Обсуждение: Per thread Connection memory

Hi,

I'm not sure why do you need to restart tomcat? Could you explain further?

Also, max_connections=1000 is way too higher for 5GB of shared buffers. You could try setting it to 300 and put a connection pooler like pgbouncer infront

Thanks!

Guten Tag Ankur Kaushik,
am Mittwoch, 27. Januar 2016 um 12:09 schrieben Sie:

> Application Using Java Program , They have there java Pool system ,
> when this Java pool memory full , tomcat need to restart in every 30- 60 min

So why do you think your problem is with Postgres? Client side
connections out of Java are not influenced by server side Postgres
configuration and Tomcat and Postgres don't share any memory.

Sounds like you should provide the exact error message you get.

> Please Let me know in Postgresql  pgbouncer  is essential use for
> handling multithread applications ?

No and you already successfully use your setup, else you wouldn't run
into problems, your only problem is with available resources and how
to balance them. But first you need to know where exactly your
bottleneck is and it sounds more like Java/Tomcat to me.

Mit freundlichen Grüßen,

Thorsten Schöning

--
Thorsten Schöning       E-Mail: Thorsten.Schoening@AM-SoFT.de
AM-SoFT IT-Systeme      http://www.AM-SoFT.de/

Telefon...........05151-  9468- 55
Fax...............05151-  9468- 88
Mobil..............0178-8 9468- 04

AM-SoFT GmbH IT-Systeme, Brandenburger Str. 7c, 31789 Hameln
AG Hannover HRB 207 694 - Geschäftsführer: Andreas Muchow

Guten Tag Ankur Kaushik,
am Freitag, 29. Januar 2016 um 13:16 schrieben Sie:

> To Kill idle connection in every 10 sec , But after 15 -30 Min top command show as below

Killing idle connections is completely pointless, because those are
kept open for performance reasons, pgbouncer is designed to keep
connections open and idle, so that applications use them faster. If
you want less idle connections, reduce the configuration for those in
pgbouncer, but in the end you defeat its purpose that way.


>   PID USER      PR  NI  VIRT  RES  SHR S     %CPU %MEM    TIME+  COMMAND
> 10454 root        20   0 7101m   2.4g  11m  S    751.3 15.4  84:47.60 java

You problem is obviously that you don't have any clue what your
problem is, therefore you're just guessing things around. Don't do
that, at least look at the logs of your Tomcat, you may even use Java
Mission Control to monitor the performance of Tomcat and so on. This
high load might even be the result of your script introducing some
kind of a race condition, where you kill a connection just in that
moment where it was given to a using process in Tomcat.

If your problem is with Tomcat consuming too much memory or having
unexplainable high CPU load, don't waste your time with advancing the
complexity of your setup by installing pgbouncer. Focus on Tomcat and
its logs first.

Mit freundlichen Grüßen,

Thorsten Schöning

--
Thorsten Schöning       E-Mail: Thorsten.Schoening@AM-SoFT.de
AM-SoFT IT-Systeme      http://www.AM-SoFT.de/

Telefon...........05151-  9468- 55
Fax...............05151-  9468- 88
Mobil..............0178-8 9468- 04

AM-SoFT GmbH IT-Systeme, Brandenburger Str. 7c, 31789 Hameln
AG Hannover HRB 207 694 - Geschäftsführer: Andreas Muchow

El vie, 29-01-2016 a las 14:18 +0100, Thorsten Schöning escribió:
> Guten Tag Ankur Kaushik,
> am Freitag, 29. Januar 2016 um 13:16 schrieben Sie:
>
> > To Kill idle connection in every 10 sec , But after 15 -30 Min top
> > command show as below
>
> Killing idle connections is completely pointless, because those are
> kept open for performance reasons, pgbouncer is designed to keep
> connections open and idle, so that applications use them faster. If
> you want less idle connections, reduce the configuration for those in
> pgbouncer, but in the end you defeat its purpose that way.

+1. If you don't want to keep many connections open in idle state, you
can configure them at pgbouncer level or pool configuration in tomcat
or application level.

>
>
> >   PID USER      PR  NI  VIRT  RES  SHR S     %CPU %MEM    TIME+
> >  COMMAND
> > 10454 root        20   0 7101m   2.4g  11m  S    751.3 15.4
> >  84:47.60 java
>
> You problem is obviously that you don't have any clue what your
> problem is, therefore you're just guessing things around. Don't do
> that, at least look at the logs of your Tomcat, you may even use Java
> Mission Control to monitor the performance of Tomcat and so on. This
> high load might even be the result of your script introducing some
> kind of a race condition, where you kill a connection just in that
> moment where it was given to a using process in Tomcat.

Java is consuming 7/8 core of cpu, so i think you your look closer into
tomcat and java processes.

How about your work_mem parameter and top stats relatives of postgresql
server ?


>
> If your problem is with Tomcat consuming too much memory or having
> unexplainable high CPU load, don't waste your time with advancing the
> complexity of your setup by installing pgbouncer. Focus on Tomcat and
> its logs first.

+1

>
> Mit freundlichen Grüßen,
>
> Thorsten Schöning
>
> --
> Thorsten Schöning       E-Mail: Thorsten.Schoening@AM-SoFT.de
> AM-SoFT IT-Systeme      http://www.AM-SoFT.de/
>
> Telefon...........05151-  9468- 55
> Fax...............05151-  9468- 88
> Mobil..............0178-8 9468- 04
>
> AM-SoFT GmbH IT-Systeme, Brandenburger Str. 7c, 31789 Hameln
> AG Hannover HRB 207 694 - Geschäftsführer: Andreas Muchow
>
>
>

> Application Using Java Program , They have there java Pool system,
> when this Java pool memory full , tomcat need to restart in every 30- 60 min

You should configure your connection pool in the Java application to _release_ those connections

e.g. if you are using the Tomcat 7 pool, I would configure something like maxIdle="10", minIdle="5" and maxActive="150"
andthen you need to define a timeout when an idle connection is closed by the pool, e.g.
minEvictableIdleTimeMillis="300000"to close an idle connection after 5 minutes.  

Depending on your workload you might want to close that faster (e.g. after 1 minute).

> Most queries found in processlist is which full the Java memory pool , Hibernate is used in Application

Those queries look suspiciously like the query that is used by DatabaseMetaData.getTables().

I think the c3p0 connection pool has a pretty stupid default configuration where the validation query (that is used to
checkif a connection is still "alive") uses exactly that call. And I also think this is configured to be called each
timea connection is returned to the pool - which imposes a huge (and unnecessary load) on the database server.   

If that is the case you should configure a different validation query that is less invasive. Something like:
preferredTestQuery="select42", once that is configure c3p0 will no longer call the (expensive)
DatabaseMetaData.getTables()method. It's one of the many reasons I don't like the c3p0 pool. I think Tomcat7's pool is
amuch better choice (I'm _not_ talking about DBCP, but about the new "native" pool introduced with Tomcat7) 

I also recommend to _not_ test the connections when the pools hands them out or if they are returned, because those
validationquery will then put too much load on the database server.  

It's better to configure the pool to validate the connections when they are idle. Again, how you configure this,
dependson the pool you are using.  


> Please Let me know in Postgresql  pgbouncer  is essential use for handling multithread applications?

No, if you already have a connection pool in your application, then you don't need another one.

pgBouncer (or pgPool) are usually used for either load balancing or if the applications themselves don't do pooling.

Thomas

El vie, 29-01-2016 a las 14:18 +0100, Thorsten Schöning escribió:
> Guten Tag Ankur Kaushik,
> am Freitag, 29. Januar 2016 um 13:16 schrieben Sie:
>
> > To Kill idle connection in every 10 sec , But after 15 -30 Min top
> > command show as below
>
> Killing idle connections is completely pointless, because those are
> kept open for performance reasons, pgbouncer is designed to keep
> connections open and idle, so that applications use them faster. If
> you want less idle connections, reduce the configuration for those in
> pgbouncer, but in the end you defeat its purpose that way.

+1. If you don't want to keep many connections open in idle state, you
can configure them at pgbouncer level or pool configuration in tomcat
or application level.

>
>
> >   PID USER      PR  NI  VIRT  RES  SHR S     %CPU %MEM    TIME+
> >  COMMAND
> > 10454 root        20   0 7101m   2.4g  11m  S    751.3 15.4
> >  84:47.60 java
>
> You problem is obviously that you don't have any clue what your
> problem is, therefore you're just guessing things around. Don't do
> that, at least look at the logs of your Tomcat, you may even use Java
> Mission Control to monitor the performance of Tomcat and so on. This
> high load might even be the result of your script introducing some
> kind of a race condition, where you kill a connection just in that
> moment where it was given to a using process in Tomcat.

Java is consuming 7/8 core of cpu, so i think you your look closer into
tomcat and java processes.

How about your work_mem parameter and top stats relatives of postgresql
server ?


>
> If your problem is with Tomcat consuming too much memory or having
> unexplainable high CPU load, don't waste your time with advancing the
> complexity of your setup by installing pgbouncer. Focus on Tomcat and
> its logs first.

+1

>
> Mit freundlichen Grüßen,
>
> Thorsten Schöning
>
> --
> Thorsten Schöning       E-Mail: Thorsten.Schoening@AM-SoFT.de
> AM-SoFT IT-Systeme      http://www.AM-SoFT.de/
>
> Telefon...........05151-  9468- 55
> Fax...............05151-  9468- 88
> Mobil..............0178-8 9468- 04
>
> AM-SoFT GmbH IT-Systeme, Brandenburger Str. 7c, 31789 Hameln
> AG Hannover HRB 207 694 - Geschäftsführer: Andreas Muchow
>
>
>
--
Jaime Soler Gómez
HOPLA Software
EnterpriseDB exclusive distributor ES/PT/IT & LatAm
| P : (+34) 616 71 44 41 | SKYPE : jaimesolergomez | W : http://www.hoplasoftware.com/