Обсуждение: Pgbouncer
We are migrating from oracle.to postgres and its done.
Now as a DBA, to improve performance say idle connecting handling or anything, if I want to recommend pgbouncer to the team, what's the compelling reason would it be?
I may ask them to handle connections from application end and they may say ok but we still.may see idle connections are not closed.
So, at point and for what compelling reason I should tell my manager, we must go-ahead and use etc.
We use pg17 and in postgres itself I know idle_session_timeount, transaction_timeout, statement_timeout. Is this enough? Will it abruptly kill queries and if so, isn't it bad?
On Thu, Sep 11, 2025 at 07:01:26PM +0530, Raj wrote: > We are migrating from oracle.to postgres and its done. > > Now as a DBA, to improve performance say idle connecting handling or > anything, if I want to recommend pgbouncer to the team, what's the > compelling reason would it be? https://www.depesz.com/2012/12/02/what-is-the-point-of-bouncing/ > I may ask them to handle connections from application end and they may say > ok but we still.may see idle connections are not closed. > > So, at point and for what compelling reason I should tell my manager, we > must go-ahead and use etc. > > We use pg17 and in postgres itself I know idle_session_timeount, > transaction_timeout, statement_timeout. Is this enough? Will it abruptly > kill queries and if so, isn't it bad? statement_timeout and transaction_timeout have the power to kill queries, yes. Best regards, depesz
I am asking should we use pgbouncer? If so, what's the compelling situation despite having postgres inbuilt timeout parameters.
On Thu, 11 Sept 2025, 19:25 hubert depesz lubaczewski, <depesz@depesz.com> wrote:
On Thu, Sep 11, 2025 at 07:01:26PM +0530, Raj wrote:
> We are migrating from oracle.to postgres and its done.
>
> Now as a DBA, to improve performance say idle connecting handling or
> anything, if I want to recommend pgbouncer to the team, what's the
> compelling reason would it be?
https://www.depesz.com/2012/12/02/what-is-the-point-of-bouncing/
> I may ask them to handle connections from application end and they may say
> ok but we still.may see idle connections are not closed.
>
> So, at point and for what compelling reason I should tell my manager, we
> must go-ahead and use etc.
>
> We use pg17 and in postgres itself I know idle_session_timeount,
> transaction_timeout, statement_timeout. Is this enough? Will it abruptly
> kill queries and if so, isn't it bad?
statement_timeout and transaction_timeout have the power to kill
queries, yes.
Best regards,
depesz
On Thu, 2025-09-11 at 22:11 +0530, Raj wrote: > I am asking should we use pgbouncer? If so, what's the compelling situation > despite having postgres inbuilt timeout parameters. Timeouts and connection pooling have little to do with each other. You want timeouts so that statements and transactions cannot take too long, hog resources and damage your database. Connection pooling has a different purpose. It is necessary, because you cannot afford to have short-lived database sessions. So you want persistent database session, but you don't want too many of them because - the more you have, the bigger the risk of overloading the database - if you have many connections, you cannot set "work_mem" too high, which will be bad for the performance of your SQL statements - switching between processes means overhead for the kernel - the more sessions you have, the longer the snapshot PostgreSQL takes at the start of each statement will take If you have a single application server, you don't need pgBouncer. Simply use the connection pooler built into your application server. Only when you start many instances of your application server (which would mean many pools) you need an external pooler like pgBouncer. If you want more reading material: https://www.cybertec-postgresql.com/estimating-connection-pool-size-with-postgresql-database-statistics/ https://www.cybertec-postgresql.com/tuning-max_connections-in-postgresql/ https://www.postgresql.eu/events/pgconfeu2022/sessions/session/3811/slides/327/1000_application_users.pdf Yours, Laurenz Albe