Track RAM usage in temptable shared block

Summary:
When decreasing temptable_max_ram system variable, I am not
seeing decrease in memory when running workloads.
create_tmp_table is still taking significant amount of memory in
temptable APIs. This is because temptable_max_ram limit does not
apply to shared block. The shared block is a TLS variable. In
multithreaded environment, this takes a significant amount of
memory (>150 MB). The shared block is unconditionally allocated from
RAM.  Also the shared_block is never deallocated once allocated in a
thread.
The fix introduces a system variable temptable_track_shared_block_ram.
When enabled, the shared block memory allocation is also tracked.
This tracking enables shared block to be allocated from mmap files
in case the total allocated memory by temptable exceeds
temptable_max_ram. Also with the fix, shared_block is deallocated
after all the chunks are destroyed. This allows us to reclaim memory
space more proactively.
With this fix, I am seeing savings of >150 MB in our workloads.

Reviewed By: yizhang82

Differential Revision: D27520132

Author: atish2196

mysql-test/r/mysqld--help-notwin.result

@@ -2585,6 +2585,9 @@ The following options may be given as the first argument:
  Maximum amount of memory (in bytes) the TempTable storage
  engine is allowed to allocate from the main memory (RAM)
  before starting to store data on disk.
+ --temptable-track-shared-block-ram 
+ Track memory consumption of TLS shared block in temptable
+ (Defaults to on; use --skip-temptable-track-shared-block-ram to disable.)
  --temptable-use-mmap 
  Use mmap files for temptables. This variable is
  deprecated and will be removed in a future release.
@@ -3485,6 +3488,7 @@ tablespace-definition-cache 256
 tc-heuristic-recover OFF
 temptable-max-mmap 1073741824
 temptable-max-ram 1073741824
+temptable-track-shared-block-ram TRUE
 temptable-use-mmap TRUE
 terminology-use-previous NONE
 thread-cache-size 9

mysql-test/r/temptable_basic.result

@@ -497,23 +497,23 @@ truncate performance_schema.memory_summary_global_by_event_name;
 select * from performance_schema.memory_summary_global_by_event_name where event_name like 'memory/temptable%';
 EVENT_NAME	COUNT_ALLOC	COUNT_FREE	SUM_NUMBER_OF_BYTES_ALLOC	SUM_NUMBER_OF_BYTES_FREE	LOW_COUNT_USED	CURRENT_COUNT_USED	HIGH_COUNT_USED	LOW_NUMBER_OF_BYTES_USED	CURRENT_NUMBER_OF_BYTES_USED	HIGH_NUMBER_OF_BYTES_USED
 memory/temptable/physical_disk	0	0	0	0	0	0	0	0	0	0
-memory/temptable/physical_ram	1	0	1048608	0	1	1	1	1048608	1048608	1048608
+memory/temptable/physical_ram	0	0	0	0	0	0	0	0	0	0
 # conn1
 show variables like '%tmp_mem_storage%';
 Variable_name	Value
 internal_tmp_mem_storage_engine	TempTable
 select * from performance_schema.memory_summary_global_by_event_name where event_name like 'memory/temptable%';
 EVENT_NAME	COUNT_ALLOC	COUNT_FREE	SUM_NUMBER_OF_BYTES_ALLOC	SUM_NUMBER_OF_BYTES_FREE	LOW_COUNT_USED	CURRENT_COUNT_USED	HIGH_COUNT_USED	LOW_NUMBER_OF_BYTES_USED	CURRENT_NUMBER_OF_BYTES_USED	HIGH_NUMBER_OF_BYTES_USED
 memory/temptable/physical_disk	0	0	0	0	0	0	0	0	0	0
-memory/temptable/physical_ram	2	0	2097216	0	1	2	2	1048608	2097216	2097216
+memory/temptable/physical_ram	1	1	1048608	1048608	0	0	1	0	0	1048608
 select * from performance_schema.memory_summary_global_by_event_name where event_name like 'memory/temptable%';
 EVENT_NAME	COUNT_ALLOC	COUNT_FREE	SUM_NUMBER_OF_BYTES_ALLOC	SUM_NUMBER_OF_BYTES_FREE	LOW_COUNT_USED	CURRENT_COUNT_USED	HIGH_COUNT_USED	LOW_NUMBER_OF_BYTES_USED	CURRENT_NUMBER_OF_BYTES_USED	HIGH_NUMBER_OF_BYTES_USED
 memory/temptable/physical_disk	0	0	0	0	0	0	0	0	0	0
-memory/temptable/physical_ram	2	0	2097216	0	1	2	2	1048608	2097216	2097216
+memory/temptable/physical_ram	1	1	1048608	1048608	0	0	1	0	0	1048608
 select * from performance_schema.memory_summary_global_by_event_name where event_name like 'memory/temptable%';
 EVENT_NAME	COUNT_ALLOC	COUNT_FREE	SUM_NUMBER_OF_BYTES_ALLOC	SUM_NUMBER_OF_BYTES_FREE	LOW_COUNT_USED	CURRENT_COUNT_USED	HIGH_COUNT_USED	LOW_NUMBER_OF_BYTES_USED	CURRENT_NUMBER_OF_BYTES_USED	HIGH_NUMBER_OF_BYTES_USED
 memory/temptable/physical_disk	0	0	0	0	0	0	0	0	0	0
-memory/temptable/physical_ram	2	1	2097216	1048608	1	1	2	1048608	1048608	2097216
+memory/temptable/physical_ram	3	3	3145824	3145824	0	0	2	0	0	2097216
 #
 # Bug #31091089 TEMPTABLE: ASSERTION `P - M_MYSQL_BUF < M_LENGTH' FAILED.
 #

mysql-test/suite/sys_vars/r/temptable_track_shared_block_ram_basic.result

@@ -0,0 +1,21 @@
+select @@global.temptable_track_shared_block_ram;
+@@global.temptable_track_shared_block_ram
+1
+select @@session.temptable_track_shared_block_ram;
+ERROR HY000: Variable 'temptable_track_shared_block_ram' is a GLOBAL variable
+show global variables like 'temptable_track_shared_block_ram';
+Variable_name	Value
+temptable_track_shared_block_ram	ON
+show session variables like 'temptable_track_shared_block_ram';
+Variable_name	Value
+temptable_track_shared_block_ram	ON
+select * from performance_schema.global_variables where variable_name='temptable_track_shared_block_ram';
+VARIABLE_NAME	VARIABLE_VALUE
+temptable_track_shared_block_ram	ON
+select * from performance_schema.session_variables where variable_name='temptable_track_shared_block_ram';
+VARIABLE_NAME	VARIABLE_VALUE
+temptable_track_shared_block_ram	ON
+set global temptable_track_shared_block_ram=1;
+ERROR HY000: Variable 'temptable_track_shared_block_ram' is a read only variable
+set session temptable_track_shared_block_ram=1;
+ERROR HY000: Variable 'temptable_track_shared_block_ram' is a read only variable

mysql-test/suite/sys_vars/t/temptable_track_shared_block_ram_basic.test

@@ -0,0 +1,21 @@
+#
+# only global
+#
+--source include/have_log_bin.inc
+select @@global.temptable_track_shared_block_ram;
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+select @@session.temptable_track_shared_block_ram;
+show global variables like 'temptable_track_shared_block_ram';
+show session variables like 'temptable_track_shared_block_ram';
+--disable_warnings
+select * from performance_schema.global_variables where variable_name='temptable_track_shared_block_ram';
+select * from performance_schema.session_variables where variable_name='temptable_track_shared_block_ram';
+--enable_warnings
+
+#
+# show that it's read-only
+#
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+set global temptable_track_shared_block_ram=1;
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+set session temptable_track_shared_block_ram=1;

mysql-test/t/all_persisted_variables.test

@@ -69,6 +69,7 @@ let $total_excluded_vars=`SELECT COUNT(*) FROM performance_schema.global_variabl
 'per_user_session_var_user_name_delimiter',
 'query_cache_size',
 'query_cache_type',
+'temptable_track_shared_block_ram',
 'tx_isolation',
 'tx_read_only',
 'sql_wsenv_tenant',

sql/mysqld.cc

@@ -1176,6 +1176,7 @@ const char *default_storage_engine;
 const char *default_tmp_storage_engine;
 ulonglong temptable_max_ram;
 ulonglong temptable_max_mmap;
+bool temptable_track_shared_block_ram = false;
 bool temptable_use_mmap;
 static char compiled_default_collation_name[] = MYSQL_DEFAULT_COLLATION_NAME;
 static bool binlog_format_used = false;

sql/mysqld.h

@@ -271,6 +271,7 @@ extern const char *default_storage_engine;
 extern const char *default_tmp_storage_engine;
 extern ulonglong temptable_max_ram;
 extern ulonglong temptable_max_mmap;
+extern bool temptable_track_shared_block_ram;
 extern bool temptable_use_mmap;
 extern bool using_udf_functions;
 extern bool locked_in_memory;

sql/sys_vars.cc

@@ -5972,6 +5972,12 @@ static Sys_var_ulonglong Sys_temptable_max_mmap(
     GLOBAL_VAR(temptable_max_mmap), CMD_LINE(REQUIRED_ARG),
     VALID_RANGE(0, ULLONG_MAX), DEFAULT(1 << 30 /* 1 GiB */), BLOCK_SIZE(1));
 
+static Sys_var_bool Sys_temptable_track_shared_block_ram(
+    "temptable_track_shared_block_ram",
+    "Track memory consumption of TLS shared block in temptable",
+    READ_ONLY NON_PERSIST GLOBAL_VAR(temptable_track_shared_block_ram),
+    CMD_LINE(OPT_ARG), DEFAULT(true));
+
 static Sys_var_bool Sys_temptable_use_mmap(
     "temptable_use_mmap",
     "Use mmap files for temptables. "

storage/temptable/include/temptable/allocator.h

@@ -552,13 +552,14 @@ inline T *Allocator<T, AllocationScheme>::allocate(size_t n_elements) {
   }
 
   Block *block;
-
   if (m_shared_block && m_shared_block->is_empty()) {
     const size_t block_size =
         AllocationScheme::block_size(0, n_bytes_requested);
-    *m_shared_block = Block(
-        block_size,
-        AllocationScheme::block_source(block_size, &m_table_resource_monitor));
+    *m_shared_block =
+        Block(block_size, temptable_track_shared_block_ram
+                              ? AllocationScheme::block_source(
+                                    block_size, &m_table_resource_monitor)
+                              : Source::RAM);
     block = m_shared_block;
   } else if (m_shared_block &&
              m_shared_block->can_accommodate(n_bytes_requested)) {
@@ -600,7 +601,14 @@ inline void Allocator<T, AllocationScheme>::deallocate(T *chunk_data,
       block.deallocate(Chunk(chunk_data), n_bytes_requested);
   if (remaining_chunks == 0) {
     if (m_shared_block && (block == *m_shared_block)) {
-      // Do nothing. Keep the last block alive.
+      // Do nothing. Keep the last block alive unless
+      // we are tracking RAM consumption of shared block.
+      if (temptable_track_shared_block_ram) {
+        if (block.type() == Source::RAM) {
+          MemoryMonitor::RAM::decrease(block.size());
+        }
+        m_shared_block->destroy();
+      }
     } else {
       assert(m_state->number_of_blocks > 0);
       if (block.type() == Source::RAM) {