From: Vladimir Davydov <vdavydov.dev@gmail.com> To: Cyrill Gorcunov <gorcunov@gmail.com> Cc: tml <tarantool-patches@freelists.org> Subject: Re: [PATCH 2/2] lib/core/fiber: Allow to extend default stack size Date: Wed, 27 Mar 2019 12:35:06 +0300 [thread overview] Message-ID: <20190327093506.zsxzchje2un76rcp@esperanza> (raw) In-Reply-To: <20190319193845.31221-3-gorcunov@gmail.com> On Tue, Mar 19, 2019 at 10:38:45PM +0300, Cyrill Gorcunov wrote: > Recently we found that 64K stack we've been using > for years is not longer enough due to third party > libraries, so we extended its size. > > Still there is no guarantee that even new size will > lasts forever and users might face same problem again, > waiting for new build where stack capacity increased. > > Instead of this ping-pong game lets allow a user to > setup default stack size via FIBER_STACK_SIZE_DEFAULT > environment variable. > > Note we fetch this value once on program startup and > because it is early init stage we can't use box > config engine. > --- > src/lib/core/fiber.c | 78 ++++++++++++++++++++++++++++++++++++++------ > 1 file changed, 68 insertions(+), 10 deletions(-) > > diff --git a/src/lib/core/fiber.c b/src/lib/core/fiber.c > index 922a0bfe8..d3cb18a15 100644 > --- a/src/lib/core/fiber.c > +++ b/src/lib/core/fiber.c > @@ -107,18 +107,25 @@ pthread_t main_thread_id; > static size_t page_size; > static bool stack_growsdown; > > +/** Indices in fiber stack sizes (FSS) */ > enum { > + FSS_IDX_MINIMAL, > + FSS_IDX_WATERMARK, > + FSS_IDX_DEFAULT, > +}; > + > +/** Fiber stack sizes */ > +static size_t fiber_stack_size[] = { > /* The minimum allowable fiber stack size in bytes */ > - FIBER_STACK_SIZE_MINIMAL = 16384, > + [FSS_IDX_MINIMAL] = 16384, > + /* Stack size watermark in bytes */ > + [FSS_IDX_WATERMARK] = 65536, > /* Default fiber stack size in bytes */ > - FIBER_STACK_SIZE_DEFAULT = 524288, > - /* Stack size watermark in bytes. */ > - FIBER_STACK_SIZE_WATERMARK = 65536, > + [FSS_IDX_DEFAULT] = 524288, > }; Please avoid unrelated changes. If you want to rename or refactor something, do it in a separate patch to ease review. Anyway, I don't see much point in this change - in Tarantool code it's common to use a enum for storing constants. > > /** Default fiber attributes */ > -static const struct fiber_attr fiber_attr_default = { > - .stack_size = FIBER_STACK_SIZE_DEFAULT, > +static struct fiber_attr fiber_attr_default = { > .flags = FIBER_DEFAULT_FLAGS > }; > > @@ -172,13 +179,13 @@ fiber_attr_delete(struct fiber_attr *fiber_attr) > int > fiber_attr_setstacksize(struct fiber_attr *fiber_attr, size_t stack_size) > { > - if (stack_size < FIBER_STACK_SIZE_MINIMAL) { > + if (stack_size < fiber_stack_size[FSS_IDX_MINIMAL]) { > errno = EINVAL; > diag_set(SystemError, "stack size is too small"); > return -1; > } > fiber_attr->stack_size = stack_size; > - if (stack_size != FIBER_STACK_SIZE_DEFAULT) { > + if (stack_size != fiber_stack_size[FSS_IDX_DEFAULT]) { > fiber_attr->flags |= FIBER_CUSTOM_STACK; > } else { > fiber_attr->flags &= ~FIBER_CUSTOM_STACK; > @@ -844,11 +851,11 @@ fiber_stack_watermark_create(struct fiber *fiber) > size_t offset = rand() % POISON_OFF * sizeof(poison_pool[0]); > if (stack_growsdown) { > fiber->stack_watermark = fiber->stack + fiber->stack_size; > - fiber->stack_watermark -= FIBER_STACK_SIZE_WATERMARK; > + fiber->stack_watermark -= fiber_stack_size[FSS_IDX_WATERMARK]; > fiber->stack_watermark += offset; > } else { > fiber->stack_watermark = fiber->stack; > - fiber->stack_watermark += FIBER_STACK_SIZE_WATERMARK; > + fiber->stack_watermark += fiber_stack_size[FSS_IDX_WATERMARK]; > fiber->stack_watermark -= page_size; > fiber->stack_watermark += offset; > } > @@ -1383,6 +1390,56 @@ cord_slab_cache(void) > return &cord()->slabc; > } > > +/** > + * Initialize early parameters of the stack > + * and setup default variables. > + */ > +static void > +fiber_stack_init(void) > +{ > + static const char default_size_name[] = "FIBER_STACK_SIZE_DEFAULT"; There's a risk it may conflict with another environment variable. At the very least, we should prefix it with TARANTOOL_. However, I don't think that using an environmental variable is a proper way to go. After all, Tarantool is an application server so we typically set system variables right from Lua code. Can we introduce a function in the fiber module for this? The tricky part is that it should change the stack size of the calling fiber, I guess.
next prev parent reply other threads:[~2019-03-27 9:35 UTC|newest] Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top 2019-03-19 19:38 [RFC 0/2] lib/core/fiber: Allow to extend stack size via env variable Cyrill Gorcunov 2019-03-19 19:38 ` [PATCH 1/2] lib/core/fiber: Rename stack_direction to stack_growsdown Cyrill Gorcunov 2019-03-27 9:35 ` Vladimir Davydov 2019-03-27 9:48 ` Cyrill Gorcunov 2019-03-27 10:20 ` Vladimir Davydov 2019-03-27 10:30 ` Cyrill Gorcunov 2019-03-19 19:38 ` [PATCH 2/2] lib/core/fiber: Allow to extend default stack size Cyrill Gorcunov 2019-03-27 9:35 ` Vladimir Davydov [this message] 2019-03-27 9:46 ` Cyrill Gorcunov 2019-03-27 10:15 ` Vladimir Davydov 2019-03-27 10:23 ` Cyrill Gorcunov 2019-04-01 17:41 ` Cyrill Gorcunov 2019-04-01 18:58 ` Konstantin Osipov 2019-04-01 19:19 ` Cyrill Gorcunov 2019-04-01 20:51 ` Konstantin Osipov 2019-04-01 22:05 ` Cyrill Gorcunov 2019-04-02 7:14 ` Konstantin Osipov 2019-04-02 8:09 ` Cyrill Gorcunov
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20190327093506.zsxzchje2un76rcp@esperanza \ --to=vdavydov.dev@gmail.com \ --cc=gorcunov@gmail.com \ --cc=tarantool-patches@freelists.org \ --subject='Re: [PATCH 2/2] lib/core/fiber: Allow to extend default stack size' \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox