Re: [Tarantool-patches] [PATCH v2 2/2] vinyl: clean-up unprocessed read views in *_build_read_views()

[Nikita Pettik <korablev@tarantool.org>]

On 19 Apr 17:31, Vladislav Shpilevoy wrote:
> Hi! Thanks for the patch!
> 
> See 2 comments below.
> 
> > diff --git a/src/box/vy_write_iterator.c b/src/box/vy_write_iterator.c
> > index 7a6a20627..910e5a062 100644
> > --- a/src/box/vy_write_iterator.c
> > +++ b/src/box/vy_write_iterator.c
> > @@ -790,8 +790,11 @@ next_lsn:
> >  	 * statement around if this is major compaction, because
> >  	 * there's no tuple it could overwrite.
> >  	 */
> > -	if (rc == 0 && stream->is_last_level &&
> > -	    stream->deferred_delete_stmt != NULL) {
> > +	if (rc != 0) {
> > +		for (int i = 0; i < stream->rv_count; ++i)
> > +			stream->read_views[i].history = NULL;
> 
> 1. Why are we sure it is safe to nullify the read view histories in
> case of a fail? Looks like they can keep tuples which should be
> unreferenced. If it would be enough to nullify them, we could
> just do the same in vy_write_iterator_stop(). But it calls
> vy_write_history_destroy() before nullification.
> 
> Why is not it called here? The same question for vy_read_view_merge().

You are absolutely right. Thanks, fixed (see updated patch in v3).
 
> > @@ -834,6 +837,15 @@ vy_read_view_merge(struct vy_write_iterator *stream, struct tuple *hint,
> >  		rv->history = NULL;
> >  		return 0;
> >  	}
> > +#ifndef NDEBUG
> > +	struct errinj *inj =
> > +		errinj(ERRINJ_VY_READ_VIEW_MERGE_FAIL, ERRINJ_BOOL);
> > +	if (inj != NULL && inj->bparam) {
> > +			inj->bparam = false;
> > +			diag_set(OutOfMemory, 666, "malloc", "struct vy_stmt");
> > +			return -1;
> > +		}
> 
> 2. Indentation is too big.

Fixed.