From: | Karl Denninger <karl(at)denninger(dot)net> |
---|---|
To: | pgsql-performance(at)postgresql(dot)org |
Subject: | Re: Big number of connections |
Date: | 2016-03-31 23:27:07 |
Message-ID: | 56FDB24B.6050605@denninger.net |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
On 3/31/2016 17:47, Alvaro Herrera wrote:
> Andrew Dunstan wrote:
>
>> On 03/31/2016 03:12 PM, Igor Neyman wrote:
>>>> We are going to build system based on PostgreSQL database for huge number of individual users (few thousands). Each user will have his own account, for authorization we will use Kerberos (MIT or Windows).
>>>> Most of users will have low activity, but for various reasons, connection should be open all the time.
>>>> I'd like to know what potential problems and limitations we can expect with such deployment.
>>>> During preliminary testing we have found that for each connection we need ~1MB RAM. Is there any way to decrease this ? Is there any risk, that such number of users will degrade performance ?
>>>> I'll be happy to hear any remarks and suggestions related to design, administration and handling of such installation.
>>> Take a look at PgBouncer.
>>> It should solve your problems.
>> If they are going to keep the client connections open, they would need to
>> run pgbouncer in statement or transaction mode.
> As I understand, in pgbouncer you cannot have connections that serve
> different users. If each individual requires its own database-level
> user, pgbouncer would not help at all.
>
> I would look seriously into getting rid of the always-open requirement
> for connections.
I'm trying to figure out where the "always open" requirement comes from;
there are very, very few instances where that's real, when you get down
to it.
--
Karl Denninger
karl(at)denninger(dot)net <mailto:karl(at)denninger(dot)net>
/The Market Ticker/
/[S/MIME encrypted email preferred]/
From | Date | Subject | |
---|---|---|---|
Next Message | Alexandre de Arruda Paes | 2016-04-01 02:44:50 | Fast HashJoin only after a cluster/recreate table |
Previous Message | Chris Cogdon | 2016-03-31 23:12:51 | Re: Big number of connections |