[Tarantool-patches] [PATCH v4 0/2] vinyl: fix uninitialized memory accesses

[Nikita Pettik]

On 12 May 14:14, Nikita Pettik wrote:
> On 10 May 21:49, Vladislav Shpilevoy wrote:
> > Hi! Thanks for the patchset!
> > 
> > LGTM.
> 
> I found that test occasionally fails. For instance:
> https://gitlab.com/tarantool/tarantool/-/jobs/548447188
> 
> So I have to push fix below to make it stable. In a nutshell we
> can't rely on range/count correlation if we want to avoid any
> possible races: ranges/runs are updated during compaction,
> meanwhile in fact we should wait till compaction is completed
> (since due to fails/error some results may turn out to be rollbacked).
> 
> It seems pretty straightforward, so hope you don't mind this fix.

Pushed to 1.10 both patches. Pushed to 2.3, 2.4 and master first
patch and a test extracted from second patch (since problem is
already fixed on master branch via 2f17c92). Updated changelogs
correspondingly. Branch is dropped. Before that I've verified that
CI statuses are OK.
 
> diff --git a/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua b/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua
> index 547ab628e..4b3c55505 100644
> --- a/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua
> +++ b/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua
> @@ -14,12 +14,18 @@ function dump(big)
>      box.snapshot()
>  end;
>  
> -function compact()
> +-- Tuple clean-up takes place after compaction is completed.
> +-- Meanwhile range count is updated during compaction process.
> +-- So instead of relying on range/run match, let's check explicitly
> +-- number of completed tasks.
> +--
> +function compact(tasks_expected)
> +    local scheduler = box.stat.vinyl().scheduler
> +    local tasks_completed = scheduler.tasks_completed
>      s.index.pk:compact()
>      repeat
>          fiber.sleep(0.001)
> -        local info = s.index.pk:stat()
> -    until info.range_count == info.run_count
> +    until box.stat.vinyl().scheduler.tasks_completed >= tasks_completed + tasks_expected
>  end;
>  test_run:cmd("setopt delimiter ''");
>  
> @@ -32,7 +38,7 @@ dump()
>  assert(s.index.pk:stat().range_count == 1)
>  assert(s.index.pk:stat().run_count == 2)
>  
> -compact()
> +compact(1)
>  assert(s.index.pk:stat().range_count == 1)
>  assert(s.index.pk:stat().run_count == 1)
>  
> @@ -46,7 +52,7 @@ errinj.set('ERRINJ_VY_STMT_ALLOC', 0)
>  -- Still split_range() fails, as a result we get one range
>  -- instead two.
>  --
> -compact()
> +compact(1)
>  assert(s.index.pk:stat().range_count == 1)
>  assert(s.index.pk:stat().run_count == 1)
>  assert(errinj.get('ERRINJ_VY_STMT_ALLOC') == -1)
> @@ -63,7 +69,7 @@ _ = s:create_index('pk', {run_count_per_level = 100, page_size = 128, range_size
>  dump(true)
>  dump()
>  
> -compact()
> +compact(1)
>  
>  dump()
>  
> @@ -72,7 +78,7 @@ errinj.set('ERRINJ_VY_STMT_ALLOC', 5)
>  -- Compaction of first range fails, so it is re-scheduled and
>  -- then successfully finishes at the second attempt.
>  --
> -compact()
> +compact(2)
>  assert(s.index.pk:stat().range_count == 2)
>  assert(s.index.pk:stat().run_count == 2)
>  assert(errinj.get('ERRINJ_VY_STMT_ALLOC') == -1)
> @@ -92,13 +98,13 @@ _ = s:create_index('pk', {run_count_per_level = 100, page_size = 128, range_size
>  dump(true)
>  dump()
>  
> -compact()
> +compact(1)
>  
>  dump()
>  
>  errinj = box.error.injection
>  errinj.set('ERRINJ_VY_READ_VIEW_MERGE_FAIL', true)
> -compact()
> +compact(2)
>  assert(s.index.pk:stat().range_count == 2)
>  assert(s.index.pk:stat().run_count == 2)
>  assert(errinj.get('ERRINJ_VY_READ_VIEW_MERGE_FAIL') == false)
> @@ -117,7 +123,7 @@ _ = s:create_index('pk', {run_count_per_level = 100, page_size = 128, range_size
>  dump(true)
>  dump()
>  
> -compact()
> +compact(1)
>  
>  dump()
>  assert(s.index.pk:stat().range_count == 1)
> @@ -125,14 +131,7 @@ assert(s.index.pk:stat().run_count == 2)
>  
>  errinj.set('ERRINJ_VY_WRITE_ITERATOR_START_FAIL', true)
>  errinj.set("ERRINJ_VY_SCHED_TIMEOUT", 0.1)
> -tasks_completed = box.stat.vinyl().scheduler.tasks_completed
> -s.index.pk:compact()
> --- Tuple clean-up takes place after compaction is completed.
> --- Meanwhile range count is updated during compaction process.
> --- So instead of relying on range/run match, let's check explicitly
> --- number of completed tasks.
> ---
> -repeat fiber.sleep(0.001) until box.stat.vinyl().scheduler.tasks_completed >= tasks_completed + 1
> +compact(2)
> 
>