Loosing connection with the database

=?UTF-8?B?UG91bCBNw7hsbGVyIEhhbnNlbg==?= <freebsd@pbnet.dk> writes:

I'm experiencing some strange problems with a Java application which I
have made.
Suddently it looses connection with the database, and in the server log
i can find:

2005-04-02 00:09:01 ERROR: invalid string enlargement request size
1358954492
2005-04-02 00:09:01 WARNING: AbortTransaction and not in in-progress
state
2005-04-02 00:09:01 FATAL: invalid frontend message type 82

This indicates the client code didn't follow the protocol properly,
ie, it sent garbage where a message length word ought to be.

This sort of thing has been seen to occur when multiple client-side
threads try to use the same database connection without proper locking
to ensure only one thread uses it at a time. See for example
http://archives.postgresql.org/pgsql-hackers/2004-09/msg00104.php

regards, tom lane

2005-04-02 00:09:01 ERROR: invalid string enlargement request size
1358954492
2005-04-02 00:09:01 WARNING: AbortTransaction and not in in-progress
state
2005-04-02 00:09:01 FATAL: invalid frontend message type 82

This indicates the client code didn't follow the protocol properly,
ie, it sent garbage where a message length word ought to be.

This sort of thing has been seen to occur when multiple client-side
threads try to use the same database connection without proper locking
to ensure only one thread uses it at a time. See for example
http://archives.postgresql.org/pgsql-hackers/2004-09/msg00104.php

regards, tom lane

This is exactly what I am doing. Must admit I haven't considered that as
an issue.
For performance reasons I suppose one database connection per client are
preferred
rather than using synchronized on the db class ?

Thanks a lot for bringing it to my attention

BR, Poul

On Sat, 2 Apr 2005, [ISO-8859-1] Poul M�ller Hansen wrote:

This sort of thing has been seen to occur when multiple client-side
threads try to use the same database connection without proper locking
to ensure only one thread uses it at a time. See for example
http://archives.postgresql.org/pgsql-hackers/2004-09/msg00104.php

This is exactly what I am doing. Must admit I haven't considered that as
an issue. For performance reasons I suppose one database connection per
client are preferred rather than using synchronized on the db class ?

The JDBC driver should be doing any synchronization necessary for multiple
threads. Could you be more clear what you are doing? What driver
version? Any chance you've got a reproducible example?

Kris Jurka

This sort of thing has been seen to occur when multiple client-side
threads try to use the same database connection without proper locking
to ensure only one thread uses it at a time. See for example
http://archives.postgresql.org/pgsql-hackers/2004-09/msg00104.php

This is exactly what I am doing. Must admit I haven't considered that as
an issue. For performance reasons I suppose one database connection per
client are preferred rather than using synchronized on the db class ?

The JDBC driver should be doing any synchronization necessary for multiple
threads. Could you be more clear what you are doing? What driver
version? Any chance you've got a reproducible example?

Kris Jurka

I have rewritten the application so every client thread is opening a new
database connection, and yesterday it happened again.
---
2005-04-11 12:27:54 ERROR: invalid string enlargement request size
1358954492
2005-04-11 12:27:54 WARNING: AbortTransaction and not in in-progress
state
2005-04-11 12:27:54 FATAL: invalid frontend message type
78
---
The application is opening a socket listener, and every client
connection opens a new connection to the database.
The clients sends a status message every 2nd minute that are written to
the database.
I'm using Postgresql version 7.4.7 and jdbc driver version
pg74.215.jdbc3.jar.
I have tried the pg80.310.jdbc3.jar but the datatype inet can't be used
with setString ??

Do you have a clue on what's going on ?

Poul

I found this: http://jdbc.postgresql.org/documentation/80/thread.html
As you are saying the jdbc driver should be thread safe, and has been
since the first version.
The

On Tue, 12 Apr 2005, [UTF-8] Poul M��ller Hansen wrote:

I have rewritten the application so every client thread is opening a new
database connection, and yesterday it happened again.
---
2005-04-11 12:27:54 ERROR: invalid string enlargement request size
1358954492
2005-04-11 12:27:54 WARNING: AbortTransaction and not in in-progress
state
2005-04-11 12:27:54 FATAL: invalid frontend message type
78
---
The application is opening a socket listener, and every client
connection opens a new connection to the database.
The clients sends a status message every 2nd minute that are written to
the database.
I'm using Postgresql version 7.4.7 and jdbc driver version
pg74.215.jdbc3.jar.

Do you have a clue on what's going on ?

No, I don't. Do you have any more information? What is your code doing
when it fails? Just issuing a regular query? Are you using any of the
less common driver features: Large objects, fastpath api, a COPY patch?
If the driver had a protocol problem I would expect it to be rather
repeatable. If the driver had a synchronization problem it should have
disappeared when you moved to a single thread model. I've attached the
test script I've used to try and beat on the driver.

Kris Jurka

Poul Møller Hansen wrote: