SLOW_QUERY
The SLOW_QUERY table provides the slow query information of the current node, which is the parsing result of the TiDB slow log file. The column names in the table are corresponding to the field names in the slow log.
For how to use this table to identify problematic statements and improve query performance, see Slow Query Log Document.
USE INFORMATION_SCHEMA;
DESC SLOW_QUERY;
The output is as follows:
+-------------------------------+---------------------+------+------+---------+-------+
| Field | Type | Null | Key | Default | Extra |
+-------------------------------+---------------------+------+------+---------+-------+
| Time | timestamp(6) | NO | PRI | NULL | |
| Txn_start_ts | bigint(20) unsigned | YES | | NULL | |
| User | varchar(64) | YES | | NULL | |
| Host | varchar(64) | YES | | NULL | |
| Conn_ID | bigint(20) unsigned | YES | | NULL | |
| Exec_retry_count | bigint(20) unsigned | YES | | NULL | |
| Exec_retry_time | double | YES | | NULL | |
| Query_time | double | YES | | NULL | |
| Parse_time | double | YES | | NULL | |
| Compile_time | double | YES | | NULL | |
| Rewrite_time | double | YES | | NULL | |
| Preproc_subqueries | bigint(20) unsigned | YES | | NULL | |
| Preproc_subqueries_time | double | YES | | NULL | |
| Optimize_time | double | YES | | NULL | |
| Wait_TS | double | YES | | NULL | |
| Prewrite_time | double | YES | | NULL | |
| Wait_prewrite_binlog_time | double | YES | | NULL | |
| Commit_time | double | YES | | NULL | |
| Get_commit_ts_time | double | YES | | NULL | |
| Commit_backoff_time | double | YES | | NULL | |
| Backoff_types | varchar(64) | YES | | NULL | |
| Resolve_lock_time | double | YES | | NULL | |
| Local_latch_wait_time | double | YES | | NULL | |
| Write_keys | bigint(22) | YES | | NULL | |
| Write_size | bigint(22) | YES | | NULL | |
| Prewrite_region | bigint(22) | YES | | NULL | |
| Txn_retry | bigint(22) | YES | | NULL | |
| Cop_time | double | YES | | NULL | |
| Process_time | double | YES | | NULL | |
| Wait_time | double | YES | | NULL | |
| Backoff_time | double | YES | | NULL | |
| LockKeys_time | double | YES | | NULL | |
| Request_count | bigint(20) unsigned | YES | | NULL | |
| Total_keys | bigint(20) unsigned | YES | | NULL | |
| Process_keys | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_delete_skipped_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_key_skipped_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_block_cache_hit_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_block_read_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_block_read_byte | bigint(20) unsigned | YES | | NULL | |
| DB | varchar(64) | YES | | NULL | |
| Index_names | varchar(100) | YES | | NULL | |
| Is_internal | tinyint(1) | YES | | NULL | |
| Digest | varchar(64) | YES | | NULL | |
| Stats | varchar(512) | YES | | NULL | |
| Cop_proc_avg | double | YES | | NULL | |
| Cop_proc_p90 | double | YES | | NULL | |
| Cop_proc_max | double | YES | | NULL | |
| Cop_proc_addr | varchar(64) | YES | | NULL | |
| Cop_wait_avg | double | YES | | NULL | |
| Cop_wait_p90 | double | YES | | NULL | |
| Cop_wait_max | double | YES | | NULL | |
| Cop_wait_addr | varchar(64) | YES | | NULL | |
| Mem_max | bigint(20) | YES | | NULL | |
| Disk_max | bigint(20) | YES | | NULL | |
| KV_total | double | YES | | NULL | |
| PD_total | double | YES | | NULL | |
| Backoff_total | double | YES | | NULL | |
| Write_sql_response_total | double | YES | | NULL | |
| Result_rows | bigint(22) | YES | | NULL | |
| Backoff_Detail | varchar(4096) | YES | | NULL | |
| Prepared | tinyint(1) | YES | | NULL | |
| Succ | tinyint(1) | YES | | NULL | |
| IsExplicitTxn | tinyint(1) | YES | | NULL | |
| IsWriteCacheTable | tinyint(1) | YES | | NULL | |
| Plan_from_cache | tinyint(1) | YES | | NULL | |
| Plan_from_binding | tinyint(1) | YES | | NULL | |
| Has_more_results | tinyint(1) | YES | | NULL | |
| Plan | longtext | YES | | NULL | |
| Plan_digest | varchar(128) | YES | | NULL | |
| Binary_plan | longtext | YES | | NULL | |
| Prev_stmt | longtext | YES | | NULL | |
| Query | longtext | YES | | NULL | |
+-------------------------------+---------------------+------+------+---------+-------+
73 rows in set (0.000 sec)
CLUSTER_SLOW_QUERY table
The CLUSTER_SLOW_QUERY table provides the slow query information of all nodes in the cluster, which is the parsing result of the TiDB slow log files. You can use the CLUSTER_SLOW_QUERY table the way you do with SLOW_QUERY. The table schema of the CLUSTER_SLOW_QUERY table differs from that of the SLOW_QUERY table in that an INSTANCE column is added to CLUSTER_SLOW_QUERY. The INSTANCE column represents the TiDB node address of the row information on the slow query.
For how to use this table to identify problematic statements and improve query performance, see Slow Query Log Document.
DESC CLUSTER_SLOW_QUERY;
The output is as follows:
+-------------------------------+---------------------+------+------+---------+-------+
| Field | Type | Null | Key | Default | Extra |
+-------------------------------+---------------------+------+------+---------+-------+
| INSTANCE | varchar(64) | YES | | NULL | |
| Time | timestamp(6) | NO | PRI | NULL | |
| Txn_start_ts | bigint(20) unsigned | YES | | NULL | |
| User | varchar(64) | YES | | NULL | |
| Host | varchar(64) | YES | | NULL | |
| Conn_ID | bigint(20) unsigned | YES | | NULL | |
| Exec_retry_count | bigint(20) unsigned | YES | | NULL | |
| Exec_retry_time | double | YES | | NULL | |
| Query_time | double | YES | | NULL | |
| Parse_time | double | YES | | NULL | |
| Compile_time | double | YES | | NULL | |
| Rewrite_time | double | YES | | NULL | |
| Preproc_subqueries | bigint(20) unsigned | YES | | NULL | |
| Preproc_subqueries_time | double | YES | | NULL | |
| Optimize_time | double | YES | | NULL | |
| Wait_TS | double | YES | | NULL | |
| Prewrite_time | double | YES | | NULL | |
| Wait_prewrite_binlog_time | double | YES | | NULL | |
| Commit_time | double | YES | | NULL | |
| Get_commit_ts_time | double | YES | | NULL | |
| Commit_backoff_time | double | YES | | NULL | |
| Backoff_types | varchar(64) | YES | | NULL | |
| Resolve_lock_time | double | YES | | NULL | |
| Local_latch_wait_time | double | YES | | NULL | |
| Write_keys | bigint(22) | YES | | NULL | |
| Write_size | bigint(22) | YES | | NULL | |
| Prewrite_region | bigint(22) | YES | | NULL | |
| Txn_retry | bigint(22) | YES | | NULL | |
| Cop_time | double | YES | | NULL | |
| Process_time | double | YES | | NULL | |
| Wait_time | double | YES | | NULL | |
| Backoff_time | double | YES | | NULL | |
| LockKeys_time | double | YES | | NULL | |
| Request_count | bigint(20) unsigned | YES | | NULL | |
| Total_keys | bigint(20) unsigned | YES | | NULL | |
| Process_keys | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_delete_skipped_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_key_skipped_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_block_cache_hit_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_block_read_count | bigint(20) unsigned | YES | | NULL | |
| Rocksdb_block_read_byte | bigint(20) unsigned | YES | | NULL | |
| DB | varchar(64) | YES | | NULL | |
| Index_names | varchar(100) | YES | | NULL | |
| Is_internal | tinyint(1) | YES | | NULL | |
| Digest | varchar(64) | YES | | NULL | |
| Stats | varchar(512) | YES | | NULL | |
| Cop_proc_avg | double | YES | | NULL | |
| Cop_proc_p90 | double | YES | | NULL | |
| Cop_proc_max | double | YES | | NULL | |
| Cop_proc_addr | varchar(64) | YES | | NULL | |
| Cop_wait_avg | double | YES | | NULL | |
| Cop_wait_p90 | double | YES | | NULL | |
| Cop_wait_max | double | YES | | NULL | |
| Cop_wait_addr | varchar(64) | YES | | NULL | |
| Mem_max | bigint(20) | YES | | NULL | |
| Disk_max | bigint(20) | YES | | NULL | |
| KV_total | double | YES | | NULL | |
| PD_total | double | YES | | NULL | |
| Backoff_total | double | YES | | NULL | |
| Write_sql_response_total | double | YES | | NULL | |
| Result_rows | bigint(22) | YES | | NULL | |
| Backoff_Detail | varchar(4096) | YES | | NULL | |
| Prepared | tinyint(1) | YES | | NULL | |
| Succ | tinyint(1) | YES | | NULL | |
| IsExplicitTxn | tinyint(1) | YES | | NULL | |
| IsWriteCacheTable | tinyint(1) | YES | | NULL | |
| Plan_from_cache | tinyint(1) | YES | | NULL | |
| Plan_from_binding | tinyint(1) | YES | | NULL | |
| Has_more_results | tinyint(1) | YES | | NULL | |
| Plan | longtext | YES | | NULL | |
| Plan_digest | varchar(128) | YES | | NULL | |
| Binary_plan | longtext | YES | | NULL | |
| Prev_stmt | longtext | YES | | NULL | |
| Query | longtext | YES | | NULL | |
+-------------------------------+---------------------+------+------+---------+-------+
74 rows in set (0.000 sec)
When the cluster system table is queried, TiDB does not obtain data from all nodes, but pushes down the related calculation to other nodes. The execution plan is as follows:
DESC SELECT COUNT(*) FROM CLUSTER_SLOW_QUERY WHERE user = 'u1';
The output is as follows:
+----------------------------+----------+-----------+--------------------------+------------------------------------------------------+
| id | estRows | task | access object | operator info |
+----------------------------+----------+-----------+--------------------------+------------------------------------------------------+
| StreamAgg_7 | 1.00 | root | | funcs:count(1)->Column#75 |
| └─TableReader_13 | 10.00 | root | | data:Selection_12 |
| └─Selection_12 | 10.00 | cop[tidb] | | eq(INFORMATION_SCHEMA.cluster_slow_query.user, "u1") |
| └─TableFullScan_11 | 10000.00 | cop[tidb] | table:CLUSTER_SLOW_QUERY | keep order:false, stats:pseudo |
+----------------------------+----------+-----------+--------------------------+------------------------------------------------------+
4 rows in set (0.00 sec)
In the preceding execution plan, the user = u1 condition is pushed down to other (cop) TiDB nodes, and the aggregate operator is also pushed down (the StreamAgg operator in the graph).
Currently, because statistics of the system tables are not collected, sometimes some aggregation operators cannot be pushed down, which results in slow execution. In this case, you can manually specify the SQL HINT to push down the aggregation operators. For example:
SELECT /*+ AGG_TO_COP() */ COUNT(*) FROM CLUSTER_SLOW_QUERY GROUP BY user;