1berry/MariaDB-server
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Bug#21074 Large query_cache freezes mysql server sporadically under heavy load

Browse Source 
Invaldating a subset of a sufficiently large query cache can take a long time.
During this time the server is efficiently frozen and no other operation can
be executed. This patch addresses this problem by moving the locks which cause
the freezing and also by temporarily disable the query cache while the 
invalidation takes place.


sql/ha_ndbcluster.cc:
  - mysql_rm_table_part2 has a new parameter to indicate if OPEN_lock mutex 
    protection is needed.
sql/lock.cc:
  - Added function for acquiring table name exclusive locks.
  - Added function for asserting that table name lock is acquired.
sql/mysql_priv.h:
  - Added function for acquiring table name exclusive locks.
  - Added function for asserting that table name lock is acquired.
  - Added parameter to mysql_rm_table_part2 to indicate whether OPEN_lock mutex
    protection is needed or not.
sql/sql_cache.cc:
  - Changed flush_in_progress-flag into a state and added a function, 
    is_flushing() to reflect on this state. A new state was needed to indicate
    that a partial invalidation was in progress.
  - An unused parameter 'under_guard' was removed.
  - The Query_cache mutex structural_guard was pushed down into one
    invalidate_table function to avoid multiple entry points which makes
    maintainens more difficult.
  - Instead of keeping the structural_guard mutex during the entire invalidation
    we set the query cache status state to TABLE_FLUSH_IN_PROGRESS to
    temporarily disable the cache and avoid locking other threads needing the
    Query_cache resource.
sql/sql_cache.h:
  - Changed flush_in_progress-flag into a state and added a function, 
    is_flushing() to reflect on this state. A new state was needed to indicate
    that a partial invalidation was in progress.
  - An unused parameter 'under_guard' was removed.
  - The Query_cache mutex structural_guard was pushed down into one
    invalidate_table function to avoid multiple entry points which makes
    maintainens more difficult.
  - Instead of keeping the structural_guard mutex during the entire invalidation
    we set the query cache status state to TABLE_FLUSH_IN_PROGRESS to
    temporarily disable the cache and avoid locking other threads needing the
    the Query_cache resource.
sql/sql_db.cc:
  - mysql_rm_table_part2_with_lock is redundant and replaced
  with mysql_rm_table_part2.
sql/sql_parse.cc:
  - Function query_cache_invalidate3 isn't protect by a lock and we have a 
    race condition.
  - Moving this function into mysql_rename_tables and make sure it is protected
    by a exclusive table name lock.
sql/sql_rename.cc:
  - Function query_cache_invalidation3 isn't protect by a lock and we have a 
    race condition.
  - Moving this function into mysql_rename_tables and make sure it is protected
    by a exclusive table name lock.
  - Instead of using LOCK_open mutex, which excludes all other threads, the lock
    is changed into exclusive table name locks instead. This prevents us from
    locking the server if a query cache invalidation would take a long time to
    complete.
sql/sql_table.cc:
  - Instead of using LOCK_open mutex, which excludes all other threads, the lock
    is changed into exclusive table name locks instead. This prevents us from
    locking the server if a query cache invalidation would take a long time to
    complete.
  - Added new parameter to mysql_rm_table_part2 to control whether OPEN_lock mutex
    needs to be aquired or not. This is currently needed by the NDB implemenation.
sql/sql_trigger.cc:
  - Table_triggers don't need to be protexted by LOCK_open mutex. This 
    patch cancel this restriction.
  - Refactored comments to doxygen style.
This commit is contained in:
unknown 2007-07-02 19:14:48 +02:00
parent 0484d44eeb
commit 289cc26c64
10 changed files with 677 additions and 372 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
sql/ha_ndbcluster.cc
View File

@ -6996,7 +6996,6 @@ int ndbcluster_find_files(handlerton *hton, THD *thd,
// Lock mutex before deleting and creating frm files // Lock mutex before deleting and creating frm files
pthread_mutex_lock(&LOCK_open); pthread_mutex_lock(&LOCK_open);
if (!global_read_lock) if (!global_read_lock)
{ {
// Delete old files // Delete old files
@ -7010,10 +7009,12 @@ int ndbcluster_find_files(handlerton *hton, THD *thd,
table_list.db= (char*) db; table_list.db= (char*) db;
table_list.alias= table_list.table_name= (char*)file_name; table_list.alias= table_list.table_name= (char*)file_name;
(void)mysql_rm_table_part2(thd, &table_list, (void)mysql_rm_table_part2(thd, &table_list,
/* if_exists */ FALSE, FALSE, /* if_exists */
/* drop_temporary */ FALSE, FALSE, /* drop_temporary */
/* drop_view */ FALSE, FALSE, /* drop_view */
/* dont_log_query*/ TRUE); TRUE, /* dont_log_query*/
FALSE); /* need lock open */
/* Clear error message that is returned when table is deleted */ /* Clear error message that is returned when table is deleted */
thd->clear_error(); thd->clear_error();
} }

96
sql/lock.cc
View File

@ -1027,6 +1027,102 @@ end:
} }
/**
@brief Lock all tables in list with an exclusive table name lock.
@param thd Thread handle.
@param table_list Names of tables to lock.
@note This function needs to be protected by LOCK_open. If we're
under LOCK TABLES, this function does not work as advertised. Namely,
it does not exclude other threads from using this table and does not
put an exclusive name lock on this table into the table cache.
@see lock_table_names
@see unlock_table_names
@retval TRUE An error occured.
@retval FALSE Name lock successfully acquired.
*/
bool lock_table_names_exclusively(THD *thd, TABLE_LIST *table_list)
{
if (lock_table_names(thd, table_list))
return TRUE;
/*
Upgrade the table name locks from semi-exclusive to exclusive locks.
*/
for (TABLE_LIST *table= table_list; table; table= table->next_global)
{
if (table->table)
table->table->open_placeholder= 1;
}
return FALSE;
}
/**
@brief Test is 'table' is protected by an exclusive name lock.
@param[in] thd The current thread handler
@param[in] table Table container containing the single table to be tested
@note Needs to be protected by LOCK_open mutex.
@return Error status code
@retval TRUE Table is protected
@retval FALSE Table is not protected
*/
bool
is_table_name_exclusively_locked_by_this_thread(THD *thd,
TABLE_LIST *table_list)
{
char key[MAX_DBKEY_LENGTH];
uint key_length;
key_length= create_table_def_key(thd, key, table_list, 0);
return is_table_name_exclusively_locked_by_this_thread(thd, (uchar *)key,
key_length);
}
/**
@brief Test is 'table key' is protected by an exclusive name lock.
@param[in] thd The current thread handler.
@param[in] table Table container containing the single table to be tested.
@note Needs to be protected by LOCK_open mutex
@retval TRUE Table is protected
@retval FALSE Table is not protected
*/
bool
is_table_name_exclusively_locked_by_this_thread(THD *thd, uchar *key,
int key_length)
{
HASH_SEARCH_STATE state;
TABLE *table;
for (table= (TABLE*) hash_first(&open_cache, key,
key_length, &state);
table ;
table= (TABLE*) hash_next(&open_cache, key,
key_length, &state))
{
if (table->in_use == thd &&
table->open_placeholder == 1 &&
table->s->version == 0)
return TRUE;
}
return FALSE;
}
/* /*
Unlock all tables in list with a name lock Unlock all tables in list with a name lock

11
sql/mysql_priv.h
View File

@ -820,10 +820,8 @@ void mysql_client_binlog_statement(THD *thd);
bool mysql_rm_table(THD *thd,TABLE_LIST *tables, my_bool if_exists, bool mysql_rm_table(THD *thd,TABLE_LIST *tables, my_bool if_exists,
my_bool drop_temporary); my_bool drop_temporary);
int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists, int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
bool drop_temporary, bool drop_view, bool log_query); bool drop_temporary, bool drop_view, bool log_query,
int mysql_rm_table_part2_with_lock(THD *thd, TABLE_LIST *tables, bool need_lock_open);
bool if_exists, bool drop_temporary,
bool log_query);
bool quick_rm_table(handlerton *base,const char *db, bool quick_rm_table(handlerton *base,const char *db,
const char *table_name, uint flags); const char *table_name, uint flags);
void close_cached_table(THD *thd, TABLE *table); void close_cached_table(THD *thd, TABLE *table);
@ -1799,6 +1797,11 @@ bool wait_for_locked_table_names(THD *thd, TABLE_LIST *table_list);
bool lock_table_names(THD *thd, TABLE_LIST *table_list); bool lock_table_names(THD *thd, TABLE_LIST *table_list);
void unlock_table_names(THD *thd, TABLE_LIST *table_list, void unlock_table_names(THD *thd, TABLE_LIST *table_list,
TABLE_LIST *last_table); TABLE_LIST *last_table);
bool lock_table_names_exclusively(THD *thd, TABLE_LIST *table_list);
bool is_table_name_exclusively_locked_by_this_thread(THD *thd,
TABLE_LIST *table_list);
bool is_table_name_exclusively_locked_by_this_thread(THD *thd, uchar *key,
int key_length);
/* old unireg functions */ /* old unireg functions */

545
sql/sql_cache.cc
View File

@ -268,6 +268,39 @@ are stored in one block.
If join_results allocated new block(s) then we need call pack_cache again. If join_results allocated new block(s) then we need call pack_cache again.
7. Interface
The query cache interfaces with the rest of the server code through 7
functions:
1. Query_cache::send_result_to_client
- Called before parsing and used to match a statement with the stored
queries hash.
If a match is found the cached result set is sent through repeated
calls to net_real_write. (note: calling thread doesn't have a regis-
tered result set writer: thd->net.query_cache_query=0)
2. Query_cache::store_query
- Called just before handle_select() and is used to register a result
set writer to the statement currently being processed
(thd->net.query_cache_query).
3. query_cache_insert
- Called from net_real_write to append a result set to a cached query
if (and only if) this query has a registered result set writer
(thd->net.query_cache_query).
4. Query_cache::invalidate
- Called from various places to invalidate query cache based on data-
base, table and myisam file name. During an on going invalidation
the query cache is temporarily disabled.
5. Query_cache::flush
- Used when a RESET QUERY CACHE is issued. This clears the entire
cache block by block.
6. Query_cache::resize
- Used to change the available memory used by the query cache. This
will also invalidate the entrie query cache in one free operation.
7. Query_cache::pack
- Used when a FLUSH QUERY CACHE is issued. This changes the order of
the used memory blocks in physical memory order and move all avail-
able memory to the 'bottom' of the memory.
TODO list: TODO list:
- Delayed till after-parsing qache answer (for column rights processing) - Delayed till after-parsing qache answer (for column rights processing)
@ -615,18 +648,25 @@ void query_cache_insert(NET *net, const char *packet, ulong length)
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
STRUCT_LOCK(&query_cache.structure_guard_mutex); STRUCT_LOCK(&query_cache.structure_guard_mutex);
bool interrupt;
if (unlikely(query_cache.query_cache_size == 0 || query_cache.wait_while_table_flush_is_in_progress(&interrupt);
query_cache.flush_in_progress)) if (interrupt)
{ {
STRUCT_UNLOCK(&query_cache.structure_guard_mutex);
return;
}
Query_cache_block *query_block= (Query_cache_block*)net->query_cache_query;
if (!query_block)
{
/*
We lost the writer and the currently processed query has been
invalidated; there is nothing left to do.
*/
STRUCT_UNLOCK(&query_cache.structure_guard_mutex); STRUCT_UNLOCK(&query_cache.structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
Query_cache_block *query_block = ((Query_cache_block*)
net->query_cache_query);
if (query_block)
{
Query_cache_query *header= query_block->query(); Query_cache_query *header= query_block->query();
Query_cache_block *result= header->result(); Query_cache_block *result= header->result();
@ -635,9 +675,9 @@ void query_cache_insert(NET *net, const char *packet, ulong length)
DBUG_PRINT("qcache", ("insert packet %lu bytes long",length)); DBUG_PRINT("qcache", ("insert packet %lu bytes long",length));
/* /*
On success STRUCT_UNLOCK(&query_cache.structure_guard_mutex) will be On success, STRUCT_UNLOCK is done by append_result_data. Otherwise, we
done by query_cache.append_result_data if success (if not we need still need structure_guard_mutex to free the query, and therefore unlock
query_cache.structure_guard_mutex locked to free query) it later in this function.
*/ */
if (!query_cache.append_result_data(&result, length, (uchar*) packet, if (!query_cache.append_result_data(&result, length, (uchar*) packet,
query_block)) query_block))
@ -651,13 +691,12 @@ void query_cache_insert(NET *net, const char *packet, ulong length)
STRUCT_UNLOCK(&query_cache.structure_guard_mutex); STRUCT_UNLOCK(&query_cache.structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
header->result(result); header->result(result);
header->last_pkt_nr= net->pkt_nr; header->last_pkt_nr= net->pkt_nr;
BLOCK_UNLOCK_WR(query_block); BLOCK_UNLOCK_WR(query_block);
DBUG_EXECUTE("check_querycache",query_cache.check_integrity(0);); DBUG_EXECUTE("check_querycache",query_cache.check_integrity(0););
}
else
STRUCT_UNLOCK(&query_cache.structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -671,17 +710,21 @@ void query_cache_abort(NET *net)
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
STRUCT_LOCK(&query_cache.structure_guard_mutex); STRUCT_LOCK(&query_cache.structure_guard_mutex);
bool interrupt;
if (unlikely(query_cache.query_cache_size == 0 || query_cache.wait_while_table_flush_is_in_progress(&interrupt);
query_cache.flush_in_progress)) if (interrupt)
{ {
STRUCT_UNLOCK(&query_cache.structure_guard_mutex); STRUCT_UNLOCK(&query_cache.structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
/*
While we were waiting another thread might have changed the status
of the writer. Make sure the writer still exists before continue.
*/
Query_cache_block *query_block= ((Query_cache_block*) Query_cache_block *query_block= ((Query_cache_block*)
net->query_cache_query); net->query_cache_query);
if (query_block) // Test if changed by other thread if (query_block)
{ {
DUMP(&query_cache); DUMP(&query_cache);
BLOCK_LOCK_WR(query_block); BLOCK_LOCK_WR(query_block);
@ -713,13 +756,22 @@ void query_cache_end_of_result(THD *thd)
STRUCT_LOCK(&query_cache.structure_guard_mutex); STRUCT_LOCK(&query_cache.structure_guard_mutex);
if (unlikely(query_cache.query_cache_size == 0 || bool interrupt;
query_cache.flush_in_progress)) query_cache.wait_while_table_flush_is_in_progress(&interrupt);
goto end; if (interrupt)
{
STRUCT_UNLOCK(&query_cache.structure_guard_mutex);
DBUG_VOID_RETURN;
}
query_block= ((Query_cache_block*) thd->net.query_cache_query); query_block= ((Query_cache_block*) thd->net.query_cache_query);
if (query_block) if (query_block)
{ {
/*
The writer is still present; finish last result block by chopping it to
suitable size if needed and setting block type. Since this is the last
block, the writer should be dropped.
*/
DUMP(&query_cache); DUMP(&query_cache);
BLOCK_LOCK_WR(query_block); BLOCK_LOCK_WR(query_block);
Query_cache_query *header= query_block->query(); Query_cache_query *header= query_block->query();
@ -746,8 +798,11 @@ void query_cache_end_of_result(THD *thd)
#endif #endif
header->found_rows(current_thd->limit_found_rows); header->found_rows(current_thd->limit_found_rows);
header->result()->type= Query_cache_block::RESULT; header->result()->type= Query_cache_block::RESULT;
/* Drop the writer. */
header->writer(0); header->writer(0);
thd->net.query_cache_query= 0; thd->net.query_cache_query= 0;
BLOCK_UNLOCK_WR(query_block); BLOCK_UNLOCK_WR(query_block);
DBUG_EXECUTE("check_querycache",query_cache.check_integrity(1);); DBUG_EXECUTE("check_querycache",query_cache.check_integrity(1););
@ -801,9 +856,9 @@ ulong Query_cache::resize(ulong query_cache_size_arg)
DBUG_ASSERT(initialized); DBUG_ASSERT(initialized);
STRUCT_LOCK(&structure_guard_mutex); STRUCT_LOCK(&structure_guard_mutex);
while (flush_in_progress) while (is_flushing())
pthread_cond_wait(&COND_flush_finished, &structure_guard_mutex); pthread_cond_wait(&COND_cache_status_changed, &structure_guard_mutex);
flush_in_progress= TRUE; m_cache_status= Query_cache::FLUSH_IN_PROGRESS;
STRUCT_UNLOCK(&structure_guard_mutex); STRUCT_UNLOCK(&structure_guard_mutex);
free_cache(); free_cache();
@ -814,8 +869,8 @@ ulong Query_cache::resize(ulong query_cache_size_arg)
DBUG_EXECUTE("check_querycache",check_integrity(0);); DBUG_EXECUTE("check_querycache",check_integrity(0););
STRUCT_LOCK(&structure_guard_mutex); STRUCT_LOCK(&structure_guard_mutex);
flush_in_progress= FALSE; m_cache_status= Query_cache::NO_FLUSH_IN_PROGRESS;
pthread_cond_signal(&COND_flush_finished); pthread_cond_signal(&COND_cache_status_changed);
STRUCT_UNLOCK(&structure_guard_mutex); STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_RETURN(new_query_cache_size); DBUG_RETURN(new_query_cache_size);
@ -910,8 +965,13 @@ def_week_frmt: %lu",
ha_release_temporary_latches(thd); ha_release_temporary_latches(thd);
STRUCT_LOCK(&structure_guard_mutex); STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size == 0 || flush_in_progress) if (query_cache_size == 0 || is_flushing())
{ {
/*
A table- or a full flush operation can potentially take a long time to
finish. We choose not to wait for them and skip caching statements
instead.
*/
STRUCT_UNLOCK(&structure_guard_mutex); STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -954,7 +1014,7 @@ def_week_frmt: %lu",
Query_cache_block *query_block; Query_cache_block *query_block;
query_block= write_block_data(tot_length, (uchar*) thd->query, query_block= write_block_data(tot_length, (uchar*) thd->query,
ALIGN_SIZE(sizeof(Query_cache_query)), ALIGN_SIZE(sizeof(Query_cache_query)),
Query_cache_block::QUERY, local_tables, 1); Query_cache_block::QUERY, local_tables);
if (query_block != 0) if (query_block != 0)
{ {
DBUG_PRINT("qcache", ("query block 0x%lx allocated, %lu", DBUG_PRINT("qcache", ("query block 0x%lx allocated, %lu",
@ -1088,13 +1148,21 @@ Query_cache::send_result_to_client(THD *thd, char *sql, uint query_length)
} }
STRUCT_LOCK(&structure_guard_mutex); STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size == 0 || flush_in_progress)
if (query_cache_size == 0)
goto err_unlock;
if (is_flushing())
{ {
/* Return; Query cache is temporarily disabled while we flush. */
DBUG_PRINT("qcache",("query cache disabled")); DBUG_PRINT("qcache",("query cache disabled"));
goto err_unlock; goto err_unlock;
} }
/* Check that we haven't forgot to reset the query cache variables */ /*
Check that we haven't forgot to reset the query cache variables;
make sure there are no attached query cache writer to this thread.
*/
DBUG_ASSERT(thd->net.query_cache_query == 0); DBUG_ASSERT(thd->net.query_cache_query == 0);
Query_cache_block *query_block; Query_cache_block *query_block;
@ -1267,7 +1335,7 @@ def_week_frmt: %lu",
("Handler require invalidation queries of %s.%s %lu-%lu", ("Handler require invalidation queries of %s.%s %lu-%lu",
table_list.db, table_list.alias, table_list.db, table_list.alias,
(ulong) engine_data, (ulong) table->engine_data())); (ulong) engine_data, (ulong) table->engine_data()));
invalidate_table((uchar *) table->db(), table->key_length()); invalidate_table(thd, (uchar *) table->db(), table->key_length());
} }
else else
thd->lex->safe_to_cache_query= 0; // Don't try to cache this thd->lex->safe_to_cache_query= 0; // Don't try to cache this
@ -1330,10 +1398,6 @@ void Query_cache::invalidate(THD *thd, TABLE_LIST *tables_used,
my_bool using_transactions) my_bool using_transactions)
{ {
DBUG_ENTER("Query_cache::invalidate (table list)"); DBUG_ENTER("Query_cache::invalidate (table list)");
STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size > 0 && !flush_in_progress)
{
DUMP(this);
using_transactions= using_transactions && using_transactions= using_transactions &&
(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)); (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN));
@ -1346,16 +1410,14 @@ void Query_cache::invalidate(THD *thd, TABLE_LIST *tables_used,
(tables_used->table->file->table_cache_type() == (tables_used->table->file->table_cache_type() ==
HA_CACHE_TBL_TRANSACT)) HA_CACHE_TBL_TRANSACT))
/* /*
Tables_used->table can't be 0 in transaction. tables_used->table can't be 0 in transaction.
Only 'drop' invalidate not opened table, but 'drop' Only 'drop' invalidate not opened table, but 'drop'
force transaction finish. force transaction finish.
*/ */
thd->add_changed_table(tables_used->table); thd->add_changed_table(tables_used->table);
else else
invalidate_table(tables_used); invalidate_table(thd, tables_used);
} }
}
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -1363,22 +1425,14 @@ void Query_cache::invalidate(THD *thd, TABLE_LIST *tables_used,
void Query_cache::invalidate(CHANGED_TABLE_LIST *tables_used) void Query_cache::invalidate(CHANGED_TABLE_LIST *tables_used)
{ {
DBUG_ENTER("Query_cache::invalidate (changed table list)"); DBUG_ENTER("Query_cache::invalidate (changed table list)");
if (tables_used) THD *thd= current_thd;
{
STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size > 0 && !flush_in_progress)
{
DUMP(this);
for (; tables_used; tables_used= tables_used->next) for (; tables_used; tables_used= tables_used->next)
{ {
invalidate_table((uchar*) tables_used->key, tables_used->key_length); invalidate_table(thd, (uchar*) tables_used->key, tables_used->key_length);
DBUG_PRINT("qcache", ("db: %s table: %s", tables_used->key, DBUG_PRINT("qcache", ("db: %s table: %s", tables_used->key,
tables_used->key+ tables_used->key+
strlen(tables_used->key)+1)); strlen(tables_used->key)+1));
} }
}
STRUCT_UNLOCK(&structure_guard_mutex);
}
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -1396,21 +1450,15 @@ void Query_cache::invalidate(CHANGED_TABLE_LIST *tables_used)
void Query_cache::invalidate_locked_for_write(TABLE_LIST *tables_used) void Query_cache::invalidate_locked_for_write(TABLE_LIST *tables_used)
{ {
DBUG_ENTER("Query_cache::invalidate_locked_for_write"); DBUG_ENTER("Query_cache::invalidate_locked_for_write");
if (tables_used)
{
STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size > 0 && !flush_in_progress)
{
DUMP(this);
for (; tables_used; tables_used= tables_used->next_local) for (; tables_used; tables_used= tables_used->next_local)
{ {
if (tables_used->lock_type & (TL_WRITE_LOW_PRIORITY | TL_WRITE) && if (tables_used->lock_type & (TL_WRITE_LOW_PRIORITY | TL_WRITE) &&
tables_used->table) tables_used->table)
invalidate_table(tables_used->table); {
THD *thd= current_thd;
invalidate_table(thd, tables_used->table);
} }
} }
STRUCT_UNLOCK(&structure_guard_mutex);
}
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -1423,18 +1471,14 @@ void Query_cache::invalidate(THD *thd, TABLE *table,
{ {
DBUG_ENTER("Query_cache::invalidate (table)"); DBUG_ENTER("Query_cache::invalidate (table)");
STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size > 0 && !flush_in_progress)
{
using_transactions= using_transactions && using_transactions= using_transactions &&
(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)); (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN));
if (using_transactions && if (using_transactions &&
(table->file->table_cache_type() == HA_CACHE_TBL_TRANSACT)) (table->file->table_cache_type() == HA_CACHE_TBL_TRANSACT))
thd->add_changed_table(table); thd->add_changed_table(table);
else else
invalidate_table(table); invalidate_table(thd, table);
}
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -1444,30 +1488,76 @@ void Query_cache::invalidate(THD *thd, const char *key, uint32 key_length,
{ {
DBUG_ENTER("Query_cache::invalidate (key)"); DBUG_ENTER("Query_cache::invalidate (key)");
STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size > 0 && !flush_in_progress)
{
using_transactions= using_transactions && using_transactions= using_transactions &&
(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)); (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN));
if (using_transactions) // used for innodb => has_transactions() is TRUE if (using_transactions) // used for innodb => has_transactions() is TRUE
thd->add_changed_table(key, key_length); thd->add_changed_table(key, key_length);
else else
invalidate_table((uchar*)key, key_length); invalidate_table(thd, (uchar*)key, key_length);
}
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
/**
@brief Synchronize the thread with any flushing operations.
This helper function is called whenever a thread needs to operate on the
query cache structure (example: during invalidation). If a table flush is in
progress this function will wait for it to stop. If a full flush is in
progress, the function will set the interrupt parameter to indicate that the
current operation is redundant and should be interrupted.
@param[out] interrupt This out-parameter will be set to TRUE if the calling
function is redundant and should be interrupted.
@return If the interrupt-parameter is TRUE then m_cache_status is set to
NO_FLUSH_IN_PROGRESS. If the interrupt-parameter is FALSE then
m_cache_status is set to FLUSH_IN_PROGRESS.
The structure_guard_mutex will in any case be locked.
*/
void Query_cache::wait_while_table_flush_is_in_progress(bool *interrupt)
{
while (is_flushing())
{
/*
If there already is a full flush in progress query cache isn't enabled
and additional flushes are redundant; just return instead.
*/
if (m_cache_status == Query_cache::FLUSH_IN_PROGRESS)
{
*interrupt= TRUE;
return;
}
/*
If a table flush is in progress; wait on cache status to change.
*/
if (m_cache_status == Query_cache::TABLE_FLUSH_IN_PROGRESS)
pthread_cond_wait(&COND_cache_status_changed, &structure_guard_mutex);
}
*interrupt= FALSE;
}
/* /*
Remove all cached queries that uses the given database Remove all cached queries that uses the given database
*/ */
void Query_cache::invalidate(char *db) void Query_cache::invalidate(char *db)
{ {
DBUG_ENTER("Query_cache::invalidate (db)"); DBUG_ENTER("Query_cache::invalidate (db)");
STRUCT_LOCK(&structure_guard_mutex); STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size > 0 && !flush_in_progress) bool interrupt;
wait_while_table_flush_is_in_progress(&interrupt);
if (interrupt)
{
STRUCT_UNLOCK(&structure_guard_mutex);
return;
}
THD *thd= current_thd;
if (query_cache_size > 0)
{ {
DUMP(this); DUMP(this);
restart_search: restart_search:
@ -1479,7 +1569,10 @@ void Query_cache::invalidate(char *db)
{ {
next= curr->next; next= curr->next;
if (strcmp(db, (char*)(curr->table()->db())) == 0) if (strcmp(db, (char*)(curr->table()->db())) == 0)
invalidate_table(curr); {
Query_cache_block_table *list_root= curr->table(0);
invalidate_query_block_list(thd,list_root);
}
/* /*
invalidate_table can freed block on which point 'next' (if invalidate_table can freed block on which point 'next' (if
table of this block used only in queries which was deleted table of this block used only in queries which was deleted
@ -1491,6 +1584,11 @@ void Query_cache::invalidate(char *db)
if (next->type == Query_cache_block::FREE) if (next->type == Query_cache_block::FREE)
goto restart_search; goto restart_search;
curr= next; curr= next;
/*
The loop will end if the circular list pointer has reached the
point where it started from, or if the current thread was signaled
to die.
*/
} while (curr != tables_blocks); } while (curr != tables_blocks);
} }
} }
@ -1504,21 +1602,12 @@ void Query_cache::invalidate_by_MyISAM_filename(const char *filename)
{ {
DBUG_ENTER("Query_cache::invalidate_by_MyISAM_filename"); DBUG_ENTER("Query_cache::invalidate_by_MyISAM_filename");
STRUCT_LOCK(&structure_guard_mutex);
if (query_cache_size > 0 && !flush_in_progress)
{
/* Calculate the key outside the lock to make the lock shorter */ /* Calculate the key outside the lock to make the lock shorter */
char key[MAX_DBKEY_LENGTH]; char key[MAX_DBKEY_LENGTH];
uint32 db_length; uint32 db_length;
uint key_length= filename_2_table_key(key, filename, &db_length); uint key_length= filename_2_table_key(key, filename, &db_length);
Query_cache_block *table_block; THD *thd= current_thd;
if ((table_block = (Query_cache_block*) hash_search(&tables, invalidate_table(thd,(uchar *)key, key_length);
(uchar*) key,
key_length)))
invalidate_table(table_block);
}
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -1540,16 +1629,43 @@ void Query_cache::flush()
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
/* Join result in cache in 1 block (if result length > join_limit) */
/**
@brief Rearrange the memory blocks and join result in cache in 1 block (if
result length > join_limit)
@param[in] join_limit If the minimum length of a result block to be joined.
@param[in] iteration_limit The maximum number of packing and joining
sequences.
*/
void Query_cache::pack(ulong join_limit, uint iteration_limit) void Query_cache::pack(ulong join_limit, uint iteration_limit)
{ {
DBUG_ENTER("Query_cache::pack"); DBUG_ENTER("Query_cache::pack");
bool interrupt;
STRUCT_LOCK(&structure_guard_mutex);
wait_while_table_flush_is_in_progress(&interrupt);
if (interrupt)
{
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN;
}
if (query_cache_size == 0)
{
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN;
}
uint i = 0; uint i = 0;
do do
{ {
pack_cache(); pack_cache();
} while ((++i < iteration_limit) && join_results(join_limit)); } while ((++i < iteration_limit) && join_results(join_limit));
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -1568,7 +1684,7 @@ void Query_cache::destroy()
free_cache(); free_cache();
STRUCT_UNLOCK(&structure_guard_mutex); STRUCT_UNLOCK(&structure_guard_mutex);
pthread_cond_destroy(&COND_flush_finished); pthread_cond_destroy(&COND_cache_status_changed);
pthread_mutex_destroy(&structure_guard_mutex); pthread_mutex_destroy(&structure_guard_mutex);
initialized = 0; initialized = 0;
} }
@ -1584,8 +1700,8 @@ void Query_cache::init()
{ {
DBUG_ENTER("Query_cache::init"); DBUG_ENTER("Query_cache::init");
pthread_mutex_init(&structure_guard_mutex,MY_MUTEX_INIT_FAST); pthread_mutex_init(&structure_guard_mutex,MY_MUTEX_INIT_FAST);
pthread_cond_init(&COND_flush_finished, NULL); pthread_cond_init(&COND_cache_status_changed, NULL);
flush_in_progress= FALSE; m_cache_status= Query_cache::NO_FLUSH_IN_PROGRESS;
initialized = 1; initialized = 1;
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -1787,9 +1903,10 @@ void Query_cache::make_disabled()
/** /**
@class Query_cache @class Query_cache
@brief Free all resources allocated by the cache. @brief Free all resources allocated by the cache.
@details This function frees all resources allocated by the cache. You
have to call init_cache() before using the cache again. This function requires This function frees all resources allocated by the cache. You
the structure_guard_mutex to be locked. have to call init_cache() before using the cache again. This function
requires the structure_guard_mutex to be locked.
*/ */
void Query_cache::free_cache() void Query_cache::free_cache()
@ -1808,24 +1925,17 @@ void Query_cache::free_cache()
*****************************************************************************/ *****************************************************************************/
/* /**
flush_cache() - flush the cache. @brief Flush the cache.
SYNOPSIS
flush_cache()
DESCRIPTION
This function will flush cache contents. It assumes we have This function will flush cache contents. It assumes we have
'structure_guard_mutex' locked. The function sets the 'structure_guard_mutex' locked. The function sets the m_cache_status flag and
flush_in_progress flag and releases the lock, so other threads may releases the lock, so other threads may proceed skipping the cache as if it
proceed skipping the cache as if it is disabled. Concurrent is disabled. Concurrent flushes are performed in turn.
flushes are performed in turn. After flush_cache() call, the cache is flushed, all the freed memory is
accumulated in bin[0], and the 'structure_guard_mutex' is locked. However,
After flush_cache() call, the cache is flushed, all the freed since we could release the mutex during execution, the rest of the cache
memory is accumulated in bin[0], and the 'structure_guard_mutex' state could have been changed, and should not be relied on.
is locked. However, since we could release the mutex during
execution, the rest of the cache state could have been changed,
and should not be relied on.
*/ */
void Query_cache::flush_cache() void Query_cache::flush_cache()
@ -1837,15 +1947,15 @@ void Query_cache::flush_cache()
Query_cache::free_cache()) depends on the fact that after the Query_cache::free_cache()) depends on the fact that after the
flush the cache is empty. flush the cache is empty.
*/ */
while (flush_in_progress) while (is_flushing())
pthread_cond_wait(&COND_flush_finished, &structure_guard_mutex); pthread_cond_wait(&COND_cache_status_changed, &structure_guard_mutex);
/* /*
Setting 'flush_in_progress' will prevent other threads from using Setting 'FLUSH_IN_PROGRESS' will prevent other threads from using
the cache while we are in the middle of the flush, and we release the cache while we are in the middle of the flush, and we release
the lock so that other threads won't block. the lock so that other threads won't block.
*/ */
flush_in_progress= TRUE; m_cache_status= Query_cache::FLUSH_IN_PROGRESS;
STRUCT_UNLOCK(&structure_guard_mutex); STRUCT_UNLOCK(&structure_guard_mutex);
my_hash_reset(&queries); my_hash_reset(&queries);
@ -1856,8 +1966,8 @@ void Query_cache::flush_cache()
} }
STRUCT_LOCK(&structure_guard_mutex); STRUCT_LOCK(&structure_guard_mutex);
flush_in_progress= FALSE; m_cache_status= Query_cache::NO_FLUSH_IN_PROGRESS;
pthread_cond_signal(&COND_flush_finished); pthread_cond_signal(&COND_cache_status_changed);
} }
/* /*
@ -2013,8 +2123,7 @@ Query_cache_block *
Query_cache::write_block_data(ulong data_len, uchar* data, Query_cache::write_block_data(ulong data_len, uchar* data,
ulong header_len, ulong header_len,
Query_cache_block::block_type type, Query_cache_block::block_type type,
TABLE_COUNTER_TYPE ntab, TABLE_COUNTER_TYPE ntab)
my_bool under_guard)
{ {
ulong all_headers_len = (ALIGN_SIZE(sizeof(Query_cache_block)) + ulong all_headers_len = (ALIGN_SIZE(sizeof(Query_cache_block)) +
ALIGN_SIZE(ntab*sizeof(Query_cache_block_table)) + ALIGN_SIZE(ntab*sizeof(Query_cache_block_table)) +
@ -2025,8 +2134,7 @@ Query_cache::write_block_data(ulong data_len, uchar* data,
DBUG_PRINT("qcache", ("data: %ld, header: %ld, all header: %ld", DBUG_PRINT("qcache", ("data: %ld, header: %ld, all header: %ld",
data_len, header_len, all_headers_len)); data_len, header_len, all_headers_len));
Query_cache_block *block= allocate_block(max(align_len, Query_cache_block *block= allocate_block(max(align_len,
min_allocation_unit), min_allocation_unit),1, 0);
1, 0, under_guard);
if (block != 0) if (block != 0)
{ {
block->type = type; block->type = type;
@ -2243,8 +2351,7 @@ my_bool Query_cache::allocate_data_chain(Query_cache_block **result_block,
if (!(new_block= allocate_block(max(min_size, align_len), if (!(new_block= allocate_block(max(min_size, align_len),
min_result_data_size == 0, min_result_data_size == 0,
all_headers_len + min_result_data_size, all_headers_len + min_result_data_size)))
1)))
{ {
DBUG_PRINT("warning", ("Can't allocate block for results")); DBUG_PRINT("warning", ("Can't allocate block for results"));
DBUG_RETURN(FALSE); DBUG_RETURN(FALSE);
@ -2286,51 +2393,94 @@ my_bool Query_cache::allocate_data_chain(Query_cache_block **result_block,
Invalidate the first table in the table_list Invalidate the first table in the table_list
*/ */
void Query_cache::invalidate_table(TABLE_LIST *table_list) void Query_cache::invalidate_table(THD *thd, TABLE_LIST *table_list)
{ {
if (table_list->table != 0) if (table_list->table != 0)
invalidate_table(table_list->table); // Table is open invalidate_table(thd, table_list->table); // Table is open
else else
{ {
char key[MAX_DBKEY_LENGTH]; char key[MAX_DBKEY_LENGTH];
uint key_length; uint key_length;
Query_cache_block *table_block;
key_length=(uint) (strmov(strmov(key,table_list->db)+1, key_length=(uint) (strmov(strmov(key,table_list->db)+1,
table_list->table_name) -key)+ 1; table_list->table_name) -key)+ 1;
// We don't store temporary tables => no key_length+=4 ... // We don't store temporary tables => no key_length+=4 ...
if ((table_block = (Query_cache_block*) invalidate_table(thd, (uchar *)key, key_length);
hash_search(&tables,(uchar*) key,key_length)))
invalidate_table(table_block);
} }
} }
void Query_cache::invalidate_table(TABLE *table) void Query_cache::invalidate_table(THD *thd, TABLE *table)
{ {
invalidate_table((uchar*) table->s->table_cache_key.str, invalidate_table(thd, (uchar*) table->s->table_cache_key.str,
table->s->table_cache_key.length); table->s->table_cache_key.length);
} }
void Query_cache::invalidate_table(uchar * key, uint32 key_length) void Query_cache::invalidate_table(THD *thd, uchar * key, uint32 key_length)
{ {
Query_cache_block *table_block; bool interrupt;
if ((table_block = ((Query_cache_block*) STRUCT_LOCK(&structure_guard_mutex);
hash_search(&tables, key, key_length)))) wait_while_table_flush_is_in_progress(&interrupt);
invalidate_table(table_block); if (interrupt)
{
STRUCT_UNLOCK(&structure_guard_mutex);
return;
} }
void Query_cache::invalidate_table(Query_cache_block *table_block) /*
Setting 'TABLE_FLUSH_IN_PROGRESS' will temporarily disable the cache
so that structural changes to cache won't block the entire server.
However, threads requesting to change the query cache will still have
to wait for the flush to finish.
*/
m_cache_status= Query_cache::TABLE_FLUSH_IN_PROGRESS;
STRUCT_UNLOCK(&structure_guard_mutex);
Query_cache_block *table_block=
(Query_cache_block*)hash_search(&tables, key, key_length);
if (query_cache_size > 0 && table_block)
{ {
Query_cache_block_table *list_root= table_block->table(0); Query_cache_block_table *list_root= table_block->table(0);
invalidate_query_block_list(thd, list_root);
}
STRUCT_LOCK(&structure_guard_mutex);
m_cache_status= Query_cache::NO_FLUSH_IN_PROGRESS;
/*
net_real_write might be waiting on a change on the m_cache_status
variable.
*/
pthread_cond_signal(&COND_cache_status_changed);
STRUCT_UNLOCK(&structure_guard_mutex);
}
/**
@brief Invalidate a linked list of query cache blocks.
Each block tries to aquire a block level lock before
free_query is a called. This function will in turn affect
related table- and result-blocks.
@param[in,out] thd Thread context.
@param[in,out] list_root A pointer to a circular list of query blocks.
*/
void
Query_cache::invalidate_query_block_list(THD *thd,
Query_cache_block_table *list_root)
{
while (list_root->next != list_root) while (list_root->next != list_root)
{ {
Query_cache_block *query_block= list_root->next->block(); Query_cache_block *query_block= list_root->next->block();
BLOCK_LOCK_WR(query_block); BLOCK_LOCK_WR(query_block);
free_query(query_block); free_query(query_block);
DBUG_EXECUTE_IF("debug_cache_locks", sleep(10););
} }
} }
/* /*
Register given table list begining with given position in tables table of Register given table list begining with given position in tables table of
block block
@ -2463,7 +2613,6 @@ my_bool Query_cache::register_all_tables(Query_cache_block *block,
if (n) if (n)
{ {
DBUG_PRINT("qcache", ("failed at table %d", (int) n));
/* Unlink the tables we allocated above */ /* Unlink the tables we allocated above */
for (Query_cache_block_table *tmp = block->table(0) ; for (Query_cache_block_table *tmp = block->table(0) ;
tmp != block_table; tmp != block_table;
@ -2473,9 +2622,13 @@ my_bool Query_cache::register_all_tables(Query_cache_block *block,
return (n); return (n);
} }
/*
Insert used tablename in cache /**
Returns 0 on error @brief Insert used table name into the cache.
@return Error status
@retval FALSE On error
@retval TRUE On success
*/ */
my_bool my_bool
@ -2489,9 +2642,10 @@ Query_cache::insert_table(uint key_len, char *key,
DBUG_PRINT("qcache", ("insert table node 0x%lx, len %d", DBUG_PRINT("qcache", ("insert table node 0x%lx, len %d",
(ulong)node, key_len)); (ulong)node, key_len));
Query_cache_block *table_block = ((Query_cache_block *) THD *thd= current_thd;
hash_search(&tables, (uchar*) key,
key_len)); Query_cache_block *table_block=
(Query_cache_block *)hash_search(&tables, (uchar*) key, key_len);
if (table_block && if (table_block &&
table_block->table()->engine_data() != engine_data) table_block->table()->engine_data() != engine_data)
@ -2506,7 +2660,11 @@ Query_cache::insert_table(uint key_len, char *key,
as far as we delete all queries with this table, table block will be as far as we delete all queries with this table, table block will be
deleted, too deleted, too
*/ */
invalidate_table(table_block); {
Query_cache_block_table *list_root= table_block->table(0);
invalidate_query_block_list(thd, list_root);
}
table_block= 0; table_block= 0;
} }
@ -2516,8 +2674,7 @@ Query_cache::insert_table(uint key_len, char *key,
(ulong) key, (int) key_len)); (ulong) key, (int) key_len));
table_block= write_block_data(key_len, (uchar*) key, table_block= write_block_data(key_len, (uchar*) key,
ALIGN_SIZE(sizeof(Query_cache_table)), ALIGN_SIZE(sizeof(Query_cache_table)),
Query_cache_block::TABLE, Query_cache_block::TABLE, 1);
1, 1);
if (table_block == 0) if (table_block == 0)
{ {
DBUG_PRINT("qcache", ("Can't write table name to cache")); DBUG_PRINT("qcache", ("Can't write table name to cache"));
@ -2526,9 +2683,18 @@ Query_cache::insert_table(uint key_len, char *key,
Query_cache_table *header= table_block->table(); Query_cache_table *header= table_block->table();
double_linked_list_simple_include(table_block, double_linked_list_simple_include(table_block,
&tables_blocks); &tables_blocks);
/*
First node in the Query_cache_block_table-chain is the table-type
block. This block will only have one Query_cache_block_table (n=0).
*/
Query_cache_block_table *list_root= table_block->table(0); Query_cache_block_table *list_root= table_block->table(0);
list_root->n= 0; list_root->n= 0;
/*
The node list is circular in nature.
*/
list_root->next= list_root->prev= list_root; list_root->next= list_root->prev= list_root;
if (my_hash_insert(&tables, (const uchar *) table_block)) if (my_hash_insert(&tables, (const uchar *) table_block))
{ {
DBUG_PRINT("qcache", ("Can't insert table to hash")); DBUG_PRINT("qcache", ("Can't insert table to hash"));
@ -2542,14 +2708,31 @@ Query_cache::insert_table(uint key_len, char *key,
header->type(cache_type); header->type(cache_type);
header->callback(callback); header->callback(callback);
header->engine_data(engine_data); header->engine_data(engine_data);
/*
We insert this table without the assumption that it isn't refrenenced by
any queries.
*/
header->m_cached_query_count= 0;
} }
/*
Table is now in the cache; link the table_block-node associated
with the currently processed query into the chain of queries depending
on the cached table.
*/
Query_cache_block_table *list_root= table_block->table(0); Query_cache_block_table *list_root= table_block->table(0);
node->next= list_root->next; node->next= list_root->next;
list_root->next= node; list_root->next= node;
node->next->prev= node; node->next->prev= node;
node->prev= list_root; node->prev= list_root;
node->parent= table_block->table(); node->parent= table_block->table();
/*
Increase the counter to keep track on how long this chain
of queries is.
*/
Query_cache_table *table_block_data= table_block->table();
table_block_data->m_cached_query_count++;
DBUG_RETURN(1); DBUG_RETURN(1);
} }
@ -2560,9 +2743,21 @@ void Query_cache::unlink_table(Query_cache_block_table *node)
node->prev->next= node->next; node->prev->next= node->next;
node->next->prev= node->prev; node->next->prev= node->prev;
Query_cache_block_table *neighbour= node->next; Query_cache_block_table *neighbour= node->next;
Query_cache_table *table_block_data= node->parent;
table_block_data->m_cached_query_count--;
DBUG_ASSERT(table_block_data->m_cached_query_count >= 0);
if (neighbour->next == neighbour) if (neighbour->next == neighbour)
{ {
// list is empty (neighbor is root of list) DBUG_ASSERT(table_block_data->m_cached_query_count == 0);
/*
If neighbor is root of list, the list is empty.
The root of the list is always a table-type block
which contain exactly one Query_cache_block_table
node object, thus we can use the block() method
to calculate the Query_cache_block address.
*/
Query_cache_block *table_block= neighbour->block(); Query_cache_block *table_block= neighbour->block();
double_linked_list_exclude(table_block, double_linked_list_exclude(table_block,
&tables_blocks); &tables_blocks);
@ -2577,12 +2772,11 @@ void Query_cache::unlink_table(Query_cache_block_table *node)
*****************************************************************************/ *****************************************************************************/
Query_cache_block * Query_cache_block *
Query_cache::allocate_block(ulong len, my_bool not_less, ulong min, Query_cache::allocate_block(ulong len, my_bool not_less, ulong min)
my_bool under_guard)
{ {
DBUG_ENTER("Query_cache::allocate_block"); DBUG_ENTER("Query_cache::allocate_block");
DBUG_PRINT("qcache", ("len %lu, not less %d, min %lu, uder_guard %d", DBUG_PRINT("qcache", ("len %lu, not less %d, min %lu",
len, not_less,min,under_guard)); len, not_less,min));
if (len >= min(query_cache_size, query_cache_limit)) if (len >= min(query_cache_size, query_cache_limit))
{ {
@ -2591,17 +2785,6 @@ Query_cache::allocate_block(ulong len, my_bool not_less, ulong min,
DBUG_RETURN(0); // in any case we don't have such piece of memory DBUG_RETURN(0); // in any case we don't have such piece of memory
} }
if (!under_guard)
{
STRUCT_LOCK(&structure_guard_mutex);
if (unlikely(query_cache.query_cache_size == 0 || flush_in_progress))
{
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_RETURN(0);
}
}
/* Free old queries until we have enough memory to store this block */ /* Free old queries until we have enough memory to store this block */
Query_cache_block *block; Query_cache_block *block;
do do
@ -2616,8 +2799,6 @@ Query_cache::allocate_block(ulong len, my_bool not_less, ulong min,
split_block(block,ALIGN_SIZE(len)); split_block(block,ALIGN_SIZE(len));
} }
if (!under_guard)
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_RETURN(block); DBUG_RETURN(block);
} }
@ -2852,9 +3033,7 @@ uint Query_cache::find_bin(ulong size)
} }
uint bin = steps[left].idx - uint bin = steps[left].idx -
(uint)((size - steps[left].size)/steps[left].increment); (uint)((size - steps[left].size)/steps[left].increment);
#ifndef DBUG_OFF
bins_dump();
#endif
DBUG_PRINT("qcache", ("bin %u step %u, size %lu step size %lu", DBUG_PRINT("qcache", ("bin %u step %u, size %lu step size %lu",
bin, left, size, steps[left].size)); bin, left, size, steps[left].size));
DBUG_RETURN(bin); DBUG_RETURN(bin);
@ -3140,18 +3319,17 @@ my_bool Query_cache::ask_handler_allowance(THD *thd,
Packing Packing
*****************************************************************************/ *****************************************************************************/
/**
@brief Rearrange all memory blocks so that free memory joins at the
'bottom' of the allocated memory block containing all cache data.
@see Query_cache::pack(ulong join_limit, uint iteration_limit)
*/
void Query_cache::pack_cache() void Query_cache::pack_cache()
{ {
DBUG_ENTER("Query_cache::pack_cache"); DBUG_ENTER("Query_cache::pack_cache");
STRUCT_LOCK(&structure_guard_mutex);
if (unlikely(query_cache_size == 0 || flush_in_progress))
{
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN;
}
DBUG_EXECUTE("check_querycache",query_cache.check_integrity(1);); DBUG_EXECUTE("check_querycache",query_cache.check_integrity(1););
uchar *border = 0; uchar *border = 0;
@ -3185,7 +3363,6 @@ void Query_cache::pack_cache()
} }
DBUG_EXECUTE("check_querycache",query_cache.check_integrity(1);); DBUG_EXECUTE("check_querycache",query_cache.check_integrity(1););
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_VOID_RETURN; DBUG_VOID_RETURN;
} }
@ -3460,8 +3637,7 @@ my_bool Query_cache::join_results(ulong join_limit)
my_bool has_moving = 0; my_bool has_moving = 0;
DBUG_ENTER("Query_cache::join_results"); DBUG_ENTER("Query_cache::join_results");
STRUCT_LOCK(&structure_guard_mutex); if (queries_blocks != 0)
if (queries_blocks != 0 && !flush_in_progress)
{ {
DBUG_ASSERT(query_cache_size > 0); DBUG_ASSERT(query_cache_size > 0);
Query_cache_block *block = queries_blocks; Query_cache_block *block = queries_blocks;
@ -3524,7 +3700,6 @@ my_bool Query_cache::join_results(ulong join_limit)
block = block->next; block = block->next;
} while ( block != queries_blocks ); } while ( block != queries_blocks );
} }
STRUCT_UNLOCK(&structure_guard_mutex);
DBUG_RETURN(has_moving); DBUG_RETURN(has_moving);
} }
@ -3760,6 +3935,14 @@ void Query_cache::tables_dump()
} }
/**
@brief Checks integrity of the various linked lists
@return Error status code
@retval FALSE Query cache is operational.
@retval TRUE Query cache is broken.
*/
my_bool Query_cache::check_integrity(bool locked) my_bool Query_cache::check_integrity(bool locked)
{ {
my_bool result = 0; my_bool result = 0;
@ -3769,14 +3952,8 @@ my_bool Query_cache::check_integrity(bool locked)
if (!locked) if (!locked)
STRUCT_LOCK(&structure_guard_mutex); STRUCT_LOCK(&structure_guard_mutex);
if (unlikely(query_cache_size == 0 || flush_in_progress)) while (is_flushing())
{ pthread_cond_wait(&COND_cache_status_changed,&structure_guard_mutex);
if (!locked)
STRUCT_UNLOCK(&query_cache.structure_guard_mutex);
DBUG_PRINT("qcache", ("Query Cache not initialized"));
DBUG_RETURN(0);
}
if (hash_check(&queries)) if (hash_check(&queries))
{ {

72
sql/sql_cache.h
View File

@ -65,17 +65,44 @@ struct Query_cache_query;
struct Query_cache_result; struct Query_cache_result;
class Query_cache; class Query_cache;
/**
@brief This class represents a node in the linked chain of queries
belonging to one table.
@note The root of this linked list is not a query-type block, but the table-
type block which all queries has in common.
*/
struct Query_cache_block_table struct Query_cache_block_table
{ {
Query_cache_block_table() {} /* Remove gcc warning */ Query_cache_block_table() {} /* Remove gcc warning */
TABLE_COUNTER_TYPE n; // numbr in table (from 0)
/**
This node holds a position in a static table list belonging
to the associated query (base 0).
*/
TABLE_COUNTER_TYPE n;
/**
Pointers to the next and previous node, linking all queries with
a common table.
*/
Query_cache_block_table *next, *prev; Query_cache_block_table *next, *prev;
/**
A pointer to the table-type block which all
linked queries has in common.
*/
Query_cache_table *parent; Query_cache_table *parent;
/**
A method to calculate the address of the query cache block
owning this node. The purpose of this calculation is to
make it easier to move the query cache block without having
to modify all the pointer addresses.
*/
inline Query_cache_block *block(); inline Query_cache_block *block();
}; };
struct Query_cache_block struct Query_cache_block
{ {
Query_cache_block() {} /* Remove gcc warning */ Query_cache_block() {} /* Remove gcc warning */
@ -151,6 +178,11 @@ struct Query_cache_table
/* data need by some engines */ /* data need by some engines */
ulonglong engine_data_buff; ulonglong engine_data_buff;
/**
The number of queries depending of this table.
*/
int32 m_cached_query_count;
inline char *db() { return (char *) data(); } inline char *db() { return (char *) data(); }
inline char *table() { return tbl; } inline char *table() { return tbl; }
inline void table(char *table_arg) { tbl= table_arg; } inline void table(char *table_arg) { tbl= table_arg; }
@ -237,9 +269,14 @@ public:
ulong free_memory, queries_in_cache, hits, inserts, refused, ulong free_memory, queries_in_cache, hits, inserts, refused,
free_memory_blocks, total_blocks, lowmem_prunes; free_memory_blocks, total_blocks, lowmem_prunes;
private: private:
pthread_cond_t COND_flush_finished; pthread_cond_t COND_cache_status_changed;
bool flush_in_progress;
enum Cache_status { NO_FLUSH_IN_PROGRESS, FLUSH_IN_PROGRESS,
TABLE_FLUSH_IN_PROGRESS };
Cache_status m_cache_status;
void free_query_internal(Query_cache_block *point); void free_query_internal(Query_cache_block *point);
@ -253,7 +290,7 @@ protected:
2. query block (for operation inside query (query block/results)) 2. query block (for operation inside query (query block/results))
Thread doing cache flush releases the mutex once it sets Thread doing cache flush releases the mutex once it sets
flush_in_progress flag, so other threads may bypass the cache as m_cache_status flag, so other threads may bypass the cache as
if it is disabled, not waiting for reset to finish. The exception if it is disabled, not waiting for reset to finish. The exception
is other threads that were going to do cache flush---they'll wait is other threads that were going to do cache flush---they'll wait
till the end of a flush operation. till the end of a flush operation.
@ -270,6 +307,7 @@ protected:
/* options */ /* options */
ulong min_allocation_unit, min_result_data_size; ulong min_allocation_unit, min_result_data_size;
uint def_query_hash_size, def_table_hash_size; uint def_query_hash_size, def_table_hash_size;
uint mem_bin_num, mem_bin_steps; // See at init_cache & find_bin uint mem_bin_num, mem_bin_steps; // See at init_cache & find_bin
my_bool initialized; my_bool initialized;
@ -295,10 +333,13 @@ protected:
ulong data_len, ulong data_len,
Query_cache_block *query_block, Query_cache_block *query_block,
my_bool first_block); my_bool first_block);
void invalidate_table(TABLE_LIST *table); void invalidate_table(THD *thd, TABLE_LIST *table);
void invalidate_table(TABLE *table); void invalidate_table(THD *thd, TABLE *table);
void invalidate_table(uchar *key, uint32 key_length); void invalidate_table(THD *thd, uchar *key, uint32 key_length);
void invalidate_table(Query_cache_block *table_block); void invalidate_table(THD *thd, Query_cache_block *table_block);
void invalidate_query_block_list(THD *thd,
Query_cache_block_table *list_root);
TABLE_COUNTER_TYPE TABLE_COUNTER_TYPE
register_tables_from_list(TABLE_LIST *tables_used, register_tables_from_list(TABLE_LIST *tables_used,
TABLE_COUNTER_TYPE counter, TABLE_COUNTER_TYPE counter,
@ -337,6 +378,8 @@ protected:
Query_cache_block *pprev); Query_cache_block *pprev);
my_bool join_results(ulong join_limit); my_bool join_results(ulong join_limit);
void wait_while_table_flush_is_in_progress(bool *interrupt);
/* /*
Following function control structure_guard_mutex Following function control structure_guard_mutex
by themself or don't need structure_guard_mutex by themself or don't need structure_guard_mutex
@ -347,8 +390,7 @@ protected:
Query_cache_block *write_block_data(ulong data_len, uchar* data, Query_cache_block *write_block_data(ulong data_len, uchar* data,
ulong header_len, ulong header_len,
Query_cache_block::block_type type, Query_cache_block::block_type type,
TABLE_COUNTER_TYPE ntab = 0, TABLE_COUNTER_TYPE ntab = 0);
my_bool under_guard=0);
my_bool append_result_data(Query_cache_block **result, my_bool append_result_data(Query_cache_block **result,
ulong data_len, uchar* data, ulong data_len, uchar* data,
Query_cache_block *parent); Query_cache_block *parent);
@ -360,8 +402,7 @@ protected:
inline ulong get_min_first_result_data_size(); inline ulong get_min_first_result_data_size();
inline ulong get_min_append_result_data_size(); inline ulong get_min_append_result_data_size();
Query_cache_block *allocate_block(ulong len, my_bool not_less, Query_cache_block *allocate_block(ulong len, my_bool not_less,
ulong min, ulong min);
my_bool under_guard=0);
/* /*
If query is cacheable return number tables in query If query is cacheable return number tables in query
(query without tables not cached) (query without tables not cached)
@ -424,6 +465,11 @@ protected:
friend void query_cache_end_of_result(THD *thd); friend void query_cache_end_of_result(THD *thd);
friend void query_cache_abort(NET *net); friend void query_cache_abort(NET *net);
bool is_flushing(void)
{
return (m_cache_status != Query_cache::NO_FLUSH_IN_PROGRESS);
}
/* /*
The following functions are only used when debugging The following functions are only used when debugging
We don't protect these with ifndef DBUG_OFF to not have to recompile We don't protect these with ifndef DBUG_OFF to not have to recompile

2
sql/sql_db.cc
View File

@ -1084,7 +1084,7 @@ static long mysql_rm_known_files(THD *thd, MY_DIR *dirp, const char *db,
} }
} }
if (thd->killed || if (thd->killed ||
(tot_list && mysql_rm_table_part2_with_lock(thd, tot_list, 1, 0, 1))) (tot_list && mysql_rm_table_part2(thd, tot_list, 1, 0, 1, 1, 1)))
goto err; goto err;
/* Remove RAID directories */ /* Remove RAID directories */

2
sql/sql_parse.cc
View File

@ -2445,7 +2445,7 @@ end_with_restore_list:
check_grant(thd, INSERT_ACL | CREATE_ACL, &new_list, 0, 1, 0))) check_grant(thd, INSERT_ACL | CREATE_ACL, &new_list, 0, 1, 0)))
goto error; goto error;
} }
query_cache_invalidate3(thd, first_table, 0);
if (end_active_trans(thd) || mysql_rename_tables(thd, first_table, 0)) if (end_active_trans(thd) || mysql_rename_tables(thd, first_table, 0))
goto error; goto error;
break; break;

14
sql/sql_rename.cc
View File

@ -144,9 +144,13 @@ bool mysql_rename_tables(THD *thd, TABLE_LIST *table_list, bool silent)
} }
} }
VOID(pthread_mutex_lock(&LOCK_open)); pthread_mutex_lock(&LOCK_open);
if (lock_table_names(thd, table_list)) if (lock_table_names_exclusively(thd, table_list))
{
pthread_mutex_unlock(&LOCK_open);
goto err; goto err;
}
pthread_mutex_unlock(&LOCK_open);
error=0; error=0;
if ((ren_table=rename_tables(thd,table_list,0))) if ((ren_table=rename_tables(thd,table_list,0)))
@ -183,10 +187,14 @@ bool mysql_rename_tables(THD *thd, TABLE_LIST *table_list, bool silent)
send_ok(thd); send_ok(thd);
} }
if (!error)
query_cache_invalidate3(thd, table_list, 0);
pthread_mutex_lock(&LOCK_open);
unlock_table_names(thd, table_list, (TABLE_LIST*) 0); unlock_table_names(thd, table_list, (TABLE_LIST*) 0);
pthread_mutex_unlock(&LOCK_open);
err: err:
pthread_mutex_unlock(&LOCK_open);
/* enable logging back if needed */ /* enable logging back if needed */
if (disable_logs) if (disable_logs)
{ {

80
sql/sql_table.cc
View File

@ -1411,14 +1411,7 @@ bool mysql_rm_table(THD *thd,TABLE_LIST *tables, my_bool if_exists,
LOCK_open during wait_if_global_read_lock(), other threads could not LOCK_open during wait_if_global_read_lock(), other threads could not
close their tables. This would make a pretty deadlock. close their tables. This would make a pretty deadlock.
*/ */
thd->mysys_var->current_mutex= &LOCK_open; error= mysql_rm_table_part2(thd, tables, if_exists, drop_temporary, 0, 0, 1);
thd->mysys_var->current_cond= &COND_refresh;
VOID(pthread_mutex_lock(&LOCK_open));
error= mysql_rm_table_part2(thd, tables, if_exists, drop_temporary, 0, 0);
pthread_mutex_unlock(&LOCK_open);
pthread_mutex_lock(&thd->mysys_var->mutex); pthread_mutex_lock(&thd->mysys_var->mutex);
thd->mysys_var->current_mutex= 0; thd->mysys_var->current_mutex= 0;
thd->mysys_var->current_cond= 0; thd->mysys_var->current_cond= 0;
@ -1433,49 +1426,6 @@ bool mysql_rm_table(THD *thd,TABLE_LIST *tables, my_bool if_exists,
DBUG_RETURN(FALSE); DBUG_RETURN(FALSE);
} }
/*
delete (drop) tables.
SYNOPSIS
mysql_rm_table_part2_with_lock()
thd Thread handle
tables List of tables to delete
if_exists If 1, don't give error if one table doesn't exists
dont_log_query Don't write query to log files. This will also not
generate warnings if the handler files doesn't exists
NOTES
Works like documented in mysql_rm_table(), but don't check
global_read_lock and don't send_ok packet to server.
RETURN
0 ok
1 error
*/
int mysql_rm_table_part2_with_lock(THD *thd,
TABLE_LIST *tables, bool if_exists,
bool drop_temporary, bool dont_log_query)
{
int error;
thd->mysys_var->current_mutex= &LOCK_open;
thd->mysys_var->current_cond= &COND_refresh;
VOID(pthread_mutex_lock(&LOCK_open));
error= mysql_rm_table_part2(thd, tables, if_exists, drop_temporary, 1,
dont_log_query);
pthread_mutex_unlock(&LOCK_open);
pthread_mutex_lock(&thd->mysys_var->mutex);
thd->mysys_var->current_mutex= 0;
thd->mysys_var->current_cond= 0;
pthread_mutex_unlock(&thd->mysys_var->mutex);
return error;
}
/* /*
Execute the drop of a normal or temporary table Execute the drop of a normal or temporary table
@ -1508,7 +1458,7 @@ int mysql_rm_table_part2_with_lock(THD *thd,
int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists, int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
bool drop_temporary, bool drop_view, bool drop_temporary, bool drop_view,
bool dont_log_query) bool dont_log_query, bool need_lock_open)
{ {
TABLE_LIST *table; TABLE_LIST *table;
char path[FN_REFLEN], *alias; char path[FN_REFLEN], *alias;
@ -1520,9 +1470,11 @@ int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
String built_query; String built_query;
DBUG_ENTER("mysql_rm_table_part2"); DBUG_ENTER("mysql_rm_table_part2");
if (need_lock_open)
pthread_mutex_lock(&LOCK_open);
LINT_INIT(alias); LINT_INIT(alias);
LINT_INIT(path_length); LINT_INIT(path_length);
safe_mutex_assert_owner(&LOCK_open);
if (thd->current_stmt_binlog_row_based && !dont_log_query) if (thd->current_stmt_binlog_row_based && !dont_log_query)
{ {
@ -1555,8 +1507,15 @@ int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
} }
} }
if (!drop_temporary && lock_table_names(thd, tables)) if (!drop_temporary && lock_table_names_exclusively(thd, tables))
{
if (need_lock_open)
pthread_mutex_unlock(&LOCK_open);
DBUG_RETURN(1); DBUG_RETURN(1);
}
if (need_lock_open)
pthread_mutex_unlock(&LOCK_open);
/* Don't give warnings for not found errors, as we already generate notes */ /* Don't give warnings for not found errors, as we already generate notes */
thd->no_warnings_for_error= 1; thd->no_warnings_for_error= 1;
@ -1567,7 +1526,7 @@ int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
handlerton *table_type; handlerton *table_type;
enum legacy_db_type frm_db_type; enum legacy_db_type frm_db_type;
mysql_ha_flush(thd, table, MYSQL_HA_CLOSE_FINAL, TRUE); mysql_ha_flush(thd, table, MYSQL_HA_CLOSE_FINAL, !need_lock_open);
if (!close_temporary_table(thd, table)) if (!close_temporary_table(thd, table))
{ {
tmp_table_deleted=1; tmp_table_deleted=1;
@ -1604,6 +1563,8 @@ int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
{ {
TABLE *locked_table; TABLE *locked_table;
abort_locked_tables(thd, db, table->table_name); abort_locked_tables(thd, db, table->table_name);
if (need_lock_open)
pthread_mutex_lock(&LOCK_open);
remove_table_from_cache(thd, db, table->table_name, remove_table_from_cache(thd, db, table->table_name,
RTFC_WAIT_OTHER_THREAD_FLAG | RTFC_WAIT_OTHER_THREAD_FLAG |
RTFC_CHECK_KILLED_FLAG); RTFC_CHECK_KILLED_FLAG);
@ -1614,6 +1575,9 @@ int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
if ((locked_table= drop_locked_tables(thd, db, table->table_name))) if ((locked_table= drop_locked_tables(thd, db, table->table_name)))
table->table= locked_table; table->table= locked_table;
if (need_lock_open)
pthread_mutex_unlock(&LOCK_open);
if (thd->killed) if (thd->killed)
{ {
thd->no_warnings_for_error= 0; thd->no_warnings_for_error= 0;
@ -1739,9 +1703,11 @@ int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
*/ */
} }
} }
if (need_lock_open)
if (!drop_temporary) pthread_mutex_lock(&LOCK_open);
unlock_table_names(thd, tables, (TABLE_LIST*) 0); unlock_table_names(thd, tables, (TABLE_LIST*) 0);
if (need_lock_open)
pthread_mutex_unlock(&LOCK_open);
thd->no_warnings_for_error= 0; thd->no_warnings_for_error= 0;
DBUG_RETURN(error); DBUG_RETURN(error);
} }

38
sql/sql_trigger.cc
View File

@ -1268,8 +1268,6 @@ bool Table_triggers_list::drop_all_triggers(THD *thd, char *db, char *name)
bzero(&table, sizeof(table)); bzero(&table, sizeof(table));
init_alloc_root(&table.mem_root, 8192, 0); init_alloc_root(&table.mem_root, 8192, 0);
safe_mutex_assert_owner(&LOCK_open);
if (Table_triggers_list::check_n_load(thd, db, name, &table, 1)) if (Table_triggers_list::check_n_load(thd, db, name, &table, 1))
{ {
result= 1; result= 1;
@ -1431,26 +1429,24 @@ Table_triggers_list::change_table_name_in_trignames(const char *db_name,
} }
/* /**
Update .TRG and .TRN files after renaming triggers' subject table. @brief Update .TRG and .TRN files after renaming triggers' subject table.
SYNOPSIS @param[in,out] thd Thread context
change_table_name() @param[in] db Old database of subject table
thd Thread context @param[in] old_table Old name of subject table
db Old database of subject table @param[in] new_db New database for subject table
old_table Old name of subject table @param[in] new_table New name of subject table
new_db New database for subject table
new_table New name of subject table
NOTE @note
This method tries to leave trigger related files in consistent state, This method tries to leave trigger related files in consistent state,
i.e. it either will complete successfully, or will fail leaving files i.e. it either will complete successfully, or will fail leaving files
in their initial state. in their initial state.
Also this method assumes that subject table is not renamed to itself. Also this method assumes that subject table is not renamed to itself.
This method needs to be called under an exclusive table name lock.
RETURN VALUE @retval FALSE Success
FALSE Success @retval TRUE Error
TRUE Error
*/ */
bool Table_triggers_list::change_table_name(THD *thd, const char *db, bool Table_triggers_list::change_table_name(THD *thd, const char *db,
@ -1466,7 +1462,19 @@ bool Table_triggers_list::change_table_name(THD *thd, const char *db,
bzero(&table, sizeof(table)); bzero(&table, sizeof(table));
init_alloc_root(&table.mem_root, 8192, 0); init_alloc_root(&table.mem_root, 8192, 0);
uchar key[MAX_DBKEY_LENGTH];
uint key_length= (uint) (strmov(strmov((char*)&key[0], db)+1,
old_table)-(char*)&key[0])+1;
/*
This method interfaces the mysql server code protected by
either LOCK_open mutex or with an exclusive table name lock.
In the future, only an exclusive table name lock will be enough.
*/
#ifndef DBUG_OFF
if (!is_table_name_exclusively_locked_by_this_thread(thd, key, key_length))
safe_mutex_assert_owner(&LOCK_open); safe_mutex_assert_owner(&LOCK_open);
#endif
DBUG_ASSERT(my_strcasecmp(table_alias_charset, db, new_db) || DBUG_ASSERT(my_strcasecmp(table_alias_charset, db, new_db) ||
my_strcasecmp(table_alias_charset, old_table, new_table)); my_strcasecmp(table_alias_charset, old_table, new_table));