smgrzeroextend() uses FileFallocate() to efficiently extend files by multiple
blocks. When extending by a small number of blocks, use FileZero() instead, as
using posix_fallocate() for small numbers of blocks is inefficient for some
file systems / operating systems. FileZero() is also used as the fallback for
FileFallocate() on platforms / filesystems that don't support fallocate.
A big advantage of using posix_fallocate() is that it typically won't cause
dirty buffers in the kernel pagecache. So far the most common pattern in our
code is that we smgrextend() a page full of zeroes and put the corresponding
page into shared buffers, from where we later write out the actual contents of
the page. If the kernel, e.g. due to memory pressure or elapsed time, already
wrote back the all-zeroes page, this can lead to doubling the amount of writes
reaching storage.
There are no users of smgrzeroextend() as of this commit. That will follow in
future commits.
Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
Reviewed-by: Heikki Linnakangas <hlinnaka@iki.fi>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Reviewed-by: David Rowley <dgrowleyml@gmail.com>
Reviewed-by: John Naylor <john.naylor@enterprisedb.com>
Discussion: https://postgr.es/m/
20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
return returnCode;
}
+/*
+ * Zero a region of the file.
+ *
+ * Returns 0 on success, -1 otherwise. In the latter case errno is set to the
+ * appropriate error.
+ */
+int
+FileZero(File file, off_t offset, off_t amount, uint32 wait_event_info)
+{
+ int returnCode;
+ ssize_t written;
+
+ Assert(FileIsValid(file));
+
+ DO_DB(elog(LOG, "FileZero: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
+ file, VfdCache[file].fileName,
+ (int64) offset, (int64) amount));
+
+ returnCode = FileAccess(file);
+ if (returnCode < 0)
+ return returnCode;
+
+ pgstat_report_wait_start(wait_event_info);
+ written = pg_pwrite_zeros(VfdCache[file].fd, amount, offset);
+ pgstat_report_wait_end();
+
+ if (written < 0)
+ return -1;
+ else if (written != amount)
+ {
+ /* if errno is unset, assume problem is no disk space */
+ if (errno == 0)
+ errno = ENOSPC;
+ return -1;
+ }
+
+ return 0;
+}
+
+/*
+ * Try to reserve file space with posix_fallocate(). If posix_fallocate() is
+ * not implemented on the operating system or fails with EINVAL / EOPNOTSUPP,
+ * use FileZero() instead.
+ *
+ * Note that at least glibc() implements posix_fallocate() in userspace if not
+ * implemented by the filesystem. That's not the case for all environments
+ * though.
+ *
+ * Returns 0 on success, -1 otherwise. In the latter case errno is set to the
+ * appropriate error.
+ */
+int
+FileFallocate(File file, off_t offset, off_t amount, uint32 wait_event_info)
+{
+#ifdef HAVE_POSIX_FALLOCATE
+ int returnCode;
+
+ Assert(FileIsValid(file));
+
+ DO_DB(elog(LOG, "FileFallocate: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
+ file, VfdCache[file].fileName,
+ (int64) offset, (int64) amount));
+
+ returnCode = FileAccess(file);
+ if (returnCode < 0)
+ return -1;
+
+ pgstat_report_wait_start(wait_event_info);
+ returnCode = posix_fallocate(VfdCache[file].fd, offset, amount);
+ pgstat_report_wait_end();
+
+ if (returnCode == 0)
+ return 0;
+
+ /* for compatibility with %m printing etc */
+ errno = returnCode;
+
+ /*
+ * Return in cases of a "real" failure, if fallocate is not supported,
+ * fall through to the FileZero() backed implementation.
+ */
+ if (returnCode != EINVAL && returnCode != EOPNOTSUPP)
+ return -1;
+#endif
+
+ return FileZero(file, offset, amount, wait_event_info);
+}
+
off_t
FileSize(File file)
{
Assert(_mdnblocks(reln, forknum, v) <= ((BlockNumber) RELSEG_SIZE));
}
+/*
+ * mdzeroextend() -- Add new zeroed out blocks to the specified relation.
+ *
+ * Similar to mdextend(), except the relation can be extended by multiple
+ * blocks at once and the added blocks will be filled with zeroes.
+ */
+void
+mdzeroextend(SMgrRelation reln, ForkNumber forknum,
+ BlockNumber blocknum, int nblocks, bool skipFsync)
+{
+ MdfdVec *v;
+ BlockNumber curblocknum = blocknum;
+ int remblocks = nblocks;
+
+ Assert(nblocks > 0);
+
+ /* This assert is too expensive to have on normally ... */
+#ifdef CHECK_WRITE_VS_EXTEND
+ Assert(blocknum >= mdnblocks(reln, forknum));
+#endif
+
+ /*
+ * If a relation manages to grow to 2^32-1 blocks, refuse to extend it any
+ * more --- we mustn't create a block whose number actually is
+ * InvalidBlockNumber or larger.
+ */
+ if ((uint64) blocknum + nblocks >= (uint64) InvalidBlockNumber)
+ ereport(ERROR,
+ (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+ errmsg("cannot extend file \"%s\" beyond %u blocks",
+ relpath(reln->smgr_rlocator, forknum),
+ InvalidBlockNumber)));
+
+ while (remblocks > 0)
+ {
+ BlockNumber segstartblock = curblocknum % ((BlockNumber) RELSEG_SIZE);
+ off_t seekpos = (off_t) BLCKSZ * segstartblock;
+ int numblocks;
+
+ if (segstartblock + remblocks > RELSEG_SIZE)
+ numblocks = RELSEG_SIZE - segstartblock;
+ else
+ numblocks = remblocks;
+
+ v = _mdfd_getseg(reln, forknum, curblocknum, skipFsync, EXTENSION_CREATE);
+
+ Assert(segstartblock < RELSEG_SIZE);
+ Assert(segstartblock + numblocks <= RELSEG_SIZE);
+
+ /*
+ * If available and useful, use posix_fallocate() (via FileAllocate())
+ * to extend the relation. That's often more efficient than using
+ * write(), as it commonly won't cause the kernel to allocate page
+ * cache space for the extended pages.
+ *
+ * However, we don't use FileAllocate() for small extensions, as it
+ * defeats delayed allocation on some filesystems. Not clear where
+ * that decision should be made though? For now just use a cutoff of
+ * 8, anything between 4 and 8 worked OK in some local testing.
+ */
+ if (numblocks > 8)
+ {
+ int ret;
+
+ ret = FileFallocate(v->mdfd_vfd,
+ seekpos, (off_t) BLCKSZ * numblocks,
+ WAIT_EVENT_DATA_FILE_EXTEND);
+ if (ret != 0)
+ {
+ ereport(ERROR,
+ errcode_for_file_access(),
+ errmsg("could not extend file \"%s\" with FileFallocate(): %m",
+ FilePathName(v->mdfd_vfd)),
+ errhint("Check free disk space."));
+ }
+ }
+ else
+ {
+ int ret;
+
+ /*
+ * Even if we don't want to use fallocate, we can still extend a
+ * bit more efficiently than writing each 8kB block individually.
+ * pg_pwrite_zeroes() (via FileZero()) uses
+ * pg_pwritev_with_retry() to avoid multiple writes or needing a
+ * zeroed buffer for the whole length of the extension.
+ */
+ ret = FileZero(v->mdfd_vfd,
+ seekpos, (off_t) BLCKSZ * numblocks,
+ WAIT_EVENT_DATA_FILE_EXTEND);
+ if (ret < 0)
+ ereport(ERROR,
+ errcode_for_file_access(),
+ errmsg("could not extend file \"%s\": %m",
+ FilePathName(v->mdfd_vfd)),
+ errhint("Check free disk space."));
+ }
+
+ if (!skipFsync && !SmgrIsTemp(reln))
+ register_dirty_segment(reln, forknum, v);
+
+ Assert(_mdnblocks(reln, forknum, v) <= ((BlockNumber) RELSEG_SIZE));
+
+ remblocks -= numblocks;
+ curblocknum += numblocks;
+ }
+}
+
/*
* mdopenfork() -- Open one fork of the specified relation.
*
bool isRedo);
void (*smgr_extend) (SMgrRelation reln, ForkNumber forknum,
BlockNumber blocknum, const void *buffer, bool skipFsync);
+ void (*smgr_zeroextend) (SMgrRelation reln, ForkNumber forknum,
+ BlockNumber blocknum, int nblocks, bool skipFsync);
bool (*smgr_prefetch) (SMgrRelation reln, ForkNumber forknum,
BlockNumber blocknum);
void (*smgr_read) (SMgrRelation reln, ForkNumber forknum,
.smgr_exists = mdexists,
.smgr_unlink = mdunlink,
.smgr_extend = mdextend,
+ .smgr_zeroextend = mdzeroextend,
.smgr_prefetch = mdprefetch,
.smgr_read = mdread,
.smgr_write = mdwrite,
reln->smgr_cached_nblocks[forknum] = InvalidBlockNumber;
}
+/*
+ * smgrzeroextend() -- Add new zeroed out blocks to a file.
+ *
+ * Similar to smgrextend(), except the relation can be extended by
+ * multiple blocks at once and the added blocks will be filled with
+ * zeroes.
+ */
+void
+smgrzeroextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
+ int nblocks, bool skipFsync)
+{
+ smgrsw[reln->smgr_which].smgr_zeroextend(reln, forknum, blocknum,
+ nblocks, skipFsync);
+
+ /*
+ * Normally we expect this to increase the fork size by nblocks, but if
+ * the cached value isn't as expected, just invalidate it so the next call
+ * asks the kernel.
+ */
+ if (reln->smgr_cached_nblocks[forknum] == blocknum)
+ reln->smgr_cached_nblocks[forknum] = blocknum + nblocks;
+ else
+ reln->smgr_cached_nblocks[forknum] = InvalidBlockNumber;
+}
+
/*
* smgrprefetch() -- Initiate asynchronous read of the specified block of a relation.
*
extern int FileRead(File file, void *buffer, size_t amount, off_t offset, uint32 wait_event_info);
extern int FileWrite(File file, const void *buffer, size_t amount, off_t offset, uint32 wait_event_info);
extern int FileSync(File file, uint32 wait_event_info);
+extern int FileZero(File file, off_t offset, off_t amount, uint32 wait_event_info);
+extern int FileFallocate(File file, off_t offset, off_t amount, uint32 wait_event_info);
+
extern off_t FileSize(File file);
extern int FileTruncate(File file, off_t offset, uint32 wait_event_info);
extern void FileWriteback(File file, off_t offset, off_t nbytes, uint32 wait_event_info);
extern void mdunlink(RelFileLocatorBackend rlocator, ForkNumber forknum, bool isRedo);
extern void mdextend(SMgrRelation reln, ForkNumber forknum,
BlockNumber blocknum, const void *buffer, bool skipFsync);
+extern void mdzeroextend(SMgrRelation reln, ForkNumber forknum,
+ BlockNumber blocknum, int nblocks, bool skipFsync);
extern bool mdprefetch(SMgrRelation reln, ForkNumber forknum,
BlockNumber blocknum);
extern void mdread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
extern void smgrdounlinkall(SMgrRelation *rels, int nrels, bool isRedo);
extern void smgrextend(SMgrRelation reln, ForkNumber forknum,
BlockNumber blocknum, const void *buffer, bool skipFsync);
+extern void smgrzeroextend(SMgrRelation reln, ForkNumber forknum,
+ BlockNumber blocknum, int nblocks, bool skipFsync);
extern bool smgrprefetch(SMgrRelation reln, ForkNumber forknum,
BlockNumber blocknum);
extern void smgrread(SMgrRelation reln, ForkNumber forknum,
projects / postgresql.git / commitdiff