rdelaage
/
postgresql
mirror of https://git.postgresql.org/git/postgresql.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Teach CLUSTER to skip writing WAL if not needed (ie, not using archiving) 

--- Simon.
Also, code review and cleanup for the previous COPY-no-WAL patches --- Tom.
This commit is contained in:
Tom Lane 2007-03-29 00:15:39 +00:00
parent 4591fb1aa8
commit fba8113c1b
12 changed files with 281 additions and 182 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

116
doc/src/sgml/perform.sgml
View File

@ -1,4 +1,4 @@
<!-- $PostgreSQL: pgsql/doc/src/sgml/perform.sgml,v 1.63 2007/02/01 19:10:24 momjian Exp $ -->
<!-- $PostgreSQL: pgsql/doc/src/sgml/perform.sgml,v 1.64 2007/03/29 00:15:36 tgl Exp $ -->
<chapter id="performance-tips">
<title>Performance Tips</title>
@ -801,7 +801,7 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
<command>EXECUTE</command> as many times as required. This avoids
some of the overhead of repeatedly parsing and planning
<command>INSERT</command>. Different interfaces provide this facility
in different ways; look for Prepared Statements in the interface
in different ways; look for <quote>prepared statements</> in the interface
documentation.
</para>
@ -815,14 +815,12 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
<para>
<command>COPY</command> is fastest when used within the same
transaction as an earlier <command>CREATE TABLE</command> or
<command>TRUNCATE</command> command. In those cases, no WAL
needs to be written because in case of an error, the files
containing the newly loaded data will be removed automatically.
<command>CREATE TABLE AS SELECT</command> is also optimized
to avoid writing WAL. <command>COPY</command> and
<command>CREATE TABLE AS SELECT</command> will write WAL
when <xref linkend="guc-archive-command"> is set and will not
therefore be optimized in that case.
<command>TRUNCATE</command> command. In such cases no WAL
needs to be written, because in case of an error, the files
containing the newly loaded data will be removed anyway.
However, this consideration does not apply when
<xref linkend="guc-archive-command"> is set, as all commands
must write WAL in that case.
</para>
</sect2>
@ -897,23 +895,51 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
<title>Turn off <varname>archive_command</varname></title>
<para>
When loading large amounts of data you might want to unset the
<xref linkend="guc-archive-command"> before loading. It might be
faster to take a new base backup once the load has completed
than to allow a large archive to accumulate.
When loading large amounts of data into an installation that uses
WAL archiving, you might want to disable archiving (unset the
<xref linkend="guc-archive-command"> configuration variable)
while loading. It might be
faster to take a new base backup after the load has completed
than to process a large amount of incremental WAL data.
</para>
<para>
This is particularly important advice because certain commands
will perform more slowly when <varname>archive_command</varname>
is set, as a result of their needing to write large amounts of WAL.
Aside from avoiding the time for the archiver to process the WAL data,
doing this will actually make certain commands faster, because they
are designed not to write WAL at all if <varname>archive_command</varname>
is unset. (They can guarantee crash safety more cheaply by doing an
<function>fsync</> at the end than by writing WAL.)
This applies to the following commands:
<command>CREATE TABLE AS SELECT</command>,
<command>CREATE INDEX</command> and also <command>COPY</command>, when
it is executed in the same transaction as a prior
<command>CREATE TABLE</command> or <command>TRUNCATE</command> command.
<itemizedlist>
<listitem>
<para>
<command>CREATE TABLE AS SELECT</command>
</para>
</listitem>
<listitem>
<para>
<command>CREATE INDEX</command> (and variants such as
<command>ALTER TABLE ADD PRIMARY KEY</command>)
</para>
</listitem>
<listitem>
<para>
<command>ALTER TABLE SET TABLESPACE</command>
</para>
</listitem>
<listitem>
<para>
<command>CLUSTER</command>
</para>
</listitem>
<listitem>
<para>
<command>COPY FROM</command>, when the target table has been
created or truncated earlier in the same transaction
</para>
</listitem>
</itemizedlist>
</para>
</sect2>
<sect2 id="populate-analyze">
@ -950,15 +976,43 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
By default, <application>pg_dump</> uses <command>COPY</>, and when
it is generating a complete schema-and-data dump, it is careful to
load data before creating indexes and foreign keys. So in this case
the first several guidelines are handled automatically. What is left
for you to do is to set appropriate (i.e., larger than normal) values
for <varname>maintenance_work_mem</varname> and
<varname>checkpoint_segments</varname>, as well as unsetting
<varname>archive_command</varname> before loading the dump script,
and then to run <command>ANALYZE</> afterwards and resetting
<varname>archive_command</varname> if required. All of the
parameters can be reset once the load has completed without needing
to restart the server, as described in <xref linkend="config-setting">.
several guidelines are handled automatically. What is left
for you to do is to:
<itemizedlist>
<listitem>
<para>
Set appropriate (i.e., larger than normal) values for
<varname>maintenance_work_mem</varname> and
<varname>checkpoint_segments</varname>.
</para>
</listitem>
<listitem>
<para>
If using WAL archiving, consider disabling it during the restore.
To do that, unset <varname>archive_command</varname> before loading the
dump script, and afterwards restore <varname>archive_command</varname>
and take a fresh base backup.
</para>
</listitem>
<listitem>
<para>
Consider whether the whole dump should be restored as a single
transaction. To do that, pass the <option>-1</> or
<option>--single-transaction</> command-line option to
<application>psql</> or <application>pg_restore</>. When using this
mode, even the smallest of errors will rollback the entire restore,
possibly discarding many hours of processing. Depending on how
interrelated the data is, that might seem preferable to manual cleanup,
or not. <command>COPY</> commands will run fastest if you use a single
transaction and have WAL archiving turned off.
</para>
</listitem>
<listitem>
<para>
Run <command>ANALYZE</> afterwards.
</para>
</listitem>
</itemizedlist>
</para>
<para>

73
src/backend/access/heap/heapam.c
View File

@ -8,7 +8,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.229 2007/03/25 19:45:13 tgl Exp $
* $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.230 2007/03/29 00:15:37 tgl Exp $
*
*
* INTERFACE ROUTINES
@ -1360,11 +1360,14 @@ heap_get_latest_tid(Relation relation,
* that all new tuples go into new pages not containing any tuples from other
* transactions, that the relation gets fsync'd before commit, and that the
* transaction emits at least one WAL record to ensure RecordTransactionCommit
* will decide to WAL-log the commit. (see heap_sync() comments also)
* will decide to WAL-log the commit. (See also heap_sync() comments)
*
* use_fsm is passed directly to RelationGetBufferForTuple, which see for
* more info.
*
* Note that use_wal and use_fsm will be applied when inserting into the
* heap's TOAST table, too, if the tuple requires any out-of-line data.
*
* The return value is the OID assigned to the tuple (either here or by the
* caller), or InvalidOid if no OID. The header fields of *tup are updated
* to match the stored tuple; in particular tup->t_self receives the actual
@ -1418,7 +1421,8 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
* into the relation; tup is the caller's original untoasted data.
*/
if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
heaptup = toast_insert_or_update(relation, tup, NULL, use_wal);
heaptup = toast_insert_or_update(relation, tup, NULL,
use_wal, use_fsm);
else
heaptup = tup;
@ -1526,8 +1530,10 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
* simple_heap_insert - insert a tuple
*
* Currently, this routine differs from heap_insert only in supplying
* a default command ID. But it should be used rather than using
* heap_insert directly in most places where we are modifying system catalogs.
* a default command ID and not allowing access to the speedup options.
*
* This should be used rather than using heap_insert directly in most places
* where we are modifying system catalogs.
*/
Oid
simple_heap_insert(Relation relation, HeapTuple tup)
@ -1535,18 +1541,6 @@ simple_heap_insert(Relation relation, HeapTuple tup)
return heap_insert(relation, tup, GetCurrentCommandId(), true, true);
}
/*
* fast_heap_insert - insert a tuple with options to improve speed
*
* Currently, this routine allows specifying additional options for speed
* in certain cases, such as WAL-avoiding COPY command
*/
Oid
fast_heap_insert(Relation relation, HeapTuple tup, bool use_wal)
{
return heap_insert(relation, tup, GetCurrentCommandId(), use_wal, use_wal);
}
/*
* heap_delete - delete a tuple
*
@ -2112,7 +2106,9 @@ l2:
*/
if (need_toast)
{
heaptup = toast_insert_or_update(relation, newtup, &oldtup, true);
/* Note we always use WAL and FSM during updates */
heaptup = toast_insert_or_update(relation, newtup, &oldtup,
true, true);
newtupsize = MAXALIGN(heaptup->t_len);
}
else
@ -3988,23 +3984,40 @@ heap2_desc(StringInfo buf, uint8 xl_info, char *rec)
appendStringInfo(buf, "UNKNOWN");
}
/* ----------------
* heap_sync - sync a heap, for use when no WAL has been written
/*
* heap_sync - sync a heap, for use when no WAL has been written
*
* ----------------
* This forces the heap contents (including TOAST heap if any) down to disk.
* If we skipped using WAL, and it's not a temp relation, we must force the
* relation down to disk before it's safe to commit the transaction. This
* requires writing out any dirty buffers and then doing a forced fsync.
*
* Indexes are not touched. (Currently, index operations associated with
* the commands that use this are WAL-logged and so do not need fsync.
* That behavior might change someday, but in any case it's likely that
* any fsync decisions required would be per-index and hence not appropriate
* to be done here.)
*/
void
heap_sync(Relation rel)
{
if (!rel->rd_istemp)
/* temp tables never need fsync */
if (rel->rd_istemp)
return;
/* main heap */
FlushRelationBuffers(rel);
/* FlushRelationBuffers will have opened rd_smgr */
smgrimmedsync(rel->rd_smgr);
/* toast heap, if any */
if (OidIsValid(rel->rd_rel->reltoastrelid))
{
/*
* If we skipped using WAL, and it's not a temp relation,
* we must force the relation down to disk before it's
* safe to commit the transaction. This requires forcing
* out any dirty buffers and then doing a forced fsync.
*/
FlushRelationBuffers(rel);
smgrimmedsync(rel->rd_smgr);
Relation toastrel;
toastrel = heap_open(rel->rd_rel->reltoastrelid, AccessShareLock);
FlushRelationBuffers(toastrel);
smgrimmedsync(toastrel->rd_smgr);
heap_close(toastrel, AccessShareLock);
}
}

22
src/backend/access/heap/tuptoaster.c
View File

@ -8,7 +8,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/access/heap/tuptoaster.c,v 1.71 2007/02/27 23:48:07 tgl Exp $
* $PostgreSQL: pgsql/src/backend/access/heap/tuptoaster.c,v 1.72 2007/03/29 00:15:37 tgl Exp $
*
*
* INTERFACE ROUTINES
@ -33,6 +33,7 @@
#include "access/genam.h"
#include "access/heapam.h"
#include "access/tuptoaster.h"
#include "access/xact.h"
#include "catalog/catalog.h"
#include "utils/fmgroids.h"
#include "utils/pg_lzcompress.h"
@ -42,7 +43,8 @@
#undef TOAST_DEBUG
static void toast_delete_datum(Relation rel, Datum value);
static Datum toast_save_datum(Relation rel, Datum value, bool use_wal);
static Datum toast_save_datum(Relation rel, Datum value,
bool use_wal, bool use_fsm);
static varattrib *toast_fetch_datum(varattrib *attr);
static varattrib *toast_fetch_datum_slice(varattrib *attr,
int32 sliceoffset, int32 length);
@ -333,6 +335,7 @@ toast_delete(Relation rel, HeapTuple oldtup)
* Inputs:
* newtup: the candidate new tuple to be inserted
* oldtup: the old row version for UPDATE, or NULL for INSERT
* use_wal, use_fsm: flags to be passed to heap_insert() for toast rows
* Result:
* either newtup if no toasting is needed, or a palloc'd modified tuple
* that is what should actually get stored
@ -342,7 +345,8 @@ toast_delete(Relation rel, HeapTuple oldtup)
* ----------
*/
HeapTuple
toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup, bool use_wal)
toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup,
bool use_wal, bool use_fsm)
{
HeapTuple result_tuple;
TupleDesc tupleDesc;
@ -618,7 +622,8 @@ toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup, bool us
i = biggest_attno;
old_value = toast_values[i];
toast_action[i] = 'p';
toast_values[i] = toast_save_datum(rel, toast_values[i], use_wal);
toast_values[i] = toast_save_datum(rel, toast_values[i],
use_wal, use_fsm);
if (toast_free[i])
pfree(DatumGetPointer(old_value));
@ -729,7 +734,8 @@ toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup, bool us
i = biggest_attno;
old_value = toast_values[i];
toast_action[i] = 'p';
toast_values[i] = toast_save_datum(rel, toast_values[i], use_wal);
toast_values[i] = toast_save_datum(rel, toast_values[i],
use_wal, use_fsm);
if (toast_free[i])
pfree(DatumGetPointer(old_value));
@ -977,7 +983,8 @@ toast_compress_datum(Datum value)
* ----------
*/
static Datum
toast_save_datum(Relation rel, Datum value, bool use_wal)
toast_save_datum(Relation rel, Datum value,
bool use_wal, bool use_fsm)
{
Relation toastrel;
Relation toastidx;
@ -985,6 +992,7 @@ toast_save_datum(Relation rel, Datum value, bool use_wal)
TupleDesc toasttupDesc;
Datum t_values[3];
bool t_isnull[3];
CommandId mycid = GetCurrentCommandId();
varattrib *result;
struct
{
@ -1063,7 +1071,7 @@ toast_save_datum(Relation rel, Datum value, bool use_wal)
if (!HeapTupleIsValid(toasttup))
elog(ERROR, "failed to build TOAST tuple");
fast_heap_insert(toastrel, toasttup, use_wal);
heap_insert(toastrel, toasttup, mycid, use_wal, use_fsm);
/*
* Create the index entry. We cheat a little here by not using

9
src/backend/catalog/index.c
View File

@ -8,7 +8,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/catalog/index.c,v 1.281 2007/03/25 19:45:14 tgl Exp $
* $PostgreSQL: pgsql/src/backend/catalog/index.c,v 1.282 2007/03/29 00:15:37 tgl Exp $
*
*
* INTERFACE ROUTINES
@ -1248,12 +1248,11 @@ setNewRelfilenode(Relation relation)
heap_close(pg_class, RowExclusiveLock);
/* Remember we did this in current transaction, to allow later optimisations */
relation->rd_newRelfilenodeSubid = GetCurrentSubTransactionId();
RelationCacheResetAtEOXact();
/* Make sure the relfilenode change is visible */
CommandCounterIncrement();
/* Mark the rel as having a new relfilenode in current transaction */
RelationCacheMarkNewRelfilenode(relation);
}

20
src/backend/commands/cluster.c
View File

@ -11,7 +11,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/commands/cluster.c,v 1.157 2007/03/13 00:33:39 tgl Exp $
* $PostgreSQL: pgsql/src/backend/commands/cluster.c,v 1.158 2007/03/29 00:15:37 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -653,6 +653,8 @@ copy_heap_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex)
char *nulls;
IndexScanDesc scan;
HeapTuple tuple;
CommandId mycid = GetCurrentCommandId();
bool use_wal;
/*
* Open the relations we need.
@ -675,6 +677,17 @@ copy_heap_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex)
nulls = (char *) palloc(natts * sizeof(char));
memset(nulls, 'n', natts * sizeof(char));
/*
* We need to log the copied data in WAL iff WAL archiving is enabled AND
* it's not a temp rel. (Since we know the target relation is new and
* can't have any FSM data, we can always tell heap_insert to ignore FSM,
* even when using WAL.)
*/
use_wal = XLogArchivingActive() && !NewHeap->rd_istemp;
/* use_wal off requires rd_targblock be initially invalid */
Assert(NewHeap->rd_targblock == InvalidBlockNumber);
/*
* Scan through the OldHeap on the OldIndex and copy each tuple into the
* NewHeap.
@ -722,7 +735,7 @@ copy_heap_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex)
if (NewHeap->rd_rel->relhasoids)
HeapTupleSetOid(copiedTuple, HeapTupleGetOid(tuple));
simple_heap_insert(NewHeap, copiedTuple);
heap_insert(NewHeap, copiedTuple, mycid, use_wal, false);
heap_freetuple(copiedTuple);
@ -734,6 +747,9 @@ copy_heap_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex)
pfree(values);
pfree(nulls);
if (!use_wal)
heap_sync(NewHeap);
index_close(OldIndex, NoLock);
heap_close(OldHeap, NoLock);
heap_close(NewHeap, NoLock);

112
src/backend/commands/copy.c
View File

@ -8,7 +8,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/commands/copy.c,v 1.278 2007/03/13 00:33:39 tgl Exp $
* $PostgreSQL: pgsql/src/backend/commands/copy.c,v 1.279 2007/03/29 00:15:38 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -1125,11 +1125,10 @@ DoCopy(const CopyStmt *stmt, const char *queryString)
cstate->copy_dest = COPY_FILE; /* default */
cstate->filename = stmt->filename;
if (is_from) /* copy from file to database */
CopyFrom(cstate);
if (is_from)
CopyFrom(cstate); /* copy from file to database */
else
/* copy from database to file */
DoCopyTo(cstate);
DoCopyTo(cstate); /* copy from database to file */
/*
* Close the relation or query. If reading, we can release the
@ -1640,7 +1639,9 @@ CopyFrom(CopyState cstate)
ExprContext *econtext; /* used for ExecEvalExpr for default atts */
MemoryContext oldcontext = CurrentMemoryContext;
ErrorContextCallback errcontext;
bool use_wal = true; /* By default, we use WAL to log db changes */
CommandId mycid = GetCurrentCommandId();
bool use_wal = true; /* by default, use WAL logging */
bool use_fsm = true; /* by default, use FSM for free space */
Assert(cstate->rel);
@ -1663,6 +1664,48 @@ CopyFrom(CopyState cstate)
RelationGetRelationName(cstate->rel))));
}
/*----------
* Check to see if we can avoid writing WAL
*
* If archive logging is not enabled *and* either
* - table was created in same transaction as this COPY
* - data is being written to relfilenode created in this transaction
* then we can skip writing WAL. It's safe because if the transaction
* doesn't commit, we'll discard the table (or the new relfilenode file).
* If it does commit, we'll have done the heap_sync at the bottom of this
* routine first.
*
* As mentioned in comments in utils/rel.h, the in-same-transaction test
* is not completely reliable, since in rare cases rd_createSubid or
* rd_newRelfilenodeSubid can be cleared before the end of the transaction.
* However this is OK since at worst we will fail to make the optimization.
*
* When skipping WAL it's entirely possible that COPY itself will write no
* WAL records at all. This is of concern because RecordTransactionCommit
* might decide it doesn't need to log our eventual commit, which we
* certainly need it to do. However, we need no special action here for
* that, because if we have a new table or new relfilenode then there
* must have been a WAL-logged pg_class update earlier in the transaction.
*
* Also, if the target file is new-in-transaction, we assume that checking
* FSM for free space is a waste of time, even if we must use WAL because
* of archiving. This could possibly be wrong, but it's unlikely.
*
* The comments for heap_insert and RelationGetBufferForTuple specify that
* skipping WAL logging is only safe if we ensure that our tuples do not
* go into pages containing tuples from any other transactions --- but this
* must be the case if we have a new table or new relfilenode, so we need
* no additional work to enforce that.
*----------
*/
if (cstate->rel->rd_createSubid != InvalidSubTransactionId ||
cstate->rel->rd_newRelfilenodeSubid != InvalidSubTransactionId)
{
use_fsm = false;
if (!XLogArchivingActive())
use_wal = false;
}
if (pipe)
{
if (whereToSendOutput == DestRemote)
@ -1832,28 +1875,6 @@ CopyFrom(CopyState cstate)
nfields = file_has_oids ? (attr_count + 1) : attr_count;
field_strings = (char **) palloc(nfields * sizeof(char *));
/*
* Check for performance optimization by avoiding WAL writes
*
* If archive logging is not be enabled *and* either
* - table is created in same transaction as this COPY
* - table data is now being written to new relfilenode
* then we can safely avoid writing WAL. Why?
* The data files for the table plus toast table/index, plus any indexes
* will all be dropped at the end of the transaction if it fails, so we
* do not need to worry about inconsistent states.
* As mentioned in comments in utils/rel.h, the in-same-transaction test is
* not completely reliable, since rd_createSubId can be reset to zero in
* certain cases before the end of the creating transaction.
* We are doing this for performance only, so we only need to know:
* if rd_createSubid != InvalidSubTransactionId then it is *always* just
* created. If we have PITR enabled, then we *must* use_wal
*/
if ((cstate->rel->rd_createSubid != InvalidSubTransactionId ||
cstate->rel->rd_newRelfilenodeSubid != InvalidSubTransactionId)
&& !XLogArchivingActive())
use_wal = false;
/* Initialize state variables */
cstate->fe_eof = false;
cstate->eol_type = EOL_UNKNOWN;
@ -2087,7 +2108,7 @@ CopyFrom(CopyState cstate)
ExecConstraints(resultRelInfo, slot, estate);
/* OK, store the tuple and create index entries for it */
fast_heap_insert(cstate->rel, tuple, use_wal);
heap_insert(cstate->rel, tuple, mycid, use_wal, use_fsm);
if (resultRelInfo->ri_NumIndices > 0)
ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
@ -2104,32 +2125,6 @@ CopyFrom(CopyState cstate)
}
}
/*
* If we skipped writing WAL for heaps, then we need to sync
*/
if (!use_wal)
{
/* main heap */
heap_sync(cstate->rel);
/* main heap indexes, if any */
/* we always use WAL for index inserts, so no need to sync */
/* toast heap, if any */
if (OidIsValid(cstate->rel->rd_rel->reltoastrelid))
{
Relation toastrel;
toastrel = heap_open(cstate->rel->rd_rel->reltoastrelid,
AccessShareLock);
heap_sync(toastrel);
heap_close(toastrel, AccessShareLock);
}
/* toast index, if toast heap */
/* we always use WAL for index inserts, so no need to sync */
}
/* Done, clean up */
error_context_stack = errcontext.previous;
@ -2164,6 +2159,13 @@ CopyFrom(CopyState cstate)
errmsg("could not read from file \"%s\": %m",
cstate->filename)));
}
/*
* If we skipped writing WAL, then we need to sync the heap (but not
* indexes since those use WAL anyway)
*/
if (!use_wal)
heap_sync(cstate->rel);
}

12
src/backend/executor/execMain.c
View File

@ -26,7 +26,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.291 2007/03/25 19:45:14 tgl Exp $
* $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.292 2007/03/29 00:15:38 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -2541,14 +2541,8 @@ CloseIntoRel(QueryDesc *queryDesc)
/* OpenIntoRel might never have gotten called */
if (estate->es_into_relation_descriptor)
{
/*
* If we skipped using WAL, and it's not a temp relation, we must
* force the relation down to disk before it's safe to commit the
* transaction. This requires forcing out any dirty buffers and then
* doing a forced fsync.
*/
if (!estate->es_into_relation_use_wal &&
!estate->es_into_relation_descriptor->rd_istemp)
/* If we skipped using WAL, must heap_sync before commit */
if (!estate->es_into_relation_use_wal)
heap_sync(estate->es_into_relation_descriptor);
/* close rel, but keep lock until commit */

75
src/backend/utils/cache/relcache.c vendored
View File

@ -8,7 +8,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/utils/cache/relcache.c,v 1.258 2007/03/19 23:38:29 wieck Exp $
* $PostgreSQL: pgsql/src/backend/utils/cache/relcache.c,v 1.259 2007/03/29 00:15:38 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -1572,7 +1572,8 @@ RelationClose(Relation relation)
#ifdef RELCACHE_FORCE_RELEASE
if (RelationHasReferenceCountZero(relation) &&
relation->rd_createSubid == InvalidSubTransactionId)
relation->rd_createSubid == InvalidSubTransactionId &&
relation->rd_newRelfilenodeSubid == InvalidSubTransactionId)
RelationClearRelation(relation, false);
#endif
}
@ -1759,11 +1760,12 @@ RelationClearRelation(Relation relation, bool rebuild)
{
/*
* When rebuilding an open relcache entry, must preserve ref count and
* rd_createSubid state. Also attempt to preserve the tupledesc and
* rewrite-rule substructures in place. (Note: the refcount mechanism
* for tupledescs may eventually ensure that we don't really need to
* preserve the tupledesc in-place, but for now there are still a lot
* of places that assume an open rel's tupledesc won't move.)
* rd_createSubid/rd_newRelfilenodeSubid state. Also attempt to
* preserve the tupledesc and rewrite-rule substructures in place.
* (Note: the refcount mechanism for tupledescs may eventually ensure
* that we don't really need to preserve the tupledesc in-place, but
* for now there are still a lot of places that assume an open rel's
* tupledesc won't move.)
*
* Note that this process does not touch CurrentResourceOwner; which
* is good because whatever ref counts the entry may have do not
@ -1839,7 +1841,7 @@ RelationFlushRelation(Relation relation)
/*
* New relcache entries are always rebuilt, not flushed; else we'd
* forget the "new" status of the relation, which is a useful
* optimization to have.
* optimization to have. Ditto for the new-relfilenode status.
*/
rebuild = true;
}
@ -1916,6 +1918,8 @@ RelationCacheInvalidateEntry(Oid relationId)
* so we do not touch new-in-transaction relations; they cannot be targets
* of cross-backend SI updates (and our own updates now go through a
* separate linked list that isn't limited by the SI message buffer size).
* Likewise, we need not discard new-relfilenode-in-transaction hints,
* since any invalidation of those would be a local event.
*
* We do this in two phases: the first pass deletes deletable items, and
* the second one rebuilds the rebuildable items. This is essential for
@ -1958,14 +1962,6 @@ RelationCacheInvalidate(void)
if (relation->rd_createSubid != InvalidSubTransactionId)
continue;
/*
* Reset newRelfilenode hint. It is never used for correctness, only
* for performance optimization. An incorrectly set hint can lead
* to data loss in some circumstances, so play safe.
*/
if (relation->rd_newRelfilenodeSubid != InvalidSubTransactionId)
relation->rd_newRelfilenodeSubid = InvalidSubTransactionId;
relcacheInvalsReceived++;
if (RelationHasReferenceCountZero(relation))
@ -2017,17 +2013,6 @@ RelationCacheInvalidate(void)
list_free(rebuildList);
}
/*
* RelationCacheResetAtEOXact
*
* Register that work will be required at main-transaction commit or abort
*/
void
RelationCacheResetAtEOXact(void)
{
need_eoxact_work = true;
}
/*
* AtEOXact_RelationCache
*
@ -2056,9 +2041,10 @@ AtEOXact_RelationCache(bool isCommit)
* the debug-only Assert checks, most transactions don't create any work
* for us to do here, so we keep a static flag that gets set if there is
* anything to do. (Currently, this means either a relation is created in
* the current xact, or an index list is forced.) For simplicity, the
* flag remains set till end of top-level transaction, even though we
* could clear it at subtransaction end in some cases.
* the current xact, or one is given a new relfilenode, or an index list
* is forced.) For simplicity, the flag remains set till end of top-level
* transaction, even though we could clear it at subtransaction end in
* some cases.
*/
if (!need_eoxact_work
#ifdef USE_ASSERT_CHECKING
@ -2111,6 +2097,10 @@ AtEOXact_RelationCache(bool isCommit)
continue;
}
}
/*
* Likewise, reset the hint about the relfilenode being new.
*/
relation->rd_newRelfilenodeSubid = InvalidSubTransactionId;
/*
@ -2173,6 +2163,10 @@ AtEOSubXact_RelationCache(bool isCommit, SubTransactionId mySubid,
continue;
}
}
/*
* Likewise, update or drop any new-relfilenode-in-subtransaction hint.
*/
if (relation->rd_newRelfilenodeSubid == mySubid)
{
if (isCommit)
@ -2194,6 +2188,23 @@ AtEOSubXact_RelationCache(bool isCommit, SubTransactionId mySubid,
}
}
/*
* RelationCacheMarkNewRelfilenode
*
* Mark the rel as having been given a new relfilenode in the current
* (sub) transaction. This is a hint that can be used to optimize
* later operations on the rel in the same transaction.
*/
void
RelationCacheMarkNewRelfilenode(Relation rel)
{
/* Mark it... */
rel->rd_newRelfilenodeSubid = GetCurrentSubTransactionId();
/* ... and now we have eoxact cleanup work to do */
need_eoxact_work = true;
}
/*
* RelationBuildLocalRelation
* Build a relcache entry for an about-to-be-created relation,
@ -2272,7 +2283,7 @@ RelationBuildLocalRelation(const char *relname,
rel->rd_newRelfilenodeSubid = InvalidSubTransactionId;
/* must flag that we have rels created in this transaction */
RelationCacheResetAtEOXact();
need_eoxact_work = true;
/* is it a temporary relation? */
rel->rd_istemp = isTempNamespace(relnamespace);
@ -2928,7 +2939,7 @@ RelationSetIndexList(Relation relation, List *indexIds, Oid oidIndex)
relation->rd_oidindex = oidIndex;
relation->rd_indexvalid = 2; /* mark list as forced */
/* must flag that we have a forced index list */
RelationCacheResetAtEOXact();
need_eoxact_work = true;
}
/*

5
src/include/access/heapam.h
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
* $PostgreSQL: pgsql/src/include/access/heapam.h,v 1.120 2007/01/25 02:17:26 momjian Exp $
* $PostgreSQL: pgsql/src/include/access/heapam.h,v 1.121 2007/03/29 00:15:39 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -178,9 +178,6 @@ extern void simple_heap_delete(Relation relation, ItemPointer tid);
extern void simple_heap_update(Relation relation, ItemPointer otid,
HeapTuple tup);
extern Oid fast_heap_insert(Relation relation, HeapTuple tup, bool use_wal);
extern void heap_markpos(HeapScanDesc scan);
extern void heap_restrpos(HeapScanDesc scan);

5
src/include/access/tuptoaster.h
View File

@ -6,7 +6,7 @@
*
* Copyright (c) 2000-2007, PostgreSQL Global Development Group
*
* $PostgreSQL: pgsql/src/include/access/tuptoaster.h,v 1.32 2007/02/05 04:22:18 tgl Exp $
* $PostgreSQL: pgsql/src/include/access/tuptoaster.h,v 1.33 2007/03/29 00:15:39 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -79,7 +79,8 @@
* ----------
*/
extern HeapTuple toast_insert_or_update(Relation rel,
HeapTuple newtup, HeapTuple oldtup, bool use_wal);
HeapTuple newtup, HeapTuple oldtup,
bool use_wal, bool use_fsm);
/* ----------
* toast_delete -

8
src/include/utils/rel.h
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
* $PostgreSQL: pgsql/src/include/utils/rel.h,v 1.99 2007/03/19 23:38:32 wieck Exp $
* $PostgreSQL: pgsql/src/include/utils/rel.h,v 1.100 2007/03/29 00:15:39 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -138,13 +138,17 @@ typedef struct RelationData
char rd_indexvalid; /* state of rd_indexlist: 0 = not valid, 1 =
* valid, 2 = temporarily forced */
SubTransactionId rd_createSubid; /* rel was created in current xact */
SubTransactionId rd_newRelfilenodeSubid; /* rel had new relfilenode in current xact */
SubTransactionId rd_newRelfilenodeSubid; /* new relfilenode assigned
* in current xact */
/*
* rd_createSubid is the ID of the highest subtransaction the rel has
* survived into; or zero if the rel was not created in the current top
* transaction. This should be relied on only for optimization purposes;
* it is possible for new-ness to be "forgotten" (eg, after CLUSTER).
* Likewise, rd_newRelfilenodeSubid is the ID of the highest subtransaction
* the relfilenode change has survived into, or zero if not changed in
* the current transaction (or we have forgotten changing it).
*/
Form_pg_class rd_rel; /* RELATION tuple */
TupleDesc rd_att; /* tuple descriptor */

6
src/include/utils/relcache.h
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
* $PostgreSQL: pgsql/src/include/utils/relcache.h,v 1.58 2007/03/03 20:08:41 momjian Exp $
* $PostgreSQL: pgsql/src/include/utils/relcache.h,v 1.59 2007/03/29 00:15:39 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -60,12 +60,12 @@ extern void RelationCacheInvalidateEntry(Oid relationId);
extern void RelationCacheInvalidate(void);
extern void RelationCacheResetAtEOXact(void);
extern void AtEOXact_RelationCache(bool isCommit);
extern void AtEOSubXact_RelationCache(bool isCommit, SubTransactionId mySubid,
SubTransactionId parentSubid);
extern void RelationCacheMarkNewRelfilenode(Relation rel);
/*
* Routines to help manage rebuilding of relcache init file
*/