use "Closes" only with the last patch in
> > the series (it makes GitHub close the issue).
>
> Strictly speaking it closes this particular issue. The things
> we gonna implement on top is rather an optimization on memory
> usage. So i would rather open another issue to address it.
Seeing "Closes #3418" a few times in the same series looks weird.
Without the third patch, this one shouldn't be committed so I'd use
"Part of".
>
> > > ---
> > > src/lib/core/fiber.c | 4 ++--
> > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/src/lib/core/fiber.c b/src/lib/core/fiber.c
> > > index abd6c6b11..f16ac873f 100644
> > > --- a/src/lib/core/fiber.c
> > > +++ b/src/lib/core/fiber.c
> > > @@ -93,9 +93,9 @@ static int stack_direction;
> > >
> > > enum {
> > > /* The minimum allowable fiber stack size in bytes */
> > > - FIBER_STACK_SIZE_MINIMAL = 16384,
> > > + FIBER_STACK_SIZE_MINIMAL = 4 << 12,
> > > /* Default fiber stack size in bytes */
> > > - FIBER_STACK_SIZE_DEFAULT = 65536
> > > + FIBER_STACK_SIZE_DEFAULT = 128 << 12
> >
> > I don't like using '<< 12' for stack sizes. Why 12? Because the page
> > size is typically 4K. I wouldn't count on that - it's userspace so it
> > looks confusing.
>
> Because madvise works with 4K pages and such record force
> a code reader to pay attention that we're counting by
But we align pointers passed to madvise by page size anyway.
Seeing <<12 in userspace is weird.
> pages. Actually once I make this values configurable
> I'll use "nr_page * page_size" record.
>
> Volodya, if these are only nits you worries then fine
> I can easily fix them :)
With this particular patch, yes only nit picks :)
More information about the Tarantool-patches
mailing list