Commit ab061a2b authored by Sergey Petrunya's avatar Sergey Petrunya

MDEV-5926, MDEV-4362 post-fixes:

- Histogram::find_bucket() should not walk off the end of the value range.
- Address review feedback in Histogram::point_selectivity(): different handling
  for zero-width buckets, and explanations.
parent dee11f96
......@@ -1378,7 +1378,7 @@ Note 1003 select `test`.`t1`.`col1` AS `col1` from `test`.`t1` where (`test`.`t1
# Must not cause fp division by zero, or produce nonsense numbers:
explain extended select * from t1 where col1 in (-1,-2,-3);
id select_type table type possible_keys key key_len ref rows filtered Extra
1 SIMPLE t1 ALL NULL NULL NULL NULL 10000 3.00 Using where
1 SIMPLE t1 ALL NULL NULL NULL NULL 10000 5.94 Using where
Warnings:
Note 1003 select `test`.`t1`.`col1` AS `col1` from `test`.`t1` where (`test`.`t1`.`col1` in (<cache>(-(1)),<cache>(-(2)),<cache>(-(3))))
explain extended select * from t1 where col1<=-1;
......
......@@ -1388,7 +1388,7 @@ Note 1003 select `test`.`t1`.`col1` AS `col1` from `test`.`t1` where (`test`.`t1
# Must not cause fp division by zero, or produce nonsense numbers:
explain extended select * from t1 where col1 in (-1,-2,-3);
id select_type table type possible_keys key key_len ref rows filtered Extra
1 SIMPLE t1 ALL NULL NULL NULL NULL 10000 3.00 Using where
1 SIMPLE t1 ALL NULL NULL NULL NULL 10000 5.94 Using where
Warnings:
Note 1003 select `test`.`t1`.`col1` AS `col1` from `test`.`t1` where (`test`.`t1`.`col1` in (<cache>(-(1)),<cache>(-(2)),<cache>(-(3))))
explain extended select * from t1 where col1<=-1;
......
......@@ -151,6 +151,7 @@ class Histogram
}
return 0;
}
/* Find the bucket which value 'pos' falls into. */
uint find_bucket(double pos, bool first)
{
......@@ -171,7 +172,7 @@ class Histogram
break;
}
if (val > get_value(i))
if (val > get_value(i) && i < (get_width() - 1))
i++;
if (val == get_value(i))
......@@ -251,6 +252,27 @@ class Histogram
@return
Expected condition selectivity (a number between 0 and 1)
@notes
[re_zero_length_buckets] If a bucket with zero value-length is in the
middle of the histogram, we will not have min==max. Example: suppose,
pos_value=0x12, and the histogram is:
#n #n+1 #n+2
... 0x10 0x12 0x12 0x14 ...
|
+------------- bucket with zero value-length
Here, we will get min=#n+1, max=#n+2, and use the multi-bucket formula.
The problem happens at the histogram ends. if pos_value=0, and the
histogram is:
0x00 0x10 ...
then min=0, max=0. This means pos_value is contained within bucket #0,
but on the other hand, histogram data says that the bucket has only one
value.
*/
double point_selectivity(double pos, double avg_sel)
......@@ -265,6 +287,16 @@ class Histogram
while (max + 1 < get_width() && get_value(max + 1) == pos_value)
max++;
/*
A special case: we're looking at a single bucket, and that bucket has
zero value-length. Use the multi-bucket formula (attempt to use
single-bucket formula will cause divison by zero).
For more details see [re_zero_length_buckets] above.
*/
if (max == min && get_value(max) == ((max==0)? 0 : get_value(max-1)))
max++;
if (max > min)
{
/*
......@@ -302,17 +334,8 @@ class Histogram
(max + 1 == get_width() ? 1.0 : (get_value(max) * inv_prec_factor)) -
(min == 0 ? 0.0 : (get_value(min-1) * inv_prec_factor));
if (current_bucket_width < 1e-16)
{
/*
A special case: we are at the first (or the last) bucket in the
histogram, the bucket's value range is a singlepoint [x,x], and
pos_value=0 (for the first bucket) or pos_value=1 (for the last).
*/
sel= avg_sel;
}
else
{
DBUG_ASSERT(current_bucket_width); /* We shouldn't get a one zero-width bucket */
/*
So:
- each bucket has the same #rows
......@@ -322,7 +345,6 @@ class Histogram
each value will have to have N times fewer rows than average.
*/
sel= avg_sel * avg_bucket_width / current_bucket_width;
}
/*
(Q: if we just follow this proportion we may end up in a situation
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment