PostgreSQL 原始碼解讀（96）- 分割槽表#3（資料插入路由#3-獲取分割槽鍵值）
摘要：
本節介紹了ExecPrepareTupleRouting->ExecFindPartition->FormPartitionKeyDatum函式，該函式獲取Tuple的分割槽鍵值。
一、資料結構
ModifyTable
通過插入、...
本節介紹了ExecPrepareTupleRouting->ExecFindPartition->FormPartitionKeyDatum函式，該函式獲取Tuple的分割槽鍵值。
一、資料結構
ModifyTable
通過插入、更新或刪除，將子計劃生成的行應用到結果表。
/* ----------------
*ModifyTable node -
*Apply rows produced by subplan(s) to result table(s),
*by inserting, updating, or deleting.
*通過插入、更新或刪除，將子計劃生成的行應用到結果表。
*
* If the originally named target table is a partitioned table, both
* nominalRelation and rootRelation contain the RT index of the partition
* root, which is not otherwise mentioned in the plan.Otherwise rootRelation
* is zero.However, nominalRelation will always be set, as it's the rel that
* EXPLAIN should claim is the INSERT/UPDATE/DELETE target.
* 如果最初命名的目標表是分割槽表，則nominalRelation和rootRelation都包含分割槽根的RT索引，計劃中沒有另外提到這個索引。
* 否則，根關係為零。但是，總是會設定名義關係，nominalRelation因為EXPLAIN應該宣告的rel是INSERT/UPDATE/DELETE目標關係。
*
* Note that rowMarks and epqParam are presumed to be valid for all the
* subplan(s); they can't contain any info that varies across subplans.
* 注意，rowMarks和epqParam被假定對所有子計劃有效;
* 它們不能包含任何在子計劃中變化的資訊。
* ----------------
*/
typedef struct ModifyTable
{
Planplan;
CmdTypeoperation;/* 操作型別;INSERT, UPDATE, or DELETE */
boolcanSetTag;/* 是否需要設定tag?do we set the command tag/es_processed? */
IndexnominalRelation;/* 用於EXPLAIN的父RT索引;Parent RT index for use of EXPLAIN */
IndexrootRelation;/* 根Root RT索引(如目標為分割槽表);Root RT index, if target is partitioned */
boolpartColsUpdated;/* 更新了層次結構中的分割槽關鍵字;some part key in hierarchy updated */
List*resultRelations;/* RT索引的整型連結串列;integer list of RT indexes */
intresultRelIndex; /* 計劃連結串列中第一個resultRel的索引;index of first resultRel in plan's list */
introotResultRelIndex; /* 分割槽表根索引;index of the partitioned table root */
List*plans;/* 生成源資料的計劃連結串列;plan(s) producing source data */
List*withCheckOptionLists;/* 每一個目標表均具備的WCO連結串列;per-target-table WCO lists */
List*returningLists; /* 每一個目標表均具備的RETURNING連結串列;per-target-table RETURNING tlists */
List*fdwPrivLists;/* 每一個目標表的FDW私有資料鏈表;per-target-table FDW private data lists */
Bitmapset*fdwDirectModifyPlans;/* FDW DM計劃索引點陣圖;indices of FDW DM plans */
List*rowMarks;/* rowMarks連結串列;PlanRowMarks (non-locking only) */
intepqParam;/* EvalPlanQual再解析使用的引數ID;ID of Param for EvalPlanQual re-eval */
OnConflictAction onConflictAction;/* ON CONFLICT action */
List*arbiterIndexes; /* 衝突仲裁器索引表;List of ON CONFLICT arbiter index OIDs*/
List*onConflictSet;/* SET for INSERT ON CONFLICT DO UPDATE */
Node*onConflictWhere;/* WHERE for ON CONFLICT UPDATE */
IndexexclRelRTI;/* RTI of the EXCLUDED pseudo relation */
List*exclRelTlist;/* 已排除偽關係的投影列連結串列;tlist of the EXCLUDED pseudo relation */
} ModifyTable;
ResultRelInfo
ResultRelInfo結構體
每當更新一個現有的關係時，我們必須更新關係上的索引，也許還需要觸發觸發器。ResultRelInfo儲存關於結果關係所需的所有信息，包括索引。
/*
* ResultRelInfo
* ResultRelInfo結構體
*
* Whenever we update an existing relation, we have to update indexes on the
* relation, and perhaps also fire triggers.ResultRelInfo holds all the
* information needed about a result relation, including indexes.
* 每當更新一個現有的關係時，我們必須更新關係上的索引，也許還需要觸發觸發器。
* ResultRelInfo儲存關於結果關係所需的所有資訊，包括索引。
*
* Normally, a ResultRelInfo refers to a table that is in the query's
* range table; then ri_RangeTableIndex is the RT index and ri_RelationDesc
* is just a copy of the relevant es_relations[] entry.But sometimes,
* in ResultRelInfos used only for triggers, ri_RangeTableIndex is zero
* and ri_RelationDesc is a separately-opened relcache pointer that needs
* to be separately closed.See ExecGetTriggerResultRel.
* 通常，ResultRelInfo是指查詢範圍表中的表;
* ri_RangeTableIndex是RT索引，而ri_RelationDesc只是相關es_relations[]條目的副本。
* 但有時，在只用於觸發器的ResultRelInfos中，ri_RangeTableIndex為零(NULL)，
*而ri_RelationDesc是一個需要單獨關閉單獨開啟的relcache指標。
*具體可參考ExecGetTriggerResultRel結構體。
*/
typedef struct ResultRelInfo
{
NodeTagtype;
/* result relation's range table index, or 0 if not in range table */
//RTE索引
Indexri_RangeTableIndex;
/* relation descriptor for result relation */
//結果/目標relation的描述符
Relationri_RelationDesc;
/* # of indices existing on result relation */
//目標關係中索引數目
intri_NumIndices;
/* array of relation descriptors for indices */
//索引的關係描述符陣列(索引視為一個relation)
RelationPtr ri_IndexRelationDescs;
/* array of key/attr info for indices */
//索引的鍵/屬性陣列
IndexInfo **ri_IndexRelationInfo;
/* triggers to be fired, if any */
//觸發的索引
TriggerDesc *ri_TrigDesc;
/* cached lookup info for trigger functions */
//觸發器函式(快取)
FmgrInfo*ri_TrigFunctions;
/* array of trigger WHEN expr states */
//WHEN表示式狀態的觸發器陣列
ExprState **ri_TrigWhenExprs;
/* optional runtime measurements for triggers */
//可選的觸發器執行期度量器
Instrumentation *ri_TrigInstrument;
/* FDW callback functions, if foreign table */
//FDW回撥函式
struct FdwRoutine *ri_FdwRoutine;
/* available to save private state of FDW */
//可用於儲存FDW的私有狀態
void*ri_FdwState;
/* true when modifying foreign table directly */
//直接更新FDW時為T
boolri_usesFdwDirectModify;
/* list of WithCheckOption's to be checked */
//WithCheckOption連結串列
List*ri_WithCheckOptions;
/* list of WithCheckOption expr states */
//WithCheckOption表示式連結串列
List*ri_WithCheckOptionExprs;
/* array of constraint-checking expr states */
//約束檢查表示式狀態陣列
ExprState **ri_ConstraintExprs;
/* for removing junk attributes from tuples */
//用於從元組中刪除junk屬性
JunkFilter *ri_junkFilter;
/* list of RETURNING expressions */
//RETURNING表示式連結串列
List*ri_returningList;
/* for computing a RETURNING list */
//用於計算RETURNING連結串列
ProjectionInfo *ri_projectReturning;
/* list of arbiter indexes to use to check conflicts */
//用於檢查衝突的仲裁器索引的列表
List*ri_onConflictArbiterIndexes;
/* ON CONFLICT evaluation state */
//ON CONFLICT解析狀態
OnConflictSetState *ri_onConflict;
/* partition check expression */
//分割槽檢查表示式連結串列
List*ri_PartitionCheck;
/* partition check expression state */
//分割槽檢查表示式狀態
ExprState*ri_PartitionCheckExpr;
/* relation descriptor for root partitioned table */
//分割槽root根表描述符
Relationri_PartitionRoot;
/* Additional information specific to partition tuple routing */
//額外的分割槽元組路由資訊
struct PartitionRoutingInfo *ri_PartitionInfo;
} ResultRelInfo;
PartitionRoutingInfo
PartitionRoutingInfo結構體
分割槽路由資訊,用於將元組路由到表分割槽的結果關係資訊。
/*
* PartitionRoutingInfo
* PartitionRoutingInfo - 分割槽路由資訊
*
* Additional result relation information specific to routing tuples to a
* table partition.
* 用於將元組路由到表分割槽的結果關係資訊。
*/
typedef struct PartitionRoutingInfo
{
/*
* Map for converting tuples in root partitioned table format into
* partition format, or NULL if no conversion is required.
* 對映，用於將根分割槽表格式的元組轉換為分割槽格式，如果不需要轉換，則轉換為NULL。
*/
TupleConversionMap *pi_RootToPartitionMap;
/*
* Map for converting tuples in partition format into the root partitioned
* table format, or NULL if no conversion is required.
* 對映，用於將分割槽格式的元組轉換為根分割槽表格式，如果不需要轉換，則轉換為NULL。
*/
TupleConversionMap *pi_PartitionToRootMap;
/*
* Slot to store tuples in partition format, or NULL when no translation
* is required between root and partition.
* 以分割槽格式儲存元組的slot.在根分割槽和分割槽之間不需要轉換時為NULL。
*/
TupleTableSlot *pi_PartitionTupleSlot;
} PartitionRoutingInfo;
TupleConversionMap
TupleConversionMap結構體,用於儲存元組轉換對映資訊.
typedef struct TupleConversionMap
{
TupleDescindesc;/* 源行型別的描述符;tupdesc for source rowtype */
TupleDescoutdesc;/* 結果行型別的描述符;tupdesc for result rowtype */
AttrNumber *attrMap;/* 輸入欄位的索引資訊,0表示NULL;indexes of input fields, or 0 for null */
Datum*invalues;/* 析構源資料的工作空間;workspace for deconstructing source */
bool*inisnull;//是否為NULL標記陣列
Datum*outvalues;/* 構造結果的工作空間;workspace for constructing result */
bool*outisnull;//null標記
} TupleConversionMap;
二、原始碼解讀
FormPartitionKeyDatum函式獲取Tuple的分割槽鍵值,返回鍵值values[]陣列和是否為null標記isnull[]陣列.
/* ----------------
*FormPartitionKeyDatum
*Construct values[] and isnull[] arrays for the partition key
*of a tuple.
*構造values[]陣列和isnull[]陣列
*
*pdPartition dispatch object of the partitioned table
*pd分割槽表的分割槽分發器(dispatch)物件
*
*slotHeap tuple from which to extract partition key
*slot從其中提前分割槽鍵的heap tuple
*
*estateexecutor state for evaluating any partition key
*expressions (must be non-NULL)
*estate解析分割槽鍵表示式(必須非NULL)的執行器狀態
*
*valuesArray of partition key Datums (output area)
*分割槽鍵Datums陣列(輸出引數)
*isnullArray of is-null indicators (output area)
*is-null標記陣列(輸出引數)
*
* the ecxt_scantuple slot of estate's per-tuple expr context must point to
* the heap tuple passed in.
* estate的per-tuple上下文的ecxt_scantuple必須指向傳入的heap tuple
* ----------------
*/
static void
FormPartitionKeyDatum(PartitionDispatch pd,
TupleTableSlot *slot,
EState *estate,
Datum *values,
bool *isnull)
{
ListCell*partexpr_item;
inti;
if (pd->key->partexprs != NIL && pd->keystate == NIL)
{
/* Check caller has set up context correctly */
//檢查呼叫者是否已正確配置記憶體上下文
Assert(estate != NULL &&
GetPerTupleExprContext(estate)->ecxt_scantuple == slot);
/* First time through, set up expression evaluation state */
//第一次進入,配置表示式解析器狀態
pd->keystate = ExecPrepareExprList(pd->key->partexprs, estate);
}
partexpr_item = list_head(pd->keystate);//獲取分割槽鍵表示式狀態
for (i = 0; i < pd->key->partnatts; i++)//迴圈遍歷分割槽鍵
{
AttrNumberkeycol = pd->key->partattrs[i];//分割槽鍵屬性編號
Datumdatum;// typedef uintptr_t Datum;sizeof(Datum) == sizeof(void *) == 4 or 8
boolisNull;//是否null
if (keycol != 0)//編號不為0
{
/* Plain column; get the value directly from the heap tuple */
//扁平列,直接從堆元組中提取值
datum = slot_getattr(slot, keycol, &isNull);
}
else
{
/* Expression; need to evaluate it */
//表示式,需要解析
if (partexpr_item == NULL)//分割槽鍵表示式狀態為NULL,報錯
elog(ERROR, "wrong number of partition key expressions");
//獲取表示式值
datum = ExecEvalExprSwitchContext((ExprState *) lfirst(partexpr_item),
GetPerTupleExprContext(estate),
&isNull);
//切換至下一個
partexpr_item = lnext(partexpr_item);
}
values[i] = datum;//賦值
isnull[i] = isNull;
}
if (partexpr_item != NULL)//引數設定有誤?報錯
elog(ERROR, "wrong number of partition key expressions");
}
/*
* slot_getattr - fetch one attribute of the slot's contents.
* slot_getattr - 提取slot中的某個屬性值
*/
static inline Datum
slot_getattr(TupleTableSlot *slot, int attnum,
bool *isnull)
{
AssertArg(attnum > 0);
if (attnum > slot->tts_nvalid)
slot_getsomeattrs(slot, attnum);
*isnull = slot->tts_isnull[attnum - 1];
return slot->tts_values[attnum - 1];
}
/*
* This function forces the entries of the slot's Datum/isnull arrays to be
* valid at least up through the attnum'th entry.
* 這個函式強制slot的Datum/isnull陣列的條目至少在attnum的第一個條目上是有效的。
*/
static inline void
slot_getsomeattrs(TupleTableSlot *slot, int attnum)
{
if (slot->tts_nvalid < attnum)
slot_getsomeattrs_int(slot, attnum);
}
/*
* slot_getsomeattrs_int - workhorse for slot_getsomeattrs()
* slot_getsomeattrs_int - slot_getsomeattrs()函式的實際實現
*/
void
slot_getsomeattrs_int(TupleTableSlot *slot, int attnum)
{
/* Check for caller errors */
//檢查呼叫者輸入引數是否有誤
Assert(slot->tts_nvalid < attnum); /* slot_getsomeattr checked */
Assert(attnum > 0);
//attnum引數判斷
if (unlikely(attnum > slot->tts_tupleDescriptor->natts))
elog(ERROR, "invalid attribute number %d", attnum);
/* Fetch as many attributes as possible from the underlying tuple. */
//從元組中獲取儘可能多的屬性。
slot->tts_ops->getsomeattrs(slot, attnum);
/*
* If the underlying tuple doesn't have enough attributes, tuple descriptor
* must have the missing attributes.
* 如果底層元組沒有足夠的屬性，那麼元組描述符必須具有缺少的屬性。
*/
if (unlikely(slot->tts_nvalid < attnum))
{
slot_getmissingattrs(slot, slot->tts_nvalid, attnum);
slot->tts_nvalid = attnum;
}
}
三、跟蹤分析
測試指令碼如下
-- Hash Partition drop table if exists t_hash_partition; create table t_hash_partition (c1 int not null,c2varchar(40),c3 varchar(40)) partition by hash(c1); create table t_hash_partition_1 partition of t_hash_partition for values with (modulus 6,remainder 0); create table t_hash_partition_2 partition of t_hash_partition for values with (modulus 6,remainder 1); create table t_hash_partition_3 partition of t_hash_partition for values with (modulus 6,remainder 2); create table t_hash_partition_4 partition of t_hash_partition for values with (modulus 6,remainder 3); create table t_hash_partition_5 partition of t_hash_partition for values with (modulus 6,remainder 4); create table t_hash_partition_6 partition of t_hash_partition for values with (modulus 6,remainder 5); insert into t_hash_partition(c1,c2,c3) VALUES(20,'HASH0','HAHS0');
啟動gdb,設定斷點
(gdb) b FormPartitionKeyDatum Breakpoint 5 at 0x6e30d2: file execPartition.c, line 1087. (gdb) b slot_getattr Breakpoint 6 at 0x489d9b: file heaptuple.c, line 1510. (gdb) c Continuing. Breakpoint 5, FormPartitionKeyDatum (pd=0x2e1bfa0, slot=0x2e1b8a0, estate=0x2e1aeb8, values=0x7fff4e2407a0, isnull=0x7fff4e240780) at execPartition.c:1087 1087if (pd->key->partexprs != NIL && pd->keystate == NIL)
迴圈,根據分割槽鍵獲取相應的鍵值
1087if (pd->key->partexprs != NIL && pd->keystate == NIL) (gdb) n 1097partexpr_item = list_head(pd->keystate); (gdb) 1098for (i = 0; i < pd->key->partnatts; i++) (gdb) 1100AttrNumberkeycol = pd->key->partattrs[i]; (gdb) 1104if (keycol != 0) (gdb) 1107datum = slot_getattr(slot, keycol, &isNull);
進入函式slot_getattr
(gdb) step Breakpoint 6, slot_getattr (slot=0x2e1b8a0, attnum=1, isnull=0x7fff4e240735) at heaptuple.c:1510 1510HeapTupletuple = slot->tts_tuple;
獲取結果,分割槽鍵值為20
... (gdb) p *isnull $31 = false (gdb) p slot->tts_values[attnum - 1] $32 = 20
返回到FormPartitionKeyDatum函式中
(gdb) n 1593} (gdb) FormPartitionKeyDatum (pd=0x2e1bfa0, slot=0x2e1b8a0, estate=0x2e1aeb8, values=0x7fff4e2407a0, isnull=0x7fff4e240780) at execPartition.c:1119 1119values[i] = datum;
完成呼叫
1119values[i] = datum; (gdb) n 1120isnull[i] = isNull; (gdb) 1098for (i = 0; i < pd->key->partnatts; i++) (gdb) 1123if (partexpr_item != NULL) (gdb) 1125} (gdb) ExecFindPartition (resultRelInfo=0x2e1b108, pd=0x2e1c5b8, slot=0x2e1b8a0, estate=0x2e1aeb8) at execPartition.c:282 282if (partdesc->nparts == 0)
DONE!
四、參考資料
PG 11.1 Source Code.
注:ofollow,noindex">doxygen 上的原始碼與PG 11.1原始碼並不一致,本節基於11.1進行分析.