mirror of
https://git.postgresql.org/git/postgresql.git
synced 2024-12-27 08:39:28 +08:00
44cac93464
There's a project policy against using plain "char buf[BLCKSZ]" local or static variables as page buffers; preferred style is to palloc or malloc each buffer to ensure it is MAXALIGN'd. However, that policy's been ignored in an increasing number of places. We've apparently got away with it so far, probably because (a) relatively few people use platforms on which misalignment causes core dumps and/or (b) the variables chance to be sufficiently aligned anyway. But this is not something to rely on. Moreover, even if we don't get a core dump, we might be paying a lot of cycles for misaligned accesses. To fix, invent new union types PGAlignedBlock and PGAlignedXLogBlock that the compiler must allocate with sufficient alignment, and use those in place of plain char arrays. I used these types even for variables where there's no risk of a misaligned access, since ensuring proper alignment should make kernel data transfers faster. I also changed some places where we had been palloc'ing short-lived buffers, for coding style uniformity and to save palloc/pfree overhead. Since this seems to be a live portability hazard (despite the lack of field reports), back-patch to all supported versions. Patch by me; thanks to Michael Paquier for review. Discussion: https://postgr.es/m/1535618100.1286.3.camel@credativ.de
206 lines
5.6 KiB
C
206 lines
5.6 KiB
C
/*-------------------------------------------------------------------------
|
|
*
|
|
* pg_prewarm.c
|
|
* prewarming utilities
|
|
*
|
|
* Copyright (c) 2010-2018, PostgreSQL Global Development Group
|
|
*
|
|
* IDENTIFICATION
|
|
* contrib/pg_prewarm/pg_prewarm.c
|
|
*
|
|
*-------------------------------------------------------------------------
|
|
*/
|
|
#include "postgres.h"
|
|
|
|
#include <sys/stat.h>
|
|
#include <unistd.h>
|
|
|
|
#include "access/heapam.h"
|
|
#include "fmgr.h"
|
|
#include "miscadmin.h"
|
|
#include "storage/bufmgr.h"
|
|
#include "storage/smgr.h"
|
|
#include "utils/acl.h"
|
|
#include "utils/builtins.h"
|
|
#include "utils/lsyscache.h"
|
|
#include "utils/rel.h"
|
|
|
|
PG_MODULE_MAGIC;
|
|
|
|
PG_FUNCTION_INFO_V1(pg_prewarm);
|
|
|
|
typedef enum
|
|
{
|
|
PREWARM_PREFETCH,
|
|
PREWARM_READ,
|
|
PREWARM_BUFFER
|
|
} PrewarmType;
|
|
|
|
static PGAlignedBlock blockbuffer;
|
|
|
|
/*
|
|
* 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_PP(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_PP(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, get_relkind_objtype(rel->rd_rel->relkind), 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.data);
|
|
++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);
|
|
}
|