RE: An I/O error occurred while sending to the backend (PG 13.4)

["ldh@xxxxxxxxxxxxxxxxxx" <ldh@xxxxxxxxxxxxxxxxxx>]

I am also starting to feel that the issue being on the database’s side is less and less likely. There is something happening in between, or possibly on the client.

 

Ranier, the only reason I was focusing on this at the PG level is that this issue started to show up several months ago shortly after I updated to PG13 from PG11. Had run PG11 for 2 years without ever seeing that issue at all. The ETL itself hasn’t changed either, except for upgrading the JDBC driver… But I did revert back to an older JDBC driver and the issue still did occur eventually.

 

Of course, other things could have changed in the client’s IT infrastructure that I am not aware of, so I am pushing that angle as well more aggressively now. I am also pushing for WireShark to monitor the network more closely. Stay tuned!

 

Thank you so much all for your support but at this time, I think the ball is in my camp and working out with it on some plan.

 

Thank you,

Laurent.

 

 

From: Ranier Vilela <ranier.vf@xxxxxxxxx>
Sent: Thursday, March 3, 2022 11:57
To: Justin Pryzby <pryzby@xxxxxxxxxxxxx>
Cc: ldh@xxxxxxxxxxxxxxxxxx; pgsql-performance@xxxxxxxxxxxxxx
Subject: Re: An I/O error occurred while sending to the backend (PG 13.4)

 

 

Em qui., 3 de mar. de 2022 às 13:46, Justin Pryzby <pryzby@xxxxxxxxxxxxx> escreveu:

On Thu, Mar 03, 2022 at 01:33:08PM -0300, Ranier Vilela wrote:
> Sorry, but this is much more on the client side.

The client is reporting the problem, as is the server.

 Are you read the server log?

" 2022-03-03 01:04:40 EST [21228] LOG:  could not receive data from client: An existing connection was forcibly closed by the remote host.
2022-03-03 01:04:40 EST [21228] LOG:  unexpected EOF on client connection with an open transaction"

 

> Following the logs, it is understood that the client is dropping the
> connection.

The logs show that the client's connection *was* dropped.
And on the server, the same.

No, the log server shows that the client dropped the connection.

 

> So most likely the error could be from Pentaho or JDBC.
>
> https://www.geeksforgeeks.org/java-net-socketexception-in-java-with-examples/
> " This *SocketException* occurs on the server-side when the client closed
> the socket connection before the response could be returned over the
> socket."
>
> I suggest moving this thread to the Pentaho or JDBC support.

We don't know the source of the problem.

Yeah, but it is much more likely to be on the client.

 

  I still doubt it's in postgres,

Everything indicates not.

 

but I
don't think it's helpful to blame the client, just because the client reported
the problem.  If the server were to disconnect abruptly, I'd expect the client
to report that, too.

Laurent would just have to start the conversation over (and probably collect
the same diagnostic information anyway).  The client projects could blame
postgres with as much rationale as there is for us to blame the client.

Please don't add confusion here. 

I just suggested, this is not an order.

 

regards,

Ranier Vilela