I wrote:
> Eric Ridge <eebbrr@gmail.com> writes:
>> As a data point, that might be interesting to know, but I'd still be
>> scratching my head about how it happened. Postgres doesn't load an
>> extension library more than once per backend session, does it?
> It's not supposed to, and AFAICS internal_load_library() will treat
> either an exact pathname match or an inode-number match as being
> "already loaded".
BTW, if it didn't occur to you already --- next time you see this,
you could try examining dfmgr.c's DynamicFileList data structure
using gdb. That might help illuminate what Postgres thinks it
loaded and why. Also note that the hang presumably occurs just
*before* internal_load_library links the new item into the file_list,
so you'd want to inspect that as a local variable.
regards, tom lane