[PATCH] recovery: remove yields from index build and format check

[PATCH] recovery: remove yields from index build and format check

[Serge Petrenko]

Iproto already listens for requests during recovery, so yielding at this
point of time allows such early requests, which arrived during recovery,
be processed while data is in unfinished state. This caused box/net.box
test failures, and is potentially harmful.
Besides, there is no need to yield during recovery.

Closes #4273
---
https://github.com/tarantool/tarantool/issues/4273
https://github.com/tarantool/tarantool/tree/sp/gh-4273-dont-yield-on-recovery

 src/box/memtx_space.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/src/box/memtx_space.c b/src/box/memtx_space.c
index 9f232803e..3e607e0e6 100644
--- a/src/box/memtx_space.c
+++ b/src/box/memtx_space.c
@@ -47,6 +47,9 @@
  * Yield every 1K tuples while building a new index or checking
  * a space format. In debug mode yield more often for testing
  * purposes.
+ * Yields do not happen during recovery. At this point of time
+ * iproto aready accepts requests, and yielding would allow them
+ * to be proccessed while data is not fully recovered.
  */
 #ifdef NDEBUG
 enum { MEMTX_DDL_YIELD_LOOPS = 1000 };
@@ -884,6 +887,7 @@ memtx_space_check_format(struct space *space, struct tuple_format *format)
 	if (it == NULL)
 		return -1;
 
+	struct memtx_engine *memtx = (struct memtx_engine *)space->engine;
 	struct memtx_ddl_state state;
 	state.format = format;
 	state.cmp_def = pk->def->key_def;
@@ -905,7 +909,7 @@ memtx_space_check_format(struct space *space, struct tuple_format *format)
 		rc = tuple_validate(format, tuple);
 		if (rc != 0)
 			break;
-		if (++count % MEMTX_DDL_YIELD_LOOPS == 0) {
+		if (++count % MEMTX_DDL_YIELD_LOOPS == 0 && memtx->state == MEMTX_OK) {
 			state.cursor = tuple;
 			tuple_ref(state.cursor);
 			fiber_sleep(0);
@@ -1033,6 +1037,7 @@ memtx_space_build_index(struct space *src_space, struct index *new_index,
 	if (it == NULL)
 		return -1;
 
+	struct memtx_engine *memtx = (struct memtx_engine *)src_space->engine;
 	struct memtx_ddl_state state;
 	state.index = new_index;
 	state.format = new_format;
@@ -1079,7 +1084,7 @@ memtx_space_build_index(struct space *src_space, struct index *new_index,
 		 */
 		if (new_index->def->iid == 0)
 			tuple_ref(tuple);
-		if (++count % MEMTX_DDL_YIELD_LOOPS == 0) {
+		if (++count % MEMTX_DDL_YIELD_LOOPS == 0 && memtx->state == MEMTX_OK) {
 			/*
 			 * Remember the latest inserted tuple to
 			 * avoid processing yet to be added tuples
-- 
2.20.1 (Apple Git-117)

Re: [PATCH] recovery: remove yields from index build and format check

[Vladimir Davydov]

On Thu, Jun 13, 2019 at 06:30:38PM +0300, Serge Petrenko wrote:
> Iproto already listens for requests during recovery, so yielding at this
> point of time allows such early requests, which arrived during recovery,
> be processed while data is in unfinished state. This caused box/net.box
> test failures, and is potentially harmful.
> Besides, there is no need to yield during recovery.
> 
> Closes #4273
> ---
> https://github.com/tarantool/tarantool/issues/4273
> https://github.com/tarantool/tarantool/tree/sp/gh-4273-dont-yield-on-recovery
> 
>  src/box/memtx_space.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)

Pushed to master with a few nitpicks (see below).

> 
> diff --git a/src/box/memtx_space.c b/src/box/memtx_space.c
> index 9f232803e..3e607e0e6 100644
> --- a/src/box/memtx_space.c
> +++ b/src/box/memtx_space.c
> @@ -47,6 +47,9 @@
>   * Yield every 1K tuples while building a new index or checking
>   * a space format. In debug mode yield more often for testing
>   * purposes.
> + * Yields do not happen during recovery. At this point of time
> + * iproto aready accepts requests, and yielding would allow them
> + * to be proccessed while data is not fully recovered.

Added a newline between paragraphs - text looks more readable that way.

>   */
>  #ifdef NDEBUG
>  enum { MEMTX_DDL_YIELD_LOOPS = 1000 };
> @@ -884,6 +887,7 @@ memtx_space_check_format(struct space *space, struct tuple_format *format)
>  	if (it == NULL)
>  		return -1;
>  
> +	struct memtx_engine *memtx = (struct memtx_engine *)space->engine;
>  	struct memtx_ddl_state state;
>  	state.format = format;
>  	state.cmp_def = pk->def->key_def;
> @@ -905,7 +909,7 @@ memtx_space_check_format(struct space *space, struct tuple_format *format)
>  		rc = tuple_validate(format, tuple);
>  		if (rc != 0)
>  			break;
> -		if (++count % MEMTX_DDL_YIELD_LOOPS == 0) {
> +		if (++count % MEMTX_DDL_YIELD_LOOPS == 0 && memtx->state == MEMTX_OK) {

The line's too long. Split.

>  			state.cursor = tuple;
>  			tuple_ref(state.cursor);
>  			fiber_sleep(0);
> @@ -1033,6 +1037,7 @@ memtx_space_build_index(struct space *src_space, struct index *new_index,
>  	if (it == NULL)
>  		return -1;
>  
> +	struct memtx_engine *memtx = (struct memtx_engine *)src_space->engine;
>  	struct memtx_ddl_state state;
>  	state.index = new_index;
>  	state.format = new_format;
> @@ -1079,7 +1084,7 @@ memtx_space_build_index(struct space *src_space, struct index *new_index,
>  		 */
>  		if (new_index->def->iid == 0)
>  			tuple_ref(tuple);
> -		if (++count % MEMTX_DDL_YIELD_LOOPS == 0) {
> +		if (++count % MEMTX_DDL_YIELD_LOOPS == 0 && memtx->state == MEMTX_OK) {

Ditto.