[PATCH] vinyl: fix vy_range_update_compaction_priority hang

[PATCH] vinyl: fix vy_range_update_compaction_priority hang

[Vladimir Davydov]

Under certain circumstances vy_slice_new() may create an empty slice,
e.g. on range split:

   |------------------ Slice ---------------|
                         |---- Run -----|
                     +
                  split key
   |---- Slice 1 ----||------ Slice 2 ------|
         ^^^^^^^
          Empty

vy_range_update_compaction_priority() uses the size of the last slice in
a range as a base for LSM tree level sizing. If the slice size happens
to be 0, it will simply hang in an infinite loop. Fix this potential
hang by using 1 if the last slice size is 0.
---
https://github.com/tarantool/tarantool/commits/dv/vy-fix-compaction-prio-calc-hang

 src/box/vy_range.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/box/vy_range.c b/src/box/vy_range.c
index 395639dc..4ff85211 100644
--- a/src/box/vy_range.c
+++ b/src/box/vy_range.c
@@ -343,7 +343,7 @@ vy_range_update_compaction_priority(struct vy_range *range,
 	uint64_t size;
 	struct vy_slice *slice;
 	slice = rlist_last_entry(&range->slices, struct vy_slice, in_range);
-	size = slice->count.bytes;
+	size = MAX(slice->count.bytes, 1);
 	slice = rlist_first_entry(&range->slices, struct vy_slice, in_range);
 	do {
 		target_run_size = size;
-- 
2.11.0

Re: [PATCH] vinyl: fix vy_range_update_compaction_priority hang

[Konstantin Osipov]

* Vladimir Davydov <vdavydov.dev@gmail.com> [19/07/08 16:38]:

Please push this asap since there is a customer waiting for  fix.

Please also fix VY_ROUND_UP type safety - let's make it static
inline instead.

This needs a test case as a follow up and a proper fix should drop
all such slices.

> Under certain circumstances vy_slice_new() may create an empty slice,
> e.g. on range split:
> 
>    |------------------ Slice ---------------|
>                          |---- Run -----|
>                      +
>                   split key
>    |---- Slice 1 ----||------ Slice 2 ------|
>          ^^^^^^^
>           Empty
> 
> vy_range_update_compaction_priority() uses the size of the last slice in
> a range as a base for LSM tree level sizing. If the slice size happens
> to be 0, it will simply hang in an infinite loop. Fix this potential
> hang by using 1 if the last slice size is 0.
> ---
> https://github.com/tarantool/tarantool/commits/dv/vy-fix-compaction-prio-calc-hang
> 
>  src/box/vy_range.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/src/box/vy_range.c b/src/box/vy_range.c
> index 395639dc..4ff85211 100644
> --- a/src/box/vy_range.c
> +++ b/src/box/vy_range.c
> @@ -343,7 +343,7 @@ vy_range_update_compaction_priority(struct vy_range *range,
>  	uint64_t size;
>  	struct vy_slice *slice;
>  	slice = rlist_last_entry(&range->slices, struct vy_slice, in_range);
> -	size = slice->count.bytes;
> +	size = MAX(slice->count.bytes, 1);
>  	slice = rlist_first_entry(&range->slices, struct vy_slice, in_range);
>  	do {
>  		target_run_size = size;
> -- 
> 2.11.0
> 

-- 
Konstantin Osipov, Moscow, Russia

Re: [PATCH] vinyl: fix vy_range_update_compaction_priority hang

[Vladimir Davydov]

On Mon, Jul 08, 2019 at 04:48:54PM +0300, Konstantin Osipov wrote:
> * Vladimir Davydov <vdavydov.dev@gmail.com> [19/07/08 16:38]:
> 
> Please push this asap since there is a customer waiting for  fix.

Pushed to master, 2.1, and 1.10.

> 
> Please also fix VY_ROUND_UP type safety - let's make it static
> inline instead.

I can't - we do need to pass double as a divident to it, because
run_size_ratio is double.

> 
> This needs a test case as a follow up and a proper fix should drop
> all such slices.

That's going to be tricky. I'll look if we can actually do that.