Postgres Pro
Downloads
Home   >   mailing lists

Re: [pgAgent][Patch] Fixing connection pool leak - Mailing list pgadmin-hackers

From Neel Patel
Subject Re: [pgAgent][Patch] Fixing connection pool leak
Date
Msg-id CACCA4P1JFmDKUVy_a5qL6QQw2W64-kAsKQCVGjpwH6Oiwden9w@mail.gmail.com
Whole thread Raw
In response to Re: [pgAgent][Patch] Fixing connection pool leak  (Ashesh Vashi <ashesh.vashi@enterprisedb.com>)
Responses Re: [pgAgent][Patch] Fixing connection pool leak  (Rob Emery <re-pgsql@codeweavers.net>)
List pgadmin-hackers
Tree view
Hi Ashesh,

I just added condition before delete the job otherwise it looks good. Correct me if I am wrong.

if (job != NULL)
{
    delete job;
    job = NULL;
}

I have created two instances of pgagent on database cluster. As of now not able to see any leak and keep you updated if found anything.

@Robert - Can you also test at your environment and keep us updated for any leak ?

Thanks,
Neel Patel

On Mon, Oct 23, 2017 at 10:30 AM, Ashesh Vashi <ashesh.vashi@enterprisedb.com> wrote:
Hi Rob,

How about this?


--

Thanks & Regards,

Ashesh Vashi
EnterpriseDB INDIA: Enterprise PostgreSQL Company


http://www.linkedin.com/in/asheshvashi


On Sat, Oct 21, 2017 at 8:36 PM, Rob Emery <re-pgsql@codeweavers.net> wrote:
Hi,

Following on from https://www.postgresql.org/message-id/CA%2BOCxoz4tONxSpd1rdU-9SPKRzucz8Bar2CXkEDnCwV6H77ZyA%40mail.gmail.com

I think I've identified and fixed the issue, please see the patch attached.

As I understand it when there are multiple pgagent instances and they clash executing a job (i.e rc != 1 on job.cpp:38), the loser of the conflict's thread will never be executed (i.e. job.cpp:418 JobThread::Entry), which is responsible for deleting the job owned by the thread, meaning that the connection is never returned to the pool. By moving the delete of the job into the destructor, we can assure that the connection is tidied up in both cases as the thread is deleted in the error case explicitly in pgAgent.cpp:185.

The only possibly unintended difference that I can see with doing this is that the log "Completed job: %s" is now output when before it wasn't, however I think this new behaviour is actually correct as the job object is completed at that time.

Thanks,
Rob






Phone: 0800 021 0888   Email: contactus@codeweavers.net
Codeweavers Ltd | Barn 4 | Dunston Business Village | Dunston | ST18 9AB
Registered in England and Wales No. 04092394 | VAT registration no. 974 9705 63 

    https://plus.google.com/b/105942302039373248738/+CodeweaversNet  https://twitter.com/CodeweaversTeam


Attachment

pgadmin-hackers by date:

Previous
From: Ashesh Vashi
Date:
Subject: Re: pgadmin4 l10n issues
Next
From: Rob Emery
Date:
Subject: Re: [pgAgent][Patch] Fixing connection pool leak