Exploring this hypothesis further, I find a suspicious place in key_or.
It has this check
and this seems wrong to me: key2 here has use_count=0, but it not the root node of the SEL_ARG graph. SEL_ARG::use_count is valid only for the root nodes.
Elsewhere in this function, one can see that the code checks key2_shared which is the value of SEL_ARG::shared of a root node.
So I make a patch which checks key2_shared instead:
@@ -9597,10 +9600,12 @@ key_or(RANGE_OPT_PARAM *param, SEL_ARG *key1,SEL_ARG *key2)
|
|
if (!tmp->next_key_part)
|
{
|
- if (key2->use_count)
|
+ SEL_ARG *key2_next= key2->next;
|
+ //if (key2->use_count)
|
+ if (key2_shared)
|
{
|
SEL_ARG *key2_cpy= new SEL_ARG(*key2);
|
- if (key2_cpy)
|
+ if (!key2_cpy)
|
return 0;
|
key2= key2_cpy;
|
}
|
The whole patch for this MDEV thus becomes : https://gist.github.com/spetrunia/3eb23e92f8ab545d3180b496ce6ad24d
and this fixes the query.
igor any thoughts?
Major MDEV-24117 Memory management problem in statistics state for queries that have a large field IN (values) part
Page Failed to load
Exploring this hypothesis further, I find a suspicious place in key_or.
It has this check
and this seems wrong to me: key2 here has use_count=0, but it not the root node of the SEL_ARG graph. SEL_ARG::use_count is valid only for the root nodes.
Elsewhere in this function, one can see that the code checks key2_shared which is the value of SEL_ARG::shared of a root node.
So I make a patch which checks key2_shared instead:
@@ -9597,10 +9600,12 @@ key_or(RANGE_OPT_PARAM *param, SEL_ARG *key1,SEL_ARG *key2)if (!tmp->next_key_part){- if (key2->use_count)+ SEL_ARG *key2_next= key2->next;+ //if (key2->use_count)+ if (key2_shared){SEL_ARG *key2_cpy= new SEL_ARG(*key2);- if (key2_cpy)+ if (!key2_cpy)return 0;key2= key2_cpy;The whole patch for this MDEV thus becomes : https://gist.github.com/spetrunia/3eb23e92f8ab545d3180b496ce6ad24d
and this fixes the query.
igor any thoughts?