Debugging query performance in postgres
Hello,
We have a situation in which we had a dml query within a procedure that was
running fine but suddenly the plan flipped and it started running longer.
It took us a good amount of time to identify the cause and fix it. So I
have below questions,
1)Do we have any data dictionary view or query available which gives us
information on what were the queryids those are executing from within a
procedure and how much time they are taking?
2)Also how to identify , if any specific queries has opted a different plan
today as compared to past executions and also response time increased
because of that?
Regards
Veem
On 9/4/25 09:57, veem v wrote:
Hello,
We have a situation in which we had a dml query within a procedure that
was running fine but suddenly the plan flipped and it started running
longer. It took us a good amount of time to identify the cause and fix
it. So I have below questions,
What was the cause?
Postgres version
Was there any major changes in the database e.g schema changes or
quantity of data?
Is autovacuum running?
1)Do we have any data dictionary view or query available which gives us
information on what were the queryids those are executing from within a
procedure and how much time they are taking?
2)Also how to identify , if any specific queries has opted a different
plan today as compared to past executions and also response time
increased because of that?
Take a look at:
https://www.postgresql.org/docs/current/pgstatstatements.html
Regards
Veem
--
Adrian Klaver
adrian.klaver@aklaver.com
On Thu, 4 Sept 2025 at 23:19, Adrian Klaver <adrian.klaver@aklaver.com>
wrote:
On 9/4/25 09:57, veem v wrote:
Hello,
We have a situation in which we had a dml query within a procedure that
was running fine but suddenly the plan flipped and it started running
longer. It took us a good amount of time to identify the cause and fix
it. So I have below questions,What was the cause?
Postgres version
Was there any major changes in the database e.g schema changes or
quantity of data?Is autovacuum running?
1)Do we have any data dictionary view or query available which gives us
information on what were the queryids those are executing from within a
procedure and how much time they are taking?
2)Also how to identify , if any specific queries has opted a different
plan today as compared to past executions and also response time
increased because of that?Take a look at:
https://www.postgresql.org/docs/current/pgstatstatements.html
We didn't get the exact root cause why the plan flipped but we have to add
additional filters to get rid of the performance issue.
The version is 16.
Yes, autovacuum running.
We don't have any major changes. But if there was a significant change in
quantity of data , Yet to check that.
The pg_stats_statement does show the execution stats of the individual
query execution , but doesnt show any historical plan deviation. Also the
individual procedure and sql statement are logged into this view but i was
wondering if there is an easy way to relate the sql text of the procedure
with the underlying sqls running within the procedure?
On 9/4/25 13:32, veem v wrote:
On Thu, 4 Sept 2025 at 23:19, Adrian Klaver <adrian.klaver@aklaver.com
We didn't get the exact root cause why the plan flipped but we have to
add additional filters to get rid of the performance issue.
Did you do an EXPLAIN ANALYZE before and after the changes?
If so did you save the results and can you share them?
The version is 16.
Yes, autovacuum running.
We don't have any major changes.
Except for the performance degradation, so something changed.
But if there was a significant change
in quantity of data , Yet to check that.The pg_stats_statement does show the execution stats of the individual
query execution , but doesnt show any historical plan deviation. Also
the individual procedure and sql statement are logged into this view but
i was wondering if there is an easy way to relate the sql text of the
procedure with the underlying sqls running within the procedure?
The pg_stat_statements view has the following fields(among others):
min_exec_time double precision
max_exec_time double precision
mean_exec_time double precision
minmax_stats_since timestamp with time zone
I could see a cron job that gathers the above data at regular intervals.
Then you could track the query performance at a rough level.
--
Adrian Klaver
adrian.klaver@aklaver.com
On Thu, Sep 4, 2025 at 12:58 PM veem v <veema0000@gmail.com> wrote:
Hello,
We have a situation in which we had a dml query within a procedure that
was running fine but suddenly the plan flipped and it started running
longer.
That sounds like something that bit me once. Took a while to figure out,
and was very puzzling.
This list explained the problem and gave me the solution. If you run the
same query multiple times in, after five executions, the PG query optimizer
decides to use a generic plan. This disables that choice to use a generic
plan:
set plan_cache_mode = force_custom_plan;
Setting that returned the procedures to their previously-rapid execution.
--
Death to <Redacted>, and butter sauce.
Don't boil me, I'm still alive.
<Redacted> lobster!