Postgresql源码分析returns setof函数oracle管道pipelined

引言

【功能】

Oracle的return pipelined管道函数可以使一次返回的集合类型，变为逐条返回pipe row(集合中的一条)给SQL层，大大减少内存的使用。
Postgresql的return setof函数并不能起到降低内存使用的效果，return next 单条数据只起到了缓存的效果，并不会把数据逐条返回SQL层处理，没有降低内存的效果。

【代码】

exec_stmt_return_next中的tupledesc从执行计划node中取出，返回值需要满足desc要求，缓存值也会按该desc保存。
return next对rec类型和row类型处理的区别
- rec类型本质上就是tuple，数据和desc都以扩展形式存放在erh中。如果需要转换为tuple，有几个标准函数提供转换功能，且支持类型转换。【转换后调用tuplestore的标准接口缓存tuple】
- row类型本质上是一个虚拟行（由一组datum位置组成），row->varnos[i]指向某一个datum，如果想把row转换为tuple，需要用exec_eval_datum算出varnos指向的datum的值，然后组装成values和nulls数组，用heap_form_tuple构造。注意这种转换过程不会有类型转换，如果需要的desc和算出来的列类型对不上，返回空。成功【转换后调用tuplestore的标准接口缓存tuple】
return next对var类型的处理：var看做单列tuple，按执行计划给的desc转换类型后构造tuple。【转换后调用tuplestore的标准接口缓存tuple】

【实用函数】

通用
- 类型转换：exec_cast_value（传入的值不能是eoh真实的头，使用前需要转成eoh存的1be头，1be指向真实头）
- 数组拼接minimaltuple：heap_form_minimal_tuple
- 有一个tuple和desc转换为另一个desc的tuple：convert_tuples_by_position、execute_attr_map_tuple
tuplestore：
- 用values数组存tuple（用tuplestore_puttuple_common拼好后传tuple）：tuplestore_putvalues
- 用HeapTuple存tuple（直接传tuple）：tuplestore_puttuple
类型
- 根据类型id和mod找出desc：lookup_rowtype_tupdesc
erh
- 从erh扩展类型拿到紧凑tuple：expanded_record_get_tuple

1 ：管道函数是什么，应用于什么场景

oracle支持pipelined函数，可以在函数定义时指定RETURN 集合类型 PIPELINED 来说明当前函数是管道函数。

管道函数最大的作用就是可以使一次返回的集合类型，变为逐条返回，大大减少内存的使用。

例如：嵌套表类型outrecset是函数f_trans的返回值，普通函数只能组装好嵌套表outrecset（全部缓存在内存），一次性返回。如果嵌套表内容较多，可能会占用较大的内存空间。

如果使用管道函数，可以通过pipe row(嵌套表中的一行)来代替return语句，函数把嵌套表逐行返回给上层处理，无需缓存，降低内存使用。

ORACLE实例：

CREATE OR REPLACE PACKAGE refcur_pkg AUTHID DEFINER IS
TYPE refcur_t IS REF CURSOR RETURN employees%ROWTYPE;
TYPE outrec_typ IS RECORD (
var_num NUMBER(6),
var_char1 VARCHAR2(30),
var_char2 VARCHAR2(30)
);
TYPE outrecset IS TABLE OF outrec_typ;
FUNCTION f_trans (p refcur_t) RETURN outrecset PIPELINED;
END refcur_pkg;
/
CREATE OR REPLACE PACKAGE BODY refcur_pkg IS
FUNCTION f_trans (p refcur_t) RETURN outrecset PIPELINED IS
out_rec outrec_typ;
in_rec p%ROWTYPE;
BEGIN
LOOP
FETCH p INTO in_rec; — input row
EXIT WHEN p%NOTFOUND;
out_rec.var_num := in_rec.employee_id;
out_rec.var_char1 := in_rec.first_name;
out_rec.var_char2 := in_rec.last_name;
PIPE ROW(out_rec); — first transformed output row
out_rec.var_char1 := in_rec.email;
out_rec.var_char2 := in_rec.phone_number;
PIPE ROW(out_rec); — second transformed output row
END LOOP;
CLOSE p;
RETURN;
END f_trans;
END refcur_pkg;
/
SELECT * FROM TABLE (
refcur_pkg.f_trans (
CURSOR (SELECT * FROM employees WHERE department_id = 60)
)
);

在PG中，普通的return语句也是需要一次性返回数据，但PG应该是参考ORACLE实现了return next的功能，也希望逐条返回数据（PG没有集合类型，已普通类型为例）：

但在内核实现中，并不是逐条返回的，return next其实只起到了缓存数据的功能，总的数据集也是一次性返回SQL层的，和直接return没有区别（只有语法上的区别）。

所以PG的return setof函数并不能起到降低内存使用的效果。下面来分析具体过程。

2 return next实现

return next目前支持三类数据的返回，var、rec、rows return next也可以不加参数，返回值按out参数列表拼接

具体处理函数：exec_stmt_return_next

static int
exec_stmt_return_next(PLpgSQL_execstate *estate,
PLpgSQL_stmt_return_next *stmt)
{
TupleDesc tupdesc;
int natts;
HeapTuple tuple;
MemoryContext oldcontext;

1 初始化tuple store

初始化总结：

1 初始化的过程就是在构造Tuplestorestate，主要动作：

给Tuplestorestate新的内存上下文ExecutorState
记录不能随机访问：eflags = EXEC_FLAG_REWIND
记录三个操作函数：copytup_heap、writetup_heap、readtup_heap

2 给estate->tuple_store_desc添加desc，desc来源：

从执行计划节点中node（Tuplestorestate）拿到后，传入ExecMakeTableFunctionResult
ExecMakeTableFunctionResult组装ReturnSetInfo挂到fcinfo->resultinfo上
plpgsql_exec_function时从fcinfo中拿出ReturnSetInfo取到desc
plpgsql_estate_setup将取到的desc存入estate->rsi = rsi

#0 plpgsql_estate_setup (estate=0x7ffd81e2f850, func=0x2419028, rsi=0x7ffd81e2fb20, simple_eval_estate=0x0, simple_eval_resowner=0x0) at pl_exec.c:3972
#1 0x00007fe0a3992064 in plpgsql_exec_function (func=0x2419028, fcinfo=0x24da5a8, simple_eval_estate=0x0, simple_eval_resowner=0x0, procedure_resowner=0x0, atomic=true) at pl_exec.c:485
#2 0x00007fe0a39ac8f9 in plpgsql_call_handler (fcinfo=0x24da5a8) at pl_handler.c:277
#3 0x0000000000738829 in ExecMakeTableFunctionResult (setexpr=0x24e0b40, econtext=0x24e0a10, argContext=0x24da490, expectedDesc=0x24e1110, randomAccess=false) at execSRF.c:235
#4 0x0000000000753eed in FunctionNext (node=0x24e0800) at nodeFunctionscan.c:95
#5 0x000000000073a081 in ExecScanFetch (node=0x24e0800, accessMtd=0x753e3b <FunctionNext>, recheckMtd=0x754242 <FunctionRecheck>) at execScan.c:133
#6 0x000000000073a0f6 in ExecScan (node=0x24e0800, accessMtd=0x753e3b <FunctionNext>, recheckMtd=0x754242 <FunctionRecheck>) at execScan.c:182
#7 0x000000000075428c in ExecFunctionScan (pstate=0x24e0800) at nodeFunctionscan.c:270
#8 0x000000000073614e in ExecProcNodeFirst (node=0x24e0800) at execProcnode.c:464
#9 0x000000000072a08a in ExecProcNode (node=0x24e0800) at ../../../src/include/executor/executor.h:262
#10 0x000000000072cb80 in ExecutePlan (estate=0x24e05d8, planstate=0x24e0800, use_parallel_mode=false, operation=CMD_SELECT, sendTuples=true, numberTuples=0, direction=ForwardScanDirection, dest=0x24d5910, execute_once=true) at execMain.c:1632
#11 0x000000000072a6d1 in standard_ExecutorRun (queryDesc=0x23f1248, direction=ForwardScanDirection, count=0, execute_once=true) at execMain.c:364
#12 0x000000000072a50b in ExecutorRun (queryDesc=0x23f1248, direction=ForwardScanDirection, count=0, execute_once=true) at execMain.c:308
#13 0x0000000000997ba9 in PortalRunSelect (portal=0x2474a28, forward=true, count=0, dest=0x24d5910) at pquery.c:924
#14 0x0000000000997867 in PortalRun (portal=0x2474a28, count=9223372036854775807, isTopLevel=true, run_once=true, dest=0x24d5910, altdest=0x24d5910, qc=0x7ffd81e300b0) at pquery.c:768
#15 0x0000000000991408 in exec_simple_query (query_string=0x23c9518 \”select * from f1(42);\”) at postgres.c:1238
#16 0x0000000000995a3e in PostgresMain (dbname=0x2400998 \”postgres\”, username=0x23c5178 \”mingjie\”) at postgres.c:4563
#17 0x00000000008d3cfe in BackendRun (port=0x23f7220) at postmaster.c:4396
#18 0x00000000008d3697 in BackendStartup (port=0x23f7220) at postmaster.c:4124
#19 0x00000000008d00b8 in ServerLoop () at postmaster.c:1791
#20 0x00000000008cf98a in PostmasterMain (argc=1, argv=0x23c3120) at postmaster.c:1463
#21 0x00000000007ada4b in main (argc=1, argv=0x23c3120) at main.c:200

分析：

if (estate->tuple_store == NULL)
exec_init_tuple_store(estate);
tupdesc = estate->tuple_store_desc;
natts = tupdesc->natts;
if (stmt->retvarno >= 0)
{
PLpgSQL_datum *retvar = estate->datums[stmt->retvarno];
switch (retvar->dtype)
{

初始化函数exec_init_tuple_store

static void
exec_init_tuple_store(PLpgSQL_execstate *estate)
{
ReturnSetInfo *rsi = estate->rsi;
MemoryContext oldcxt;
ResourceOwner oldowner;
// 从\”SPI Proc\”切换到\”ExecutorState\”
oldcxt = MemoryContextSwitchTo(estate->tuple_store_cxt);
// 从“Portal”切换到\”Portal\”
oldowner = CurrentResourceOwner;
CurrentResourceOwner = estate->tuple_store_owner;
// 进入tuplestore_begin_heap函数
estate->tuple_store =
tuplestore_begin_heap(rsi->allowedModes & SFRM_Materialize_Random, false, work_mem);
CurrentResourceOwner = oldowner;
MemoryContextSwitchTo(oldcxt);
// 给estate添加DESC，rsi->expectedDesc的来源？
estate->tuple_store_desc = rsi->expectedDesc;
}

进入tuplestore_begin_heap

Tuplestorestate *
tuplestore_begin_heap(bool randomAccess, bool interXact, int maxKBytes)
{
// 输入false不允许随机访问、false、8192
Tuplestorestate *state;
int eflags;
// eflags = EXEC_FLAG_REWIND
eflags = randomAccess ?
(EXEC_FLAG_BACKWARD | EXEC_FLAG_REWIND) :
(EXEC_FLAG_REWIND);
// 进入tuple store模块开始初始化返回Tuplestorestate，注意他会直接拿当前的memcontext
state = tuplestore_begin_common(eflags, interXact, maxKBytes);
// 返回的Tuplestorestate状态：
// state = {status = TSS_INMEM, eflags = 2, backward = false, interXact = false,
// truncated = false, availMem = 8372200, allowedMem = 8388608, tuples = 0,
// myfile = 0x0, context = \”ExecutorState\”, resowner = \”Portal\”, copytup = 0x0,
// writetup = 0x0, readtup = 0x0, memtuples = 0x24f0d88, memtupdeleted = 0,
// memtupcount = 0, memtupsize = 2048, growmemtuples = true, readptrs = 0x24e7a70,
// activeptr = 0, readptrcount = 1, readptrsize = 8, writepos_file = 0,writepos_offset = 0}
state->copytup = copytup_heap;
state->writetup = writetup_heap;
state->readtup = readtup_heap;
return state;
}

后面根据返回值的不同，进入几个分支。

在进入前，desc已经获取到了： tupdesc = estate->tuple_store_desc; natts = tupdesc->natts;

场景一：return next返回var类型

case PLPGSQL_DTYPE_VAR:
{
PLpgSQL_var *var = (PLpgSQL_var *) retvar;
Datum retval = var->value;
bool isNull = var->isnull;
Form_pg_attribute attr = TupleDescAttr(tupdesc, 0);
if (natts != 1)
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg(\”wrong result type supplied in RETURN NEXT\”)));
// retval是一个eoh的头，后续处理需要一个1be的头（1be的data部分指向eoh）
retval = MakeExpandedObjectReadOnly(retval, isNull, var->datatype->typlen);
// 转成需要的类型
retval = exec_cast_value(estate,
retval,
&isNull,
var->datatype->typoid,
var->datatype->atttypmod,
attr->atttypid,
attr->atttypmod);
tuplestore_putvalues(estate->tuple_store, tupdesc, &retval, &isNull);
}
break;

执行tuplestore_putvalues保存元组

void
tuplestore_putvalues(Tuplestorestate *state, TupleDesc tdesc,
Datum *values, bool *isnull)
{
MinimalTuple tuple;
MemoryContext oldcxt = MemoryContextSwitchTo(state->context);
tuple = heap_form_minimal_tuple(tdesc, values, isnull);
// 记录使用了多少空间，修改state->availMem
USEMEM(state, GetMemoryChunkSpace(tuple));
tuplestore_puttuple_common(state, (void *) tuple);
MemoryContextSwitchTo(oldcxt);
}
static void
tuplestore_puttuple_common(Tuplestorestate *state, void *tuple)
{
TSReadPointer *readptr;
int i;
ResourceOwner oldowner;
state->tuples++;
switch (state->status)
{

内存态直接用数组缓存tuple，tuple使用的内存是在外层函数切换上下文申请的。

case TSS_INMEM:
readptr = state->readptrs;
for (i = 0; i < state->readptrcount; readptr++, i++)
{
if (readptr->eof_reached && i != state->activeptr)
{
readptr->eof_reached = false;
readptr->current = state->memtupcount;
}
}
if (state->memtupcount >= state->memtupsize – 1)
{
(void) grow_memtuples(state);
}
state->memtuples[state->memtupcount++] = tuple;
if (state->memtupcount < state->memtupsize && !LACKMEM(state))
return;
PrepareTempTablespaces();
oldowner = CurrentResourceOwner;
CurrentResourceOwner = state->resowner;
state->myfile = BufFileCreateTemp(state->interXact);
CurrentResourceOwner = oldowner;
state->backward = (state->eflags & EXEC_FLAG_BACKWARD) != 0;
state->status = TSS_WRITEFILE;
dumptuples(state);
break;
…

场景二：return next返回record类型

case PLPGSQL_DTYPE_REC:
{
PLpgSQL_rec *rec = (PLpgSQL_rec *) retvar;
TupleDesc rec_tupdesc;
TupleConversionMap *tupmap;

拿到record：

{dtype = PLPGSQL_DTYPE_REC, dno = 1, refname = 0x24db608 "r", lineno = 3, isconst = false, notnull = false, default_val = 0x0, datatype = {typname='foo'}, rectypeid = 17117, firstfield = -1, erh = 0x2509708}

数据和desc都在erh中，列名在firstfield指向的位置。
数据类型在datatype中：foo
数据类型oid在rectypeid中：17117->foo

if (rec->erh == NULL)
instantiate_empty_record_variable(estate, rec);
if (ExpandedRecordIsEmpty(rec->erh))
deconstruct_expanded_record(rec->erh);
// \”SPI Proc\”切到\”ExprContext\”
oldcontext = MemoryContextSwitchTo(get_eval_mcontext(estate));
// return erh->er_tupdesc;
rec_tupdesc = expanded_record_get_tupdesc(rec->erh);
// 从保存的desc：rec_tupdesc转换到输出的desc：tupdesc，第一步：生成转换map
tupmap = convert_tuples_by_position(rec_tupdesc,
tupdesc,
gettext_noop(\”wrong record type supplied in RETURN NEXT\”));
tuple = expanded_record_get_tuple(rec->erh);
if (tupmap)
// 从保存的desc：rec_tupdesc转换到输出的desc：tupdesc，第二步：用map生成转换后的元组
tuple = execute_attr_map_tuple(tuple, tupmap);
// 缓存元组
tuplestore_puttuple(estate->tuple_store, tuple);
MemoryContextSwitchTo(oldcontext);
}
break;

场景三：return next返回row类型

必须是两列以上的out参数，直接return next空，才会使用这段逻辑。

case PLPGSQL_DTYPE_ROW:
{
PLpgSQL_row *row = (PLpgSQL_row *) retvar;
oldcontext = MemoryContextSwitchTo(get_eval_mcontext(estate));
// 必须严格匹配tupdesc的类型，对不上则转换失败
tuple = make_tuple_from_row(estate, row, tupdesc);
if (tuple == NULL)
ereport(ERROR,…)
tuplestore_puttuple(estate->tuple_store, tuple);
MemoryContextSwitchTo(oldcontext);
}
break;
default:
elog(ERROR, \”unrecognized dtype: %d\”, retvar->dtype);
break;
}
}

3 用例

drop function f1;
create or replace function f1(in i int, out j int) returns setof int as $$
begin
j := i+1;
return next;
j := i+2;
return next;
return;
end$$ language plpgsql;
select * from f1(42);
—-
CREATE TABLE foo (fooid INT, foosubid INT, fooname TEXT);
INSERT INTO foo VALUES (1, 2, \’three\’);
INSERT INTO foo VALUES (4, 5, \’six\’);
CREATE OR REPLACE FUNCTION get_all_foo() RETURNS SETOF foo AS
$BODY$
DECLARE
r foo%rowtype;
BEGIN
FOR r IN
SELECT * FROM foo WHERE fooid > 0
LOOP
— can do some processing here
RETURN NEXT r; — return current row of SELECT
END LOOP;
RETURN;
END;
$BODY$
LANGUAGE plpgsql;
SELECT * FROM get_all_foo();
——–
drop function f1(int);
create function f1(in i int, out j int, out k text) returns setof record as $$
begin
j := i+1;
k := \’foo\’;
return next;
j := j+1;
k := \’foot\’;
return next;
return;
end$$ language plpgsql;
select * from f1(42);