patch for large queries

Hi Hackers,

I don't like last minute patches before the final freeze, but I believe that
this one could be useful for people experiencing out-of-memory crashes while
executing queries which retrieve or use a very large number of tuples.

The problem happens when storage is allocated for functions results used in
a large query, for example:
 select upper(name) from big_table; select big_table.array[1] from big_table; select count(upper(name)) from
big_table;

This patch is a dirty hack that fixes the out-of-memory problem for the most
common cases, like the above ones. It is not the final solution for the
problem but it can work for some people, so I'm posting it.

The patch should be safe because all changes are under #ifdef. Furthermore
the feature can be enabled or disabled at runtime by the `free_tuple_memory'
options in the pg_options file. The option is disabled by default and must
be explicitly enabled at runtime to have any effect.

To enable the patch add the follwing line to Makefile.custom:

CUSTOM_COPT += -DFREE_TUPLE_MEMORY

To enable the option at runtime add the following line to pg_option:

free_tuple_memory=1

Here is the patch:

*** src/include/utils/portal.h.orig    Wed May 26 09:07:16 1999
--- src/include/utils/portal.h    Fri Jun 11 18:06:07 1999
***************
*** 80,85 ****
--- 80,89 ---- extern PortalVariableMemory PortalGetVariableMemory(Portal portal); extern PortalHeapMemory
PortalGetHeapMemory(Portalportal); 
 
+ #ifdef FREE_TUPLE_MEMORY
+ bool   PortalHeapMemoryIsValid(MemoryContext context, Pointer pointer);
+ #endif
+  /* estimate of the maximum number of open portals a user would have,  * used in initially sizing the PortalHashTable
in   EnablePortalManager()  */
 
*** src/backend/utils/mmgr/portalmem.c.orig    Wed May 26 09:06:02 1999
--- src/backend/utils/mmgr/portalmem.c    Fri Jun 11 18:06:28 1999
***************
*** 289,294 ****
--- 289,312 ----     } } 
+ #ifdef FREE_TUPLE_MEMORY
+ /*
+  * PortalHeapMemoryIsValid --
+  *
+  * Check if a pointer is allocated in a memory context.
+  *
+  */
+ bool
+ PortalHeapMemoryIsValid(MemoryContext context, Pointer pointer)
+ {
+     HeapMemoryBlock block = HEAPMEMBLOCK((PortalHeapMemory) context);
+ 
+     AssertState(PointerIsValid(block));
+ 
+     return (AllocSetContains(&block->setData, pointer));
+ }
+ #endif
+  /* ----------------  *        PortalHeapMemoryRealloc  * ----------------
*** src/include/utils/trace.h.orig    Sat Jun  5 09:00:40 1999
--- src/include/utils/trace.h    Fri Jun 11 19:01:30 1999
***************
*** 64,69 ****
--- 64,72 ----     OPT_SYSLOG,                    /* use syslog for error messages */     OPT_HOSTLOOKUP,
/* enable hostname lookup in ps_status */     OPT_SHOWPORTNUMBER,            /* show port number in ps_status */
 
+ #ifdef FREE_TUPLE_MEMORY
+     OPT_FREE_TUPLE_MEMORY,        /* try to pfree memory for each tuple */
+ #endif      NUM_PG_OPTIONS                /* must be the last item of enum */ };
***************
*** 83,91 **** #endif     /* TRACE_H */  /*
!  * Local variables:
!  *    tab-width: 4
!  *    c-indent-level: 4
!  *    c-basic-offset: 4  * End:  */
--- 86,94 ---- #endif     /* TRACE_H */  /*
!  * Local Variables:
!  *  tab-width: 4
!  *  c-indent-level: 4
!  *  c-basic-offset: 4  * End:  */
*** src/backend/utils/misc/trace.c.orig    Sat Jun  5 09:00:37 1999
--- src/backend/utils/misc/trace.c    Fri Jun 11 19:01:31 1999
***************
*** 73,78 ****
--- 73,81 ----     "syslog",                    /* use syslog for error messages */     "hostlookup",                /*
enablehostname lookup in ps_status */     "showportnumber",            /* show port number in ps_status */
 
+ #ifdef FREE_TUPLE_MEMORY
+     "free_tuple_memory",        /* try to pfree memory for each tuple */
+ #endif      /* NUM_PG_OPTIONS */        /* must be the last item of enum */ };
***************
*** 404,412 **** }  /*
!  * Local variables:
!  *    tab-width: 4
!  *    c-indent-level: 4
!  *    c-basic-offset: 4  * End:  */
--- 407,415 ---- }  /*
!  * Local Variables:
!  *  tab-width: 4
!  *  c-indent-level: 4
!  *  c-basic-offset: 4  * End:  */
*** src/backend/access/common/heaptuple.c.orig    Wed May 26 09:01:59 1999
--- src/backend/access/common/heaptuple.c    Fri Jun 11 19:08:33 1999
***************
*** 27,32 ****
--- 27,37 ---- #include <storage/bufpage.h> #include <utils/memutils.h> 
+ #ifdef FREE_TUPLE_MEMORY
+ #include <utils/portal.h>
+ #include <utils/trace.h>
+ #endif
+  #ifndef HAVE_MEMMOVE #include <regex/utils.h> #else
***************
*** 93,98 ****
--- 98,106 ----     int            i;     int            numberOfAttributes = tupleDesc->natts;     Form_pg_attribute
*att= tupleDesc->attrs;
 
+ #ifdef FREE_TUPLE_MEMORY
+     bool        free_tuple_memory = pg_options[OPT_FREE_TUPLE_MEMORY];
+ #endif      if (bit != NULL)     {
***************
*** 131,136 ****
--- 139,152 ----                 *infomask |= HEAP_HASVARLENA;                 data_length =
VARSIZE(DatumGetPointer(value[i]));                memmove(data, DatumGetPointer(value[i]), data_length);
 
+ #ifdef FREE_TUPLE_MEMORY
+                 /* try to pfree value[i] - dz */
+                 if (free_tuple_memory &&
+                     PortalHeapMemoryIsValid(CurrentMemoryContext,
+                                             (Pointer) value[i])) {
+                     pfree(value[i]);
+                 }
+ #endif                 break;             case sizeof(char):                 *data = att[i]->attbyval ?
***************
*** 147,154 ****                                    *((int32 *) value[i]));                 break;
default:
!                 memmove(data, DatumGetPointer(value[i]),
!                         att[i]->attlen);                 break;         }         data = (char *)
att_addlength((long)data, att[i]->attlen, value[i]);
 
--- 163,177 ----                                    *((int32 *) value[i]));                 break;
default:
!                 memmove(data, DatumGetPointer(value[i]), att[i]->attlen);
! #ifdef FREE_TUPLE_MEMORY
!                 /* try to pfree value[i] - dz */
!                 if (free_tuple_memory &&
!                     PortalHeapMemoryIsValid(CurrentMemoryContext,
!                                             (Pointer) value[i])) {
!                     pfree(value[i]);
!                 }
! #endif                 break;         }         data = (char *) att_addlength((long) data, att[i]->attlen,
value[i]);
*** src/backend/executor/nodeAgg.c.orig    Wed May 26 09:02:57 1999
--- src/backend/executor/nodeAgg.c    Fri Jun 11 19:10:27 1999
***************
*** 31,36 ****
--- 31,41 ---- #include "utils/syscache.h" #include "optimizer/clauses.h" 
+ #ifdef FREE_TUPLE_MEMORY
+ #include <utils/portal.h>
+ #include <utils/trace.h>
+ #endif
+  /*  * AggFuncInfo -  *      keeps the transition functions information around
***************
*** 113,119 ****                 isNull1 = FALSE,                 isNull2 = FALSE;     bool        qual_result;
!       /* ---------------------      *    get state info from node
--- 118,126 ----                 isNull1 = FALSE,                 isNull2 = FALSE;     bool        qual_result;
! #ifdef FREE_TUPLE_MEMORY
!     bool        free_tuple_memory = pg_options[OPT_FREE_TUPLE_MEMORY];
! #endif      /* ---------------------      *    get state info from node
***************
*** 241,246 ****
--- 248,257 ----         for (;;)         {             TupleTableSlot *outerslot;
+ #ifdef FREE_TUPLE_MEMORY
+             Oid                valueType;
+             bool            isByValue = 0;
+ #endif              isNull = isNull1 = isNull2 = 0;             outerslot = ExecProcNode(outerPlan, (Plan *) node);
***************
*** 293,298 ****
--- 304,334 ----                     newVal = ExecEvalExpr(aggref->target, econtext,
      &isNull, &isDone);                 }
 
+ #ifdef FREE_TUPLE_MEMORY
+                 if (free_tuple_memory) {
+                     switch (nodeTag(aggref->target)) {
+                     case T_Const:
+                         isByValue = ((Const*) (aggref->target))->constbyval;
+                         break;
+                     case T_Var:
+                         valueType = ((Var*) (aggref->target))->vartype;
+                         isByValue = typeByVal(typeidType(valueType));
+                         break;
+                     case T_Array:
+                         isByValue = ((Array*)(aggref->target))->arrayelembyval;
+                         break;
+                     case T_ArrayRef:
+                         isByValue =((ArrayRef*)(aggref->target))->refelembyval;
+                         break;
+                     case T_Expr:
+                         valueType = ((Expr*) (aggref->target))->typeOid;
+                         isByValue = typeByVal(typeidType(valueType));
+                         break;
+                     default:
+                         break;
+                     }
+                 }
+ #endif                  if (isNull && !aggref->usenulls)                     continue;    /* ignore this tuple for
thisagg */
 
***************
*** 353,358 ****
--- 389,404 ----                                           (FmgrValues *) args, &isNull2);
Assert(!isNull2);                }
 
+ 
+ #ifdef FREE_TUPLE_MEMORY
+                 /* try to pfree newVal if not isByValue - dz */
+                 if (free_tuple_memory && !isByValue && 
+                     PortalHeapMemoryIsValid(CurrentMemoryContext,
+                                             (Pointer) newVal))
+                 {
+                     pfree(newVal);
+                 }
+ #endif             }              /*


-- 
Massimo Dal Zotto

+----------------------------------------------------------------------+
|  Massimo Dal Zotto               email: dz@cs.unitn.it               |
|  Via Marconi, 141                phone: ++39-0461534251              |
|  38057 Pergine Valsugana (TN)      www: http://www.cs.unitn.it/~dz/  |
|  Italy                             pgp: finger dz@tango.cs.unitn.it  |
+----------------------------------------------------------------------+