Re: Large object loading stalls

Tom Lane wrote:

Michael Akinde <michael.akinde@met.no> writes: 

Anyway - the situation now is that just the loading process is hanging 
on the server, with an <IDLE> in transaction. But it is definitely the 
loading program that is hanging, not the Postgres server.   

What the stack traces seem to show is that both the client and the
server are waiting for each other to send some data.  Which means
somebody's bollixed the protocol.

In the past we've seen this type of thing caused by multithreaded
client programs in which more than one thread tried to use the same
PGconn object without adequate interlocking.  libpq itself does not
provide any threading guards --- if you want more than one thread
accessing a PGconn then it's up to you to use a mutex or something
to serialize them.  Is it possible this case applies here? 

My apologies for the delayed response.

Our application is single-threaded, so it seems unlikely that we are running into a problem with that.

The only thing I can think of right now, is that we are running a Postgres 8.3 on Debian Etch (so a backported debian package), whereas the libraries linked into our application are older library version (libpq4 and libpqxx 2.6.8). I'll try to upgrade the OS to a version with "native" support for 8.3 and up to date (or at least more up to date) versions of pq, pqxx and check whether the tests still break down.

Regards,

Michael A.