STATEMENT
Obtains information about executed statements (normalized SQL statements) on the current node. You can view all statistics about normalized SQL statements received by the CN, whereas you can view only the statistics about normalized SQL statements executed on the current DN.
The unique_sql_id generated by different savepoint_name values is different. When a large number of savepoint_name is used, the number of unique_sql_id values generated in the system increases rapidly. If the number of unique_sql_id values is greater than the number of instr_unique_sql_count values, the newly generated unique_sql_id information is not counted.
Name |
Type |
Description |
---|---|---|
node_name |
name |
Node name |
node_id |
integer |
Node ID (node_id in pgxc_node) |
user_name |
name |
Username |
user_id |
oid |
OID of the user |
unique_sql_id |
bigint |
ID of the normalized SQL statement |
query |
text |
Normalized SQL statement Note: The length is controlled by track_activity_query_size. |
n_calls |
bigint |
Number of calls |
min_elapse_time |
bigint |
Minimum execution time of the SQL statement in the kernel (unit: μs) |
max_elapse_time |
bigint |
Maximum execution time of the SQL statement in the kernel (unit: μs) |
total_elapse_time |
bigint |
Total execution time of the SQL statement in the kernel (unit: μs) |
n_returned_rows |
bigint |
Number of rows in the result set returned by the SELECT statement |
n_tuples_fetched |
bigint |
Number of rows randomly scanned |
n_tuples_returned |
bigint |
Number of rows sequentially scanned |
n_tuples_inserted |
bigint |
Number of rows inserted |
n_tuples_updated |
bigint |
Number of rows updated |
n_tuples_deleted |
bigint |
Number of rows deleted |
n_blocks_fetched |
bigint |
Number of buffer block access times |
n_blocks_hit |
bigint |
Number of buffer block hits |
n_soft_parse |
bigint |
Number of soft parsing times. The value of n_soft_parse plus the value of n_hard_parse may be greater than the value of n_calls because the number of subqueries are not counted in the value of n_calls. |
n_hard_parse |
bigint |
Number of hard parsing times. The value of n_soft_parse plus the value of n_hard_parse may be greater than the value of n_calls because the number of subqueries are not counted in the value of n_calls. |
db_time |
bigint |
Valid DB time, which is accumulated if multiple threads are involved (unit: μs) |
cpu_time |
bigint |
CPU time (unit: μs) |
execution_time |
bigint |
Execution time in the executor (unit: μs) |
parse_time |
bigint |
SQL parsing time (unit: μs) |
plan_time |
bigint |
SQL plan generation time (unit: μs) |
rewrite_time |
bigint |
SQL rewriting time (unit: μs) |
pl_execution_time |
bigint |
Execution time of PL/pgSQL (unit: μs) |
pl_compilation_time |
bigint |
Compilation time of PL/pgSQL (unit: μs) |
data_io_time |
bigint |
I/O time (unit: μs) |
net_send_info |
text |
Network status of messages sent through a physical connection, including the time (unit: μs), number of calls, and throughput (unit: byte). In a distributed database, CNs communicate with each other, CNs communicate with customer service ends, and CNs communicate with DNs through physical connections. This column can be used to analyze the network overhead of SQL statements in a distributed system. Example: {"time":xxx, "n_calls":xxx, "size":xxx}. |
net_recv_info |
text |
Network status of messages received through a physical connection, including the time (unit: μs), number of calls, and throughput (unit: byte). In a distributed database, CNs communicate with each other, CNs communicate with customer service ends, and CNs communicate with DNs through physical connections. This column can be used to analyze the network overhead of SQL statements in a distributed system. Example: {"time":xxx, "n_calls":xxx, "size":xxx}. |
net_stream_send_info |
text |
Network status of messages sent through a logical connection, including the time (unit: μs), number of calls, and throughput (unit: byte). In a distributed database, DNs of different shards communicate with each other through logical connections. This column can be used to analyze the network overhead of SQL statements in a distributed system. Example: {"time":xxx, "n_calls":xxx, "size":xxx}. |
net_stream_recv_info |
text |
Network status of messages received through a logical connection, including the time (unit: μs), number of calls, and throughput (unit: byte). In a distributed database, DNs of different shards communicate with each other through logical connections. This column can be used to analyze the network overhead of SQL statements in a distributed system. Example: {"time":xxx, "n_calls":xxx, "size":xxx}. |
last_updated |
timestamp with time zone |
Last time when the statement was updated |
sort_count |
bigint |
Sorting count |
sort_time |
bigint |
Sorting duration (unit: μs) |
sort_mem_used |
bigint |
Size of work memory used during sorting (unit: KB) |
sort_spill_count |
bigint |
Count of file writing when data is flushed to disks during sorting |
sort_spill_size |
bigint |
File size used when data is flushed to disks during sorting (unit: KB) |
hash_count |
bigint |
Hashing count |
hash_time |
bigint |
Hashing duration (unit: μs) |
hash_mem_used |
bigint |
Size of work memory used during hashing (unit: KB) |
hash_spill_count |
bigint |
Count of file writing when data is flushed to disks during hashing |
hash_spill_size |
bigint |
File size used when data is flushed to disks during hashing (unit: KB) |
parent_unique_sql_id |
bigint |
Unique ID of the parent SQL statement. The value is 0 for a non-stored procedure substatement. |
n_calls indicates the actual number of calling times. For the FETCH statement in the stored procedure, the actual number of triggering times of the FETCH statement is the increase times of n_calls of the statement actually executed by the cursor.
Feedback
Was this page helpful?
Provide feedbackThank you very much for your feedback. We will continue working to improve the documentation.See the reply and handling status in My Cloud VOC.
For any further questions, feel free to contact us through the chatbot.
Chatbot