Re: New SQL counter statistics view (pg_stat_sql)

On 2016/10/12 12:21, Haribabu Kommi wrote:

I tried changing the pg_stat_sql into row mode and ran the regress suite to 

add different type of SQL commands to the view and ran the pgbench test

on my laptop to find out any performance impact with this patch.

                                HEAD      PATCH

pgbench - select      828          816

Here I attached the pg_stat_sql patch to keep track of all SQL commands

based on the commandTag and their counts. I attached the result of this

view that is run on the database after "make installcheck" just for reference.

Some comments:
I think we can use pgstat_* instead of pgstat* for code consistency.
 
+static HTAB *pgStatBackendSql = NULL;
How about *pgstat_backend_sql

+HTAB       *pgStatSql = NULL;
How about *pgstat_sql

+static void pgstat_recv_sqlstmt(PgStat_MsgSqlstmt * msg, int len);
How about PgStat_MsgSqlstmt *msg instead of PgStat_MsgSqlstmt * msg

+typedef struct PgStatSqlstmtEntry
How about PgStat_SqlstmtEntry

+typedef struct PgStatMsgSqlstmt
How about PgStat_MsgSqlstmt

I have observed below behavior.
I have SET track_sql to ON and then execute the SELECT command and it return 0 rows.
It start counting from the second command.
postgres=# SET track_sql TO ON;
SET
postgres=# SELECT * FROM pg_stat_sql where tag='SELECT';
 tag | count
-----+-------
(0 rows)

postgres=# SELECT * FROM pg_stat_sql where tag='SELECT';
  tag   | count
--------+-------
 SELECT |     1
(1 row)
Is this a correct behavior?