Re: Commit problem in read-commited isolation level

On 08/16/2013 10:38 PM, S H wrote:
> > 1) What is your exact Postgres version i.e 8.1.5.?
> V - 8.1.18
>
> > 3) You also say it works fine in two environments, but not one.
> >
> > What are the environments?
> > OS and version, memory, Postgres versions, etc.
>
> It is working on most of the production system ( more than 100) but
> failed once in one of the system ( not always).

Not sure I understand failed once, not always. Not always would seem to
mean it failed more than once. Could you elaborate?

Always the same machine or does that change?

If the same machine, is there something different about it?

>
>> OS and version, memory, Postgres versions, etc.
> RHEL 5.2 , 32 GB , 8.1.18

All the systems are exactly the same?

> > You say at the top the new value is 0, but show 5 as the new value at
> > the bottom.
> >
>
> Sorry it is returning new value ie 0 after some time. Corrected the
> above description.
>

So the commit happens, it just takes time. First guess would be some
sort of disk I/O contention problem. Honestly though there is not enough
information available to do more than guess. We would need a clearer
picture of what happens during these periods before it would be possible
to formulate a clearer answer. In other words log information from
Postgres, the application and the OS and ideally load information from
the system. Also are you talking about a specific procedure that causes
this, or seemingly random event? Basically, at this point it would be
better to provide as much information as possible.

--
Adrian Klaver
adrian(dot)klaver(at)gmail(dot)com