Re: parallel query in 9.6.6 + oracle_fdw -> ERROR: invalid cache ID: 41

Chris Mair <chris(at)1006(dot)org> writes:
> Whenever a session has performed a query on a foreign table, any subsequent
> query on a local table big enough to use the parallel query feature exits with
> an error:
> ERROR: invalid cache ID: 41
> CONTEXT: parallel worker

Hm, syscache 41 is PROCNAMEARGSNSP in 9.6, if I counted right.

> (gdb) bt
> #0 0x00007f16a0f4d1f7 in raise () from /lib64/libc.so.6
> #1 0x00007f16a0f4e8e8 in abort () from /lib64/libc.so.6
> #2 0x00000000008094b4 in errfinish (dummy=dummy(at)entry=0) at elog.c:557
> #3 0x000000000080aea2 in elog_finish (elevel=elevel(at)entry=22, fmt=fmt(at)entry=0x9d9965 "invalid cache ID: %d") at elog.c:1378
> #4 0x00000000007ffd82 in SearchSysCacheList (cacheId=cacheId(at)entry=41, nkeys=nkeys(at)entry=2, key1=key1(at)entry=139734905138463, key2=<optimized out>, key3=key3(at)entry=0, key4=key4(at)entry=0) at syscache.c:1210
> #5 0x00007f169161a59f in _PG_init () at oracle_fdw.c:709
> #6 0x000000000080c476 in internal_load_library (libname=<optimized out>, libname(at)entry=0x7f16a1bfdde8 <Address 0x7f16a1bfdde8 out of bounds>) at dfmgr.c:276
> #7 0x000000000080c708 in RestoreLibraryState (start_address=0x7f16a1bfdde8 <Address 0x7f16a1bfdde8 out of bounds>) at dfmgr.c:741
> #8 0x00000000004e72cf in ParallelWorkerMain (main_arg=<optimized out>) at parallel.c:1069

Apparently, oracle_fdw is trying to do a procedure lookup in its _PG_init
function. This is a horrible idea: it assumes that _PG_init is invoked
inside a transaction, which is wrong if the library is preloaded, for
example. (I'd bet that adding oracle_fdw to shared_preload_libraries
would fail badly, though perhaps not with this exact error message.)

So I'd call this an oracle_fdw bug. It needs to postpone what it's
doing here to the first normal FDW function call in a session.

regards, tom lane