/*
* 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);
}
/*
* Remove the visibility map fork for a relation. If there turn out to be
* any bugs in the visibility map code that require rebuilding the VM, this
* provides users with a way to do it that is cleaner than shutting down the
* server and removing files by hand.
*
* This is a cut-down version of RelationTruncate.
*/
Datum
pg_truncate_visibility_map(PG_FUNCTION_ARGS)
{
Oid relid = PG_GETARG_OID(0);
Relation rel;
rel = relation_open(relid, AccessExclusiveLock);
if (rel->rd_rel->relkind != RELKIND_RELATION &&
rel->rd_rel->relkind != RELKIND_MATVIEW &&
rel->rd_rel->relkind != RELKIND_TOASTVALUE)
ereport(ERROR,
(errcode(ERRCODE_WRONG_OBJECT_TYPE),
errmsg("\"%s\" is not a table, materialized view, or TOAST table",
RelationGetRelationName(rel))));
RelationOpenSmgr(rel);
rel->rd_smgr->smgr_vm_nblocks = InvalidBlockNumber;
visibilitymap_truncate(rel, 0);
if (RelationNeedsWAL(rel))
{
xl_smgr_truncate xlrec;
xlrec.blkno = 0;
xlrec.rnode = rel->rd_node;
xlrec.flags = SMGR_TRUNCATE_VM;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, sizeof(xlrec));
XLogInsert(RM_SMGR_ID, XLOG_SMGR_TRUNCATE | XLR_SPECIAL_REL_UPDATE);
}
/*
* Release the lock right away, not at commit time.
*
* It would be a problem to release the lock prior to commit if this
* truncate operation sends any transactional invalidation messages. Other
* backends would potentially be able to lock the relation without
* processing them in the window of time between when we release the lock
* here and when we sent the messages at our eventual commit. However,
* we're currently only sending a non-transactional smgr invalidation,
* which will have been posted to shared memory immediately from within
* visibilitymap_truncate. Therefore, there should be no race here.
*
* The reason why it's desirable to release the lock early here is because
* of the possibility that someone will need to use this to blow away many
* visibility map forks at once. If we can't release the lock until
* commit time, the transaction doing this will accumulate
* AccessExclusiveLocks on all of those relations at the same time, which
* is undesirable. However, if this turns out to be unsafe we may have no
* choice...
*/
relation_close(rel, AccessExclusiveLock);
/* Nothing to return. */
PG_RETURN_VOID();
}
/*
* Ensure that the visibility map fork is at least vm_nblocks long, extending
* it if necessary with zeroed pages.
*/
static void
vm_extend(Relation rel, BlockNumber vm_nblocks)
{
BlockNumber vm_nblocks_now;
Page pg;
pg = (Page) palloc(BLCKSZ);
PageInit(pg, BLCKSZ, 0);
/*
* We use the relation extension lock to lock out other backends trying to
* extend the visibility map at the same time. It also locks out extension
* of the main fork, unnecessarily, but extending the visibility map
* happens seldom enough that it doesn't seem worthwhile to have a
* separate lock tag type for it.
*
* Note that another backend might have extended or created the relation
* by the time we get the lock.
*/
LockRelationForExtension(rel, ExclusiveLock);
/* Might have to re-open if a cache flush happened */
RelationOpenSmgr(rel);
/*
* Create the file first if it doesn't exist. If smgr_vm_nblocks is
* positive then it must exist, no need for an smgrexists call.
*/
if ((rel->rd_smgr->smgr_vm_nblocks == 0 ||
rel->rd_smgr->smgr_vm_nblocks == InvalidBlockNumber) &&
!smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
smgrcreate(rel->rd_smgr, VISIBILITYMAP_FORKNUM, false);
vm_nblocks_now = smgrnblocks(rel->rd_smgr, VISIBILITYMAP_FORKNUM);
/* Now extend the file */
while (vm_nblocks_now < vm_nblocks)
{
smgrextend(rel->rd_smgr, VISIBILITYMAP_FORKNUM, vm_nblocks_now,
(char *) pg, false);
vm_nblocks_now++;
}
/*
* Send a shared-inval message to force other backends to close any smgr
* references they may have for this rel, which we are about to change.
* This is a useful optimization because it means that backends don't have
* to keep checking for creation or extension of the file, which happens
* infrequently.
*/
CacheInvalidateSmgr(rel->rd_smgr->smgr_rnode);
/* Update local cache with the up-to-date size */
rel->rd_smgr->smgr_vm_nblocks = vm_nblocks_now;
UnlockRelationForExtension(rel, ExclusiveLock);
pfree(pg);
}
/*
* FreeSpaceMapTruncateRel - adjust for truncation of a relation.
*
* The caller must hold AccessExclusiveLock on the relation, to ensure that
* other backends receive the smgr invalidation event that this function sends
* before they access the FSM again.
*
* nblocks is the new___ size of the heap.
*/
void
FreeSpaceMapTruncateRel(Relation rel, BlockNumber nblocks)
{
BlockNumber new_nfsmblocks;
FSMAddress first_removed_address;
uint16 first_removed_slot;
Buffer buf;
RelationOpenSmgr(rel);
/*
* If no FSM has been created yet for this relation, there's nothing to
* truncate.
*/
if (!smgrexists(rel->rd_smgr, FSM_FORKNUM))
return;
/* Get the location in the FSM of the first removed heap block */
first_removed_address = fsm_get_location(nblocks, &first_removed_slot);
/*
* Zero out the tail of the last remaining FSM page. If the slot
* representing the first removed heap block is at a page boundary, as the
* first slot on the FSM page that first_removed_address points to, we can
* just truncate that page altogether.
*/
if (first_removed_slot > 0)
{
buf = fsm_readbuf(rel, first_removed_address, false);
if (!BufferIsValid(buf))
return; /* nothing to do; the FSM was already smaller */
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
fsm_truncate_avail(BufferGetPage(buf), first_removed_slot);
MarkBufferDirtyHint(buf, false);
UnlockReleaseBuffer(buf);
new_nfsmblocks = fsm_logical_to_physical(first_removed_address) + 1;
}
else
{
new_nfsmblocks = fsm_logical_to_physical(first_removed_address);
if (smgrnblocks(rel->rd_smgr, FSM_FORKNUM) <= new_nfsmblocks)
return; /* nothing to do; the FSM was already smaller */
}
/* Truncate the unused FSM pages, and send smgr inval message */
smgrtruncate(rel->rd_smgr, FSM_FORKNUM, new_nfsmblocks);
/*
* We might as well update the local smgr_fsm_nblocks setting.
* smgrtruncate sent an smgr cache inval message, which will cause other
* backends to invalidate their copy of smgr_fsm_nblocks, and this one too
* at the next command boundary. But this ensures it isn't outright wrong
* until then.
*/
if (rel->rd_smgr)
rel->rd_smgr->smgr_fsm_nblocks = new_nfsmblocks;
}
/*
* emit a completed btree page, and release the working storage.
*/
static void
_bt_blwritepage(BTWriteState *wstate, Page page, BlockNumber blkno)
{
/* Ensure rd_smgr is open (could have been closed by relcache flush!) */
RelationOpenSmgr(wstate->index);
/* XLOG stuff */
if (wstate->btws_use_wal)
{
/* We use the heap NEWPAGE record type for this */
log_newpage(&wstate->index->rd_node, MAIN_FORKNUM, blkno, page);
}
else
{
/* Leave the page LSN zero if not WAL-logged, but set TLI anyway */
PageSetTLI(page, ThisTimeLineID);
}
/*
* If we have to write pages nonsequentially, fill in the space with
* zeroes until we come back and overwrite. This is not logically
* necessary on standard Unix filesystems (unwritten space will read as
* zeroes anyway), but it should help to avoid fragmentation. The dummy
* pages aren't WAL-logged though.
*/
while (blkno > wstate->btws_pages_written)
{
if (!wstate->btws_zeropage)
wstate->btws_zeropage = (Page) palloc0(BLCKSZ);
smgrextend(wstate->index->rd_smgr, MAIN_FORKNUM,
wstate->btws_pages_written++,
(char *) wstate->btws_zeropage,
true);
}
/*
* Now write the page. We say isTemp = true even if it's not a temp
* index, because there's no need for smgr to schedule an fsync for this
* write; we'll do it ourselves before ending the build.
*/
if (blkno == wstate->btws_pages_written)
{
/* extending the file... */
smgrextend(wstate->index->rd_smgr, MAIN_FORKNUM, blkno,
(char *) page, true);
wstate->btws_pages_written++;
}
else
{
/* overwriting a block we zero-filled before */
smgrwrite(wstate->index->rd_smgr, MAIN_FORKNUM, blkno,
(char *) page, true);
}
pfree(page);
}
/*
* End a rewrite.
*
* state and any other resources are freed.
*/
void
end_heap_rewrite(RewriteState state)
{
HASH_SEQ_STATUS seq_status;
UnresolvedTup unresolved;
/*
* Write any remaining tuples in the UnresolvedTups table. If we have any
* left, they should in fact be dead, but let's err on the safe side.
*/
hash_seq_init(&seq_status, state->rs_unresolved_tups);
while ((unresolved = hash_seq_search(&seq_status)) != NULL)
{
ItemPointerSetInvalid(&unresolved->tuple->t_data->t_ctid);
raw_heap_insert(state, unresolved->tuple);
}
/* Write the last page, if any */
if (state->rs_buffer_valid)
{
if (state->rs_use_wal)
log_newpage(&state->rs_new_rel->rd_node,
MAIN_FORKNUM,
state->rs_blockno,
state->rs_buffer,
true);
RelationOpenSmgr(state->rs_new_rel);
PageSetChecksumInplace(state->rs_buffer, state->rs_blockno);
smgrextend(state->rs_new_rel->rd_smgr, MAIN_FORKNUM, state->rs_blockno,
(char *) state->rs_buffer, true);
}
/*
* If the rel is WAL-logged, must fsync before commit. We use heap_sync
* to ensure that the toast table gets fsync'd too.
*
* It's obvious that we must do this when not WAL-logging. It's less
* obvious that we have to do it even if we did WAL-log the pages. The
* reason is the same as in tablecmds.c's copy_relation_data(): we're
* writing data that's not in shared buffers, and so a CHECKPOINT
* occurring during the rewriteheap operation won't have fsync'd data we
* wrote before the checkpoint.
*/
if (RelationNeedsWAL(state->rs_new_rel))
heap_sync(state->rs_new_rel);
/* Deleting the context frees everything */
MemoryContextDelete(state->rs_cxt);
}
/*
* Ensure that the FSM fork is at least fsm_nblocks long, extending
* it if necessary with empty pages. And by empty, I mean pages filled
* with zeros, meaning there's no free space.
*/
static void
fsm_extend(Relation rel, BlockNumber fsm_nblocks)
{
BlockNumber fsm_nblocks_now;
Page pg;
pg = (Page) palloc(BLCKSZ);
PageInit(pg, BLCKSZ, 0);
/*
* We use the relation extension lock to lock out other backends trying to
* extend the FSM at the same time. It also locks out extension of the
* main fork, unnecessarily, but extending the FSM happens seldom enough
* that it doesn't seem worthwhile to have a separate lock tag type for
* it.
*
* Note that another backend might have extended or created the relation
* by the time we get the lock.
*/
LockRelationForExtension(rel, ExclusiveLock);
/* Might have to re-open if a cache flush happened */
RelationOpenSmgr(rel);
/*
* Create the FSM file first if it doesn't exist. If smgr_fsm_nblocks is
* positive then it must exist, no need for an smgrexists call.
*/
if ((rel->rd_smgr->smgr_fsm_nblocks == 0 ||
rel->rd_smgr->smgr_fsm_nblocks == InvalidBlockNumber) &&
!smgrexists(rel->rd_smgr, FSM_FORKNUM))
smgrcreate(rel->rd_smgr, FSM_FORKNUM, false);
fsm_nblocks_now = smgrnblocks(rel->rd_smgr, FSM_FORKNUM);
while (fsm_nblocks_now < fsm_nblocks)
{
PageSetChecksumInplace(pg, fsm_nblocks_now);
smgrextend(rel->rd_smgr, FSM_FORKNUM, fsm_nblocks_now,
(char *) pg, false);
fsm_nblocks_now++;
}
/* Update local cache with the up-to-date size */
rel->rd_smgr->smgr_fsm_nblocks = fsm_nblocks_now;
UnlockRelationForExtension(rel, ExclusiveLock);
pfree(pg);
}
/*
* For heaps, we prevent creation of the FSM unless the number of pages
* exceeds HEAP_FSM_CREATION_THRESHOLD. For tables that don't already have
* a FSM, this will save an inode and a few kB of space.
*
* XXX The API is a little awkward -- if the caller passes a valid nblocks
* value, it can avoid invoking a system call. If the caller passes
* InvalidBlockNumber and receives a false return value, it can get an
* up-to-date relation size from get_nblocks. This saves a few cycles in
* the caller, which would otherwise need to get the relation size by itself.
*/
static bool
fsm_allow_writes(Relation rel, BlockNumber heapblk,
BlockNumber nblocks, BlockNumber *get_nblocks)
{
bool skip_get_nblocks;
if (heapblk >= HEAP_FSM_CREATION_THRESHOLD)
return true;
/* Non-heap rels can always create a FSM. */
if (rel->rd_rel->relkind != RELKIND_RELATION &&
rel->rd_rel->relkind != RELKIND_TOASTVALUE)
return true;
/*
* If the caller knows nblocks, we can avoid a system call later. If it
* doesn't, maybe we have relpages from a previous VACUUM. Since the table
* may have extended since then, we still have to count the pages later if
* we can't return now.
*/
if (nblocks != InvalidBlockNumber)
{
if (nblocks > HEAP_FSM_CREATION_THRESHOLD)
return true;
else
skip_get_nblocks = true;
}
else
{
if (rel->rd_rel->relpages != InvalidBlockNumber &&
rel->rd_rel->relpages > HEAP_FSM_CREATION_THRESHOLD)
return true;
else
skip_get_nblocks = false;
}
RelationOpenSmgr(rel);
if (smgrexists(rel->rd_smgr, FSM_FORKNUM))
return true;
if (skip_get_nblocks)
return false;
/* last resort */
*get_nblocks = RelationGetNumberOfBlocks(rel);
if (*get_nblocks > HEAP_FSM_CREATION_THRESHOLD)
return true;
else
return false;
}
/*
* Read a visibility map page.
*
* If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
* true, the visibility map file is extended.
*/
static Buffer
vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
{
Buffer buf;
/*
* We might not have opened the relation at the smgr level yet, or we
* might have been forced to close it by a sinval message. The code below
* won't necessarily notice relation extension immediately when extend =
* false, so we rely on sinval messages to ensure that our ideas about the
* size of the map aren't too far out of date.
*/
RelationOpenSmgr(rel);
/*
* If we haven't cached the size of the visibility map fork yet, check it
* first.
*/
if (rel->rd_smgr->smgr_vm_nblocks == InvalidBlockNumber)
{
if (smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
rel->rd_smgr->smgr_vm_nblocks = smgrnblocks(rel->rd_smgr,
VISIBILITYMAP_FORKNUM);
else
rel->rd_smgr->smgr_vm_nblocks = 0;
}
/* Handle requests beyond EOF */
if (blkno >= rel->rd_smgr->smgr_vm_nblocks)
{
if (extend)
vm_extend(rel, blkno + 1);
else
return InvalidBuffer;
}
/*
* Use ZERO_ON_ERROR mode, and initialize the page if necessary. It's
* always safe to clear bits, so it's better to clear corrupt pages than
* error out.
*/
buf = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, blkno,
RBM_ZERO_ON_ERROR, NULL);
if (PageIsNew(BufferGetPage(buf)))
PageInit(BufferGetPage(buf), BLCKSZ, 0);
return buf;
}
/*
* _hash_alloc_buckets -- allocate a new splitpoint's worth of bucket pages
*
* This does not need to initialize the new bucket pages; we'll do that as
* each one is used by _hash_expandtable(). But we have to extend the logical
* EOF to the end of the splitpoint; this keeps smgr's idea of the EOF in
* sync with ours, so that we don't get complaints from smgr.
*
* We do this by writing a page of zeroes at the end of the splitpoint range.
* We expect that the filesystem will ensure that the intervening pages read
* as zeroes too. On many filesystems this "hole" will not be allocated
* immediately, which means that the index file may end up more fragmented
* than if we forced it all to be allocated now; but since we don't scan
* hash indexes sequentially anyway, that probably doesn't matter.
*
* XXX It's annoying that this code is executed with the metapage lock held.
* We need to interlock against _hash_addovflpage() adding a new overflow page
* concurrently, but it'd likely be better to use LockRelationForExtension
* for the purpose. OTOH, adding a splitpoint is a very infrequent operation,
* so it may not be worth worrying about.
*
* Returns TRUE if successful, or FALSE if allocation failed due to
* BlockNumber overflow.
*/
static bool
_hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
{
BlockNumber lastblock;
char zerobuf[BLCKSZ];
Page page;
HashPageOpaque ovflopaque;
lastblock = firstblock + nblocks - 1;
/*
* Check for overflow in block number calculation; if so, we cannot extend
* the index anymore.
*/
if (lastblock < firstblock || lastblock == InvalidBlockNumber)
return false;
page = (Page) zerobuf;
/*
* Initialize the page. Just zeroing the page won't work; see
* _hash_freeovflpage for similar usage. We take care to make the special
* space valid for the benefit of tools such as pageinspect.
*/
_hash_pageinit(page, BLCKSZ);
ovflopaque = (HashPageOpaque) PageGetSpecialPointer(page);
ovflopaque->hasho_prevblkno = InvalidBlockNumber;
ovflopaque->hasho_nextblkno = InvalidBlockNumber;
ovflopaque->hasho_bucket = -1;
ovflopaque->hasho_flag = LH_UNUSED_PAGE;
ovflopaque->hasho_page_id = HASHO_PAGE_ID;
if (RelationNeedsWAL(rel))
log_newpage(&rel->rd_node,
MAIN_FORKNUM,
lastblock,
zerobuf,
true);
RelationOpenSmgr(rel);
smgrextend(rel->rd_smgr, MAIN_FORKNUM, lastblock, zerobuf, false);
return true;
}
/*
* Read a FSM page.
*
* If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
* true, the FSM file is extended.
*/
static Buffer
fsm_readbuf(Relation rel, FSMAddress addr, bool extend)
{
BlockNumber blkno = fsm_logical_to_physical(addr);
Buffer buf;
RelationOpenSmgr(rel);
/*
* If we haven't cached the size of the FSM yet, check it first. Also
* recheck if the requested block seems to be past end, since our cached
* value might be stale. (We send smgr inval messages on truncation, but
* not on extension.)
*/
if (rel->rd_smgr->smgr_fsm_nblocks == InvalidBlockNumber ||
blkno >= rel->rd_smgr->smgr_fsm_nblocks)
{
if (smgrexists(rel->rd_smgr, FSM_FORKNUM))
rel->rd_smgr->smgr_fsm_nblocks = smgrnblocks(rel->rd_smgr,
FSM_FORKNUM);
else
rel->rd_smgr->smgr_fsm_nblocks = 0;
}
/* Handle requests beyond EOF */
if (blkno >= rel->rd_smgr->smgr_fsm_nblocks)
{
if (extend)
fsm_extend(rel, blkno + 1);
else
return InvalidBuffer;
}
/*
* Use ZERO_ON_ERROR mode, and initialize the page if necessary. The FSM
* information is not accurate anyway, so it's better to clear corrupt
* pages than error out. Since the FSM changes are not WAL-logged, the
* so-called torn page problem on crash can lead to pages with corrupt
* headers, for example.
*/
buf = ReadBufferExtended(rel, FSM_FORKNUM, blkno, RBM_ZERO_ON_ERROR, NULL);
if (PageIsNew(BufferGetPage(buf)))
PageInit(BufferGetPage(buf), BLCKSZ, 0);
return buf;
}
/*
* Read a visibility map page.
*
* If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
* true, the visibility map file is extended.
*/
static Buffer
vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
{
Buffer buf;
RelationOpenSmgr(rel);
/*
* If we haven't cached the size of the visibility map fork yet, check it
* first. Also recheck if the requested block seems to be past end, since
* our cached value might be stale. (We send smgr inval messages on
* truncation, but not on extension.)
*/
if (rel->rd_smgr->smgr_vm_nblocks == InvalidBlockNumber ||
blkno >= rel->rd_smgr->smgr_vm_nblocks)
{
if (smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
rel->rd_smgr->smgr_vm_nblocks = smgrnblocks(rel->rd_smgr,
VISIBILITYMAP_FORKNUM);
else
rel->rd_smgr->smgr_vm_nblocks = 0;
}
/* Handle requests beyond EOF */
if (blkno >= rel->rd_smgr->smgr_vm_nblocks)
{
if (extend)
vm_extend(rel, blkno + 1);
else
return InvalidBuffer;
}
/*
* Use ZERO_ON_ERROR mode, and initialize the page if necessary. It's
* always safe to clear bits, so it's better to clear corrupt pages than
* error out.
*/
buf = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, blkno,
RBM_ZERO_ON_ERROR, NULL);
if (PageIsNew(BufferGetPage(buf)))
PageInit(BufferGetPage(buf), BLCKSZ, 0);
return buf;
}
/*
* Read a visibility map page.
*
* If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
* true, the visibility map file is extended.
*/
static Buffer
vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
{
Buffer buf;
RelationOpenSmgr(rel);
/*
* The current size of the visibility map fork is kept in relcache, to
* avoid reading beyond EOF. If we haven't cached the size of the map yet,
* do that first.
*/
if (rel->rd_vm_nblocks == InvalidBlockNumber)
{
if (smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
rel->rd_vm_nblocks = smgrnblocks(rel->rd_smgr,
VISIBILITYMAP_FORKNUM);
else
rel->rd_vm_nblocks = 0;
}
/* Handle requests beyond EOF */
if (blkno >= rel->rd_vm_nblocks)
{
if (extend)
vm_extend(rel, blkno + 1);
else
return InvalidBuffer;
}
/*
* Use ZERO_ON_ERROR mode, and initialize the page if necessary. It's
* always safe to clear bits, so it's better to clear corrupt pages than
* error out.
*/
buf = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, blkno,
RBM_ZERO_ON_ERROR, NULL);
if (PageIsNew(BufferGetPage(buf)))
PageInit(BufferGetPage(buf), BLCKSZ, 0);
return buf;
}
/*
* _hash_alloc_buckets -- allocate a new splitpoint's worth of bucket pages
*
* This does not need to initialize the new bucket pages; we'll do that as
* each one is used by _hash_expandtable(). But we have to extend the logical
* EOF to the end of the splitpoint; this keeps smgr's idea of the EOF in
* sync with ours, so that we don't get complaints from smgr.
*
* We do this by writing a page of zeroes at the end of the splitpoint range.
* We expect that the filesystem will ensure that the intervening pages read
* as zeroes too. On many filesystems this "hole" will not be allocated
* immediately, which means that the index file may end up more fragmented
* than if we forced it all to be allocated now; but since we don't scan
* hash indexes sequentially anyway, that probably doesn't matter.
*
* XXX It's annoying that this code is executed with the metapage lock held.
* We need to interlock against _hash_getovflpage() adding a new overflow page
* concurrently, but it'd likely be better to use LockRelationForExtension
* for the purpose. OTOH, adding a splitpoint is a very infrequent operation,
* so it may not be worth worrying about.
*
* Returns TRUE if successful, or FALSE if allocation failed due to
* BlockNumber overflow.
*/
static bool
_hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
{
BlockNumber lastblock;
char zerobuf[BLCKSZ];
lastblock = firstblock + nblocks - 1;
/*
* Check for overflow in block number calculation; if so, we cannot extend
* the index anymore.
*/
if (lastblock < firstblock || lastblock == InvalidBlockNumber)
return false;
MemSet(zerobuf, 0, sizeof(zerobuf));
RelationOpenSmgr(rel);
smgrextend(rel->rd_smgr, MAIN_FORKNUM, lastblock, zerobuf, false);
return true;
}
/*
* _hash_alloc_buckets -- allocate a new splitpoint's worth of bucket pages
*
* This does not need to initialize the new bucket pages; we'll do that as
* each one is used by _hash_expandtable(). But we have to extend the logical
* EOF to the end of the splitpoint; otherwise the first overflow page
* allocated beyond the splitpoint will represent a noncontiguous access,
* which can confuse md.c (and will probably be forbidden by future changes
* to md.c).
*
* We do this by writing a page of zeroes at the end of the splitpoint range.
* We expect that the filesystem will ensure that the intervening pages read
* as zeroes too. On many filesystems this "hole" will not be allocated
* immediately, which means that the index file may end up more fragmented
* than if we forced it all to be allocated now; but since we don't scan
* hash indexes sequentially anyway, that probably doesn't matter.
*
* XXX It's annoying that this code is executed with the metapage lock held.
* We need to interlock against _hash_getovflpage() adding a new overflow page
* concurrently, but it'd likely be better to use LockRelationForExtension
* for the purpose. OTOH, adding a splitpoint is a very infrequent operation,
* so it may not be worth worrying about.
*
* Returns TRUE if successful, or FALSE if allocation failed due to
* BlockNumber overflow.
*/
static bool
_hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
{
BlockNumber lastblock;
BlockNumber endblock;
char zerobuf[BLCKSZ];
lastblock = firstblock + nblocks - 1;
/*
* Check for overflow in block number calculation; if so, we cannot
* extend the index anymore.
*/
if (lastblock < firstblock || lastblock == InvalidBlockNumber)
return false;
MemSet(zerobuf, 0, sizeof(zerobuf));
RelationOpenSmgr(rel);
/*
* XXX If the extension results in creation of new segment files,
* we have to make sure that each non-last file is correctly filled out to
* RELSEG_SIZE blocks. This ought to be done inside mdextend, but
* changing the smgr API seems best left for development cycle not late
* beta. Temporary fix for bug #2737.
*/
#ifndef LET_OS_MANAGE_FILESIZE
for (endblock = firstblock | (RELSEG_SIZE - 1);
endblock < lastblock;
endblock += RELSEG_SIZE)
smgrextend(rel->rd_smgr, endblock, zerobuf, rel->rd_istemp);
#endif
smgrextend(rel->rd_smgr, lastblock, zerobuf, rel->rd_istemp);
return true;
}
static void
ReadBlocks(int filenum)
{
FILE *file;
char record_type;
char *dbname;
Oid record_filenode;
ForkNumber record_forknum;
BlockNumber record_blocknum;
BlockNumber record_range;
int log_level = DEBUG3;
Oid relOid = InvalidOid;
Relation rel = NULL;
bool skip_relation = false;
bool skip_fork = false;
bool skip_block = false;
BlockNumber nblocks = 0;
BlockNumber blocks_restored = 0;
const char *filepath;
/*
* If this condition changes, then this code, and the code in the writer
* will need to be changed; especially the format specifiers in log and
* error messages.
*/
StaticAssertStmt(MaxBlockNumber == 0xFFFFFFFE, "Code may need review.");
filepath = getSavefileName(filenum);
file = fileOpen(filepath, PG_BINARY_R);
dbname = readDBName(file, filepath);
/*
* When restoring global objects, the dbname is zero-length string, and non-
* zero length otherwise. And filenum is never expected to be smaller than 1.
*/
Assert(filenum >= 1);
Assert(filenum == 1 ? strlen(dbname) == 0 : strlen(dbname) > 0);
/* To restore the global objects, use default database */
BackgroundWorkerInitializeConnection(filenum == 1 ? guc_default_database : dbname, NULL);
SetCurrentStatementStartTimestamp();
StartTransactionCommand();
SPI_connect();
PushActiveSnapshot(GetTransactionSnapshot());
pgstat_report_activity(STATE_RUNNING, "restoring buffers");
/*
* Note that in case of a read error, we will leak relcache entry that we may
* currently have open. In case of EOF, we close the relation after the loop.
*/
while (fileRead(&record_type, 1, file, true, filepath))
{
/*
* If we want to process the signals, this seems to be the best place
* to do it. Generally the backends refrain from processing config file
* while in transaction, but that's more for the fear of allowing GUC
* changes to affect expression evaluation, causing different results
* for the same expression in a transaction. Since this worker is not
* processing any queries, it is okay to process the config file here.
*
* Even though it's okay to process SIGHUP here, doing so doesn't add
* any value. The only reason we might want to process config file here
* would be to allow the user to interrupt the BlockReader's operation
* by changing this extenstion's GUC parameter. But the user can do that
* anyway, using SIGTERM or pg_terminate_backend().
*/
/* Stop processing the save-file if the Postmaster wants us to die. */
if (got_sigterm)
break;
ereport(log_level,
(errmsg("record type %x - %c", record_type, record_type)));
switch (record_type)
{
case 'r':
{
/* Close the previous relation, if any. */
if (rel)
{
relation_close(rel, AccessShareLock);
rel = NULL;
}
record_forknum = InvalidForkNumber;
record_blocknum = InvalidBlockNumber;
nblocks = 0;
fileRead(&record_filenode, sizeof(Oid), file, false, filepath);
relOid = GetRelOid(record_filenode);
ereport(log_level, (errmsg("processing filenode %u, relation %u",
record_filenode, relOid)));
/*
* If the relation has been rewritten/dropped since we saved it,
* just skip it and process the next relation.
*/
if (relOid == InvalidOid)
skip_relation = true;
else
{
skip_relation = false;
/* Open the relation */
rel = relation_open(relOid, AccessShareLock);
RelationOpenSmgr(rel);
}
}
break;
case 'f':
{
record_blocknum = InvalidBlockNumber;
nblocks = 0;
fileRead(&record_forknum, sizeof(ForkNumber), file, false, filepath);
if (skip_relation)
continue;
if (rel == NULL)
ereport(ERROR,
(errmsg("found a fork record without a preceeding relation record")));
ereport(log_level, (errmsg("processing fork %d", record_forknum)));
if (!smgrexists(rel->rd_smgr, record_forknum))
skip_fork = true;
else
{
skip_fork = false;
nblocks = RelationGetNumberOfBlocksInFork(rel, record_forknum);
}
}
break;
case 'b':
{
if (record_forknum == InvalidForkNumber)
ereport(ERROR,
(errmsg("found a block record without a preceeding fork record")));
fileRead(&record_blocknum, sizeof(BlockNumber), file, false, filepath);
if (skip_relation || skip_fork)
continue;
/*
* Don't try to read past the file; the file may have been shrunk
* by a vaccum/truncate operation.
*/
if (record_blocknum >= nblocks)
{
ereport(log_level,
(errmsg("reader %d skipping block filenode %u forknum %d blocknum %u",
filenum, record_filenode, record_forknum, record_blocknum)));
skip_block = true;
continue;
}
else
{
Buffer buf;
skip_block = false;
ereport(log_level,
(errmsg("reader %d reading block filenode %u forknum %d blocknum %u",
filenum, record_filenode, record_forknum, record_blocknum)));
buf = ReadBufferExtended(rel, record_forknum, record_blocknum, RBM_NORMAL, NULL);
ReleaseBuffer(buf);
++blocks_restored;
}
}
break;
case 'N':
{
BlockNumber block;
Assert(record_blocknum != InvalidBlockNumber);
if (record_blocknum == InvalidBlockNumber)
ereport(ERROR,
(errmsg("found a block range record without a preceeding block record")));
fileRead(&record_range, sizeof(int), file, false, filepath);
if (skip_relation || skip_fork || skip_block)
continue;
ereport(log_level,
(errmsg("reader %d reading range filenode %u forknum %d blocknum %u range %u",
filenum, record_filenode, record_forknum, record_blocknum, record_range)));
for (block = record_blocknum + 1; block <= (record_blocknum + record_range); ++block)
{
Buffer buf;
/*
* Don't try to read past the file; the file may have been
* shrunk by a vaccum operation.
*/
if (block >= nblocks)
{
ereport(log_level,
(errmsg("reader %d skipping block range filenode %u forknum %d start %u end %u",
filenum, record_filenode, record_forknum,
block, record_blocknum + record_range)));
break;
}
buf = ReadBufferExtended(rel, record_forknum, block, RBM_NORMAL, NULL);
ReleaseBuffer(buf);
++blocks_restored;
}
}
break;
default:
{
ereport(ERROR,
(errmsg("found unexpected save-file marker %x - %c)", record_type, record_type)));
Assert(false);
}
break;
}
}
if (rel)
relation_close(rel, AccessShareLock);
ereport(LOG,
(errmsg("Block Reader %d: restored %u blocks",
filenum, blocks_restored)));
SPI_finish();
PopActiveSnapshot();
CommitTransactionCommand();
pgstat_report_activity(STATE_IDLE, NULL);
fileClose(file, filepath);
/* Remove the save-file */
if (remove(filepath) != 0)
ereport(ERROR,
(errcode_for_file_access(),
errmsg("error removing file \"%s\" : %m", filepath)));
}
/*
* visibilitymap_truncate - truncate the visibility map
*
* The caller must hold AccessExclusiveLock on the relation, to ensure that
* other backends receive the smgr invalidation event that this function sends
* before they access the VM again.
*
* nheapblocks is the new size of the heap.
*/
void
visibilitymap_truncate(Relation rel, BlockNumber nheapblocks)
{
BlockNumber newnblocks;
/* last remaining block, byte, and bit */
BlockNumber truncBlock = HEAPBLK_TO_MAPBLOCK(nheapblocks);
uint32 truncByte = HEAPBLK_TO_MAPBYTE(nheapblocks);
uint8 truncBit = HEAPBLK_TO_MAPBIT(nheapblocks);
#ifdef TRACE_VISIBILITYMAP
elog(DEBUG1, "vm_truncate %s %d", RelationGetRelationName(rel), nheapblocks);
#endif
RelationOpenSmgr(rel);
/*
* If no visibility map has been created yet for this relation, there's
* nothing to truncate.
*/
if (!smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
return;
/*
* Unless the new size is exactly at a visibility map page boundary, the
* tail bits in the last remaining map page, representing truncated heap
* blocks, need to be cleared. This is not only tidy, but also necessary
* because we don't get a chance to clear the bits if the heap is extended
* again.
*/
if (truncByte != 0 || truncBit != 0)
{
Buffer mapBuffer;
Page page;
char *map;
newnblocks = truncBlock + 1;
mapBuffer = vm_readbuf(rel, truncBlock, false);
if (!BufferIsValid(mapBuffer))
{
/* nothing to do, the file was already smaller */
return;
}
page = BufferGetPage(mapBuffer);
map = PageGetContents(page);
LockBuffer(mapBuffer, BUFFER_LOCK_EXCLUSIVE);
/* Clear out the unwanted bytes. */
MemSet(&map[truncByte + 1], 0, MAPSIZE - (truncByte + 1));
/*
* Mask out the unwanted bits of the last remaining byte.
*
* ((1 << 0) - 1) = 00000000 ((1 << 1) - 1) = 00000001 ... ((1 << 6) -
* 1) = 00111111 ((1 << 7) - 1) = 01111111
*/
map[truncByte] &= (1 << truncBit) - 1;
MarkBufferDirty(mapBuffer);
UnlockReleaseBuffer(mapBuffer);
}
else
newnblocks = truncBlock;
if (smgrnblocks(rel->rd_smgr, VISIBILITYMAP_FORKNUM) <= newnblocks)
{
/* nothing to do, the file was already smaller than requested size */
return;
}
/* Truncate the unused VM pages, and send smgr inval message */
smgrtruncate(rel->rd_smgr, VISIBILITYMAP_FORKNUM, newnblocks);
/*
* We might as well update the local smgr_vm_nblocks setting. smgrtruncate
* sent an smgr cache inval message, which will cause other backends to
* invalidate their copy of smgr_vm_nblocks, and this one too at the next
* command boundary. But this ensures it isn't outright wrong until then.
*/
if (rel->rd_smgr)
rel->rd_smgr->smgr_vm_nblocks = newnblocks;
}
/*
* 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);
}
}
/*
* Insert a tuple to the new relation. This has to track heap_insert
* and its subsidiary functions!
*
* t_self of the tuple is set to the new TID of the tuple. If t_ctid of the
* tuple is invalid on entry, it's replaced with the new TID as well (in
* the inserted data only, not in the caller's copy).
*/
static void
raw_heap_insert(RewriteState state, HeapTuple tup)
{
Page page = state->rs_buffer;
Size pageFreeSpace,
saveFreeSpace;
Size len;
OffsetNumber newoff;
HeapTuple heaptup;
/*
* If the new tuple is too big for storage or contains already toasted
* out-of-line attributes from some other relation, invoke the toaster.
*
* Note: below this point, heaptup is the data we actually intend to store
* into the relation; tup is the caller's original untoasted data.
*/
if (state->rs_new_rel->rd_rel->relkind == RELKIND_TOASTVALUE)
{
/* toast table entries should never be recursively toasted */
Assert(!HeapTupleHasExternal(tup));
heaptup = tup;
}
else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
heaptup = toast_insert_or_update(state->rs_new_rel, tup, NULL,
HEAP_INSERT_SKIP_FSM |
(state->rs_use_wal ?
0 : HEAP_INSERT_SKIP_WAL));
else
heaptup = tup;
len = MAXALIGN(heaptup->t_len); /* be conservative */
/*
* If we're gonna fail for oversize tuple, do it right away
*/
if (len > MaxHeapTupleSize)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("row is too big: size %zu, maximum size %zu",
len, MaxHeapTupleSize)));
/* Compute desired extra freespace due to fillfactor option */
saveFreeSpace = RelationGetTargetPageFreeSpace(state->rs_new_rel,
HEAP_DEFAULT_FILLFACTOR);
/* Now we can check to see if there's enough free space already. */
if (state->rs_buffer_valid)
{
pageFreeSpace = PageGetHeapFreeSpace(page);
if (len + saveFreeSpace > pageFreeSpace)
{
/* Doesn't fit, so write out the existing page */
/* XLOG stuff */
if (state->rs_use_wal)
log_newpage(&state->rs_new_rel->rd_node,
MAIN_FORKNUM,
state->rs_blockno,
page,
true);
/*
* Now write the page. We say isTemp = true even if it's not a
* temp table, because there's no need for smgr to schedule an
* fsync for this write; we'll do it ourselves in
* end_heap_rewrite.
*/
RelationOpenSmgr(state->rs_new_rel);
PageSetChecksumInplace(page, state->rs_blockno);
smgrextend(state->rs_new_rel->rd_smgr, MAIN_FORKNUM,
state->rs_blockno, (char *) page, true);
state->rs_blockno++;
state->rs_buffer_valid = false;
}
}
if (!state->rs_buffer_valid)
{
/* Initialize a new empty page */
PageInit(page, BLCKSZ, 0);
state->rs_buffer_valid = true;
}
/* And now we can insert the tuple into the page */
newoff = PageAddItem(page, (Item) heaptup->t_data, heaptup->t_len,
InvalidOffsetNumber, false, true);
if (newoff == InvalidOffsetNumber)
elog(ERROR, "failed to add tuple");
/* Update caller's t_self to the actual position where it was stored */
ItemPointerSet(&(tup->t_self), state->rs_blockno, newoff);
/*
* Insert the correct position into CTID of the stored tuple, too, if the
* caller didn't supply a valid CTID.
*/
if (!ItemPointerIsValid(&tup->t_data->t_ctid))
{
ItemId newitemid;
HeapTupleHeader onpage_tup;
newitemid = PageGetItemId(page, newoff);
onpage_tup = (HeapTupleHeader) PageGetItem(page, newitemid);
onpage_tup->t_ctid = tup->t_self;
}
/* If heaptup is a private copy, release it. */
if (heaptup != tup)
heap_freetuple(heaptup);
}
/*
* emit a completed btree page, and release the working storage.
*/
static void
_bt_blwritepage(BTWriteState *wstate, Page page, BlockNumber blkno)
{
// Fetch gp_persistent_relation_node information that will be added to XLOG record.
RelationFetchGpRelationNodeForXLog(wstate->index);
/* Ensure rd_smgr is open (could have been closed by relcache flush!) */
RelationOpenSmgr(wstate->index);
/* XLOG stuff */
if (wstate->btws_use_wal)
{
_bt_lognewpage(wstate->index, page, blkno);
}
else
{
/* Leave the page LSN zero if not WAL-logged, but set TLI anyway */
PageSetTLI(page, ThisTimeLineID);
}
/*
* If we have to write pages nonsequentially, fill in the space with
* zeroes until we come back and overwrite. This is not logically
* necessary on standard Unix filesystems (unwritten space will read as
* zeroes anyway), but it should help to avoid fragmentation. The dummy
* pages aren't WAL-logged though.
*/
while (blkno > wstate->btws_pages_written)
{
if (!wstate->btws_zeropage)
wstate->btws_zeropage = (Page) palloc0(BLCKSZ);
// -------- MirroredLock ----------
// UNDONE: Unfortunately, I think we write temp relations to the mirror...
LWLockAcquire(MirroredLock, LW_SHARED);
smgrextend(wstate->index->rd_smgr, wstate->btws_pages_written++,
(char *) wstate->btws_zeropage,
true);
LWLockRelease(MirroredLock);
// -------- MirroredLock ----------
}
// -------- MirroredLock ----------
// UNDONE: Unfortunately, I think we write temp relations to the mirror...
LWLockAcquire(MirroredLock, LW_SHARED);
/*
* Now write the page. We say isTemp = true even if it's not a temp
* index, because there's no need for smgr to schedule an fsync for this
* write; we'll do it ourselves before ending the build.
*/
if (blkno == wstate->btws_pages_written)
{
/* extending the file... */
smgrextend(wstate->index->rd_smgr, blkno, (char *) page, true);
wstate->btws_pages_written++;
}
else
{
/* overwriting a block we zero-filled before */
smgrwrite(wstate->index->rd_smgr, blkno, (char *) page, true);
}
LWLockRelease(MirroredLock);
// -------- MirroredLock ----------
pfree(page);
}
/*
* Open a relation during XLOG replay
*
* Note: this once had an API that allowed NULL return on failure, but it
* no longer does; any failure results in elog().
*/
Relation
XLogOpenRelation(RelFileNode rnode)
{
XLogRelDesc *res;
XLogRelCacheEntry *hentry;
bool found;
hentry = (XLogRelCacheEntry *)
hash_search(_xlrelcache, (void *) &rnode, HASH_FIND, NULL);
if (hentry)
{
res = hentry->rdesc;
res->lessRecently->moreRecently = res->moreRecently;
res->moreRecently->lessRecently = res->lessRecently;
}
else
{
res = _xl_new_reldesc();
sprintf(RelationGetRelationName(&(res->reldata)), "%u", rnode.relNode);
res->reldata.rd_node = rnode;
/*
* We set up the lockRelId in case anything tries to lock the dummy
* relation. Note that this is fairly bogus since relNode may be
* different from the relation's OID. It shouldn't really matter
* though, since we are presumably running by ourselves and can't have
* any lock conflicts ...
*/
res->reldata.rd_lockInfo.lockRelId.dbId = rnode.dbNode;
res->reldata.rd_lockInfo.lockRelId.relId = rnode.relNode;
hentry = (XLogRelCacheEntry *)
hash_search(_xlrelcache, (void *) &rnode, HASH_ENTER, &found);
if (found)
elog(PANIC, "xlog relation already present on insert into cache");
hentry->rdesc = res;
res->reldata.rd_targblock = InvalidBlockNumber;
res->reldata.rd_smgr = NULL;
RelationOpenSmgr(&(res->reldata));
/*
* Create the target file if it doesn't already exist. This lets us
* cope if the replay sequence contains writes to a relation that is
* later deleted. (The original coding of this routine would instead
* return NULL, causing the writes to be suppressed. But that seems
* like it risks losing valuable data if the filesystem loses an inode
* during a crash. Better to write the data until we are actually
* told to delete the file.)
*/
smgrcreate(res->reldata.rd_smgr, res->reldata.rd_istemp, true);
}
res->moreRecently = &(_xlrelarr[0]);
res->lessRecently = _xlrelarr[0].lessRecently;
_xlrelarr[0].lessRecently = res;
res->lessRecently->moreRecently = res;
return &(res->reldata);
}
/*
* FreeSpaceMapTruncateRel - adjust for truncation of a relation.
*
* The caller must hold AccessExclusiveLock on the relation, to ensure that
* other backends receive the smgr invalidation event that this function sends
* before they access the FSM again.
*
* nblocks is the new size of the heap.
*/
void
FreeSpaceMapTruncateRel(Relation rel, BlockNumber nblocks)
{
BlockNumber new_nfsmblocks;
FSMAddress first_removed_address;
uint16 first_removed_slot;
Buffer buf;
RelationOpenSmgr(rel);
/*
* If no FSM has been created yet for this relation, there's nothing to
* truncate.
*/
if (!smgrexists(rel->rd_smgr, FSM_FORKNUM))
return;
/* Get the location in the FSM of the first removed heap block */
first_removed_address = fsm_get_location(nblocks, &first_removed_slot);
/*
* Zero out the tail of the last remaining FSM page. If the slot
* representing the first removed heap block is at a page boundary, as the
* first slot on the FSM page that first_removed_address points to, we can
* just truncate that page altogether.
*/
if (first_removed_slot > 0)
{
buf = fsm_readbuf(rel, first_removed_address, false);
if (!BufferIsValid(buf))
return; /* nothing to do; the FSM was already smaller */
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
/* NO EREPORT(ERROR) from here till changes are logged */
START_CRIT_SECTION();
fsm_truncate_avail(BufferGetPage(buf), first_removed_slot);
/*
* Truncation of a relation is WAL-logged at a higher-level, and we
* will be called at WAL replay. But if checksums are enabled, we need
* to still write a WAL record to protect against a torn page, if the
* page is flushed to disk before the truncation WAL record. We cannot
* use MarkBufferDirtyHint here, because that will not dirty the page
* during recovery.
*/
MarkBufferDirty(buf);
if (!InRecovery && RelationNeedsWAL(rel) && XLogHintBitIsNeeded())
log_newpage_buffer(buf, false);
END_CRIT_SECTION();
UnlockReleaseBuffer(buf);
new_nfsmblocks = fsm_logical_to_physical(first_removed_address) + 1;
}
else
{
new_nfsmblocks = fsm_logical_to_physical(first_removed_address);
if (smgrnblocks(rel->rd_smgr, FSM_FORKNUM) <= new_nfsmblocks)
return; /* nothing to do; the FSM was already smaller */
}
/* Truncate the unused FSM pages, and send smgr inval message */
smgrtruncate(rel->rd_smgr, FSM_FORKNUM, new_nfsmblocks);
/*
* We might as well update the local smgr_fsm_nblocks setting.
* smgrtruncate sent an smgr cache inval message, which will cause other
* backends to invalidate their copy of smgr_fsm_nblocks, and this one too
* at the next command boundary. But this ensures it isn't outright wrong
* until then.
*/
if (rel->rd_smgr)
rel->rd_smgr->smgr_fsm_nblocks = new_nfsmblocks;
/*
* Update upper-level FSM pages to account for the truncation. This is
* important because the just-truncated pages were likely marked as
* all-free, and would be preferentially selected.
*/
FreeSpaceMapVacuumRange(rel, nblocks, InvalidBlockNumber);
}
/*
* Read a FSM page.
*
* If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
* true, the FSM file is extended.
*/
static Buffer
fsm_readbuf(Relation rel, FSMAddress addr, bool extend)
{
BlockNumber blkno = fsm_logical_to_physical(addr);
Buffer buf;
RelationOpenSmgr(rel);
/*
* If we haven't cached the size of the FSM yet, check it first. Also
* recheck if the requested block seems to be past end, since our cached
* value might be stale. (We send smgr inval messages on truncation, but
* not on extension.)
*/
if (rel->rd_smgr->smgr_fsm_nblocks == InvalidBlockNumber ||
blkno >= rel->rd_smgr->smgr_fsm_nblocks)
{
if (smgrexists(rel->rd_smgr, FSM_FORKNUM))
rel->rd_smgr->smgr_fsm_nblocks = smgrnblocks(rel->rd_smgr,
FSM_FORKNUM);
else
rel->rd_smgr->smgr_fsm_nblocks = 0;
}
/* Handle requests beyond EOF */
if (blkno >= rel->rd_smgr->smgr_fsm_nblocks)
{
if (extend)
fsm_extend(rel, blkno + 1);
else
return InvalidBuffer;
}
/*
* Use ZERO_ON_ERROR mode, and initialize the page if necessary. The FSM
* information is not accurate anyway, so it's better to clear corrupt
* pages than error out. Since the FSM changes are not WAL-logged, the
* so-called torn page problem on crash can lead to pages with corrupt
* headers, for example.
*
* The initialize-the-page part is trickier than it looks, because of the
* possibility of multiple backends doing this concurrently, and our
* desire to not uselessly take the buffer lock in the normal path where
* the page is OK. We must take the lock to initialize the page, so
* recheck page newness after we have the lock, in case someone else
* already did it. Also, because we initially check PageIsNew with no
* lock, it's possible to fall through and return the buffer while someone
* else is still initializing the page (i.e., we might see pd_upper as set
* but other page header fields are still zeroes). This is harmless for
* callers that will take a buffer lock themselves, but some callers
* inspect the page without any lock at all. The latter is OK only so
* long as it doesn't depend on the page header having correct contents.
* Current usage is safe because PageGetContents() does not require that.
*/
buf = ReadBufferExtended(rel, FSM_FORKNUM, blkno, RBM_ZERO_ON_ERROR, NULL);
if (PageIsNew(BufferGetPage(buf)))
{
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (PageIsNew(BufferGetPage(buf)))
PageInit(BufferGetPage(buf), BLCKSZ, 0);
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
}
return buf;
}
/*
* _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);
}
/*
* Open a relation during XLOG replay
*
* Note: this once had an API that allowed NULL return on failure, but it
* no longer does; any failure results in elog().
*/
Relation
XLogOpenRelation(RelFileNode rnode)
{
XLogRelDesc *res;
XLogRelCacheEntry *hentry;
bool found;
hentry = (XLogRelCacheEntry *)
hash_search(_xlrelcache, (void *) &rnode, HASH_FIND, NULL);
if (hentry)
{
res = hentry->rdesc;
res->lessRecently->moreRecently = res->moreRecently;
res->moreRecently->lessRecently = res->lessRecently;
}
else
{
/*
* We need to fault in the database directory on the standby.
*/
if (rnode.spcNode != GLOBALTABLESPACE_OID && IsStandbyMode())
{
char *primaryFilespaceLocation = NULL;
char *dbPath;
if (IsBuiltinTablespace(rnode.spcNode))
{
/*
* No filespace to fetch.
*/
}
else
{
char *mirrorFilespaceLocation = NULL;
/*
* Investigate whether the containing directories exist to give more detail.
*/
PersistentTablespace_GetPrimaryAndMirrorFilespaces(
rnode.spcNode,
&primaryFilespaceLocation,
&mirrorFilespaceLocation);
if (primaryFilespaceLocation == NULL ||
strlen(primaryFilespaceLocation) == 0)
{
elog(ERROR, "Empty primary filespace directory location");
}
if (mirrorFilespaceLocation != NULL)
{
pfree(mirrorFilespaceLocation);
mirrorFilespaceLocation = NULL;
}
}
dbPath = (char*)palloc(MAXPGPATH + 1);
FormDatabasePath(
dbPath,
primaryFilespaceLocation,
rnode.spcNode,
rnode.dbNode);
if (primaryFilespaceLocation != NULL)
{
pfree(primaryFilespaceLocation);
primaryFilespaceLocation = NULL;
}
if (mkdir(dbPath, 0700) == 0)
{
if (Debug_persistent_recovery_print)
{
elog(PersistentRecovery_DebugPrintLevel(),
"XLogOpenRelation: Re-created database directory \"%s\"",
dbPath);
}
}
else
{
/*
* Allowed to already exist.
*/
if (errno != EEXIST)
{
elog(ERROR, "could not create database directory \"%s\": %m",
dbPath);
}
else
{
if (Debug_persistent_recovery_print)
{
elog(PersistentRecovery_DebugPrintLevel(),
"XLogOpenRelation: Database directory \"%s\" already exists",
dbPath);
}
}
}
pfree(dbPath);
}
res = _xl_new_reldesc();
sprintf(RelationGetRelationName(&(res->reldata)), "%u", rnode.relNode);
res->reldata.rd_node = rnode;
/*
* We set up the lockRelId in case anything tries to lock the dummy
* relation. Note that this is fairly bogus since relNode may be
* different from the relation's OID. It shouldn't really matter
* though, since we are presumably running by ourselves and can't have
* any lock conflicts ...
*/
res->reldata.rd_lockInfo.lockRelId.dbId = rnode.dbNode;
res->reldata.rd_lockInfo.lockRelId.relId = rnode.relNode;
hentry = (XLogRelCacheEntry *)
hash_search(_xlrelcache, (void *) &rnode, HASH_ENTER, &found);
if (found)
elog(PANIC, "xlog relation already present on insert into cache");
hentry->rdesc = res;
res->reldata.rd_targblock = InvalidBlockNumber;
res->reldata.rd_smgr = NULL;
RelationOpenSmgr(&(res->reldata));
/*
* Create the target file if it doesn't already exist. This lets us
* cope if the replay sequence contains writes to a relation that is
* later deleted. (The original coding of this routine would instead
* return NULL, causing the writes to be suppressed. But that seems
* like it risks losing valuable data if the filesystem loses an inode
* during a crash. Better to write the data until we are actually
* told to delete the file.)
*/
// NOTE: We no longer re-create files automatically because
// new FileRep persistent objects will ensure files exist.
// UNDONE: Can't remove this block of code yet until boot time calls to this routine are analyzed...
{
MirrorDataLossTrackingState mirrorDataLossTrackingState;
int64 mirrorDataLossTrackingSessionNum;
bool mirrorDataLossOccurred;
// UNDONE: What about the persistent rel files table???
// UNDONE: This condition should not occur anymore.
// UNDONE: segmentFileNum and AO?
mirrorDataLossTrackingState =
FileRepPrimary_GetMirrorDataLossTrackingSessionNum(
&mirrorDataLossTrackingSessionNum);
smgrcreate(
res->reldata.rd_smgr,
res->reldata.rd_isLocalBuf,
/* relationName */ NULL, // Ok to be NULL -- we don't know the name here.
mirrorDataLossTrackingState,
mirrorDataLossTrackingSessionNum,
/* ignoreAlreadyExists */ true,
&mirrorDataLossOccurred);
}
}
res->moreRecently = &(_xlrelarr[0]);
res->lessRecently = _xlrelarr[0].lessRecently;
_xlrelarr[0].lessRecently = res;
res->lessRecently->moreRecently = res;
Assert(&(res->reldata) != NULL); // Assert what it says in the interface -- we don't return NULL anymore.
return &(res->reldata);
}
/*
* RelationTruncate
* Physically truncate a relation to the specified number of blocks.
*
* This includes getting rid of any buffers for the blocks that are to be
* dropped.
*/
void
RelationTruncate(Relation rel, BlockNumber nblocks)
{
bool fsm;
bool vm;
/* Open it at the smgr level if not already done */
RelationOpenSmgr(rel);
/*
* Make sure smgr_targblock etc aren't pointing somewhere past new end
*/
rel->rd_smgr->smgr_targblock = InvalidBlockNumber;
rel->rd_smgr->smgr_fsm_nblocks = InvalidBlockNumber;
rel->rd_smgr->smgr_vm_nblocks = InvalidBlockNumber;
/* Truncate the FSM first if it exists */
fsm = smgrexists(rel->rd_smgr, FSM_FORKNUM);
if (fsm)
FreeSpaceMapTruncateRel(rel, nblocks);
/* Truncate the visibility map too if it exists. */
vm = smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM);
if (vm)
visibilitymap_truncate(rel, nblocks);
/*
* We WAL-log the truncation before actually truncating, which means
* trouble if the truncation fails. If we then crash, the WAL replay
* likely isn't going to succeed in the truncation either, and cause a
* PANIC. It's tempting to put a critical section here, but that cure
* would be worse than the disease. It would turn a usually harmless
* failure to truncate, that might spell trouble at WAL replay, into a
* certain PANIC.
*/
if (!rel->rd_istemp)
{
/*
* Make an XLOG entry reporting the file truncation.
*/
XLogRecPtr lsn;
XLogRecData rdata;
xl_smgr_truncate xlrec;
xlrec.blkno = nblocks;
xlrec.rnode = rel->rd_node;
rdata.data = (char *) &xlrec;
rdata.len = sizeof(xlrec);
rdata.buffer = InvalidBuffer;
rdata.next = NULL;
lsn = XLogInsert(RM_SMGR_ID, XLOG_SMGR_TRUNCATE, &rdata);
/*
* Flush, because otherwise the truncation of the main relation might
* hit the disk before the WAL record, and the truncation of the FSM
* or visibility map. If we crashed during that window, we'd be left
* with a truncated heap, but the FSM or visibility map would still
* contain entries for the non-existent heap pages.
*/
if (fsm || vm)
XLogFlush(lsn);
}
/* Do the real work */
smgrtruncate(rel->rd_smgr, MAIN_FORKNUM, nblocks, rel->rd_istemp);
}
/*
* 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);
}
}
/*
* pg_prewarm(regclass, mode text, fork text,
* first_block int8, last_block int8)
*
* The first argument is the relation to be prewarmed; the second controls
* how prewarming is done; legal options are 'prefetch', 'read', and 'buffer'.
* The third is the name of the relation fork to be prewarmed. The fourth
* and fifth arguments specify the first and last block to be prewarmed.
* If the fourth argument is NULL, it will be taken as 0; if the fifth argument
* is NULL, it will be taken as the number of blocks in the relation. The
* return value is the number of blocks successfully prewarmed.
*/
Datum
pg_prewarm(PG_FUNCTION_ARGS)
{
Oid relOid;
text *forkName;
text *type;
int64 first_block;
int64 last_block;
int64 nblocks;
int64 blocks_done = 0;
int64 block;
Relation rel;
ForkNumber forkNumber;
char *forkString;
char *ttype;
PrewarmType ptype;
AclResult aclresult;
/* Basic sanity checking. */
if (PG_ARGISNULL(0))
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("relation cannot be null")));
relOid = PG_GETARG_OID(0);
if (PG_ARGISNULL(1))
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
(errmsg("prewarm type cannot be null"))));
type = PG_GETARG_TEXT_P(1);
ttype = text_to_cstring(type);
if (strcmp(ttype, "prefetch") == 0)
ptype = PREWARM_PREFETCH;
else if (strcmp(ttype, "read") == 0)
ptype = PREWARM_READ;
else if (strcmp(ttype, "buffer") == 0)
ptype = PREWARM_BUFFER;
else
{
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("invalid prewarm type"),
errhint("Valid prewarm types are \"prefetch\", \"read\", and \"buffer\".")));
PG_RETURN_INT64(0); /* Placate compiler. */
}
if (PG_ARGISNULL(2))
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
(errmsg("relation fork cannot be null"))));
forkName = PG_GETARG_TEXT_P(2);
forkString = text_to_cstring(forkName);
forkNumber = forkname_to_number(forkString);
/* Open relation and check privileges. */
rel = relation_open(relOid, AccessShareLock);
aclresult = pg_class_aclcheck(relOid, GetUserId(), ACL_SELECT);
if (aclresult != ACLCHECK_OK)
aclcheck_error(aclresult, ACL_KIND_CLASS, get_rel_name(relOid));
/* Check that the fork exists. */
RelationOpenSmgr(rel);
if (!smgrexists(rel->rd_smgr, forkNumber))
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("fork \"%s\" does not exist for this relation",
forkString)));
/* Validate block numbers, or handle nulls. */
nblocks = RelationGetNumberOfBlocksInFork(rel, forkNumber);
if (PG_ARGISNULL(3))
first_block = 0;
else
{
first_block = PG_GETARG_INT64(3);
if (first_block < 0 || first_block >= nblocks)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("starting block number must be between 0 and " INT64_FORMAT,
nblocks - 1)));
}
if (PG_ARGISNULL(4))
last_block = nblocks - 1;
else
{
last_block = PG_GETARG_INT64(4);
if (last_block < 0 || last_block >= nblocks)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("ending block number must be between 0 and " INT64_FORMAT,
nblocks - 1)));
}
/* Now we're ready to do the real work. */
if (ptype == PREWARM_PREFETCH)
{
#ifdef USE_PREFETCH
/*
* In prefetch mode, we just hint the OS to read the blocks, but we
* don't know whether it really does it, and we don't wait for it to
* finish.
*
* It would probably be better to pass our prefetch requests in chunks
* of a megabyte or maybe even a whole segment at a time, but there's
* no practical way to do that at present without a gross modularity
* violation, so we just do this.
*/
for (block = first_block; block <= last_block; ++block)
{
CHECK_FOR_INTERRUPTS();
PrefetchBuffer(rel, forkNumber, block);
++blocks_done;
}
#else
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("prefetch is not supported by this build")));
#endif
}
else if (ptype == PREWARM_READ)
{
/*
* In read mode, we actually read the blocks, but not into shared
* buffers. This is more portable than prefetch mode (it works
* everywhere) and is synchronous.
*/
for (block = first_block; block <= last_block; ++block)
{
CHECK_FOR_INTERRUPTS();
smgrread(rel->rd_smgr, forkNumber, block, blockbuffer);
++blocks_done;
}
}
else if (ptype == PREWARM_BUFFER)
{
/*
* In buffer mode, we actually pull the data into shared_buffers.
*/
for (block = first_block; block <= last_block; ++block)
{
Buffer buf;
CHECK_FOR_INTERRUPTS();
buf = ReadBufferExtended(rel, forkNumber, block, RBM_NORMAL, NULL);
ReleaseBuffer(buf);
++blocks_done;
}
}
/* Close relation, release lock. */
relation_close(rel, AccessShareLock);
PG_RETURN_INT64(blocks_done);
}
Relation DirectOpen_Open(
DirectOpen *direct,
Oid relationId,
Oid tablespace,
Oid database,
Oid relfilenode,
FormData_pg_class *pgClass,
FormData_pg_attribute *attrArray,
FormData_pg_am *pgAm,
FormData_pg_index *pgIndex,
int2 *indKeyArray,
Oid *indClassArray,
bool relHasOid)
{
int natts;
int i;
Assert(pgClass != NULL);
natts = pgClass->relnatts;
if (relationId == -1)
relationId = pgClass->relfilenode; // Assume it is ok to use the relfilenode as the relationId in our limited usage.
if (relfilenode == -1)
relfilenode = pgClass->relfilenode;
if (!direct->isInit)
{
/*
* Lots of Hard-coded construction of the gp_persistent* RelationS and
* dependent objects like tuple descriptors, etc.
*/
direct->relationData.rd_refcnt = 0;
direct->relationData.rd_isvalid = true;
direct->relationData.rd_id = relationId;
direct->relationData.rd_rel = pgClass;
if (pgIndex != NULL)
{
int pgIndexFixedLen = offsetof(FormData_pg_index, indkey);
int indKeyVectorLen = Int2VectorSize(natts);
int2vector *indKeyVector;
oidvector *indClassVector;
uint16 amstrategies;
uint16 amsupport;
Oid *operator;
RegProcedure *support;
FmgrInfo *supportinfo;
Assert(pgAm != NULL);
Assert(indKeyArray != NULL);
Assert(indClassArray != NULL);
/*
* Allocate Formdata_pg_index with fields through indkey
* where indkey is a variable length int2vector with indKeyArray values.
*/
direct->relationData.rd_index =
(FormData_pg_index*)palloc(
pgIndexFixedLen + indKeyVectorLen);
memcpy(direct->relationData.rd_index, pgIndex, pgIndexFixedLen);
indKeyVector = buildint2vector(
indKeyArray,
natts);
memcpy(
&direct->relationData.rd_index->indkey,
indKeyVector,
indKeyVectorLen);
pfree(indKeyVector);
direct->relationData.rd_am = pgAm;
amstrategies = pgAm->amstrategies;
amsupport = pgAm->amsupport;
direct->relationData.rd_indexcxt = TopMemoryContext;
/*
* Allocate arrays to hold data
*/
direct->relationData.rd_aminfo = (RelationAmInfo *)
MemoryContextAllocZero(TopMemoryContext, sizeof(RelationAmInfo));
direct->relationData.rd_opfamily = (Oid *)
MemoryContextAllocZero(TopMemoryContext, natts * sizeof(Oid));
direct->relationData.rd_opcintype = (Oid *)
MemoryContextAllocZero(TopMemoryContext, natts * sizeof(Oid));
if (amstrategies > 0)
operator = (Oid *)
MemoryContextAllocZero(TopMemoryContext,
natts * amstrategies * sizeof(Oid));
else
operator = NULL;
if (amsupport > 0)
{
int nsupport = natts * amsupport;
support = (RegProcedure *)
MemoryContextAllocZero(TopMemoryContext, nsupport * sizeof(RegProcedure));
supportinfo = (FmgrInfo *)
MemoryContextAllocZero(TopMemoryContext, nsupport * sizeof(FmgrInfo));
}
else
{
support = NULL;
supportinfo = NULL;
}
direct->relationData.rd_operator = operator;
direct->relationData.rd_support = support;
direct->relationData.rd_supportinfo = supportinfo;
direct->relationData.rd_indoption = (int16 *)
MemoryContextAllocZero(TopMemoryContext, natts * sizeof(int16));
/*
* Create oidvector in rd_indclass with values from indClassArray.
*/
indClassVector = buildoidvector(indClassArray, natts);
/*
* Fill the operator and support procedure OID arrays. (aminfo and
* supportinfo are left as zeroes, and are filled on-the-fly when used)
*/
IndexSupportInitialize(indClassVector,
operator, support,
direct->relationData.rd_opfamily,
direct->relationData.rd_opcintype,
amstrategies, amsupport, natts);
/*
* expressions and predicate cache will be filled later.
*/
direct->relationData.rd_indexprs = NIL;
direct->relationData.rd_indpred = NIL;
direct->relationData.rd_amcache = NULL;
}
// Not much in terms of contraints.
direct->constrData.has_not_null = true;
/*
* Setup tuple descriptor for columns.
*/
direct->descData.natts = pgClass->relnatts;
// Make the array of pointers.
direct->descData.attrs =
(Form_pg_attribute*)
MemoryContextAllocZero(
TopMemoryContext,
sizeof(Form_pg_attribute*) * pgClass->relnatts);
for (i = 0; i < pgClass->relnatts; i++)
{
direct->descData.attrs[i] =
(Form_pg_attribute)
MemoryContextAllocZero(
TopMemoryContext,
sizeof(FormData_pg_attribute));
memcpy(direct->descData.attrs[i], &(attrArray[i]), sizeof(FormData_pg_attribute));
// Patch up relation id.
direct->descData.attrs[i]->attrelid = relationId;
}
direct->descData.constr = &direct->constrData;
direct->descData.tdtypeid = pgClass->reltype;
direct->descData.tdtypmod = -1;
direct->descData.tdqdtypmod = -1;
direct->descData.tdhasoid = relHasOid;
direct->descData.tdrefcount = 1;
direct->relationData.rd_att = &direct->descData;
direct->pgStat.t_id = relationId;
direct->pgStat.t_shared = 1;
direct->relationData.pgstat_info = &direct->pgStat;
direct->isInit = true;
}
// UNDONE: Should verify for NON-SHARED relations we don't open relations in different databases / or
// UNDONE: open different relations in same database at same time !!!
direct->relationData.rd_node.spcNode = tablespace;
direct->relationData.rd_node.dbNode = database;
direct->relationData.rd_node.relNode = relfilenode;
direct->relationData.rd_targblock = InvalidBlockNumber;
for (i = 0; i < direct->relationData.rd_rel->relnatts; i++)
{
Assert(direct->descData.attrs[i] != NULL);
// Patch up relation id.
direct->descData.attrs[i]->attrelid = direct->relationData.rd_id;
}
direct->relationData.rd_refcnt++;
RelationOpenSmgr(&direct->relationData);
return &direct->relationData;
}
