The memory seems to be allocated here:

void
dict_stats_recalc_pool_init()
/=========================/
{
ut_ad(!srv_read_only_mode);

recalc_pool.reserve(RECALC_POOL_INITIAL_SLOTS);
}

recalc_pool is defined above as:

typedef std::vector<table_id_t> recalc_pool_t;
static recalc_pool_t recalc_pool;

There is a corresponding deinit function: dict_stats_recalc_pool_deinit()

If I put printouts into dict_stats_recalc_pool_init() and dict_stats_recalc_pool_deinit(), the error log shows that _deinit() call is made by the test. Memory leak is still reported.

Exploration patch:

=== modified file 'storage/innobase/dict/dict0stats_bg.cc'
— storage/innobase/dict/dict0stats_bg.cc 2013-03-25 22:03:13 +0000
+++ storage/innobase/dict/dict0stats_bg.cc 2013-07-04 06:49:06 +0000
@@ -50,7 +50,22 @@ static const ulint RECALC_POOL_INITIAL_S

/** The multitude of tables whose stats are to be automatically
recalculated - an STL vector */
-typedef std::vector<table_id_t> recalc_pool_t;
+typedef std::vector<table_id_t> recalc_pool_t_parent;
+
+class recalc_pool_t : public recalc_pool_t_parent
+{
+public:
+ recalc_pool_t()
+

+ ~recalc_pool_t()
+

+
+};
+
static recalc_pool_t recalc_pool;

typedef recalc_pool_t::iterator recalc_pool_iterator_t;
@@ -63,6 +78,7 @@ dict_stats_recalc_pool_init()
/=========================/

@@ -76,6 +92,7 @@ dict_stats_recalc_pool_deinit()
/===========================/

Running with the exploration patch produces this:

recalc_pool_t ctor
...
dict_stats_recalc_pool_init
...
dict_stats_recalc_pool_deinit
...
Warning: 1024 bytes lost at 0x2134780, allocated by T@0 at mysys/my_new.cc:33, ext/new_allocator.h:90, bits/stl_vector.h:140, bits/stl_vector.h:967, bits/vector.tcc:74, dict/dict0stats_bg.cc:84, dict/dict0stats_bg.cc:264, srv/srv0start.cc:2183
Memory lost: 1024 bytes in 1 chunks
...
recalc_pool_t dtor
...

Apparently, the memory is freed in the end, but that happens after the mem leak detector is run.

Fix pushed into 10.0-monty