Re: [Tarantool-patches] [PATCH 3/3] lua: expose temporary Lua state for iproto calls

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

Thanks for the patch!

See 3 comments below.

On 01/06/2020 20:10, Alexander Turenko wrote:
> There is code that may save some time and resources for creating a new
> Lua state when it is present in the fiber storage of a current fiber.
> There are not so much of them: running a Lua trigger and construction of
> a next tuple in a merge source.
> 
> Before the patch, fiber->storage.lua.stack is filled only for the main
> fiber and fibers created from Lua using fiber.create() or fiber.new(),
> but not for background fibers (which serve binary protocol requests).
> 
> This patch fills fiber->storage.lua.stack for background fibers that
> serve a Lua call or eval: we already have this state and nothing prevent
> us from exposing it via the fiber storage.
> 
> Follows up #4954
> ---
>  src/box/lua/call.c   | 27 +++++++++++++++++++++++++++
>  src/lib/core/fiber.h |  8 ++++++--
>  2 files changed, 33 insertions(+), 2 deletions(-)
> 
> diff --git a/src/box/lua/call.c b/src/box/lua/call.c
> index 6588ec2fa..f69d1adbe 100644
> --- a/src/box/lua/call.c
> +++ b/src/box/lua/call.c
> @@ -537,12 +537,39 @@ box_process_lua(lua_CFunction handler, struct execute_lua_ctx *ctx,
>  	port_lua_create(ret, L);
>  	((struct port_lua *) ret)->ref = coro_ref;
>  
> +	/*
> +	 * A code that need a temporary fiber-local Lua state may
> +	 * save some time and resources for creating a new state
> +	 * and use this one.
> +	 */
> +	bool has_lua_stack = fiber()->storage.lua.stack != NULL;
> +	if (! has_lua_stack)
> +		fiber()->storage.lua.stack = L;

1. According to recent code style changes, we should omit single
whitespace after unary operators.

> +
>  	lua_pushcfunction(L, handler);
>  	lua_pushlightuserdata(L, ctx);
>  	if (luaT_call(L, 1, LUA_MULTRET) != 0) {
> +		if (! has_lua_stack)
> +			fiber()->storage.lua.stack = NULL;
>  		port_lua_destroy(ret);
>  		return -1;
>  	}
> +
> +	/*
> +	 * Since this field is optional we're not obligated to
> +	 * keep it until the Lua state will be unreferenced in
> +	 * port_lua_destroy().
> +	 *
> +	 * There is no much sense to keep it beyond the Lua call,
> +	 * so let's zap now.
> +	 *
> +	 * But: keep the stack if it was present before the call,
> +	 * because it would be counter-intuitive if the existing
> +	 * state pointer would be zapped after this function call.

2. It is not just counter-intuitive. It would break Lua-born fibers,
since they would suddenly loose their stack.

> +	 */
> +	if (! has_lua_stack)
> +		fiber()->storage.lua.stack = NULL;
> +
>  	return 0;
>  }
>  
> diff --git a/src/lib/core/fiber.h b/src/lib/core/fiber.h
> index cd9346a55..db68fb47e 100644
> --- a/src/lib/core/fiber.h
> +++ b/src/lib/core/fiber.h
> @@ -496,8 +496,12 @@ struct fiber {
>  		struct credentials *credentials;
>  		struct txn *txn;
>  		/**
> -		 * Lua stack and the optional
> -		 * fiber.storage Lua reference.
> +		 * Optional Lua state (may be NULL). Useful as a
> +		 * temporary Lua state to save time and resources
> +		 * on creating it. Should not be used in other
> +		 * fibers.
> +		 *
> +		 * Optional fiber.storage Lua reference.

3. You change the comment anyway, so it would be better to make it
right and put the member-related comments just above the members.
One comment for the stack, and separate comment for the ref.

>  		 */
>  		struct {
>  			struct lua_State *stack;
>