Thread: When is a shared library unloaded?
I have a shared library loaded into postgresql using CREATE FUNCTION medbevent() RETURNS trigger as '/usr/lib/pgsql/libmedbevent.so' language 'C'; I'm randomly having issues with my triggers not firing (it turns out they are getting removed) and when i turned INFO logs on i saw the following 2007-08-14 13:41:44.740305500 LOCATION: _fini, medbevent_init.c:129 2007-08-14 13:41:45.790850500 INFO: 00000: medbeventlib -Unloading library _fini() is called Could someone please explain when a shared library gets loaded/unloaded and how this all works? I'm getting the above logs when i wouldn't expect to see the library be unloaded which is causing me problems. Could the library get unloaded in this manner due to an internal error in the library? Thanks, Jon.
"Jon Horsman" <horshaq@gmail.com> writes:
> I'm randomly having issues with my triggers not firing (it turns out
> they are getting removed) and when i turned INFO logs on i saw the
> following
> 2007-08-14 13:41:44.740305500 LOCATION: _fini, medbevent_init.c:129
> 2007-08-14 13:41:45.790850500 INFO: 00000: medbeventlib -Unloading
> library _fini() is called
> Could someone please explain when a shared library gets
> loaded/unloaded and how this all works?
Are you sure the above doesn't just occur during server process
termination? AFAIK Postgres never unloads a library, except when you
specifically command it to load an updated version via the LOAD command.
regards, tom lane
> "Jon Horsman" <horshaq@gmail.com> writes: > > I'm randomly having issues with my triggers not firing (it turns out > > they are getting removed) and when i turned INFO logs on i saw the > > following > > > 2007-08-14 13:41:44.740305500 LOCATION: _fini, medbevent_init.c:129 > > 2007-08-14 13:41:45.790850500 INFO: 00000: medbeventlib -Unloading > > library _fini() is called > > > Could someone please explain when a shared library gets > > loaded/unloaded and how this all works? > > Are you sure the above doesn't just occur during server process > termination? AFAIK Postgres never unloads a library, except when you > specifically command it to load an updated version via the LOAD command. > > regards, tom lane Yup, the server process is still up and going, i have other clients connected to postgres that still function when this happens. I'm not using the LOAD command anywhere so thats not the problem. One of my colleagues was doing some reading and said he found something that stating that the fini() function is "unreliable", i'm not sure where he read that though... Jon.
Here is a little more information about my problem. There is a library( that has stored procedures) for processing trigger events. The library is loaded on first access( i.e when there is a trigger event to be sent). Normally, it gets unloaded during a postgres server shutdown but the library seems to be getting unloaded occasionally. We have logs in the _init( ) and _fini( ) of the library and we see that there are times when _fini( ) is getting called when you wouldn't expect it to be. Has anyone else had issues similar to this? Thanks, Jon. On 8/15/07, Jon Horsman <horshaq@gmail.com> wrote: > > "Jon Horsman" <horshaq@gmail.com> writes: > > > I'm randomly having issues with my triggers not firing (it turns out > > > they are getting removed) and when i turned INFO logs on i saw the > > > following > > > > > 2007-08-14 13:41:44.740305500 LOCATION: _fini, medbevent_init.c:129 > > > 2007-08-14 13:41:45.790850500 INFO: 00000: medbeventlib -Unloading > > > library _fini() is called > > > > > Could someone please explain when a shared library gets > > > loaded/unloaded and how this all works? > > > > Are you sure the above doesn't just occur during server process > > termination? AFAIK Postgres never unloads a library, except when you > > specifically command it to load an updated version via the LOAD command. > > > > regards, tom lane > > Yup, the server process is still up and going, i have other clients > connected to postgres that still function when this happens. I'm not > using the LOAD command anywhere so thats not the problem. One of my > colleagues was doing some reading and said he found something that > stating that the fini() function is "unreliable", i'm not sure where > he read that though... > > Jon. >