/*
* Fills in the relation statistics for an append-only relation.
*
* This information is used to update the reltuples and relpages information
* in pg_class. reltuples is the same as "pg_aoseg_<oid>:tupcount"
* column and we simulate relpages by subdividing the eof value
* ("pg_aoseg_<oid>:eof") over the defined page size.
*/
void
vacuum_appendonly_fill_stats(Relation aorel, Snapshot snapshot,
BlockNumber *rel_pages, double *rel_tuples,
bool *relhasindex)
{
FileSegTotals *fstotal;
BlockNumber nblocks;
char *relname;
double num_tuples;
double totalbytes;
double eof;
int64 hidden_tupcount;
AppendOnlyVisimap visimap;
Assert(RelationIsAoRows(aorel) || RelationIsAoCols(aorel));
relname = RelationGetRelationName(aorel);
/* get updated statistics from the pg_aoseg table */
if (RelationIsAoRows(aorel))
{
fstotal = GetSegFilesTotals(aorel, snapshot);
}
else
{
Assert(RelationIsAoCols(aorel));
fstotal = GetAOCSSSegFilesTotals(aorel, snapshot);
}
/* calculate the values we care about */
eof = (double)fstotal->totalbytes;
num_tuples = (double)fstotal->totaltuples;
totalbytes = eof;
nblocks = (uint32)RelationGuessNumberOfBlocks(totalbytes);
AppendOnlyVisimap_Init(&visimap,
aorel->rd_appendonly->visimaprelid,
aorel->rd_appendonly->visimapidxid,
AccessShareLock,
snapshot);
hidden_tupcount = AppendOnlyVisimap_GetRelationHiddenTupleCount(&visimap);
num_tuples -= hidden_tupcount;
Assert(num_tuples > -1.0);
AppendOnlyVisimap_Finish(&visimap, AccessShareLock);
elogif (Debug_appendonly_print_compaction, LOG,
"Gather statistics after vacuum for append-only relation %s: "
"page count %d, tuple count %f",
relname,
nblocks, num_tuples);
*rel_pages = nblocks;
*rel_tuples = num_tuples;
*relhasindex = aorel->rd_rel->relhasindex;
ereport(elevel,
(errmsg("\"%s\": found %.0f rows in %u pages.",
relname, num_tuples, nblocks)));
pfree(fstotal);
}
/*
* Returns true if the relation has no tuples. Prepare phase of
* compaction invokes this function on each QE.
*
* Examples of empty tables:
* 1. parent of a partitioned table
* 2. table that is created but no tuples have been inserted yet
* 3. table from which all existing tuples are deleted and the table
* is vacuumed. This is a special case in which pg_aoseg_<oid> has
* non-zero number of rows but tupcount value is zero for all rows.
*/
bool
AppendOnlyCompaction_IsRelationEmpty(Relation aorel)
{
AppendOnlyEntry *aoEntry;
Relation pg_aoseg_rel;
TupleDesc pg_aoseg_dsc;
HeapTuple tuple;
HeapScanDesc aoscan;
int Anum_tupcount;
bool empty = true;
Assert(RelationIsAoRows(aorel) || RelationIsAoCols(aorel));
aoEntry = GetAppendOnlyEntry(RelationGetRelid(aorel), SnapshotNow);
pg_aoseg_rel = heap_open(aoEntry->segrelid, AccessShareLock);
pg_aoseg_dsc = RelationGetDescr(pg_aoseg_rel);
aoscan = heap_beginscan(pg_aoseg_rel, SnapshotNow, 0, NULL);
Anum_tupcount = RelationIsAoRows(aorel)? Anum_pg_aoseg_tupcount: Anum_pg_aocs_tupcount;
while ((tuple = heap_getnext(aoscan, ForwardScanDirection)) != NULL &&
empty)
{
if (0 < fastgetattr(tuple, Anum_tupcount,
pg_aoseg_dsc, NULL))
empty = false;
}
heap_endscan(aoscan);
heap_close(pg_aoseg_rel, AccessShareLock);
return empty;
}
static void gp_statistics_estimate_reltuples_relpages_ao_rows(Relation rel, float4 *reltuples, float4 *relpages)
{
FileSegTotals *fstotal;
AppendOnlyEntry *aoEntry;
AppendOnlyVisimap visimap;
int64 hidden_tupcount = 0;
/**
* Ensure that the right kind of relation with the right type of storage is passed to us.
*/
Assert(rel->rd_rel->relkind == RELKIND_RELATION);
Assert(RelationIsAoRows(rel));
fstotal = GetSegFilesTotals(rel, SnapshotNow);
Assert(fstotal);
/**
* The planner doesn't understand AO's blocks, so need this method to try to fudge up a number for
* the planner.
*/
*relpages = RelationGuessNumberOfBlocks((double)fstotal->totalbytes);
aoEntry = GetAppendOnlyEntry(RelationGetRelid(rel), SnapshotNow);
AppendOnlyVisimap_Init(&visimap, aoEntry->visimaprelid, aoEntry->visimapidxid, AccessShareLock, SnapshotNow);
hidden_tupcount = AppendOnlyVisimap_GetRelationHiddenTupleCount(&visimap);
AppendOnlyVisimap_Finish(&visimap, AccessShareLock);
/**
* The number of tuples in AO table is known accurately. Therefore, we just utilize this value.
*/
*reltuples = (double)(fstotal->totaltuples - hidden_tupcount);
pfree(fstotal);
pfree(aoEntry);
return;
}
static void LockSegfilesOnMasterForSingleRel(Relation rel, int32 segno) {
Insist(Gp_role == GP_ROLE_DISPATCH);
/*
* do not lock segfile with content id = -1
*/
/*
for (i = 1; i < rel->rd_segfile0_count; ++i)
{
if (RelationIsAoRows(rel) || RelationIsParquet(rel))
{
LockRelationAppendOnlySegmentFile(&rel->rd_node, segno,
AccessExclusiveLock, false, i - 1);
}
}
*/
{
if (RelationIsAoRows(rel) || RelationIsParquet(rel)) {
LockRelationAppendOnlySegmentFile(&rel->rd_node, segno,
AccessExclusiveLock,
false);
}
}
}
/*
* calculate size of (one fork of) a relation
*
* Iterator over all files belong to the relation and do stat.
* The obviously better way is to use glob. For whatever reason,
* glob is extremely slow if there are lots of relations in the
* database. So we handle all cases, instead.
*
* Note: we can safely apply this to temp tables of other sessions, so there
* is no check here or at the call sites for that.
*/
static int64
calculate_relation_size(Relation rel, ForkNumber forknum)
{
int64 totalsize = 0;
char *relationpath;
char pathname[MAXPGPATH];
unsigned int segcount = 0;
relationpath = relpathbackend(rel->rd_node, rel->rd_backend, forknum);
if (RelationIsHeap(rel))
{
/* Ordinary relation, including heap and index.
* They take form of relationpath, or relationpath.%d
* There will be no holes, therefore, we can stop when
* we reach the first non-existing file.
*/
for (segcount = 0;; segcount++)
{
struct stat fst;
CHECK_FOR_INTERRUPTS();
if (segcount == 0)
snprintf(pathname, MAXPGPATH, "%s",
relationpath);
else
snprintf(pathname, MAXPGPATH, "%s.%u",
relationpath, segcount);
if (stat(pathname, &fst) < 0)
{
if (errno == ENOENT)
break;
else
ereport(ERROR,
(errcode_for_file_access(),
errmsg("could not stat file %s: %m", pathname)));
}
totalsize += fst.st_size;
}
}
/* AO tables don't have any extra forks. */
else if (forknum == MAIN_FORKNUM)
{
if (RelationIsAoRows(rel))
{
totalsize = GetAOTotalBytes(rel, GetActiveSnapshot());
}
else if (RelationIsAoCols(rel))
{
totalsize = GetAOCSTotalBytes(rel, GetActiveSnapshot(), true);
}
}
/* RELSTORAGE_VIRTUAL has no space usage */
return totalsize;
}
/**
* Given the oid of a relation, this method calculates reltuples, relpages. This only looks up
* local information (on master or segments). It produces meaningful values for AO and
* heap tables and returns [0.0,0.0] for all other relations.
* Input:
* relationoid
* Output:
* array of two values [reltuples,relpages]
*/
Datum
gp_statistics_estimate_reltuples_relpages_oid(PG_FUNCTION_ARGS)
{
float4 relpages = 0.0;
float4 reltuples = 0.0;
Oid relOid = PG_GETARG_OID(0);
Datum values[2];
ArrayType *result;
Relation rel = try_relation_open(relOid, AccessShareLock, false);
if (rel != NULL)
{
if (rel->rd_rel->relkind == RELKIND_RELATION)
{
if (RelationIsHeap(rel))
{
gp_statistics_estimate_reltuples_relpages_heap(rel, &reltuples, &relpages);
}
else if (RelationIsAoRows(rel))
{
gp_statistics_estimate_reltuples_relpages_ao_rows(rel, &reltuples, &relpages);
}
else if (RelationIsAoCols(rel))
{
gp_statistics_estimate_reltuples_relpages_ao_cs(rel, &reltuples, &relpages);
}
}
else if (rel->rd_rel->relkind == RELKIND_INDEX)
{
reltuples = 1.0;
relpages = RelationGetNumberOfBlocks(rel);
}
else
{
/**
* Should we silently return [0.0,0.0] or error out? Currently, we choose option 1.
*/
}
relation_close(rel, AccessShareLock);
}
else
{
/**
* Should we silently return [0.0,0.0] or error out? Currently, we choose option 1.
*/
}
values[0] = Float4GetDatum(reltuples);
values[1] = Float4GetDatum(relpages);
result = construct_array(values, 2,
FLOAT4OID,
sizeof(float4), true, 'i');
PG_RETURN_ARRAYTYPE_P(result);
}
/*
* AppendOnlySegmentFileTruncateToEOF()
*
* Assumes that the segment file lock is already held.
*
* For the segment file is truncates to the eof.
*/
static void
AppendOnlySegmentFileTruncateToEOF(Relation aorel,
FileSegInfo *fsinfo)
{
const char* relname = RelationGetRelationName(aorel);
MirroredAppendOnlyOpen mirroredOpened;
int32 fileSegNo;
char filenamepath[MAXPGPATH];
int segno;
int64 segeof;
Assert(fsinfo);
Assert(RelationIsAoRows(aorel));
segno = fsinfo->segno;
relname = RelationGetRelationName(aorel);
segeof = (int64)fsinfo->eof;
/* Open and truncate the relation segfile beyond its eof */
MakeAOSegmentFileName(aorel, segno, -1, &fileSegNo, filenamepath);
elogif(Debug_appendonly_print_compaction, LOG,
"Opening AO relation \"%s.%s\", relation id %u, relfilenode %u (physical segment file #%d, logical EOF " INT64_FORMAT ")",
get_namespace_name(RelationGetNamespace(aorel)),
relname,
aorel->rd_id,
aorel->rd_node.relNode,
segno,
segeof);
if (OpenAOSegmentFile(aorel, filenamepath, fileSegNo, segeof, &mirroredOpened))
{
TruncateAOSegmentFile(&mirroredOpened, aorel, segeof, ERROR);
CloseAOSegmentFile(&mirroredOpened);
elogif(Debug_appendonly_print_compaction, LOG,
"Successfully truncated AO ROL relation \"%s.%s\", relation id %u, relfilenode %u (physical segment file #%d, logical EOF " INT64_FORMAT ")",
get_namespace_name(RelationGetNamespace(aorel)),
relname,
aorel->rd_id,
aorel->rd_node.relNode,
segno,
segeof);
}
else
{
elogif(Debug_appendonly_print_compaction, LOG,
"No gp_relation_node entry for AO ROW relation \"%s.%s\", relation id %u, relfilenode %u (physical segment file #%d, logical EOF " INT64_FORMAT ")",
get_namespace_name(RelationGetNamespace(aorel)),
relname,
aorel->rd_id,
aorel->rd_node.relNode,
segno,
segeof);
}
}
/*
* calculate size of a relation
*
* Iterator over all files belong to the relation and do stat.
* The obviously better way is to use glob. For whatever reason,
* glob is extremely slow if there are lots of relations in the
* database. So we handle all cases, instead.
*/
int64
calculate_relation_size(Relation rel)
{
int64 totalsize = 0;
char *relationpath;
char pathname[MAXPGPATH];
struct stat fst;
int i;
relationpath = relpath(rel->rd_node);
if(RelationIsHeap(rel))
{
/* Ordinary relation, including heap and index.
* They take form of relationpath, or relationpath.%d
* There will be no holes, therefore, we can stop we
* we reach the first non-exist file.
*/
for(i=0; ; ++i)
{
if (i==0)
snprintf(pathname, MAXPGPATH, "%s", relationpath);
else
snprintf(pathname, MAXPGPATH, "%s.%d", relationpath, i);
if (stat(pathname, &fst) >= 0)
totalsize += fst.st_size;
else
{
if (errno == ENOENT)
break;
else
ereport(ERROR, (errcode_for_file_access(),
errmsg("could not stat file %s: %m", pathname)
));
}
}
}
else if (RelationIsAoRows(rel))
totalsize = GetAOTotalBytes(rel, SnapshotNow);
else if (RelationIsParquet(rel))
totalsize = GetParquetTotalBytes(rel, SnapshotNow);
/* RELSTORAGE_VIRTUAL has no space usage */
return totalsize;
}
/*
* Has the same signature as RelationGetAttributeCompressionFuncs() even though
* we don't actually need the full Relation data structure. I deem consistency
* of API more important in this case.
*/
PGFunction *
RelationGetRelationCompressionFuncs(Relation rel)
{
AppendOnlyEntry *aoentry;
char *comptype = NULL;
PGFunction *compFuncs;
if(RelationIsAoRows(rel) || RelationIsParquet(rel)){
aoentry = GetAppendOnlyEntry(RelationGetRelid(rel), SnapshotNow);
comptype = aoentry->compresstype;
}
compFuncs = get_funcs_for_compression(comptype);
return compFuncs;
}
/*
* Compute the on-disk size of files for the relation according to the
* stat function, including heap data, index data, toast data, aoseg data,
* aoblkdir data, and aovisimap data.
*/
static int64
calculate_total_relation_size(Oid Relid)
{
Relation heapRel;
Oid toastOid;
AppendOnlyEntry *aoEntry = NULL;
int64 size;
ListCell *cell;
heapRel = try_relation_open(Relid, AccessShareLock, false);
if (!RelationIsValid(heapRel))
return 0;
toastOid = heapRel->rd_rel->reltoastrelid;
if (RelationIsAoRows(heapRel) || RelationIsAoCols(heapRel))
aoEntry = GetAppendOnlyEntry(Relid, SnapshotNow);
/* Get the heap size */
if (Relid == 0 || heapRel->rd_node.relNode == 0)
size = 0;
else
size = calculate_relation_size(heapRel);
/* Include any dependent indexes */
if (heapRel->rd_rel->relhasindex)
{
List *index_oids = RelationGetIndexList(heapRel);
foreach(cell, index_oids)
{
Oid idxOid = lfirst_oid(cell);
Relation iRel;
iRel = try_relation_open(idxOid, AccessShareLock, false);
if (RelationIsValid(iRel))
{
size += calculate_relation_size(iRel);
relation_close(iRel, AccessShareLock);
}
}
/*
* getTableType
* Return the table type for a given relation.
*/
int
getTableType(Relation rel)
{
Assert(rel != NULL && rel->rd_rel != NULL);
if (RelationIsHeap(rel))
{
return TableTypeHeap;
}
if (RelationIsAoRows(rel))
{
return TableTypeAppendOnly;
}
if (RelationIsParquet(rel))
{
return TableTypeParquet;
}
elog(ERROR, "undefined table type for storage format: %c", rel->rd_rel->relstorage);
return TableTypeInvalid;
}
/*
* InitScanStateRelationDetails
* Opens a relation and sets various relation specific ScanState fields.
*/
void
InitScanStateRelationDetails(ScanState *scanState, Plan *plan, EState *estate)
{
Assert(NULL != scanState);
PlanState *planState = &scanState->ps;
/* Initialize child expressions */
planState->targetlist = (List *)ExecInitExpr((Expr *)plan->targetlist, planState);
planState->qual = (List *)ExecInitExpr((Expr *)plan->qual, planState);
Relation currentRelation = ExecOpenScanRelation(estate, ((Scan *)plan)->scanrelid);
scanState->ss_currentRelation = currentRelation;
if (RelationIsAoRows(currentRelation) || RelationIsParquet(currentRelation))
{
scanState->splits = GetFileSplitsOfSegment(estate->es_plannedstmt->scantable_splits,
currentRelation->rd_id, GetQEIndex());
}
ExecAssignScanType(scanState, RelationGetDescr(currentRelation));
ExecAssignScanProjectionInfo(scanState);
scanState->tableType = getTableType(scanState->ss_currentRelation);
}
/*
* vacuum_appendonly_rel() -- vaccum an append-only relation
*
* This procedure will be what gets executed both for VACUUM
* and VACUUM FULL (and also ANALYZE or any other thing that
* needs the pg_class stats updated).
*
* The function can compact append-only segment files or just
* truncating the segment file to its existing eof.
*
* Afterwards, the reltuples and relpages information in pg_class
* are updated. reltuples is the same as "pg_aoseg_<oid>:tupcount"
* column and we simulate relpages by subdividing the eof value
* ("pg_aoseg_<oid>:eof") over the defined page size.
*
*
* There are txn ids, hint bits, free space, dead tuples,
* etc. these are all irrelevant in the append only relation context.
*
*/
void
vacuum_appendonly_rel(Relation aorel, VacuumStmt *vacstmt)
{
char *relname;
PGRUsage ru0;
Assert(RelationIsAoRows(aorel) || RelationIsAoCols(aorel));
Assert(!vacummStatement_IsInAppendOnlyCleanupPhase(vacstmt));
pg_rusage_init(&ru0);
relname = RelationGetRelationName(aorel);
ereport(elevel,
(errmsg("vacuuming \"%s.%s\"",
get_namespace_name(RelationGetNamespace(aorel)),
relname)));
if (Gp_role == GP_ROLE_DISPATCH)
{
return;
}
Assert(list_length(vacstmt->appendonly_compaction_insert_segno) <= 1);
if (vacstmt->appendonly_compaction_insert_segno == NULL)
{
elogif(Debug_appendonly_print_compaction, LOG,
"Vacuum drop phase %s", RelationGetRelationName(aorel));
if (RelationIsAoRows(aorel))
{
AppendOnlyDrop(aorel, vacstmt->appendonly_compaction_segno);
}
else
{
Assert(RelationIsAoCols(aorel));
AOCSDrop(aorel, vacstmt->appendonly_compaction_segno);
}
}
else
{
int insert_segno = linitial_int(vacstmt->appendonly_compaction_insert_segno);
if (insert_segno == APPENDONLY_COMPACTION_SEGNO_INVALID)
{
elogif(Debug_appendonly_print_compaction, LOG,
"Vacuum pseudo-compaction phase %s", RelationGetRelationName(aorel));
}
else
{
elogif(Debug_appendonly_print_compaction, LOG,
"Vacuum compaction phase %s", RelationGetRelationName(aorel));
if (RelationIsAoRows(aorel))
{
AppendOnlyCompact(aorel,
vacstmt->appendonly_compaction_segno,
insert_segno, vacstmt->full);
}
else
{
Assert(RelationIsAoCols(aorel));
AOCSCompact(aorel,
vacstmt->appendonly_compaction_segno,
insert_segno, vacstmt->full);
}
}
}
}
/* ----------------------------------------------------------------
* ExecInsert
*
* INSERTs have to add the tuple into
* the base relation and insert appropriate tuples into the
* index relations.
* Insert can be part of an update operation when
* there is a preceding SplitUpdate node.
* ----------------------------------------------------------------
*/
void
ExecInsert(TupleTableSlot *slot,
DestReceiver *dest,
EState *estate,
PlanGenerator planGen,
bool isUpdate)
{
void *tuple = NULL;
ResultRelInfo *resultRelInfo = NULL;
Relation resultRelationDesc = NULL;
Oid newId = InvalidOid;
TupleTableSlot *partslot = NULL;
AOTupleId aoTupleId = AOTUPLEID_INIT;
bool rel_is_heap = false;
bool rel_is_aorows = false;
bool rel_is_aocols = false;
bool rel_is_external = false;
/*
* get information on the (current) result relation
*/
if (estate->es_result_partitions)
{
resultRelInfo = slot_get_partition(slot, estate);
/* Check whether the user provided the correct leaf part only if required */
if (!dml_ignore_target_partition_check)
{
Assert(NULL != estate->es_result_partitions->part &&
NULL != resultRelInfo->ri_RelationDesc);
List *resultRelations = estate->es_plannedstmt->resultRelations;
/*
* Only inheritance can generate multiple result relations and inheritance
* is not compatible with partitions. As we are in inserting in partitioned
* table, we should not have more than one resultRelation
*/
Assert(list_length(resultRelations) == 1);
/* We only have one resultRelations entry where the user originally intended to insert */
int rteIdxForUserRel = linitial_int(resultRelations);
Assert (rteIdxForUserRel > 0);
Oid userProvidedRel = InvalidOid;
if (1 == rteIdxForUserRel)
{
/* Optimization for typical case */
userProvidedRel = ((RangeTblEntry *) estate->es_plannedstmt->rtable->head->data.ptr_value)->relid;
}
else
{
userProvidedRel = getrelid(rteIdxForUserRel, estate->es_plannedstmt->rtable);
}
/* Error out if user provides a leaf partition that does not match with our calculated partition */
if (userProvidedRel != estate->es_result_partitions->part->parrelid &&
userProvidedRel != resultRelInfo->ri_RelationDesc->rd_id)
{
ereport(ERROR,
(errcode(ERRCODE_CHECK_VIOLATION),
errmsg("Trying to insert row into wrong partition"),
errdetail("Expected partition: %s, provided partition: %s",
resultRelInfo->ri_RelationDesc->rd_rel->relname.data,
estate->es_result_relation_info->ri_RelationDesc->rd_rel->relname.data)));
}
}
estate->es_result_relation_info = resultRelInfo;
}
else
{
resultRelInfo = estate->es_result_relation_info;
}
Assert (!resultRelInfo->ri_projectReturning);
resultRelationDesc = resultRelInfo->ri_RelationDesc;
rel_is_heap = RelationIsHeap(resultRelationDesc);
rel_is_aocols = RelationIsAoCols(resultRelationDesc);
rel_is_aorows = RelationIsAoRows(resultRelationDesc);
rel_is_external = RelationIsExternal(resultRelationDesc);
partslot = reconstructMatchingTupleSlot(slot, resultRelInfo);
if (rel_is_heap)
{
tuple = ExecFetchSlotHeapTuple(partslot);
}
else if (rel_is_aorows)
{
tuple = ExecFetchSlotMemTuple(partslot, false);
}
else if (rel_is_external)
{
if (estate->es_result_partitions &&
estate->es_result_partitions->part->parrelid != 0)
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Insert into external partitions not supported.")));
return;
}
else
{
tuple = ExecFetchSlotHeapTuple(partslot);
}
}
else
{
Assert(rel_is_aocols);
tuple = ExecFetchSlotMemTuple(partslot, true);
}
Assert(partslot != NULL && tuple != NULL);
/* Execute triggers in Planner-generated plans */
if (planGen == PLANGEN_PLANNER)
{
/* BEFORE ROW INSERT Triggers */
if (resultRelInfo->ri_TrigDesc &&
resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
{
HeapTuple newtuple;
/* NYI */
if(rel_is_aocols)
elog(ERROR, "triggers are not supported on tables that use column-oriented storage");
newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
if (newtuple == NULL) /* "do nothing" */
{
return;
}
if (newtuple != tuple) /* modified by Trigger(s) */
{
/*
* Put the modified tuple into a slot for convenience of routines
* below. We assume the tuple was allocated in per-tuple memory
* context, and therefore will go away by itself. The tuple table
* slot should not try to clear it.
*/
TupleTableSlot *newslot = estate->es_trig_tuple_slot;
if (newslot->tts_tupleDescriptor != partslot->tts_tupleDescriptor)
ExecSetSlotDescriptor(newslot, partslot->tts_tupleDescriptor);
ExecStoreGenericTuple(newtuple, newslot, false);
newslot->tts_tableOid = partslot->tts_tableOid; /* for constraints */
tuple = newtuple;
partslot = newslot;
}
}
}
/*
* Check the constraints of the tuple
*/
if (resultRelationDesc->rd_att->constr &&
planGen == PLANGEN_PLANNER)
{
ExecConstraints(resultRelInfo, partslot, estate);
}
/*
* insert the tuple
*
* Note: heap_insert returns the tid (location) of the new tuple in the
* t_self field.
*
* NOTE: for append-only relations we use the append-only access methods.
*/
if (rel_is_aorows)
{
if (resultRelInfo->ri_aoInsertDesc == NULL)
{
/* Set the pre-assigned fileseg number to insert into */
ResultRelInfoSetSegno(resultRelInfo, estate->es_result_aosegnos);
resultRelInfo->ri_aoInsertDesc =
appendonly_insert_init(resultRelationDesc,
ActiveSnapshot,
resultRelInfo->ri_aosegno,
false);
}
appendonly_insert(resultRelInfo->ri_aoInsertDesc, tuple, &newId, &aoTupleId);
}
else if (rel_is_aocols)
{
if (resultRelInfo->ri_aocsInsertDesc == NULL)
{
ResultRelInfoSetSegno(resultRelInfo, estate->es_result_aosegnos);
resultRelInfo->ri_aocsInsertDesc = aocs_insert_init(resultRelationDesc,
resultRelInfo->ri_aosegno, false);
}
newId = aocs_insert(resultRelInfo->ri_aocsInsertDesc, partslot);
aoTupleId = *((AOTupleId*)slot_get_ctid(partslot));
}
else if (rel_is_external)
{
/* Writable external table */
if (resultRelInfo->ri_extInsertDesc == NULL)
resultRelInfo->ri_extInsertDesc = external_insert_init(resultRelationDesc);
newId = external_insert(resultRelInfo->ri_extInsertDesc, tuple);
}
else
{
Insist(rel_is_heap);
newId = heap_insert(resultRelationDesc,
tuple,
estate->es_snapshot->curcid,
true, true, GetCurrentTransactionId());
}
IncrAppended();
(estate->es_processed)++;
(resultRelInfo->ri_aoprocessed)++;
estate->es_lastoid = newId;
partslot->tts_tableOid = RelationGetRelid(resultRelationDesc);
if (rel_is_aorows || rel_is_aocols)
{
/*
* insert index entries for AO Row-Store tuple
*/
if (resultRelInfo->ri_NumIndices > 0)
ExecInsertIndexTuples(partslot, (ItemPointer)&aoTupleId, estate, false);
}
else
{
/* Use parttuple for index update in case this is an indexed heap table. */
TupleTableSlot *xslot = partslot;
void *xtuple = tuple;
setLastTid(&(((HeapTuple) xtuple)->t_self));
/*
* insert index entries for tuple
*/
if (resultRelInfo->ri_NumIndices > 0)
ExecInsertIndexTuples(xslot, &(((HeapTuple) xtuple)->t_self), estate, false);
}
if (planGen == PLANGEN_PLANNER)
{
/* AFTER ROW INSERT Triggers */
ExecARInsertTriggers(estate, resultRelInfo, tuple);
}
}
void
AlterTableCreateAoBlkdirTable(Oid relOid, bool is_part_child)
{
Relation rel;
TupleDesc tupdesc;
IndexInfo *indexInfo;
Oid classObjectId[3];
int16 coloptions[3];
/*
* Grab an exclusive lock on the target table, which we will NOT release
* until end of transaction. (This is probably redundant in all present
* uses...)
*/
if (is_part_child)
rel = heap_open(relOid, NoLock);
else
rel = heap_open(relOid, AccessExclusiveLock);
if (!RelationIsAoRows(rel) && !RelationIsAoCols(rel)) {
heap_close(rel, NoLock);
return;
}
/* Create a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(4, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"segno",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"columngroup_no",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"first_row_no",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 4,
"minipage",
VARBITOID,
-1, 0);
/*
* We don't want any toast columns here.
*/
tupdesc->attrs[0]->attstorage = 'p';
tupdesc->attrs[1]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
/* TODO (dmeister): In the next line, the index should have been 3.
* Therefore the minipage might be toasted.
*/
tupdesc->attrs[2]->attstorage = 'p';
/*
* Create index on segno, first_row_no.
*/
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = 3;
indexInfo->ii_KeyAttrNumbers[0] = 1;
indexInfo->ii_KeyAttrNumbers[1] = 2;
indexInfo->ii_KeyAttrNumbers[2] = 3;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = NIL;
indexInfo->ii_PredicateState = NIL;
indexInfo->ii_Unique = true;
indexInfo->ii_Concurrent = false;
classObjectId[0] = INT4_BTREE_OPS_OID;
classObjectId[1] = INT4_BTREE_OPS_OID;
classObjectId[2] = INT8_BTREE_OPS_OID;
coloptions[0] = 0;
coloptions[1] = 0;
coloptions[2] = 0;
(void) CreateAOAuxiliaryTable(rel,
"pg_aoblkdir",
RELKIND_AOBLOCKDIR,
tupdesc, indexInfo, classObjectId, coloptions);
heap_close(rel, NoLock);
}
/* ----------------------------------------------------------------
* ExecDelete
*
* DELETE is like UPDATE, except that we delete the tuple and no
* index modifications are needed.
* DELETE can be part of an update operation when
* there is a preceding SplitUpdate node.
*
* ----------------------------------------------------------------
*/
void
ExecDelete(ItemPointer tupleid,
TupleTableSlot *planSlot,
DestReceiver *dest,
EState *estate,
PlanGenerator planGen,
bool isUpdate)
{
ResultRelInfo *resultRelInfo;
Relation resultRelationDesc;
HTSU_Result result;
ItemPointerData update_ctid;
TransactionId update_xmax;
/*
* Get information on the (current) result relation.
*/
if (estate->es_result_partitions && planGen == PLANGEN_OPTIMIZER)
{
Assert(estate->es_result_partitions->part->parrelid);
#ifdef USE_ASSERT_CHECKING
Oid parent = estate->es_result_partitions->part->parrelid;
#endif
/* Obtain part for current tuple. */
resultRelInfo = slot_get_partition(planSlot, estate);
estate->es_result_relation_info = resultRelInfo;
#ifdef USE_ASSERT_CHECKING
Oid part = RelationGetRelid(resultRelInfo->ri_RelationDesc);
#endif
Assert(parent != part);
}
else
{
resultRelInfo = estate->es_result_relation_info;
}
resultRelationDesc = resultRelInfo->ri_RelationDesc;
Assert (!resultRelInfo->ri_projectReturning);
if (planGen == PLANGEN_PLANNER)
{
/* BEFORE ROW DELETE Triggers */
if (resultRelInfo->ri_TrigDesc &&
resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_DELETE] > 0)
{
bool dodelete;
dodelete = ExecBRDeleteTriggers(estate, resultRelInfo, tupleid,
estate->es_snapshot->curcid);
if (!dodelete) /* "do nothing" */
return;
}
}
bool isHeapTable = RelationIsHeap(resultRelationDesc);
bool isAORowsTable = RelationIsAoRows(resultRelationDesc);
bool isAOColsTable = RelationIsAoCols(resultRelationDesc);
bool isExternalTable = RelationIsExternal(resultRelationDesc);
if (isExternalTable && estate->es_result_partitions &&
estate->es_result_partitions->part->parrelid != 0)
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Delete from external partitions not supported.")));
return;
}
/*
* delete the tuple
*
* Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
* the row to be deleted is visible to that snapshot, and throw a can't-
* serialize error if not. This is a special-case behavior needed for
* referential integrity updates in serializable transactions.
*/
ldelete:;
if (isHeapTable)
{
result = heap_delete(resultRelationDesc, tupleid,
&update_ctid, &update_xmax,
estate->es_snapshot->curcid,
estate->es_crosscheck_snapshot,
true /* wait for commit */ );
}
else if (isAORowsTable)
{
if (IsXactIsoLevelSerializable)
{
if (!isUpdate)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Deletes on append-only tables are not supported in serializable transactions.")));
else
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Updates on append-only tables are not supported in serializable transactions.")));
}
if (resultRelInfo->ri_deleteDesc == NULL)
{
resultRelInfo->ri_deleteDesc =
appendonly_delete_init(resultRelationDesc, ActiveSnapshot);
}
AOTupleId* aoTupleId = (AOTupleId*)tupleid;
result = appendonly_delete(resultRelInfo->ri_deleteDesc, aoTupleId);
}
else if (isAOColsTable)
{
if (IsXactIsoLevelSerializable)
{
if (!isUpdate)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Deletes on append-only tables are not supported in serializable transactions.")));
else
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Updates on append-only tables are not supported in serializable transactions.")));
}
if (resultRelInfo->ri_deleteDesc == NULL)
{
resultRelInfo->ri_deleteDesc =
aocs_delete_init(resultRelationDesc);
}
AOTupleId* aoTupleId = (AOTupleId*)tupleid;
result = aocs_delete(resultRelInfo->ri_deleteDesc, aoTupleId);
}
else
{
Insist(0);
}
switch (result)
{
case HeapTupleSelfUpdated:
/* already deleted by self; nothing to do */
/*
* In an scenario in which R(a,b) and S(a,b) have
* R S
* ________ ________
* (1, 1) (1, 2)
* (1, 7)
*
* An update query such as:
* UPDATE R SET a = S.b FROM S WHERE R.b = S.a;
*
* will have an non-deterministic output. The tuple in R
* can be updated to (2,1) or (7,1).
* Since the introduction of SplitUpdate, these queries will
* send multiple requests to delete the same tuple. Therefore,
* in order to avoid a non-deterministic output,
* an error is reported in such scenario.
*/
if (isUpdate)
{
ereport(ERROR,
(errcode(ERRCODE_IN_FAILED_SQL_TRANSACTION ),
errmsg("multiple updates to a row by the same query is not allowed")));
}
return;
case HeapTupleMayBeUpdated:
break;
case HeapTupleUpdated:
if (IsXactIsoLevelSerializable)
ereport(ERROR,
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("could not serialize access due to concurrent update")));
else if (!ItemPointerEquals(tupleid, &update_ctid))
{
TupleTableSlot *epqslot;
epqslot = EvalPlanQual(estate,
resultRelInfo->ri_RangeTableIndex,
&update_ctid,
update_xmax,
estate->es_snapshot->curcid);
if (!TupIsNull(epqslot))
{
*tupleid = update_ctid;
goto ldelete;
}
}
/* tuple already deleted; nothing to do */
return;
default:
elog(ERROR, "unrecognized heap_delete status: %u", result);
return;
}
if (!isUpdate)
{
IncrDeleted();
(estate->es_processed)++;
/*
* To notify master if tuples deleted or not, to update mod_count.
*/
(resultRelInfo->ri_aoprocessed)++;
}
/*
* Note: Normally one would think that we have to delete index tuples
* associated with the heap tuple now...
*
* ... but in POSTGRES, we have no need to do this because VACUUM will
* take care of it later. We can't delete index tuples immediately
* anyway, since the tuple is still visible to other transactions.
*/
if (planGen == PLANGEN_PLANNER)
{
/* AFTER ROW DELETE Triggers */
ExecARDeleteTriggers(estate, resultRelInfo, tupleid);
}
}
/*
* Returns true iff the given segment file should be compacted.
*/
bool
AppendOnlyCompaction_ShouldCompact(
Relation aoRelation,
AppendOnlyEntry *aoEntry,
int segno,
int64 segmentTotalTupcount,
bool isFull)
{
bool result;
AppendOnlyVisimap visiMap;
int64 hiddenTupcount;
int hideRatio;
Assert(RelationIsAoRows(aoRelation) || RelationIsAoCols(aoRelation));
if (!gp_appendonly_compaction)
{
ereport(LOG,
(errmsg("Append-only compaction skipped on relation %s, segment file num %d",
RelationGetRelationName(aoRelation),
segno),
errdetail("Compaction is disabled")));
/* Disable compaction by global guc. */
return false;
}
AppendOnlyVisimap_Init(&visiMap,
aoEntry->visimaprelid,
aoEntry->visimapidxid,
ShareLock,
SnapshotNow);
hiddenTupcount = AppendOnlyVisimap_GetSegmentFileHiddenTupleCount(
&visiMap, segno);
result = true;
if (isFull && hiddenTupcount > 0)
{
/*
* if it is a full vacuum and there is any obsolete data, do a compaction
*/
result = true;
}
else
{
hideRatio = AppendOnlyCompaction_GetHideRatio(hiddenTupcount, segmentTotalTupcount);
if (hideRatio <= gp_appendonly_compaction_threshold || gp_appendonly_compaction_threshold == 0)
{
if (hiddenTupcount > 0)
{
ereportif(Debug_appendonly_print_compaction, LOG,
(errmsg("Append-only compaction skipped on relation %s, segment file num %d, "
"hidden tupcount " INT64_FORMAT ", total tupcount " INT64_FORMAT ", "
"hide ratio %d%%, threshold %d%%",
RelationGetRelationName(aoRelation),
segno,
hiddenTupcount, segmentTotalTupcount,
hideRatio, gp_appendonly_compaction_threshold)));
ereport(LOG,
(errmsg("Append-only compaction skipped on relation %s, segment file num %d",
RelationGetRelationName(aoRelation),
segno),
errdetail("Ratio of obsolete tuples below threshold (%d%% vs %d%%)",
hideRatio, gp_appendonly_compaction_threshold)));
}
else
{
ereportif(Debug_appendonly_print_compaction, LOG,
(errmsg("Append-only compaction skipped on relation %s, segment file num %d, "
"hidden tupcount " INT64_FORMAT ", total tupcount " INT64_FORMAT ", "
"hide ratio %d%%, threshold %d%%",
RelationGetRelationName(aoRelation),
segno,
hiddenTupcount, segmentTotalTupcount,
hideRatio, gp_appendonly_compaction_threshold)));
}
result = false;
}
elogif(Debug_appendonly_print_compaction, LOG,
"Schedule compaction: "
"segno %d, "
"hidden tupcount " INT64_FORMAT ", total tupcount " INT64_FORMAT ", "
"hide ratio %d%%, threshold %d%%",
segno,
hiddenTupcount, segmentTotalTupcount,
hideRatio, gp_appendonly_compaction_threshold);
}
AppendOnlyVisimap_Finish(&visiMap, ShareLock);
return result;
}
/*
* Create append-only auxiliary relations for target relation rel.
* Returns true if they are newly created. If pg_appendonly has already
* known those tables, don't create them and returns false.
*/
bool
CreateAOAuxiliaryTable(
Relation rel,
const char *auxiliaryNamePrefix,
char relkind,
TupleDesc tupledesc,
IndexInfo *indexInfo,
Oid *classObjectId,
int16 *coloptions)
{
char aoauxiliary_relname[NAMEDATALEN];
char aoauxiliary_idxname[NAMEDATALEN];
bool shared_relation;
Oid relOid, aoauxiliary_relid = InvalidOid;
Oid aoauxiliary_idxid = InvalidOid;
ObjectAddress baseobject;
ObjectAddress aoauxiliaryobject;
Assert(RelationIsValid(rel));
Assert(RelationIsAoRows(rel) || RelationIsAoCols(rel));
Assert(auxiliaryNamePrefix);
Assert(tupledesc);
Assert(classObjectId);
if (relkind != RELKIND_AOSEGMENTS)
Assert(indexInfo);
shared_relation = rel->rd_rel->relisshared;
/*
* We cannot allow creating an auxiliary table for a shared relation
* after initdb (because there's no way to let other databases know
* this visibility map.
*/
if (shared_relation && !IsBootstrapProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared tables cannot have append-only auxiliary relations after initdb")));
relOid = RelationGetRelid(rel);
switch(relkind)
{
case RELKIND_AOVISIMAP:
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow, NULL,
NULL, NULL, &aoauxiliary_relid, &aoauxiliary_idxid);
break;
case RELKIND_AOBLOCKDIR:
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow, NULL,
&aoauxiliary_relid, &aoauxiliary_idxid, NULL, NULL);
break;
case RELKIND_AOSEGMENTS:
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow,
&aoauxiliary_relid,
NULL, NULL, NULL, NULL);
break;
default:
elog(ERROR, "unsupported auxiliary relkind '%c'", relkind);
}
/*
* Does it have the auxiliary relation?
*/
if (OidIsValid(aoauxiliary_relid))
{
return false;
}
snprintf(aoauxiliary_relname, sizeof(aoauxiliary_relname),
"%s_%u", auxiliaryNamePrefix, relOid);
snprintf(aoauxiliary_idxname, sizeof(aoauxiliary_idxname),
"%s_%u_index", auxiliaryNamePrefix, relOid);
/*
* We place auxiliary relation in the pg_aoseg namespace
* even if its master relation is a temp table. There cannot be
* any naming collision, and the auxiliary relation will be
* destroyed when its master is, so there is no need to handle
* the aovisimap relation as temp.
*/
aoauxiliary_relid = heap_create_with_catalog(aoauxiliary_relname,
PG_AOSEGMENT_NAMESPACE,
rel->rd_rel->reltablespace,
InvalidOid,
rel->rd_rel->relowner,
tupledesc,
/* relam */ InvalidOid,
relkind,
RELSTORAGE_HEAP,
shared_relation,
true,
/* bufferPoolBulkLoad */ false,
0,
ONCOMMIT_NOOP,
NULL, /* GP Policy */
(Datum) 0,
true,
/* valid_opts */ false,
/* persistentTid */ NULL,
/* persistentSerialNum */ NULL);
/* Make this table visible, else index creation will fail */
CommandCounterIncrement();
/* Create an index on AO auxiliary tables (like visimap) except for pg_aoseg table */
if (relkind != RELKIND_AOSEGMENTS)
{
aoauxiliary_idxid = index_create(aoauxiliary_relid,
aoauxiliary_idxname,
InvalidOid,
indexInfo,
BTREE_AM_OID,
rel->rd_rel->reltablespace,
classObjectId, coloptions, (Datum) 0,
true, false, true, false,
false, NULL);
/* Unlock target table -- no one can see it */
UnlockRelationOid(aoauxiliary_relid, ShareLock);
/* Unlock the index -- no one can see it anyway */
UnlockRelationOid(aoauxiliary_idxid, AccessExclusiveLock);
}
/*
* Store the auxiliary table's OID in the parent relation's pg_appendonly row.
* TODO (How to generalize this?)
*/
switch (relkind)
{
case RELKIND_AOVISIMAP:
UpdateAppendOnlyEntryAuxOids(relOid, InvalidOid,
InvalidOid, InvalidOid,
aoauxiliary_relid, aoauxiliary_idxid);
break;
case RELKIND_AOBLOCKDIR:
UpdateAppendOnlyEntryAuxOids(relOid, InvalidOid,
aoauxiliary_relid, aoauxiliary_idxid,
InvalidOid, InvalidOid);
break;
case RELKIND_AOSEGMENTS:
UpdateAppendOnlyEntryAuxOids(relOid,
aoauxiliary_relid,
InvalidOid, InvalidOid,
InvalidOid, InvalidOid);
break;
default:
elog(ERROR, "unsupported auxiliary relkind '%c'", relkind);
}
/*
* Register dependency from the auxiliary table to the master, so that the
* aoseg table will be deleted if the master is.
*/
baseobject.classId = RelationRelationId;
baseobject.objectId = relOid;
baseobject.objectSubId = 0;
aoauxiliaryobject.classId = RelationRelationId;
aoauxiliaryobject.objectId = aoauxiliary_relid;
aoauxiliaryobject.objectSubId = 0;
recordDependencyOn(&aoauxiliaryobject, &baseobject, DEPENDENCY_INTERNAL);
/*
* Make changes visible
*/
CommandCounterIncrement();
return true;
}
void
AlterTableCreateAoVisimapTable(Oid relOid, bool is_part_child)
{
Relation rel;
IndexInfo *indexInfo;
TupleDesc tupdesc;
Oid classObjectId[2];
int16 coloptions[2];
elogif(Debug_appendonly_print_visimap, LOG,
"Create visimap for relation %d",
relOid);
/*
* Grab an exclusive lock on the target table, which we will NOT release
* until end of transaction. (This is probably redundant in all present
* uses...)
*/
if (is_part_child)
rel = heap_open(relOid, NoLock);
else
rel = heap_open(relOid, AccessExclusiveLock);
if (!RelationIsAoRows(rel) && !RelationIsAoCols(rel))
{
heap_close(rel, NoLock);
return;
}
/* Create a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(Natts_pg_aovisimap, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"segno",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"first_row_no",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"visimap",
BYTEAOID,
-1, 0);
/*
* We don't want any toast columns here.
*/
tupdesc->attrs[0]->attstorage = 'p';
tupdesc->attrs[1]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
/*
* Create index on segno, first_row_no.
*/
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = 2;
indexInfo->ii_KeyAttrNumbers[0] = 1;
indexInfo->ii_KeyAttrNumbers[1] = 2;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = NIL;
indexInfo->ii_PredicateState = NIL;
indexInfo->ii_Unique = true;
indexInfo->ii_Concurrent = false;
classObjectId[0] = INT4_BTREE_OPS_OID;
classObjectId[1] = INT8_BTREE_OPS_OID;
coloptions[0] = 0;
coloptions[1] = 0;
(void) CreateAOAuxiliaryTable(rel,
"pg_aovisimap",
RELKIND_AOVISIMAP,
tupdesc, indexInfo, classObjectId, coloptions);
heap_close(rel, NoLock);
}
/*
* lazy_vacuum_rel() -- perform LAZY VACUUM for one heap relation
*
* This routine vacuums a single heap, cleans out its indexes, and
* updates its relpages and reltuples statistics.
*
* At entry, we have already established a transaction and opened
* and locked the relation.
*/
void
lazy_vacuum_rel(Relation onerel, VacuumStmt *vacstmt, List *updated_stats)
{
LVRelStats *vacrelstats;
Relation *Irel;
int nindexes;
BlockNumber possibly_freeable;
if (vacstmt->verbose)
elevel = INFO;
else
elevel = DEBUG2;
if (Gp_role == GP_ROLE_DISPATCH)
elevel = DEBUG2; /* vacuum and analyze messages aren't interesting from the QD */
#ifdef FAULT_INJECTOR
if (vacuumStatement_IsInAppendOnlyDropPhase(vacstmt))
{
FaultInjector_InjectFaultIfSet(
CompactionBeforeSegmentFileDropPhase,
DDLNotSpecified,
"", // databaseName
""); // tableName
}
if (vacummStatement_IsInAppendOnlyCleanupPhase(vacstmt))
{
FaultInjector_InjectFaultIfSet(
CompactionBeforeCleanupPhase,
DDLNotSpecified,
"", // databaseName
""); // tableName
}
#endif
/*
* MPP-23647. Update xid limits for heap as well as appendonly
* relations. This allows setting relfrozenxid to correct value
* for an appendonly (AO/CO) table.
*/
vacuum_set_xid_limits(vacstmt, onerel->rd_rel->relisshared,
&OldestXmin, &FreezeLimit);
/*
* Execute the various vacuum operations. Appendonly tables are treated
* differently.
*/
if (RelationIsAoRows(onerel) || RelationIsAoCols(onerel))
{
lazy_vacuum_aorel(onerel, vacstmt, updated_stats);
return;
}
vacrelstats = (LVRelStats *) palloc0(sizeof(LVRelStats));
/* heap relation */
/* Set threshold for interesting free space = average request size */
/* XXX should we scale it up or down? Adjust vacuum.c too, if so */
vacrelstats->threshold = GetAvgFSMRequestSize(&onerel->rd_node);
/* Open all indexes of the relation */
vac_open_indexes(onerel, RowExclusiveLock, &nindexes, &Irel);
vacrelstats->hasindex = (nindexes > 0);
/* Do the vacuuming */
lazy_scan_heap(onerel, vacrelstats, Irel, nindexes, updated_stats, vacstmt->extra_oids);
/* Done with indexes */
vac_close_indexes(nindexes, Irel, NoLock);
/*
* Optionally truncate the relation.
*
* Don't even think about it unless we have a shot at releasing a goodly
* number of pages. Otherwise, the time taken isn't worth it.
*/
possibly_freeable = vacrelstats->rel_pages - vacrelstats->nonempty_pages;
if (possibly_freeable >= REL_TRUNCATE_MINIMUM ||
possibly_freeable >= vacrelstats->rel_pages / REL_TRUNCATE_FRACTION)
lazy_truncate_heap(onerel, vacrelstats);
/* Update shared free space map with final free space info */
lazy_update_fsm(onerel, vacrelstats);
/* Update statistics in pg_class */
vac_update_relstats(onerel,
vacrelstats->rel_pages,
vacrelstats->rel_tuples,
vacrelstats->hasindex,
FreezeLimit,
updated_stats);
/* report results to the stats collector, too */
pgstat_report_vacuum(RelationGetRelid(onerel), onerel->rd_rel->relisshared,
true /*vacrelstats->scanned_all*/,
vacstmt->analyze, vacrelstats->rel_tuples);
}
/* ----------------------------------------------------------------
* ExecUpdate
*
* note: we can't run UPDATE queries with transactions
* off because UPDATEs are actually INSERTs and our
* scan will mistakenly loop forever, updating the tuple
* it just inserted.. This should be fixed but until it
* is, we don't want to get stuck in an infinite loop
* which corrupts your database..
* ----------------------------------------------------------------
*/
void
ExecUpdate(TupleTableSlot *slot,
ItemPointer tupleid,
TupleTableSlot *planSlot,
DestReceiver *dest,
EState *estate)
{
void* tuple;
ResultRelInfo *resultRelInfo;
Relation resultRelationDesc;
HTSU_Result result;
ItemPointerData update_ctid;
TransactionId update_xmax;
AOTupleId aoTupleId = AOTUPLEID_INIT;
TupleTableSlot *partslot = NULL;
/*
* abort the operation if not running transactions
*/
if (IsBootstrapProcessingMode())
elog(ERROR, "cannot UPDATE during bootstrap");
/*
* get information on the (current) result relation
*/
resultRelInfo = estate->es_result_relation_info;
resultRelationDesc = resultRelInfo->ri_RelationDesc;
bool rel_is_heap = RelationIsHeap(resultRelationDesc);
bool rel_is_aorows = RelationIsAoRows(resultRelationDesc);
bool rel_is_aocols = RelationIsAoCols(resultRelationDesc);
bool rel_is_external = RelationIsExternal(resultRelationDesc);
/*
* get the heap tuple out of the tuple table slot, making sure we have a
* writable copy
*/
if (rel_is_heap)
{
partslot = slot;
tuple = ExecFetchSlotHeapTuple(partslot);
}
else if (rel_is_aorows || rel_is_aocols)
{
/*
* It is necessary to reconstruct a logically compatible tuple to
* a phyiscally compatible tuple. The slot's tuple descriptor comes
* from the projection target list, which doesn't indicate dropped
* columns, and MemTuple cannot deal with cases without converting
* the target list back into the original relation's tuple desc.
*/
partslot = reconstructMatchingTupleSlot(slot, resultRelInfo);
/*
* We directly inline toasted columns here as update with toasted columns
* would create two references to the same toasted value.
*/
tuple = ExecFetchSlotMemTuple(partslot, true);
}
else if (rel_is_external)
{
if (estate->es_result_partitions &&
estate->es_result_partitions->part->parrelid != 0)
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Update external partitions not supported.")));
return;
}
else
{
partslot = slot;
tuple = ExecFetchSlotHeapTuple(partslot);
}
}
else
{
Insist(false);
}
/* see if this update would move the tuple to a different partition */
if (estate->es_result_partitions)
checkPartitionUpdate(estate, partslot, resultRelInfo);
/* BEFORE ROW UPDATE Triggers */
if (resultRelInfo->ri_TrigDesc &&
resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_UPDATE] > 0)
{
HeapTuple newtuple;
newtuple = ExecBRUpdateTriggers(estate, resultRelInfo,
tupleid, tuple,
estate->es_snapshot->curcid);
if (newtuple == NULL) /* "do nothing" */
return;
if (newtuple != tuple) /* modified by Trigger(s) */
{
/*
* Put the modified tuple into a slot for convenience of routines
* below. We assume the tuple was allocated in per-tuple memory
* context, and therefore will go away by itself. The tuple table
* slot should not try to clear it.
*/
TupleTableSlot *newslot = estate->es_trig_tuple_slot;
if (newslot->tts_tupleDescriptor != partslot->tts_tupleDescriptor)
ExecSetSlotDescriptor(newslot, partslot->tts_tupleDescriptor);
ExecStoreGenericTuple(newtuple, newslot, false);
newslot->tts_tableOid = partslot->tts_tableOid; /* for constraints */
partslot = newslot;
tuple = newtuple;
}
}
/*
* Check the constraints of the tuple
*
* If we generate a new candidate tuple after EvalPlanQual testing, we
* must loop back here and recheck constraints. (We don't need to redo
* triggers, however. If there are any BEFORE triggers then trigger.c
* will have done heap_lock_tuple to lock the correct tuple, so there's no
* need to do them again.)
*/
lreplace:;
if (resultRelationDesc->rd_att->constr)
ExecConstraints(resultRelInfo, partslot, estate);
if (!GpPersistent_IsPersistentRelation(resultRelationDesc->rd_id))
{
/*
* Normal UPDATE path.
*/
/*
* replace the heap tuple
*
* Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
* the row to be updated is visible to that snapshot, and throw a can't-
* serialize error if not. This is a special-case behavior needed for
* referential integrity updates in serializable transactions.
*/
if (rel_is_heap)
{
result = heap_update(resultRelationDesc, tupleid, tuple,
&update_ctid, &update_xmax,
estate->es_snapshot->curcid,
estate->es_crosscheck_snapshot,
true /* wait for commit */ );
}
else if (rel_is_aorows)
{
if (IsXactIsoLevelSerializable)
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Updates on append-only tables are not supported in serializable transactions.")));
}
if (resultRelInfo->ri_updateDesc == NULL)
{
ResultRelInfoSetSegno(resultRelInfo, estate->es_result_aosegnos);
resultRelInfo->ri_updateDesc = (AppendOnlyUpdateDesc)
appendonly_update_init(resultRelationDesc, ActiveSnapshot, resultRelInfo->ri_aosegno);
}
result = appendonly_update(resultRelInfo->ri_updateDesc,
tuple, (AOTupleId *) tupleid, &aoTupleId);
}
else if (rel_is_aocols)
{
if (IsXactIsoLevelSerializable)
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Updates on append-only tables are not supported in serializable transactions.")));
}
if (resultRelInfo->ri_updateDesc == NULL)
{
ResultRelInfoSetSegno(resultRelInfo, estate->es_result_aosegnos);
resultRelInfo->ri_updateDesc = (AppendOnlyUpdateDesc)
aocs_update_init(resultRelationDesc, resultRelInfo->ri_aosegno);
}
result = aocs_update(resultRelInfo->ri_updateDesc,
partslot, (AOTupleId *) tupleid, &aoTupleId);
}
else
{
Assert(!"We should not be here");
}
switch (result)
{
case HeapTupleSelfUpdated:
/* already deleted by self; nothing to do */
return;
case HeapTupleMayBeUpdated:
break;
case HeapTupleUpdated:
if (IsXactIsoLevelSerializable)
ereport(ERROR,
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("could not serialize access due to concurrent update")));
else if (!ItemPointerEquals(tupleid, &update_ctid))
{
TupleTableSlot *epqslot;
epqslot = EvalPlanQual(estate,
resultRelInfo->ri_RangeTableIndex,
&update_ctid,
update_xmax,
estate->es_snapshot->curcid);
if (!TupIsNull(epqslot))
{
*tupleid = update_ctid;
partslot = ExecFilterJunk(estate->es_junkFilter, epqslot);
tuple = ExecFetchSlotHeapTuple(partslot);
goto lreplace;
}
}
/* tuple already deleted; nothing to do */
return;
default:
elog(ERROR, "unrecognized heap_update status: %u", result);
return;
}
}
else
{
HeapTuple persistentTuple;
/*
* Persistent metadata path.
*/
persistentTuple = heap_copytuple(tuple);
persistentTuple->t_self = *tupleid;
frozen_heap_inplace_update(resultRelationDesc, persistentTuple);
heap_freetuple(persistentTuple);
}
IncrReplaced();
(estate->es_processed)++;
(resultRelInfo->ri_aoprocessed)++;
/*
* Note: instead of having to update the old index tuples associated with
* the heap tuple, all we do is form and insert new index tuples. This is
* because UPDATEs are actually DELETEs and INSERTs, and index tuple
* deletion is done later by VACUUM (see notes in ExecDelete). All we do
* here is insert new index tuples. -cim 9/27/89
*/
/*
* insert index entries for tuple
*
* Note: heap_update returns the tid (location) of the new tuple in the
* t_self field.
*/
if (rel_is_aorows || rel_is_aocols)
{
if (resultRelInfo->ri_NumIndices > 0)
ExecInsertIndexTuples(partslot, (ItemPointer)&aoTupleId, estate, false);
}
else
{
if (resultRelInfo->ri_NumIndices > 0)
ExecInsertIndexTuples(partslot, &(((HeapTuple) tuple)->t_self), estate, false);
}
/* AFTER ROW UPDATE Triggers */
ExecARUpdateTriggers(estate, resultRelInfo, tupleid, tuple);
}
static Datum
gp_aovisimap_entry_internal(PG_FUNCTION_ARGS, Oid aoRelOid)
{
Datum values[4];
bool nulls[4];
HeapTuple tuple;
Datum result;
typedef struct Context
{
AppendOnlyVisimap visiMap;
Relation parentRelation;
IndexScanDesc indexScan;
text *bitmapBuffer;
} Context;
FuncCallContext *funcctx;
Context *context;
if (SRF_IS_FIRSTCALL())
{
TupleDesc tupdesc;
MemoryContext oldcontext;
/* create a function context for cross-call persistence */
funcctx = SRF_FIRSTCALL_INIT();
/*
* switch to memory context appropriate for multiple function
* calls
*/
oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
/* build tupdesc for result tuples */
tupdesc = CreateTemplateTupleDesc(4, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1, "segno",
INT4OID, -1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2, "first_row_num",
INT8OID, -1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3, "hidden_tupcount",
INT4OID, -1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 4, "bitmap",
TEXTOID, -1, 0);
funcctx->tuple_desc = BlessTupleDesc(tupdesc);
/*
* Collect all the locking information that we will format and send
* out as a result set.
*/
context = (Context *) palloc0(sizeof(Context));
context->parentRelation = heap_open(aoRelOid, AccessShareLock);
if (!(RelationIsAoRows(context->parentRelation) || RelationIsAoCols(context->parentRelation)))
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Function not supported on relation")));
}
AppendOnlyVisimap_Init(&context->visiMap,
context->parentRelation->rd_appendonly->visimaprelid,
context->parentRelation->rd_appendonly->visimapidxid,
AccessShareLock,
SnapshotNow);
context->indexScan = AppendOnlyVisimapStore_BeginScan(&
context->visiMap.visimapStore, 0, NULL);
context->bitmapBuffer = palloc0(VARHDRSZ + APPENDONLY_VISIMAP_MAX_RANGE + 1);
funcctx->user_fctx = (void *) context;
MemoryContextSwitchTo(oldcontext);
}
funcctx = SRF_PERCALL_SETUP();
context = (Context *) funcctx->user_fctx;
if (AppendOnlyVisimapStore_GetNext(&context->visiMap.visimapStore,
context->indexScan,
ForwardScanDirection,
&context->visiMap.visimapEntry,
NULL))
{
AppendOnlyVisimapEntry *visimapEntry = &context->visiMap.visimapEntry;
MemSet(values, 0, sizeof(values));
MemSet(nulls, false, sizeof(nulls));
values[0] = Int32GetDatum(visimapEntry->segmentFileNum);
values[1] = Int64GetDatum(visimapEntry->firstRowNum);
values[2] = Int32GetDatum(
(int32)AppendOnlyVisimapEntry_GetHiddenTupleCount(visimapEntry));
gp_aovisimap_encode_bitmap(VARDATA(context->bitmapBuffer),
visimapEntry->bitmap);
SET_VARSIZE(context->bitmapBuffer, APPENDONLY_VISIMAP_MAX_RANGE);
values[3] = PointerGetDatum(context->bitmapBuffer);
tuple = heap_form_tuple(funcctx->tuple_desc, values, nulls);
result = HeapTupleGetDatum(tuple);
SRF_RETURN_NEXT(funcctx, result);
}
AppendOnlyVisimapStore_EndScan(&context->visiMap.visimapStore,
context->indexScan);
AppendOnlyVisimap_Finish(&context->visiMap, AccessShareLock);
heap_close(context->parentRelation, AccessShareLock);
pfree(context->bitmapBuffer);
pfree(context);
funcctx->user_fctx = NULL;
SRF_RETURN_DONE(funcctx);
}
/*
* Assumes that the segment file lock is already held.
* Assumes that the segment file should be compacted.
*
*/
static void
AppendOnlySegmentFileFullCompaction(Relation aorel,
AppendOnlyEntry *aoEntry,
AppendOnlyInsertDesc insertDesc,
FileSegInfo* fsinfo)
{
const char* relname;
AppendOnlyVisimap visiMap;
AppendOnlyScanDesc scanDesc;
TupleDesc tupDesc;
MemTuple tuple;
TupleTableSlot *slot;
MemTupleBinding *mt_bind;
int compact_segno;
int64 movedTupleCount = 0;
ResultRelInfo *resultRelInfo;
EState *estate;
AOTupleId *aoTupleId;
int64 tupleCount = 0;
int64 tuplePerPage = INT_MAX;
Assert(Gp_role == GP_ROLE_EXECUTE || Gp_role == GP_ROLE_UTILITY);
Assert(RelationIsAoRows(aorel));
Assert(insertDesc);
compact_segno = fsinfo->segno;
if (fsinfo->varblockcount > 0)
{
tuplePerPage = fsinfo->total_tupcount / fsinfo->varblockcount;
}
relname = RelationGetRelationName(aorel);
AppendOnlyVisimap_Init(&visiMap,
aoEntry->visimaprelid,
aoEntry->visimapidxid,
ShareUpdateExclusiveLock,
SnapshotNow);
elogif(Debug_appendonly_print_compaction,
LOG, "Compact AO segno %d, relation %s, insert segno %d",
compact_segno, relname, insertDesc->storageWrite.segmentFileNum);
/*
* Todo: We need to limit the scan to one file and we need to avoid to
* lock the file again.
*
* We use SnapshotAny to get visible and invisible tuples.
*/
scanDesc = appendonly_beginrangescan(aorel,
SnapshotAny, SnapshotNow,
&compact_segno, 1, 0, NULL);
tupDesc = RelationGetDescr(aorel);
slot = MakeSingleTupleTableSlot(tupDesc);
mt_bind = create_memtuple_binding(tupDesc);
/*
* We need a ResultRelInfo and an EState so we can use the regular
* executor's index-entry-making machinery.
*/
estate = CreateExecutorState();
resultRelInfo = makeNode(ResultRelInfo);
resultRelInfo->ri_RangeTableIndex = 1; /* dummy */
resultRelInfo->ri_RelationDesc = aorel;
resultRelInfo->ri_TrigDesc = NULL; /* we don't fire triggers */
ExecOpenIndices(resultRelInfo);
estate->es_result_relations = resultRelInfo;
estate->es_num_result_relations = 1;
estate->es_result_relation_info = resultRelInfo;
/*
* Go through all visible tuples and move them to a new segfile.
*/
while ((tuple = appendonly_getnext(scanDesc, ForwardScanDirection, slot)) != NULL)
{
/* Check interrupts as this may take time. */
CHECK_FOR_INTERRUPTS();
aoTupleId = (AOTupleId*)slot_get_ctid(slot);
if (AppendOnlyVisimap_IsVisible(&scanDesc->visibilityMap, aoTupleId))
{
AppendOnlyMoveTuple(tuple,
slot,
mt_bind,
insertDesc,
resultRelInfo,
estate);
movedTupleCount++;
}
else
{
/* Tuple is invisible and needs to be dropped */
AppendOnlyThrowAwayTuple(aorel,
tuple,
slot,
mt_bind);
}
/*
* Check for vacuum delay point after approximatly a var block
*/
tupleCount++;
if (VacuumCostActive && tupleCount % tuplePerPage == 0)
{
vacuum_delay_point();
}
}
SetFileSegInfoState(aorel, aoEntry, compact_segno, AOSEG_STATE_AWAITING_DROP);
AppendOnlyVisimap_DeleteSegmentFile(&visiMap, compact_segno);
/* Delete all mini pages of the segment files if block directory exists */
if (OidIsValid(aoEntry->blkdirrelid))
{
AppendOnlyBlockDirectory_DeleteSegmentFile(
aoEntry,
SnapshotNow,
compact_segno,
0);
}
elogif(Debug_appendonly_print_compaction, LOG,
"Finished compaction: "
"AO segfile %d, relation %s, moved tuple count " INT64_FORMAT,
compact_segno, relname, movedTupleCount);
AppendOnlyVisimap_Finish(&visiMap, NoLock);
ExecCloseIndices(resultRelInfo);
FreeExecutorState(estate);
ExecDropSingleTupleTableSlot(slot);
destroy_memtuple_binding(mt_bind);
appendonly_endscan(scanDesc);
}
/*
* create_aoblkdir_table
*
* rel is already opened and exclusive-locked.
* comptypeOid is InvalidOid.
*/
static bool
create_aoblkdir_table(Relation rel, Oid aoblkdirOid,
Oid aoblkdirIndexOid, Oid *comptypeOid)
{
Oid relOid = RelationGetRelid(rel);
Oid aoblkdir_relid;
Oid aoblkdir_idxid;
bool shared_relation = rel->rd_rel->relisshared;
char aoblkdir_relname[NAMEDATALEN];
char aoblkdir_idxname[NAMEDATALEN];
TupleDesc tupdesc;
IndexInfo *indexInfo;
Oid classObjectId[3];
ObjectAddress baseobject;
ObjectAddress aoblkdirobject;
Oid tablespaceOid = ChooseTablespaceForLimitedObject(rel->rd_rel->reltablespace);
if (!RelationIsAoRows(rel))
return false;
/*
* We cannot allow creating a block directory for a shared relation
* after initdb (because there's no way to let other databases know
* this block directory.
*/
if (shared_relation && !IsBootstrapProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared tables cannot have block directory after initdb")));
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow, NULL,NULL, &aoblkdir_relid, &aoblkdir_idxid);
/*
* Does it have a block directory?
*/
if (aoblkdir_relid != InvalidOid)
{
return false;
}
snprintf(aoblkdir_relname, sizeof(aoblkdir_relname),
"pg_aoblkdir_%u", relOid);
snprintf(aoblkdir_idxname, sizeof(aoblkdir_idxname),
"pg_aoblkdir_%u_index", relOid);
/* Create a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(4, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"segno",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"columngroup_no",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"first_row_no",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 4,
"minipage",
VARBITOID,
-1, 0);
/*
* We don't want any toast columns here.
*/
tupdesc->attrs[0]->attstorage = 'p';
tupdesc->attrs[1]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
/*
* We place aoblkdir relation in the pg_aoseg namespace
* even if its master relation is a temp table. There cannot be
* any naming collision, and the aoblkdir relation will be
* destroyed when its master is, so there is no need to handle
* the aoblkdir relation as temp.
*/
aoblkdir_relid = heap_create_with_catalog(aoblkdir_relname,
PG_AOSEGMENT_NAMESPACE,
tablespaceOid,
aoblkdirOid,
rel->rd_rel->relowner,
tupdesc,
/* relam */ InvalidOid,
RELKIND_AOBLOCKDIR,
RELSTORAGE_HEAP,
shared_relation,
true,
/* bufferPoolBulkLoad */ false,
0,
ONCOMMIT_NOOP,
NULL, /* GP Policy */
(Datum) 0,
true,
comptypeOid,
/* persistentTid */ NULL,
/* persistentSerialNum */ NULL);
/* Make this table visible, else index creation will fail */
CommandCounterIncrement();
/*
* Create index on segno, first_row_no.
*/
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = 3;
indexInfo->ii_KeyAttrNumbers[0] = 1;
indexInfo->ii_KeyAttrNumbers[1] = 2;
indexInfo->ii_KeyAttrNumbers[2] = 3;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = NIL;
indexInfo->ii_PredicateState = NIL;
indexInfo->ii_Unique = false;
indexInfo->ii_Concurrent = false;
classObjectId[0] = INT4_BTREE_OPS_OID;
classObjectId[1] = INT4_BTREE_OPS_OID;
classObjectId[2] = INT8_BTREE_OPS_OID;
aoblkdir_idxid = index_create(aoblkdirOid, aoblkdir_idxname, aoblkdirIndexOid,
indexInfo,
BTREE_AM_OID,
tablespaceOid,
classObjectId, (Datum) 0,
true, false, (Oid *) NULL, true, false, false, NULL);
/* Unlock target table -- no one can see it */
UnlockRelationOid(aoblkdirOid, ShareLock);
/* Unlock the index -- no one can see it anyway */
UnlockRelationOid(aoblkdirIndexOid, AccessExclusiveLock);
/*
* Store the aoblkdir table's OID in the parent relation's pg_appendonly row.
*/
UpdateAppendOnlyEntryAuxOids(relOid, InvalidOid, InvalidOid,
aoblkdir_relid, aoblkdir_idxid);
/*
* Register dependency from the aoseg table to the master, so that the
* aoseg table will be deleted if the master is.
*/
baseobject.classId = RelationRelationId;
baseobject.objectId = relOid;
baseobject.objectSubId = 0;
aoblkdirobject.classId = RelationRelationId;
aoblkdirobject.objectId = aoblkdirOid;
aoblkdirobject.objectSubId = 0;
recordDependencyOn(&aoblkdirobject, &baseobject, DEPENDENCY_INTERNAL);
/*
* Make changes visible
*/
CommandCounterIncrement();
return true;
}
/*
* Performs a compaction of an append-only relation.
*
* In non-utility mode, all compaction segment files should be
* marked as in-use/in-compaction in the appendonlywriter.c code.
*
*/
void
AppendOnlyDrop(Relation aorel,
List* compaction_segno)
{
const char* relname;
int total_segfiles;
FileSegInfo** segfile_array;
int i, segno;
FileSegInfo* fsinfo;
Assert (Gp_role == GP_ROLE_EXECUTE || Gp_role == GP_ROLE_UTILITY);
Assert (RelationIsAoRows(aorel));
relname = RelationGetRelationName(aorel);
AppendOnlyEntry *aoEntry = GetAppendOnlyEntry(RelationGetRelid(aorel), SnapshotNow);
elogif (Debug_appendonly_print_compaction, LOG,
"Drop AO relation %s", relname);
/* Get information about all the file segments we need to scan */
segfile_array = GetAllFileSegInfo(aorel, aoEntry, SnapshotNow, &total_segfiles);
for(i = 0 ; i < total_segfiles ; i++)
{
segno = segfile_array[i]->segno;
if (list_find_int(compaction_segno, segno) < 0)
{
continue;
}
/*
* Try to get the transaction write-lock for the Append-Only segment file.
*
* NOTE: This is a transaction scope lock that must be held until commit / abort.
*/
LockRelationAppendOnlySegmentFile(
&aorel->rd_node,
segfile_array[i]->segno,
AccessExclusiveLock,
false);
/* Re-fetch under the write lock to get latest committed eof. */
fsinfo = GetFileSegInfo(aorel, aoEntry, SnapshotNow, segno);
if (fsinfo->state == AOSEG_STATE_AWAITING_DROP)
{
Assert(HasLockForSegmentFileDrop(aorel));
Assert(!HasSerializableBackends(false));
AppendOnlyCompaction_DropSegmentFile(aorel, segno);
ClearFileSegInfo(aorel, aoEntry, segno,
AOSEG_STATE_DEFAULT);
}
pfree(fsinfo);
}
pfree(aoEntry);
if (segfile_array)
{
FreeAllSegFileInfo(segfile_array, total_segfiles);
pfree(segfile_array);
}
}
/*
* lazy_vacuum_rel() -- perform LAZY VACUUM for one heap relation
*
* This routine vacuums a single heap, cleans out its indexes, and
* updates its relpages and reltuples statistics.
*
* At entry, we have already established a transaction and opened
* and locked the relation.
*
* The return value indicates whether this function has held off
* interrupts -- caller must RESUME_INTERRUPTS() after commit if true.
*/
bool
lazy_vacuum_rel(Relation onerel, VacuumStmt *vacstmt,
BufferAccessStrategy bstrategy, List *updated_stats)
{
LVRelStats *vacrelstats;
Relation *Irel;
int nindexes;
BlockNumber possibly_freeable;
PGRUsage ru0;
TimestampTz starttime = 0;
bool heldoff = false;
pg_rusage_init(&ru0);
/* measure elapsed time iff autovacuum logging requires it */
if (IsAutoVacuumWorkerProcess() && Log_autovacuum_min_duration > 0)
starttime = GetCurrentTimestamp();
if (vacstmt->verbose)
elevel = INFO;
else
elevel = DEBUG2;
if (Gp_role == GP_ROLE_DISPATCH)
elevel = DEBUG2; /* vacuum and analyze messages aren't interesting from the QD */
#ifdef FAULT_INJECTOR
if (vacuumStatement_IsInAppendOnlyDropPhase(vacstmt))
{
FaultInjector_InjectFaultIfSet(
CompactionBeforeSegmentFileDropPhase,
DDLNotSpecified,
"", // databaseName
""); // tableName
}
if (vacummStatement_IsInAppendOnlyCleanupPhase(vacstmt))
{
FaultInjector_InjectFaultIfSet(
CompactionBeforeCleanupPhase,
DDLNotSpecified,
"", // databaseName
""); // tableName
}
#endif
/*
* MPP-23647. Update xid limits for heap as well as appendonly
* relations. This allows setting relfrozenxid to correct value
* for an appendonly (AO/CO) table.
*/
vac_strategy = bstrategy;
vacuum_set_xid_limits(vacstmt->freeze_min_age, onerel->rd_rel->relisshared,
&OldestXmin, &FreezeLimit);
/*
* Execute the various vacuum operations. Appendonly tables are treated
* differently.
*/
if (RelationIsAoRows(onerel) || RelationIsAoCols(onerel))
{
lazy_vacuum_aorel(onerel, vacstmt, updated_stats);
return false;
}
vacrelstats = (LVRelStats *) palloc0(sizeof(LVRelStats));
/* heap relation */
/* Set threshold for interesting free space = average request size */
/* XXX should we scale it up or down? Adjust vacuum.c too, if so */
vacrelstats->threshold = GetAvgFSMRequestSize(&onerel->rd_node);
vacrelstats->num_index_scans = 0;
/* Open all indexes of the relation */
vac_open_indexes(onerel, RowExclusiveLock, &nindexes, &Irel);
vacrelstats->hasindex = (nindexes > 0);
/* Do the vacuuming */
lazy_scan_heap(onerel, vacrelstats, Irel, nindexes, updated_stats);
/* Done with indexes */
vac_close_indexes(nindexes, Irel, NoLock);
/*
* Optionally truncate the relation.
*
* Don't even think about it unless we have a shot at releasing a goodly
* number of pages. Otherwise, the time taken isn't worth it.
*
* Note that after we've truncated the heap, it's too late to abort the
* transaction; doing so would lose the sinval messages needed to tell
* the other backends about the table being shrunk. We prevent interrupts
* in that case; caller is responsible for re-enabling them after
* committing the transaction.
*/
possibly_freeable = vacrelstats->rel_pages - vacrelstats->nonempty_pages;
if (possibly_freeable > 0 &&
(possibly_freeable >= REL_TRUNCATE_MINIMUM ||
possibly_freeable >= vacrelstats->rel_pages / REL_TRUNCATE_FRACTION))
{
HOLD_INTERRUPTS();
heldoff = true;
lazy_truncate_heap(onerel, vacrelstats);
}
/* Update shared free space map with final free space info */
lazy_update_fsm(onerel, vacrelstats);
if (vacrelstats->tot_free_pages > MaxFSMPages)
ereport(WARNING,
(errmsg("relation \"%s.%s\" contains more than \"max_fsm_pages\" pages with useful free space",
get_namespace_name(RelationGetNamespace(onerel)),
RelationGetRelationName(onerel)),
/* Only suggest VACUUM FULL if > 20% free */
(vacrelstats->tot_free_pages > vacrelstats->rel_pages * 0.20) ?
errhint("Consider using VACUUM FULL on this relation or increasing the configuration parameter \"max_fsm_pages\".") :
errhint("Consider increasing the configuration parameter \"max_fsm_pages\".")));
/* Update statistics in pg_class */
vac_update_relstats_from_list(onerel,
vacrelstats->rel_pages,
vacrelstats->rel_tuples,
vacrelstats->hasindex,
FreezeLimit,
updated_stats);
/* report results to the stats collector, too */
pgstat_report_vacuum(RelationGetRelid(onerel), onerel->rd_rel->relisshared,
true /*vacrelstats->scanned_all*/,
vacstmt->analyze, vacrelstats->rel_tuples);
if (gp_indexcheck_vacuum == INDEX_CHECK_ALL ||
(gp_indexcheck_vacuum == INDEX_CHECK_SYSTEM &&
PG_CATALOG_NAMESPACE == RelationGetNamespace(onerel)))
{
int i;
for (i = 0; i < nindexes; i++)
{
if (Irel[i]->rd_rel->relam == BTREE_AM_OID)
_bt_validate_vacuum(Irel[i], onerel, OldestXmin);
}
}
/* and log the action if appropriate */
if (IsAutoVacuumWorkerProcess() && Log_autovacuum_min_duration >= 0)
{
if (Log_autovacuum_min_duration == 0 ||
TimestampDifferenceExceeds(starttime, GetCurrentTimestamp(),
Log_autovacuum_min_duration))
ereport(LOG,
(errmsg("automatic vacuum of table \"%s.%s.%s\": index scans: %d\n"
"pages: %d removed, %d remain\n"
"tuples: %.0f removed, %.0f remain\n"
"system usage: %s",
get_database_name(MyDatabaseId),
get_namespace_name(RelationGetNamespace(onerel)),
RelationGetRelationName(onerel),
vacrelstats->num_index_scans,
vacrelstats->pages_removed, vacrelstats->rel_pages,
vacrelstats->tuples_deleted, vacrelstats->rel_tuples,
pg_rusage_show(&ru0))));
}
return heldoff;
}
/*
* lazy_vacuum_aorel -- perform LAZY VACUUM for one Append-only relation.
*/
static void
lazy_vacuum_aorel(Relation onerel, VacuumStmt *vacstmt, List *updated_stats)
{
LVRelStats *vacrelstats;
bool update_relstats = true;
vacrelstats = (LVRelStats *) palloc0(sizeof(LVRelStats));
if (vacuumStatement_IsInAppendOnlyPreparePhase(vacstmt))
{
elogif(Debug_appendonly_print_compaction, LOG,
"Vacuum prepare phase %s", RelationGetRelationName(onerel));
vacuum_appendonly_indexes(onerel, vacstmt, updated_stats);
if (RelationIsAoRows(onerel))
AppendOnlyTruncateToEOF(onerel);
else
AOCSTruncateToEOF(onerel);
/*
* MPP-23647. For empty tables, we skip compaction phase
* and cleanup phase. Therefore, we update the stats
* (specifically, relfrozenxid) in prepare phase if the
* table is empty. Otherwise, the stats will be updated in
* the cleanup phase, when we would have computed the
* correct values for stats.
*/
if (vacstmt->appendonly_relation_empty)
{
update_relstats = true;
/*
* For an empty relation, the only stats we care about
* is relfrozenxid and relhasindex. We need to be
* mindful of correctly setting relhasindex here.
* relfrozenxid is already taken care of above by
* calling vacuum_set_xid_limits().
*/
vacrelstats->hasindex = onerel->rd_rel->relhasindex;
}
else
{
/*
* For a non-empty relation, follow the usual
* compaction phases and do not update stats in
* prepare phase.
*/
update_relstats = false;
}
}
else if (!vacummStatement_IsInAppendOnlyCleanupPhase(vacstmt))
{
vacuum_appendonly_rel(onerel, vacstmt);
update_relstats = false;
}
else
{
elogif(Debug_appendonly_print_compaction, LOG,
"Vacuum cleanup phase %s", RelationGetRelationName(onerel));
vacuum_appendonly_fill_stats(onerel, ActiveSnapshot,
&vacrelstats->rel_pages,
&vacrelstats->rel_tuples,
&vacrelstats->hasindex);
/* reset the remaining LVRelStats values */
vacrelstats->nonempty_pages = 0;
vacrelstats->num_dead_tuples = 0;
vacrelstats->max_dead_tuples = 0;
vacrelstats->tuples_deleted = 0;
vacrelstats->tot_free_pages = 0;
vacrelstats->fs_is_heap = false;
vacrelstats->num_free_pages = 0;
vacrelstats->max_free_pages = 0;
vacrelstats->pages_removed = 0;
}
if (update_relstats)
{
/* Update statistics in pg_class */
vac_update_relstats_from_list(onerel,
vacrelstats->rel_pages,
vacrelstats->rel_tuples,
vacrelstats->hasindex,
FreezeLimit,
updated_stats);
/* report results to the stats collector, too */
pgstat_report_vacuum(RelationGetRelid(onerel),
onerel->rd_rel->relisshared,
true /*vacrelstats->scanned_all*/,
vacstmt->analyze, vacrelstats->rel_tuples);
}
}
/* ----------------------------------------------------------------
* ExecInsert
*
* INSERTs have to add the tuple into
* the base relation and insert appropriate tuples into the
* index relations.
* Insert can be part of an update operation when
* there is a preceding SplitUpdate node.
* ----------------------------------------------------------------
*/
void
ExecInsert(TupleTableSlot *slot,
DestReceiver *dest,
EState *estate,
PlanGenerator planGen,
bool isUpdate)
{
void *tuple = NULL;
ResultRelInfo *resultRelInfo = NULL;
Relation resultRelationDesc = NULL;
Oid newId = InvalidOid;
TupleTableSlot *partslot = NULL;
AOTupleId aoTupleId = AOTUPLEID_INIT;
bool rel_is_heap = false;
bool rel_is_aorows = false;
bool rel_is_external = false;
bool rel_is_parquet = false;
/*
* get information on the (current) result relation
*/
if (estate->es_result_partitions)
{
resultRelInfo = slot_get_partition(slot, estate);
estate->es_result_relation_info = resultRelInfo;
if (NULL != resultRelInfo->ri_parquetSendBack)
{
/*
* The Parquet part we are about to insert into
* has sendBack information. This means we're inserting into the
* part twice, which is not supported. Error out (GPSQL-2291)
*/
Assert(gp_parquet_insert_sort);
ereport(ERROR, (errcode(ERRCODE_CDB_FEATURE_NOT_YET),
errmsg("Cannot insert out-of-order tuples in parquet partitions"),
errhint("Sort the data on the partitioning key(s) before inserting"),
errOmitLocation(true)));
}
/*
* Check if we need to close the last parquet partition we
* inserted into (GPSQL-2291).
*/
Oid new_part_oid = resultRelInfo->ri_RelationDesc->rd_id;
if (gp_parquet_insert_sort &&
PLANGEN_OPTIMIZER == planGen &&
InvalidOid != estate->es_last_parq_part &&
new_part_oid != estate->es_last_parq_part)
{
Assert(NULL != estate->es_partition_state->result_partition_hash);
ResultPartHashEntry *entry = hash_search(estate->es_partition_state->result_partition_hash,
&estate->es_last_parq_part,
HASH_FIND,
NULL /* found */);
Assert(NULL != entry);
Assert(entry->offset < estate->es_num_result_relations);
ResultRelInfo *oldResultRelInfo = & estate->es_result_relations[entry->offset];
elog(DEBUG1, "Switching from old part oid=%d name=[%s] to new part oid=%d name=[%s]",
estate->es_last_parq_part,
oldResultRelInfo->ri_RelationDesc->rd_rel->relname.data,
new_part_oid,
resultRelInfo->ri_RelationDesc->rd_rel->relname.data);
/*
* We are opening a new partition, and the last partition we
* inserted into was a Parquet part. Let's close the old
* parquet insert descriptor to free the memory before
* opening the new one.
*/
ParquetInsertDescData *oldInsertDesc = oldResultRelInfo->ri_parquetInsertDesc;
/*
* We need to preserve the "sendback" information that needs to be
* sent back to the QD process from this part.
* Compute it here, and store it for later use.
*/
QueryContextDispatchingSendBack sendback =
CreateQueryContextDispatchingSendBack(1);
sendback->relid = RelationGetRelid(oldResultRelInfo->ri_RelationDesc);
oldInsertDesc->sendback = sendback;
parquet_insert_finish(oldInsertDesc);
/* Store the sendback information in the resultRelInfo for this part */
oldResultRelInfo->ri_parquetSendBack = sendback;
/* Record in the resultRelInfo that we closed the parquet insert descriptor */
oldResultRelInfo->ri_parquetInsertDesc = NULL;
/* Reset the last parquet part Oid, it's now closed */
estate->es_last_parq_part = InvalidOid;
}
}
else
{
resultRelInfo = estate->es_result_relation_info;
}
Assert (!resultRelInfo->ri_projectReturning);
resultRelationDesc = resultRelInfo->ri_RelationDesc;
rel_is_heap = RelationIsHeap(resultRelationDesc);
rel_is_aorows = RelationIsAoRows(resultRelationDesc);
rel_is_external = RelationIsExternal(resultRelationDesc);
rel_is_parquet = RelationIsParquet(resultRelationDesc);
/* Validate that insert is not part of an non-allowed update operation. */
if (isUpdate && (rel_is_aorows || rel_is_parquet))
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Append-only tables are not updatable. Operation not permitted."),
errOmitLocation(true)));
}
partslot = reconstructMatchingTupleSlot(slot, resultRelInfo);
if (rel_is_heap || rel_is_external)
{
tuple = ExecFetchSlotHeapTuple(partslot);
}
else if (rel_is_aorows)
{
tuple = ExecFetchSlotMemTuple(partslot, false);
}
else if (rel_is_parquet)
{
tuple = NULL;
}
Assert( partslot != NULL );
Assert( rel_is_parquet || (tuple != NULL));
/* Execute triggers in Planner-generated plans */
if (planGen == PLANGEN_PLANNER)
{
/* BEFORE ROW INSERT Triggers */
if (resultRelInfo->ri_TrigDesc &&
resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
{
HeapTuple newtuple;
/* NYI */
if(rel_is_parquet)
elog(ERROR, "triggers are not supported on tables that use column-oriented storage");
newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
if (newtuple == NULL) /* "do nothing" */
{
return;
}
if (newtuple != tuple) /* modified by Trigger(s) */
{
/*
* Put the modified tuple into a slot for convenience of routines
* below. We assume the tuple was allocated in per-tuple memory
* context, and therefore will go away by itself. The tuple table
* slot should not try to clear it.
*/
TupleTableSlot *newslot = estate->es_trig_tuple_slot;
if (newslot->tts_tupleDescriptor != partslot->tts_tupleDescriptor)
ExecSetSlotDescriptor(newslot, partslot->tts_tupleDescriptor);
ExecStoreGenericTuple(newtuple, newslot, false);
newslot->tts_tableOid = partslot->tts_tableOid; /* for constraints */
tuple = newtuple;
partslot = newslot;
}
}
}
/*
* Check the constraints of the tuple
*/
if (resultRelationDesc->rd_att->constr &&
planGen == PLANGEN_PLANNER)
{
ExecConstraints(resultRelInfo, partslot, estate);
}
/*
* insert the tuple
*
* Note: heap_insert returns the tid (location) of the new tuple in the
* t_self field.
*
* NOTE: for append-only relations we use the append-only access methods.
*/
if (rel_is_aorows)
{
if (resultRelInfo->ri_aoInsertDesc == NULL)
{
ResultRelSegFileInfo *segfileinfo = NULL;
/* Set the pre-assigned fileseg number to insert into */
ResultRelInfoSetSegFileInfo(resultRelInfo, estate->es_result_segfileinfos);
segfileinfo = (ResultRelSegFileInfo *)list_nth(resultRelInfo->ri_aosegfileinfos, GetQEIndex());
resultRelInfo->ri_aoInsertDesc =
appendonly_insert_init(resultRelationDesc,
segfileinfo);
}
appendonly_insert(resultRelInfo->ri_aoInsertDesc, tuple, &newId, &aoTupleId);
}
else if (rel_is_external)
{
/* Writable external table */
if (resultRelInfo->ri_extInsertDesc == NULL)
resultRelInfo->ri_extInsertDesc = external_insert_init(
resultRelationDesc, 0);
newId = external_insert(resultRelInfo->ri_extInsertDesc, tuple);
}
else if(rel_is_parquet)
{
/* If there is no parquet insert descriptor, create it now. */
if (resultRelInfo->ri_parquetInsertDesc == NULL)
{
ResultRelSegFileInfo *segfileinfo = NULL;
ResultRelInfoSetSegFileInfo(resultRelInfo, estate->es_result_segfileinfos);
segfileinfo = (ResultRelSegFileInfo *)list_nth(resultRelInfo->ri_aosegfileinfos, GetQEIndex());
resultRelInfo->ri_parquetInsertDesc = parquet_insert_init(resultRelationDesc, segfileinfo);
/*
* Just opened a new parquet partition for insert. Save the Oid
* in estate, so that we can close it when switching to a
* new partition (GPSQL-2291)
*/
elog(DEBUG1, "Saving es_last_parq_part. Old=%d, new=%d", estate->es_last_parq_part, resultRelationDesc->rd_id);
estate->es_last_parq_part = resultRelationDesc->rd_id;
}
newId = parquet_insert(resultRelInfo->ri_parquetInsertDesc, partslot);
}
else
{
Insist(rel_is_heap);
newId = heap_insert(resultRelationDesc,
tuple,
estate->es_snapshot->curcid,
true, true, GetCurrentTransactionId());
}
IncrAppended();
(estate->es_processed)++;
(resultRelInfo->ri_aoprocessed)++;
estate->es_lastoid = newId;
partslot->tts_tableOid = RelationGetRelid(resultRelationDesc);
if (rel_is_aorows || rel_is_parquet)
{
/* NOTE: Current version does not support index upon parquet table. */
/*
* insert index entries for AO Row-Store tuple
*/
if (resultRelInfo->ri_NumIndices > 0 && !rel_is_parquet)
ExecInsertIndexTuples(partslot, (ItemPointer)&aoTupleId, estate, false);
}
else
{
/* Use parttuple for index update in case this is an indexed heap table. */
TupleTableSlot *xslot = partslot;
void *xtuple = tuple;
setLastTid(&(((HeapTuple) xtuple)->t_self));
/*
* insert index entries for tuple
*/
if (resultRelInfo->ri_NumIndices > 0)
ExecInsertIndexTuples(xslot, &(((HeapTuple) xtuple)->t_self), estate, false);
}
if (planGen == PLANGEN_PLANNER)
{
/* AFTER ROW INSERT Triggers */
ExecARInsertTriggers(estate, resultRelInfo, tuple);
}
}
void
AlterTableCreateAoSegTable(Oid relOid, bool is_part_child, bool is_part_parent)
{
TupleDesc tupdesc;
Relation rel;
const char *prefix;
/*
* Grab an exclusive lock on the target table, which we will NOT release
* until end of transaction. (This is probably redundant in all present
* uses...)
*/
if (is_part_child)
rel = heap_open(relOid, NoLock);
else
rel = heap_open(relOid, AccessExclusiveLock);
if(RelationIsAoRows(rel))
{
prefix = "pg_aoseg";
/* this is pretty painful... need a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(8, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"segno",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"eof",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"tupcount",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 4,
"varblockcount",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 5,
"eofuncompressed",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 6,
"modcount",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 7,
"formatversion",
INT2OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 8,
"state",
INT2OID,
-1, 0);
}
else if (RelationIsAoCols(rel))
{
prefix = "pg_aocsseg";
/*
* XXX
* At this moment, we hardwire the rel aocs info.
* Essentially, we assume total vertical partition, and
* we do not do datatype specific compression.
*
* In order to make things right, we need to first fix
* the DefineRelation, so that we store the per column
* info, then, we need to open the catalog, pull out
* info here.
*/
/*
* XXX We do not handle add/drop column etc nicely yet.
*/
/*
* Assuming full vertical partition, we want to include
* the following in the seg table.
*
* segno int, -- whatever purpose ao use it
* tupcount bigint -- total tup
* varblockcount bigint, -- total varblock
* vpinfo varbinary(max) -- vertical partition info encoded in
* binary. NEEDS TO BE REFACTORED
* INTO MULTIPLE COLUMNS!!
* state (smallint) -- state of the segment file
*/
tupdesc = CreateTemplateTupleDesc(7, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"segno",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"tupcount",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"varblockcount",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 4,
"vpinfo",
BYTEAOID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 5,
"modcount",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 6,
"formatversion",
INT2OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 7,
"state",
INT2OID,
-1, 0);
}
else
{
heap_close(rel, NoLock);
return;
}
(void) CreateAOAuxiliaryTable(rel, prefix, RELKIND_AOSEGMENTS,
tupdesc,
NULL, NIL, NULL, NULL, is_part_parent);
heap_close(rel, NoLock);
}
/*
* Truncates each segment file to the AO relation to its EOF.
* If we cannot get a lock on the segment file (because e.g. a concurrent insert)
* the segment file is skipped.
*/
void
AppendOnlyTruncateToEOF(Relation aorel)
{
const char* relname;
int total_segfiles;
FileSegInfo** segfile_array;
int i, segno;
LockAcquireResult acquireResult;
FileSegInfo* fsinfo;
Assert (RelationIsAoRows(aorel));
relname = RelationGetRelationName(aorel);
AppendOnlyEntry *aoEntry = GetAppendOnlyEntry(RelationGetRelid(aorel), SnapshotNow);
elogif (Debug_appendonly_print_compaction, LOG,
"Compact AO relation %s", relname);
/* Get information about all the file segments we need to scan */
segfile_array = GetAllFileSegInfo(aorel, aoEntry, SnapshotNow, &total_segfiles);
for(i = 0 ; i < total_segfiles ; i++)
{
segno = segfile_array[i]->segno;
/*
* Try to get the transaction write-lock for the Append-Only segment file.
*
* NOTE: This is a transaction scope lock that must be held until commit / abort.
*/
acquireResult = LockRelationAppendOnlySegmentFile(
&aorel->rd_node,
segfile_array[i]->segno,
AccessExclusiveLock,
/* dontWait */ true);
if (acquireResult == LOCKACQUIRE_NOT_AVAIL)
{
elog(DEBUG5, "truncate skips AO segfile %d, "
"relation %s", segfile_array[i]->segno, relname);
continue;
}
/* Re-fetch under the write lock to get latest committed eof. */
fsinfo = GetFileSegInfo(aorel, aoEntry, SnapshotNow, segno);
/*
* This should not occur since this segfile info was found by the
* "all" method, but better to catch for trouble shooting
* (possibly index corruption?)
*/
if (fsinfo == NULL)
elog(ERROR, "file seginfo for AO relation %s %u/%u/%u (segno=%u) is missing",
relname,
aorel->rd_node.spcNode,
aorel->rd_node.dbNode,
aorel->rd_node.relNode,
segno);
AppendOnlySegmentFileTruncateToEOF(aorel, fsinfo);
pfree(fsinfo);
}
pfree(aoEntry);
if (segfile_array)
{
FreeAllSegFileInfo(segfile_array, total_segfiles);
pfree(segfile_array);
}
}