/*
* btbuildempty() -- build an empty btree index in the initialization fork
*/
void
btbuildempty(Relation index)
{
Page metapage;
/* Construct metapage. */
metapage = (Page) palloc(BLCKSZ);
_bt_initmetapage(metapage, P_NONE, 0);
/*
* Write the page and log it. It might seem that an immediate sync would
* be sufficient to guarantee that the file exists on disk, but recovery
* itself might remove it while replaying, for example, an
* XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we need
* this even when wal_level=minimal.
*/
PageSetChecksumInplace(metapage, BTREE_METAPAGE);
smgrwrite(index->rd_smgr, INIT_FORKNUM, BTREE_METAPAGE,
(char *) metapage, true);
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
BTREE_METAPAGE, metapage, true);
/*
* An immediate sync is required even if we xlog'd the page, because the
* write did not go through shared_buffers and therefore a concurrent
* checkpoint may have moved the redo pointer past our xlog record.
*/
smgrimmedsync(index->rd_smgr, INIT_FORKNUM);
}
/*
* Build an empty SPGiST index in the initialization fork
*/
Datum
spgbuildempty(PG_FUNCTION_ARGS)
{
Relation index = (Relation) PG_GETARG_POINTER(0);
Page page;
/* Construct metapage. */
page = (Page) palloc(BLCKSZ);
SpGistInitMetapage(page);
/* Write the page. If archiving/streaming, XLOG it. */
smgrwrite(index->rd_smgr, INIT_FORKNUM, SPGIST_METAPAGE_BLKNO,
(char *) page, true);
if (XLogIsNeeded())
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
SPGIST_METAPAGE_BLKNO, page);
/* Likewise for the root page. */
SpGistInitPage(page, SPGIST_LEAF);
smgrwrite(index->rd_smgr, INIT_FORKNUM, SPGIST_HEAD_BLKNO,
(char *) page, true);
if (XLogIsNeeded())
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
SPGIST_HEAD_BLKNO, page);
/*
* An immediate sync is required even if we xlog'd the pages, because the
* writes did not go through shared buffers and therefore a concurrent
* checkpoint may have moved the redo pointer past our xlog record.
*/
smgrimmedsync(index->rd_smgr, INIT_FORKNUM);
PG_RETURN_VOID();
}
/*
* btbuildempty() -- build an empty btree index in the initialization fork
*/
Datum
btbuildempty(PG_FUNCTION_ARGS)
{
Relation index = (Relation) PG_GETARG_POINTER(0);
Page metapage;
/* Construct metapage. */
metapage = (Page) palloc(BLCKSZ);
_bt_initmetapage(metapage, P_NONE, 0);
/* Write the page. If archiving/streaming, XLOG it. */
smgrwrite(index->rd_smgr, INIT_FORKNUM, BTREE_METAPAGE,
(char *) metapage, true);
if (XLogIsNeeded())
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
BTREE_METAPAGE, metapage);
/*
* An immediate sync is require even if we xlog'd the page, because the
* write did not go through shared_buffers and therefore a concurrent
* checkpoint may have move the redo pointer past our xlog record.
*/
smgrimmedsync(index->rd_smgr, INIT_FORKNUM);
PG_RETURN_VOID();
}
/*
* SetMatViewToPopulated
* Indicate that the materialized view has been populated by its query.
*
* NOTE: The heap starts out in a state that doesn't look scannable, and can
* only transition from there to scannable at the time a new heap is created.
*
* NOTE: caller must be holding an appropriate lock on the relation.
*/
void
SetMatViewToPopulated(Relation relation)
{
Page page;
Assert(relation->rd_rel->relkind == RELKIND_MATVIEW);
Assert(relation->rd_ispopulated == false);
page = (Page) palloc(BLCKSZ);
PageInit(page, BLCKSZ, 0);
if (RelationNeedsWAL(relation))
log_newpage(&(relation->rd_node), MAIN_FORKNUM, 0, page);
RelationOpenSmgr(relation);
PageSetChecksumInplace(page, 0);
smgrextend(relation->rd_smgr, MAIN_FORKNUM, 0, (char *) page, true);
pfree(page);
smgrimmedsync(relation->rd_smgr, MAIN_FORKNUM);
RelationCacheInvalidateEntry(relation->rd_id);
}
/*
* Build an empty SPGiST index in the initialization fork
*/
void
spgbuildempty(Relation index)
{
Page page;
/* Construct metapage. */
page = (Page) palloc(BLCKSZ);
SpGistInitMetapage(page);
/* Write the page. If archiving/streaming, XLOG it. */
PageSetChecksumInplace(page, SPGIST_METAPAGE_BLKNO);
smgrwrite(index->rd_smgr, INIT_FORKNUM, SPGIST_METAPAGE_BLKNO,
(char *) page, true);
if (XLogIsNeeded())
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
SPGIST_METAPAGE_BLKNO, page, false);
/* Likewise for the root page. */
SpGistInitPage(page, SPGIST_LEAF);
PageSetChecksumInplace(page, SPGIST_ROOT_BLKNO);
smgrwrite(index->rd_smgr, INIT_FORKNUM, SPGIST_ROOT_BLKNO,
(char *) page, true);
if (XLogIsNeeded())
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
SPGIST_ROOT_BLKNO, page, true);
/* Likewise for the null-tuples root page. */
SpGistInitPage(page, SPGIST_LEAF | SPGIST_NULLS);
PageSetChecksumInplace(page, SPGIST_NULL_BLKNO);
smgrwrite(index->rd_smgr, INIT_FORKNUM, SPGIST_NULL_BLKNO,
(char *) page, true);
if (XLogIsNeeded())
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
SPGIST_NULL_BLKNO, page, true);
/*
* An immediate sync is required even if we xlog'd the pages, because the
* writes did not go through shared buffers and therefore a concurrent
* checkpoint may have moved the redo pointer past our xlog record.
*/
smgrimmedsync(index->rd_smgr, INIT_FORKNUM);
}
/*
* btbuildempty() -- build an empty btree index in the initialization fork
*/
void
btbuildempty(Relation index)
{
Page metapage;
/* Construct metapage. */
metapage = (Page) palloc(BLCKSZ);
_bt_initmetapage(metapage, P_NONE, 0);
/* Write the page. If archiving/streaming, XLOG it. */
PageSetChecksumInplace(metapage, BTREE_METAPAGE);
smgrwrite(index->rd_smgr, INIT_FORKNUM, BTREE_METAPAGE,
(char *) metapage, true);
if (XLogIsNeeded())
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
BTREE_METAPAGE, metapage, false);
/*
* An immediate sync is required even if we xlog'd the page, because the
* write did not go through shared_buffers and therefore a concurrent
* checkpoint may have moved the redo pointer past our xlog record.
*/
smgrimmedsync(index->rd_smgr, INIT_FORKNUM);
}
/*
* Read tuples in correct sort order from tuplesort, and load them into
* btree leaves.
*/
static void
_bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
{
BTPageState *state = NULL;
bool merge = (btspool2 != NULL);
IndexTuple itup,
itup2 = NULL;
bool load1;
TupleDesc tupdes = RelationGetDescr(wstate->index);
int i,
keysz = RelationGetNumberOfAttributes(wstate->index);
ScanKey indexScanKey = NULL;
SortSupport sortKeys;
if (merge)
{
/*
* Another BTSpool for dead tuples exists. Now we have to merge
* btspool and btspool2.
*/
/* the preparation of merge */
itup = tuplesort_getindextuple(btspool->sortstate, true);
itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
indexScanKey = _bt_mkscankey_nodata(wstate->index);
/* Prepare SortSupport data for each column */
sortKeys = (SortSupport) palloc0(keysz * sizeof(SortSupportData));
for (i = 0; i < keysz; i++)
{
SortSupport sortKey = sortKeys + i;
ScanKey scanKey = indexScanKey + i;
int16 strategy;
sortKey->ssup_cxt = CurrentMemoryContext;
sortKey->ssup_collation = scanKey->sk_collation;
sortKey->ssup_nulls_first =
(scanKey->sk_flags & SK_BT_NULLS_FIRST) != 0;
sortKey->ssup_attno = scanKey->sk_attno;
/* Abbreviation is not supported here */
sortKey->abbreviate = false;
AssertState(sortKey->ssup_attno != 0);
strategy = (scanKey->sk_flags & SK_BT_DESC) != 0 ?
BTGreaterStrategyNumber : BTLessStrategyNumber;
PrepareSortSupportFromIndexRel(wstate->index, strategy, sortKey);
}
_bt_freeskey(indexScanKey);
for (;;)
{
load1 = true; /* load BTSpool next ? */
if (itup2 == NULL)
{
if (itup == NULL)
break;
}
else if (itup != NULL)
{
for (i = 1; i <= keysz; i++)
{
SortSupport entry;
Datum attrDatum1,
attrDatum2;
bool isNull1,
isNull2;
int32 compare;
entry = sortKeys + i - 1;
attrDatum1 = index_getattr(itup, i, tupdes, &isNull1);
attrDatum2 = index_getattr(itup2, i, tupdes, &isNull2);
compare = ApplySortComparator(attrDatum1, isNull1,
attrDatum2, isNull2,
entry);
if (compare > 0)
{
load1 = false;
break;
}
else if (compare < 0)
break;
}
}
else
load1 = false;
/* When we see first tuple, create first index page */
if (state == NULL)
state = _bt_pagestate(wstate, 0);
if (load1)
{
_bt_buildadd(wstate, state, itup);
itup = tuplesort_getindextuple(btspool->sortstate, true);
}
else
{
_bt_buildadd(wstate, state, itup2);
itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
}
}
pfree(sortKeys);
}
else
{
/* merge is unnecessary */
while ((itup = tuplesort_getindextuple(btspool->sortstate,
true)) != NULL)
{
/* When we see first tuple, create first index page */
if (state == NULL)
state = _bt_pagestate(wstate, 0);
_bt_buildadd(wstate, state, itup);
}
}
/* Close down final pages and write the metapage */
_bt_uppershutdown(wstate, state);
/*
* If the index is WAL-logged, we must fsync it down to disk before it's
* safe to commit the transaction. (For a non-WAL-logged index we don't
* care since the index will be uninteresting after a crash anyway.)
*
* It's obvious that we must do this when not WAL-logging the build. It's
* less obvious that we have to do it even if we did WAL-log the index
* pages. The reason is that since we're building outside shared buffers,
* a CHECKPOINT occurring during the build has no way to flush the
* previously written data to disk (indeed it won't know the index even
* exists). A crash later on would replay WAL from the checkpoint,
* therefore it wouldn't replay our earlier WAL entries. If we do not
* fsync those pages here, they might still not be on disk when the crash
* occurs.
*/
if (RelationNeedsWAL(wstate->index))
{
RelationOpenSmgr(wstate->index);
smgrimmedsync(wstate->index->rd_smgr, MAIN_FORKNUM);
}
}
/*
* _bt_mergeload - Merge two streams of index tuples into new index files.
*/
static void
_bt_mergeload(Spooler *self, BTWriteState *wstate, BTSpool *btspool, BTReader *btspool2, Relation heapRel)
{
BTPageState *state = NULL;
IndexTuple itup,
itup2;
bool should_free = false;
TupleDesc tupdes = RelationGetDescr(wstate->index);
int keysz = RelationGetNumberOfAttributes(wstate->index);
ScanKey indexScanKey;
ON_DUPLICATE on_duplicate = self->on_duplicate;
Assert(btspool != NULL);
/* the preparation of merge */
itup = BTSpoolGetNextItem(btspool, NULL, &should_free);
itup2 = BTReaderGetNextItem(btspool2);
indexScanKey = _bt_mkscankey_nodata(wstate->index);
for (;;)
{
bool load1 = true; /* load BTSpool next ? */
bool hasnull;
int32 compare;
if (self->dup_old + self->dup_new > self->max_dup_errors)
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("Maximum duplicate error count exceeded")));
if (itup2 == NULL)
{
if (itup == NULL)
break;
}
else if (itup != NULL)
{
compare = compare_indextuple(itup, itup2, indexScanKey,
keysz, tupdes, &hasnull);
if (compare == 0 && !hasnull && btspool->isunique)
{
ItemPointerData t_tid2;
/*
* t_tid is update by heap_is_visible(), because use it for an
* index, t_tid backup
*/
ItemPointerCopy(&itup2->t_tid, &t_tid2);
/* The tuple pointed by the old index should not be visible. */
if (!heap_is_visible(heapRel, &itup->t_tid))
{
itup = BTSpoolGetNextItem(btspool, itup, &should_free);
}
else if (!heap_is_visible(heapRel, &itup2->t_tid))
{
itup2 = BTReaderGetNextItem(btspool2);
}
else
{
if (on_duplicate == ON_DUPLICATE_KEEP_NEW)
{
self->dup_old++;
remove_duplicate(self, heapRel, itup2,
RelationGetRelationName(wstate->index));
itup2 = BTReaderGetNextItem(btspool2);
}
else
{
ItemPointerCopy(&t_tid2, &itup2->t_tid);
self->dup_new++;
remove_duplicate(self, heapRel, itup,
RelationGetRelationName(wstate->index));
itup = BTSpoolGetNextItem(btspool, itup, &should_free);
}
}
continue;
}
else if (compare > 0)
load1 = false;
}
else
load1 = false;
BULKLOAD_PROFILE(&prof_merge_unique);
/* When we see first tuple, create first index page */
if (state == NULL)
state = _bt_pagestate(wstate, 0);
if (load1)
{
IndexTuple next_itup = NULL;
bool next_should_free = false;
for (;;)
{
/* get next item */
next_itup = BTSpoolGetNextItem(btspool, next_itup,
&next_should_free);
if (!btspool->isunique || next_itup == NULL)
break;
compare = compare_indextuple(itup, next_itup, indexScanKey,
keysz, tupdes, &hasnull);
if (compare < 0 || hasnull)
break;
if (compare > 0)
{
/* shouldn't happen */
elog(ERROR, "faild in tuplesort_performsort");
}
/*
* If tupple is deleted by other unique indexes, not visible
*/
if (!heap_is_visible(heapRel, &next_itup->t_tid))
{
continue;
}
if (!heap_is_visible(heapRel, &itup->t_tid))
{
if (should_free)
pfree(itup);
itup = next_itup;
should_free = next_should_free;
next_should_free = false;
continue;
}
/* not unique between input files */
self->dup_new++;
remove_duplicate(self, heapRel, next_itup,
RelationGetRelationName(wstate->index));
if (self->dup_old + self->dup_new > self->max_dup_errors)
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("Maximum duplicate error count exceeded")));
}
_bt_buildadd(wstate, state, itup);
if (should_free)
pfree(itup);
itup = next_itup;
should_free = next_should_free;
}
else
{
_bt_buildadd(wstate, state, itup2);
itup2 = BTReaderGetNextItem(btspool2);
}
BULKLOAD_PROFILE(&prof_merge_insert);
}
_bt_freeskey(indexScanKey);
/* Close down final pages and write the metapage */
_bt_uppershutdown(wstate, state);
/*
* If the index isn't temp, we must fsync it down to disk before it's safe
* to commit the transaction. (For a temp index we don't care since the
* index will be uninteresting after a crash anyway.)
*
* It's obvious that we must do this when not WAL-logging the build. It's
* less obvious that we have to do it even if we did WAL-log the index
* pages. The reason is that since we're building outside shared buffers,
* a CHECKPOINT occurring during the build has no way to flush the
* previously written data to disk (indeed it won't know the index even
* exists). A crash later on would replay WAL from the checkpoint,
* therefore it wouldn't replay our earlier WAL entries. If we do not
* fsync those pages here, they might still not be on disk when the crash
* occurs.
*/
if (!RELATION_IS_LOCAL(wstate->index))
{
RelationOpenSmgr(wstate->index);
smgrimmedsync(wstate->index->rd_smgr, MAIN_FORKNUM);
}
BULKLOAD_PROFILE(&prof_merge_term);
}
static int
FileRepPrimary_ResyncBufferPoolIncrementalWrite(ChangeTrackingRequest *request)
{
int status = STATUS_OK;
Page page;
Buffer buf;
BlockNumber numBlocks = 0;
SMgrRelation smgr_relation = NULL;
char relidstr[OIDCHARS + 1 + OIDCHARS + 1 + OIDCHARS + 1];
int ii;
XLogRecPtr loc;
XLogRecPtr loc1;
int count = 0;
int thresholdCount = 0;
bool mirrorDataLossOccurred = FALSE;
int NumberOfRelations = request->count;
FileRepResyncHashEntry_s entry;
ChangeTrackingResult *result = NULL;
while (1)
{
/* allow flushing buffers from buffer pool during scan */
FileRepResync_SetReadBufferRequest();
if ((result = ChangeTracking_GetChanges(request)) != NULL)
{
FileRepResync_ResetReadBufferRequest();
for (ii = 0; ii < result->count; ii++)
{
if (smgr_relation == NULL)
{
NumberOfRelations--;
smgr_relation = smgropen(result->entries[ii].relFileNode);
snprintf(relidstr, sizeof(relidstr), "%u/%u/%u",
smgr_relation->smgr_rnode.spcNode,
smgr_relation->smgr_rnode.dbNode,
smgr_relation->smgr_rnode.relNode);
numBlocks = smgrnblocks(smgr_relation);
if (Debug_filerep_print)
elog(LOG, "resynchronize buffer pool relation '%u/%u/%u' "
"number of blocks:'%u' ",
smgr_relation->smgr_rnode.spcNode,
smgr_relation->smgr_rnode.dbNode,
smgr_relation->smgr_rnode.relNode,
numBlocks);
thresholdCount = Min(numBlocks, 1024);
}
loc1 = result->entries[ii].lsn_end;
/*
* if relation was truncated then block_num from change tracking can be beyond numBlocks
*/
if (result->entries[ii].block_num >= numBlocks)
{
ereport(LOG,
(errmsg("could not resynchonize buffer pool relation '%s' block '%d' (maybe due to truncate), "
"lsn change tracking '%s(%u/%u)' "
"number of blocks '%d' ",
relidstr,
result->entries[ii].block_num,
XLogLocationToString(&loc1),
loc1.xlogid,
loc1.xrecoff,
numBlocks),
FileRep_errcontext()));
goto flush_check;
}
/* allow flushing buffers from buffer pool during scan */
FileRepResync_SetReadBufferRequest();
buf = ReadBuffer_Resync(smgr_relation,
result->entries[ii].block_num,
relidstr);
FileRepResync_ResetReadBufferRequest();
Assert(result->entries[ii].block_num < numBlocks);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
page = BufferGetPage(buf);
loc = PageGetLSN(page);
if(Debug_filerep_print)
{
elog(LOG,
"incremental resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn page '%s(%u/%u)' "
"lsn end change tracking '%s(%u/%u)' ",
relidstr,
numBlocks,
result->entries[ii].block_num,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&loc1),
result->entries[ii].lsn_end.xlogid,
result->entries[ii].lsn_end.xrecoff);
}
else
{
char tmpBuf[FILEREP_MAX_LOG_DESCRIPTION_LEN];
snprintf(tmpBuf, sizeof(tmpBuf),
"incremental resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn page '%s(%u/%u)' ",
relidstr,
numBlocks,
result->entries[ii].block_num,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
snprintf(tmpBuf, sizeof(tmpBuf),
"incremental resync buffer pool identifier '%s' lsn end change tracking '%s(%u/%u)' ",
relidstr,
XLogLocationToString(&loc1),
result->entries[ii].lsn_end.xlogid,
result->entries[ii].lsn_end.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
}
if (XLByteLE(result->entries[ii].lsn_end, PageGetLSN(page)))
{
if (! XLByteEQ(PageGetLSN(page), result->entries[ii].lsn_end))
{
ereport(LOG,
(errmsg("Resynchonize buffer pool relation '%s' block '%d' has page lsn less than CT lsn, "
"lsn end change tracking '%s(%u/%u)' lsn page '%s(%u/%u)' "
"number of blocks '%d'",
relidstr,
result->entries[ii].block_num,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&loc1),
loc1.xlogid,
loc1.xrecoff,
numBlocks),
FileRep_errcontext()));
}
/*
* It's safe and better to perform write of the page to mirror,
* for this case, as primary and mirror data pages should always
* be same. So, we might do some extra work but definitely won't
* loose out blocks, or error out and need to perform full recovery.
* Need to cover for this case as there are some known scenarios where
* CT file can have extra records which should have been discarded,
* but as we loose out information of xlog LSN cannot be discarded.
* One such case is when CT_TRANSIENT being compacted to CT_COMPACT
* with specific xlog LSN (to discard extra records) in CT mode gets
* interrupted by resync. Compaction during Resync collects all the
* CT records and doesn't have xlog LSN information to discard any
* extra records from CT_TRANSIENT.
*/
smgrwrite(smgr_relation,
result->entries[ii].block_num,
(char *)BufferGetBlock(buf),
FALSE);
}
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorker,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
UnlockReleaseBuffer(buf);
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorker,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
flush_check:
if (((ii + 1) == result->count) ||
! (result->entries[ii].relFileNode.spcNode == result->entries[ii+1].relFileNode.spcNode &&
result->entries[ii].relFileNode.dbNode == result->entries[ii+1].relFileNode.dbNode &&
result->entries[ii].relFileNode.relNode == result->entries[ii+1].relFileNode.relNode))
{
if (result->ask_for_more == false)
{
smgrimmedsync(smgr_relation);
smgrclose(smgr_relation);
smgr_relation = NULL;
FileRep_GetRelationPath(
entry.fileName,
result->entries[ii].relFileNode,
0 /* segment file number is always 0 for Buffer Pool */);
status = FileRepResync_UpdateEntry(&entry);
if (status != STATUS_OK)
{
break;
}
}
}
if (count > thresholdCount)
{
count = 0;
FileRepSubProcess_ProcessSignals();
if (! (FileRepSubProcess_GetState() == FileRepStateReady &&
dataState == DataStateInResync))
{
mirrorDataLossOccurred = TRUE;
break;
}
}
else
count++;
} // for (ii = 0; ii < result->count; ii++)
} // if ((result = ChangeTracking_GetChanges(request)) != NULL)
FileRepResync_ResetReadBufferRequest();
if (result != NULL && result->ask_for_more == true)
{
Assert(request->count == 1);
request->entries[0].lsn_start = result->next_start_lsn;
}
else
{
break;
}
} // while(1)
ChangeTracking_FreeRequest(request);
ChangeTracking_FreeResult(result);
Insist(NumberOfRelations == 0);
if (mirrorDataLossOccurred)
status = STATUS_ERROR;
return status;
}
static int
FileRepPrimary_ResyncWrite(FileRepResyncHashEntry_s *entry)
{
int status = STATUS_OK;
Page page;
Buffer buf;
BlockNumber numBlocks;
BlockNumber blkno;
SMgrRelation smgr_relation;
char relidstr[OIDCHARS + 1 + OIDCHARS + 1 + OIDCHARS + 1];
XLogRecPtr loc;
int count = 0;
int thresholdCount = 0;
bool mirrorDataLossOccurred = FALSE;
switch (entry->relStorageMgr)
{
case PersistentFileSysRelStorageMgr_BufferPool:
switch (entry->mirrorDataSynchronizationState)
{
case MirroredRelDataSynchronizationState_BufferPoolScanIncremental:
case MirroredRelDataSynchronizationState_FullCopy:
smgr_relation = smgropen(entry->relFileNode);
numBlocks = smgrnblocks(smgr_relation);
snprintf(relidstr, sizeof(relidstr), "%u/%u/%u",
smgr_relation->smgr_rnode.spcNode,
smgr_relation->smgr_rnode.dbNode,
smgr_relation->smgr_rnode.relNode);
if (Debug_filerep_print)
elog(LOG, "resync buffer pool relation '%s' number of blocks '%d' ",
relidstr, numBlocks);
thresholdCount = Min(numBlocks, 1024);
/*
* required in order to report how many blocks were synchronized
* if gp_persistent_relation_node does not return that information
*/
if (entry->mirrorBufpoolResyncChangedPageCount == 0)
{
entry->mirrorBufpoolResyncChangedPageCount = numBlocks - entry->mirrorBufpoolResyncCkptBlockNum;
}
for (blkno = entry->mirrorBufpoolResyncCkptBlockNum; blkno < numBlocks; blkno++)
{
XLogRecPtr endResyncLSN = (isFullResync() ?
FileRepResync_GetEndFullResyncLSN() :
FileRepResync_GetEndIncrResyncLSN());
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorkerRead,
DDLNotSpecified,
"", //databaseName
""); // tableName
#endif
FileRepResync_SetReadBufferRequest();
buf = ReadBuffer_Resync(smgr_relation, blkno, relidstr);
FileRepResync_ResetReadBufferRequest();
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
page = BufferGetPage(buf);
loc = PageGetLSN(page);
if (Debug_filerep_print)
{
elog(LOG,
"full resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn begin change tracking '%s(%u/%u)' "
"lsn page '%s(%u/%u)' lsn end change tracking '%s(%u/%u)' ",
relidstr,
numBlocks,
blkno,
XLogLocationToString(&entry->mirrorBufpoolResyncCkptLoc),
entry->mirrorBufpoolResyncCkptLoc.xlogid,
entry->mirrorBufpoolResyncCkptLoc.xrecoff,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&endResyncLSN),
endResyncLSN.xlogid,
endResyncLSN.xrecoff);
}
else
{
char tmpBuf[FILEREP_MAX_LOG_DESCRIPTION_LEN];
snprintf(tmpBuf, sizeof(tmpBuf),
"full resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn begin change tracking '%s(%u/%u)' ",
relidstr,
numBlocks,
blkno,
XLogLocationToString(&entry->mirrorBufpoolResyncCkptLoc),
entry->mirrorBufpoolResyncCkptLoc.xlogid,
entry->mirrorBufpoolResyncCkptLoc.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
snprintf(tmpBuf, sizeof(tmpBuf),
"full resync buffer pool identifier '%s' lsn page '%s(%u/%u)' lsn end change tracking '%s(%u/%u)' ",
relidstr,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&endResyncLSN),
endResyncLSN.xlogid,
endResyncLSN.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
}
if (XLByteLE(PageGetLSN(page), endResyncLSN) &&
XLByteLE(entry->mirrorBufpoolResyncCkptLoc, PageGetLSN(page)))
{
smgrwrite(smgr_relation,
blkno,
(char *)BufferGetBlock(buf),
FALSE);
}
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorker,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
UnlockReleaseBuffer(buf);
if (count > thresholdCount)
{
count = 0;
FileRepSubProcess_ProcessSignals();
if (! (FileRepSubProcess_GetState() == FileRepStateReady &&
dataState == DataStateInResync))
{
mirrorDataLossOccurred = TRUE;
break;
}
}
else
count++;
}
if (mirrorDataLossOccurred)
break;
if (entry->mirrorDataSynchronizationState != MirroredRelDataSynchronizationState_FullCopy)
{
LockRelationForResyncExtension(&smgr_relation->smgr_rnode, ExclusiveLock);
numBlocks = smgrnblocks(smgr_relation);
smgrtruncate(smgr_relation,
numBlocks,
TRUE /* isTemp, TRUE means to not record in XLOG */,
FALSE /* isLocalBuf */,
&entry->persistentTid,
entry->persistentSerialNum);
UnlockRelationForResyncExtension(&smgr_relation->smgr_rnode, ExclusiveLock);
}
smgrimmedsync(smgr_relation);
smgrclose(smgr_relation);
smgr_relation = NULL;
break;
case MirroredRelDataSynchronizationState_None:
case MirroredRelDataSynchronizationState_DataSynchronized:
break;
default:
ereport(LOG,
(errmsg("could not resynchronize relation '%u/%u/%u' "
"mirror synchronization state:'%s(%d)' ",
entry->relFileNode.relNode,
entry->relFileNode.spcNode,
entry->relFileNode.dbNode,
MirroredRelDataSynchronizationState_Name(entry->mirrorDataSynchronizationState),
entry->mirrorDataSynchronizationState)));
break;
}
break;
case PersistentFileSysRelStorageMgr_AppendOnly:
{
MirroredAppendOnlyOpen mirroredOpen;
int primaryError;
bool mirrorDataLossOccurred;
char *buffer = NULL;
int64 endOffset = entry->mirrorAppendOnlyNewEof;
int64 startOffset = entry->mirrorAppendOnlyLossEof;
int32 bufferLen = 0;
int retval = 0;
switch (entry->mirrorDataSynchronizationState)
{
case MirroredRelDataSynchronizationState_AppendOnlyCatchup:
case MirroredRelDataSynchronizationState_FullCopy:
/*
* required in order to report how many blocks were synchronized
* if gp_persistent_relation_node does not return that information
*/
if (entry->mirrorBufpoolResyncChangedPageCount == 0)
{
entry->mirrorBufpoolResyncChangedPageCount = (endOffset - startOffset) / BLCKSZ;
}
/*
* The MirroredAppendOnly_OpenResynchonize routine knows we are a resynch worker and
* will open BOTH, but write only the MIRROR!!!
*/
MirroredAppendOnly_OpenResynchonize(
&mirroredOpen,
&entry->relFileNode,
entry->segmentFileNum,
startOffset,
&primaryError,
&mirrorDataLossOccurred);
if (primaryError != 0)
{
ereport(ERROR,
(errcode_for_file_access(),
errmsg("could not open file %u/%u/%u.%u : %s",
entry->relFileNode.dbNode,
entry->relFileNode.spcNode,
entry->relFileNode.relNode,
entry->segmentFileNum,
strerror(primaryError))));
break;
}
if (mirrorDataLossOccurred)
break;
/* AO and CO Data Store writes 64k size by default */
bufferLen = (Size) Min(2*BLCKSZ, endOffset - startOffset);
buffer = (char*) palloc(bufferLen);
if (buffer == NULL)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
(errmsg("not enough memory for resynchronization"))));
MemSet(buffer, 0, bufferLen);
while (startOffset < endOffset)
{
retval = MirroredAppendOnly_Read(
&mirroredOpen,
buffer,
bufferLen);
if (retval != bufferLen)
{
ereport(ERROR,
(errcode_for_file_access(),
errmsg("could not read from position:" INT64_FORMAT " in file %u/%u/%u.%u : %m",
startOffset,
entry->relFileNode.dbNode,
entry->relFileNode.spcNode,
entry->relFileNode.relNode,
entry->segmentFileNum)));
break;
}
MirroredAppendOnly_Append(
&mirroredOpen,
buffer,
bufferLen,
&primaryError,
&mirrorDataLossOccurred);
if (mirrorDataLossOccurred)
break;
Assert(primaryError == 0); // No primary writes as resync worker.
startOffset += bufferLen;
/* AO and CO Data Store writes 64k size by default */
bufferLen = (Size) Min(2*BLCKSZ, endOffset - startOffset);
}
if (buffer)
{
pfree(buffer);
buffer = NULL;
}
if (mirrorDataLossOccurred)
break;
/* Flush written data on Mirror */
MirroredAppendOnly_Flush(
&mirroredOpen,
&primaryError,
&mirrorDataLossOccurred);
if (mirrorDataLossOccurred)
break;
Assert(primaryError == 0); // Not flushed on primary as resync worker.
/* Close Primary and Mirror */
MirroredAppendOnly_Close(
&mirroredOpen,
&mirrorDataLossOccurred);
break;
case MirroredRelDataSynchronizationState_None:
case MirroredRelDataSynchronizationState_DataSynchronized:
break;
default:
ereport(LOG,
(errmsg("could not resynchronize relation '%u/%u/%u' "
"mirror synchronization state:'%s(%d)' ",
entry->relFileNode.relNode,
entry->relFileNode.spcNode,
entry->relFileNode.dbNode,
MirroredRelDataSynchronizationState_Name(entry->mirrorDataSynchronizationState),
entry->mirrorDataSynchronizationState)));
break;
}
break;
} //case
default:
Assert(0);
break;
} //switch
if (mirrorDataLossOccurred)
status = STATUS_ERROR;
return status;
}
/*
* Read tuples in correct sort order from tuplesort, and load them into
* btree leaves.
*/
static void
_bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
{
BTPageState *state = NULL;
bool merge = (btspool2 != NULL);
IndexTuple itup,
itup2 = NULL;
bool should_free,
should_free2,
load1;
TupleDesc tupdes = RelationGetDescr(wstate->index);
int i,
keysz = RelationGetNumberOfAttributes(wstate->index);
ScanKey indexScanKey = NULL;
if (merge)
{
/*
* Another BTSpool for dead tuples exists. Now we have to merge
* btspool and btspool2.
*/
/* the preparation of merge */
itup = tuplesort_getindextuple(btspool->sortstate,
true, &should_free);
itup2 = tuplesort_getindextuple(btspool2->sortstate,
true, &should_free2);
indexScanKey = _bt_mkscankey_nodata(wstate->index);
for (;;)
{
load1 = true; /* load BTSpool next ? */
if (itup2 == NULL)
{
if (itup == NULL)
break;
}
else if (itup != NULL)
{
for (i = 1; i <= keysz; i++)
{
ScanKey entry;
Datum attrDatum1,
attrDatum2;
bool isNull1,
isNull2;
int32 compare;
entry = indexScanKey + i - 1;
attrDatum1 = index_getattr(itup, i, tupdes, &isNull1);
attrDatum2 = index_getattr(itup2, i, tupdes, &isNull2);
if (isNull1)
{
if (isNull2)
compare = 0; /* NULL "=" NULL */
else if (entry->sk_flags & SK_BT_NULLS_FIRST)
compare = -1; /* NULL "<" NOT_NULL */
else
compare = 1; /* NULL ">" NOT_NULL */
}
else if (isNull2)
{
if (entry->sk_flags & SK_BT_NULLS_FIRST)
compare = 1; /* NOT_NULL ">" NULL */
else
compare = -1; /* NOT_NULL "<" NULL */
}
else
{
compare =
DatumGetInt32(FunctionCall2Coll(&entry->sk_func,
entry->sk_collation,
attrDatum1,
attrDatum2));
if (entry->sk_flags & SK_BT_DESC)
compare = -compare;
}
if (compare > 0)
{
load1 = false;
break;
}
else if (compare < 0)
break;
}
}
else
load1 = false;
/* When we see first tuple, create first index page */
if (state == NULL)
state = _bt_pagestate(wstate, 0);
if (load1)
{
_bt_buildadd(wstate, state, itup);
if (should_free)
pfree(itup);
itup = tuplesort_getindextuple(btspool->sortstate,
true, &should_free);
}
else
{
_bt_buildadd(wstate, state, itup2);
if (should_free2)
pfree(itup2);
itup2 = tuplesort_getindextuple(btspool2->sortstate,
true, &should_free2);
}
}
_bt_freeskey(indexScanKey);
}
else
{
/* merge is unnecessary */
while ((itup = tuplesort_getindextuple(btspool->sortstate,
true, &should_free)) != NULL)
{
/* When we see first tuple, create first index page */
if (state == NULL)
state = _bt_pagestate(wstate, 0);
_bt_buildadd(wstate, state, itup);
if (should_free)
pfree(itup);
}
}
/* Close down final pages and write the metapage */
_bt_uppershutdown(wstate, state);
/*
* If the index is WAL-logged, we must fsync it down to disk before it's
* safe to commit the transaction. (For a non-WAL-logged index we don't
* care since the index will be uninteresting after a crash anyway.)
*
* It's obvious that we must do this when not WAL-logging the build. It's
* less obvious that we have to do it even if we did WAL-log the index
* pages. The reason is that since we're building outside shared buffers,
* a CHECKPOINT occurring during the build has no way to flush the
* previously written data to disk (indeed it won't know the index even
* exists). A crash later on would replay WAL from the checkpoint,
* therefore it wouldn't replay our earlier WAL entries. If we do not
* fsync those pages here, they might still not be on disk when the crash
* occurs.
*/
if (RelationNeedsWAL(wstate->index))
{
RelationOpenSmgr(wstate->index);
smgrimmedsync(wstate->index->rd_smgr, MAIN_FORKNUM);
}
}
void PersistentStore_UpdateTuple(
PersistentStoreData *storeData,
PersistentStoreSharedData *storeSharedData,
ItemPointer persistentTid,
/* TID of the stored tuple. */
Datum *values,
bool flushToXLog)
/* When true, the XLOG record for this change will be flushed to disk. */
{
Relation persistentRel;
bool *nulls;
HeapTuple persistentTuple = NULL;
XLogRecPtr xlogUpdateEndLoc;
#ifdef USE_ASSERT_CHECKING
if (storeSharedData == NULL ||
!PersistentStoreSharedData_EyecatcherIsValid(storeSharedData))
elog(ERROR, "Persistent store shared-memory not valid");
#endif
if (Debug_persistent_store_print)
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_ReplaceTuple: Going to update whole tuple at TID %s ('%s', shared data %p)",
ItemPointerToString(persistentTid),
storeData->tableName,
storeSharedData);
persistentRel = (*storeData->openRel)();
/*
* In order to keep the tuples the exact same size to enable direct reuse of
* free tuples, we do not use NULLs.
*/
nulls = (bool*)palloc0(storeData->numAttributes * sizeof(bool));
/*
* Form the tuple.
*/
persistentTuple = heap_form_tuple(persistentRel->rd_att, values, nulls);
if (!HeapTupleIsValid(persistentTuple))
elog(ERROR, "Failed to build persistent tuple ('%s')",
storeData->tableName);
persistentTuple->t_self = *persistentTid;
frozen_heap_inplace_update(persistentRel, persistentTuple);
/*
* Return the XLOG location of the UPDATE tuple's XLOG record.
*/
xlogUpdateEndLoc = XLogLastInsertEndLoc();
heap_freetuple(persistentTuple);
#ifdef FAULT_INJECTOR
if (FaultInjector_InjectFaultIfSet(SyncPersistentTable,
DDLNotSpecified,
"" /* databaseName */,
"" /* tableName */)== FaultInjectorTypeSkip)
{
FlushRelationBuffers(persistentRel);
smgrimmedsync(persistentRel->rd_smgr);
}
#endif
(*storeData->closeRel)(persistentRel);
if (Debug_persistent_store_print)
{
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_UpdateTuple: Updated whole tuple at TID %s ('%s')",
ItemPointerToString(persistentTid),
storeData->tableName);
(*storeData->printTupleCallback)(
PersistentStore_DebugPrintLevel(),
"STORE UPDATED TUPLE",
persistentTid,
values);
}
if (flushToXLog)
{
XLogFlush(xlogUpdateEndLoc);
XLogRecPtr_Zero(&nowaitXLogEndLoc);
}
else
nowaitXLogEndLoc = xlogUpdateEndLoc;
}
static void PersistentStore_InsertTuple(
PersistentStoreData *storeData,
PersistentStoreSharedData *storeSharedData,
Datum *values,
bool flushToXLog,
/* When true, the XLOG record for this change will be flushed to disk. */
ItemPointer persistentTid)
/* TID of the stored tuple. */
{
Relation persistentRel;
#ifdef USE_ASSERT_CHECKING
if (storeSharedData == NULL ||
!PersistentStoreSharedData_EyecatcherIsValid(storeSharedData))
elog(ERROR, "Persistent store shared-memory not valid");
#endif
if (Debug_persistent_store_print)
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_InsertTuple: Going to insert new tuple ('%s', shared data %p)",
storeData->tableName,
storeSharedData);
persistentRel = (*storeData->openRel)();
PersistentStore_DoInsertTuple(
storeData,
storeSharedData,
persistentRel,
values,
flushToXLog,
persistentTid);
#ifdef FAULT_INJECTOR
if (FaultInjector_InjectFaultIfSet(SyncPersistentTable,
DDLNotSpecified,
"" /* databaseName */,
"" /* tableName */)== FaultInjectorTypeSkip)
{
FlushRelationBuffers(persistentRel);
smgrimmedsync(persistentRel->rd_smgr);
}
#endif
(*storeData->closeRel)(persistentRel);
if (Debug_persistent_store_print)
{
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_InsertTuple: Inserted new tuple at TID %s ('%s')",
ItemPointerToString(persistentTid),
storeData->tableName);
(*storeData->printTupleCallback)(
PersistentStore_DebugPrintLevel(),
"STORE INSERT TUPLE",
persistentTid,
values);
}
}
static int64
PersistentBuild_TruncateAllGpRelationNode(void)
{
Relation pg_database;
HeapScanDesc scan;
HeapTuple tuple;
int64 count;
pg_database = heap_open(
DatabaseRelationId,
AccessShareLock);
/*
* Truncate gp_relation_node and its index in each database.
*/
scan = heap_beginscan(pg_database, SnapshotNow, 0, NULL);
count = 0;
while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
Form_pg_database form_pg_database =
(Form_pg_database)GETSTRUCT(tuple);
Oid dbOid;
Oid dattablespace;
RelFileNode relFileNode;
SMgrRelation smgrRelation;
Page btree_metapage;
dbOid = HeapTupleGetOid(tuple);
dattablespace = form_pg_database->dattablespace;
if (dbOid == HcatalogDbOid)
continue;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"PersistentBuild_TruncateAllGpRelationNode: dbOid %u, '%s'",
dbOid,
form_pg_database->datname.data);
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Truncating gp_relation_node %u/%u/%u in database oid %u ('%s')",
relFileNode.spcNode,
relFileNode.dbNode,
relFileNode.relNode,
dbOid,
form_pg_database->datname.data);
relFileNode.spcNode = dattablespace;
relFileNode.dbNode = dbOid;
relFileNode.relNode = GpRelfileNodeRelationId;
/*
* Truncate WITHOUT generating an XLOG record (i.e. pretend it is a temp relation).
*/
PersistentBuild_NonTransactionTruncate(&relFileNode);
count++;
/*
* And, the index. Unfortunately, the relfilenode OID can change due to a
* REINDEX {TABLE|INDEX} command.
*/
PersistentBuild_FindGpRelationNodeIndex(
dbOid,
dattablespace,
&relFileNode);
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Truncating gp_relation_node_index %u/%u/%u in database oid %u ('%s'). relfilenode different %s, tablespace different %s",
relFileNode.spcNode,
relFileNode.dbNode,
relFileNode.relNode,
dbOid,
form_pg_database->datname.data,
((relFileNode.relNode != GpRelfileNodeOidIndexId) ? "true" : "false"),
((relFileNode.spcNode != dattablespace) ? "true" : "false"));
PersistentBuild_NonTransactionTruncate(&relFileNode);
// The BTree needs an empty meta-data block.
smgrRelation = smgropen(relFileNode);
btree_metapage = (Page)palloc(BLCKSZ);
_bt_initmetapage(btree_metapage, P_NONE, 0);
smgrwrite(
smgrRelation,
/* blockNum */ 0,
(char*)btree_metapage,
/* isTemp */ false);
smgrimmedsync(smgrRelation);
pfree(btree_metapage);
smgrclose(smgrRelation);
count++;
}
heap_endscan(scan);
heap_close(pg_database, AccessShareLock);
return count;
}
