postgresql/src/backend/replication/basebackup_zstd.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

302 lines
8.2 KiB
C
Raw Normal View History

/*-------------------------------------------------------------------------
*
* basebackup_zstd.c
* Basebackup sink implementing zstd compression.
*
* Portions Copyright (c) 2010-2020, PostgreSQL Global Development Group
*
* IDENTIFICATION
* src/backend/replication/basebackup_zstd.c
*
*-------------------------------------------------------------------------
*/
#include "postgres.h"
#ifdef USE_ZSTD
#include <zstd.h>
#endif
#include "replication/basebackup_sink.h"
#ifdef USE_ZSTD
typedef struct bbsink_zstd
{
/* Common information for all types of sink. */
bbsink base;
/* Compression level */
int compresslevel;
ZSTD_CCtx *cctx;
ZSTD_outBuffer zstd_outBuf;
} bbsink_zstd;
static void bbsink_zstd_begin_backup(bbsink *sink);
static void bbsink_zstd_begin_archive(bbsink *sink, const char *archive_name);
static void bbsink_zstd_archive_contents(bbsink *sink, size_t avail_in);
static void bbsink_zstd_manifest_contents(bbsink *sink, size_t len);
static void bbsink_zstd_end_archive(bbsink *sink);
static void bbsink_zstd_cleanup(bbsink *sink);
static void bbsink_zstd_end_backup(bbsink *sink, XLogRecPtr endptr,
TimeLineID endtli);
const bbsink_ops bbsink_zstd_ops = {
.begin_backup = bbsink_zstd_begin_backup,
.begin_archive = bbsink_zstd_begin_archive,
.archive_contents = bbsink_zstd_archive_contents,
.end_archive = bbsink_zstd_end_archive,
.begin_manifest = bbsink_forward_begin_manifest,
.manifest_contents = bbsink_zstd_manifest_contents,
.end_manifest = bbsink_forward_end_manifest,
.end_backup = bbsink_zstd_end_backup,
.cleanup = bbsink_zstd_cleanup
};
#endif
/*
Replace BASE_BACKUP COMPRESSION_LEVEL option with COMPRESSION_DETAIL. There are more compression parameters that can be specified than just an integer compression level, so rename the new COMPRESSION_LEVEL option to COMPRESSION_DETAIL before it gets released. Introduce a flexible syntax for that option to allow arbitrary options to be specified without needing to adjust the main replication grammar, and common code to parse it that is shared between the client and the server. This commit doesn't actually add any new compression parameters, so the only user-visible change is that you can now type something like pg_basebackup --compress gzip:level=5 instead of writing just pg_basebackup --compress gzip:5. However, it should make it easy to add new options. If for example gzip starts offering fries, we can support pg_basebackup --compress gzip:level=5,fries=true for the benefit of users who want fries with that. Along the way, this fixes a few things in pg_basebackup so that the pg_basebackup can be used with a server-side compression algorithm that pg_basebackup itself does not understand. For example, pg_basebackup --compress server-lz4 could still succeed even if only the server and not the client has LZ4 support, provided that the other options to pg_basebackup don't require the client to decompress the archive. Patch by me. Reviewed by Justin Pryzby and Dagfinn Ilmari Mannsåker. Discussion: http://postgr.es/m/CA+TgmoYvpetyRAbbg1M8b3-iHsaN4nsgmWPjOENu5-doHuJ7fA@mail.gmail.com
2022-03-23 14:19:14 +01:00
* Create a new basebackup sink that performs zstd compression.
*/
bbsink *
Replace BASE_BACKUP COMPRESSION_LEVEL option with COMPRESSION_DETAIL. There are more compression parameters that can be specified than just an integer compression level, so rename the new COMPRESSION_LEVEL option to COMPRESSION_DETAIL before it gets released. Introduce a flexible syntax for that option to allow arbitrary options to be specified without needing to adjust the main replication grammar, and common code to parse it that is shared between the client and the server. This commit doesn't actually add any new compression parameters, so the only user-visible change is that you can now type something like pg_basebackup --compress gzip:level=5 instead of writing just pg_basebackup --compress gzip:5. However, it should make it easy to add new options. If for example gzip starts offering fries, we can support pg_basebackup --compress gzip:level=5,fries=true for the benefit of users who want fries with that. Along the way, this fixes a few things in pg_basebackup so that the pg_basebackup can be used with a server-side compression algorithm that pg_basebackup itself does not understand. For example, pg_basebackup --compress server-lz4 could still succeed even if only the server and not the client has LZ4 support, provided that the other options to pg_basebackup don't require the client to decompress the archive. Patch by me. Reviewed by Justin Pryzby and Dagfinn Ilmari Mannsåker. Discussion: http://postgr.es/m/CA+TgmoYvpetyRAbbg1M8b3-iHsaN4nsgmWPjOENu5-doHuJ7fA@mail.gmail.com
2022-03-23 14:19:14 +01:00
bbsink_zstd_new(bbsink *next, bc_specification *compress)
{
#ifndef USE_ZSTD
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("zstd compression is not supported by this build")));
return NULL; /* keep compiler quiet */
#else
bbsink_zstd *sink;
int compresslevel;
Assert(next != NULL);
Replace BASE_BACKUP COMPRESSION_LEVEL option with COMPRESSION_DETAIL. There are more compression parameters that can be specified than just an integer compression level, so rename the new COMPRESSION_LEVEL option to COMPRESSION_DETAIL before it gets released. Introduce a flexible syntax for that option to allow arbitrary options to be specified without needing to adjust the main replication grammar, and common code to parse it that is shared between the client and the server. This commit doesn't actually add any new compression parameters, so the only user-visible change is that you can now type something like pg_basebackup --compress gzip:level=5 instead of writing just pg_basebackup --compress gzip:5. However, it should make it easy to add new options. If for example gzip starts offering fries, we can support pg_basebackup --compress gzip:level=5,fries=true for the benefit of users who want fries with that. Along the way, this fixes a few things in pg_basebackup so that the pg_basebackup can be used with a server-side compression algorithm that pg_basebackup itself does not understand. For example, pg_basebackup --compress server-lz4 could still succeed even if only the server and not the client has LZ4 support, provided that the other options to pg_basebackup don't require the client to decompress the archive. Patch by me. Reviewed by Justin Pryzby and Dagfinn Ilmari Mannsåker. Discussion: http://postgr.es/m/CA+TgmoYvpetyRAbbg1M8b3-iHsaN4nsgmWPjOENu5-doHuJ7fA@mail.gmail.com
2022-03-23 14:19:14 +01:00
if ((compress->options & BACKUP_COMPRESSION_OPTION_LEVEL) == 0)
compresslevel = 0;
else
{
compresslevel = compress->level;
Assert(compresslevel >= 1 && compresslevel <= 22);
}
sink = palloc0(sizeof(bbsink_zstd));
*((const bbsink_ops **) &sink->base.bbs_ops) = &bbsink_zstd_ops;
sink->base.bbs_next = next;
sink->compresslevel = compresslevel;
return &sink->base;
#endif
}
#ifdef USE_ZSTD
/*
* Begin backup.
*/
static void
bbsink_zstd_begin_backup(bbsink *sink)
{
bbsink_zstd *mysink = (bbsink_zstd *) sink;
size_t output_buffer_bound;
mysink->cctx = ZSTD_createCCtx();
if (!mysink->cctx)
elog(ERROR, "could not create zstd compression context");
ZSTD_CCtx_setParameter(mysink->cctx, ZSTD_c_compressionLevel,
mysink->compresslevel);
/*
* We need our own buffer, because we're going to pass different data to
* the next sink than what gets passed to us.
*/
mysink->base.bbs_buffer = palloc(mysink->base.bbs_buffer_length);
/*
* Make sure that the next sink's bbs_buffer is big enough to accommodate
* the compressed input buffer.
*/
output_buffer_bound = ZSTD_compressBound(mysink->base.bbs_buffer_length);
/*
* The buffer length is expected to be a multiple of BLCKSZ, so round up.
*/
output_buffer_bound = output_buffer_bound + BLCKSZ -
(output_buffer_bound % BLCKSZ);
bbsink_begin_backup(sink->bbs_next, sink->bbs_state, output_buffer_bound);
}
/*
* Prepare to compress the next archive.
*/
static void
bbsink_zstd_begin_archive(bbsink *sink, const char *archive_name)
{
bbsink_zstd *mysink = (bbsink_zstd *) sink;
char *zstd_archive_name;
/*
* At the start of each archive we reset the state to start a new
* compression operation. The parameters are sticky and they will stick
* around as we are resetting with option ZSTD_reset_session_only.
*/
ZSTD_CCtx_reset(mysink->cctx, ZSTD_reset_session_only);
mysink->zstd_outBuf.dst = mysink->base.bbs_next->bbs_buffer;
mysink->zstd_outBuf.size = mysink->base.bbs_next->bbs_buffer_length;
mysink->zstd_outBuf.pos = 0;
/* Add ".zst" to the archive name. */
zstd_archive_name = psprintf("%s.zst", archive_name);
Assert(sink->bbs_next != NULL);
bbsink_begin_archive(sink->bbs_next, zstd_archive_name);
pfree(zstd_archive_name);
}
/*
* Compress the input data to the output buffer until we run out of input
* data. Each time the output buffer falls below the compression bound for
* the input buffer, invoke the archive_contents() method for the next sink.
*
* Note that since we're compressing the input, it may very commonly happen
* that we consume all the input data without filling the output buffer. In
* that case, the compressed representation of the current input data won't
* actually be sent to the next bbsink until a later call to this function,
* or perhaps even not until bbsink_zstd_end_archive() is invoked.
*/
static void
bbsink_zstd_archive_contents(bbsink *sink, size_t len)
{
bbsink_zstd *mysink = (bbsink_zstd *) sink;
ZSTD_inBuffer inBuf = {mysink->base.bbs_buffer, len, 0};
while (inBuf.pos < inBuf.size)
{
size_t yet_to_flush;
size_t max_needed = ZSTD_compressBound(inBuf.size - inBuf.pos);
/*
* If the out buffer is not left with enough space, send the output
* buffer to the next sink, and reset it.
*/
if (mysink->zstd_outBuf.size - mysink->zstd_outBuf.pos < max_needed)
{
bbsink_archive_contents(mysink->base.bbs_next,
mysink->zstd_outBuf.pos);
mysink->zstd_outBuf.dst = mysink->base.bbs_next->bbs_buffer;
mysink->zstd_outBuf.size =
mysink->base.bbs_next->bbs_buffer_length;
mysink->zstd_outBuf.pos = 0;
}
yet_to_flush = ZSTD_compressStream2(mysink->cctx, &mysink->zstd_outBuf,
&inBuf, ZSTD_e_continue);
if (ZSTD_isError(yet_to_flush))
elog(ERROR,
"could not compress data: %s",
ZSTD_getErrorName(yet_to_flush));
}
}
/*
* There might be some data inside zstd's internal buffers; we need to get that
* flushed out, also end the zstd frame and then get that forwarded to the
* successor sink as archive content.
*
* Then we can end processing for this archive.
*/
static void
bbsink_zstd_end_archive(bbsink *sink)
{
bbsink_zstd *mysink = (bbsink_zstd *) sink;
size_t yet_to_flush;
do
{
ZSTD_inBuffer in = {NULL, 0, 0};
size_t max_needed = ZSTD_compressBound(0);
/*
* If the out buffer is not left with enough space, send the output
* buffer to the next sink, and reset it.
*/
if (mysink->zstd_outBuf.size - mysink->zstd_outBuf.pos < max_needed)
{
bbsink_archive_contents(mysink->base.bbs_next,
mysink->zstd_outBuf.pos);
mysink->zstd_outBuf.dst = mysink->base.bbs_next->bbs_buffer;
mysink->zstd_outBuf.size =
mysink->base.bbs_next->bbs_buffer_length;
mysink->zstd_outBuf.pos = 0;
}
yet_to_flush = ZSTD_compressStream2(mysink->cctx,
&mysink->zstd_outBuf,
&in, ZSTD_e_end);
if (ZSTD_isError(yet_to_flush))
elog(ERROR, "could not compress data: %s",
ZSTD_getErrorName(yet_to_flush));
} while (yet_to_flush > 0);
/* Make sure to pass any remaining bytes to the next sink. */
if (mysink->zstd_outBuf.pos > 0)
bbsink_archive_contents(mysink->base.bbs_next,
mysink->zstd_outBuf.pos);
/* Pass on the information that this archive has ended. */
bbsink_forward_end_archive(sink);
}
/*
* Free the resources and context.
*/
static void
bbsink_zstd_end_backup(bbsink *sink, XLogRecPtr endptr,
TimeLineID endtli)
{
bbsink_zstd *mysink = (bbsink_zstd *) sink;
/* Release the context. */
if (mysink->cctx)
{
ZSTD_freeCCtx(mysink->cctx);
mysink->cctx = NULL;
}
bbsink_forward_end_backup(sink, endptr, endtli);
}
/*
* Manifest contents are not compressed, but we do need to copy them into
* the successor sink's buffer, because we have our own.
*/
static void
bbsink_zstd_manifest_contents(bbsink *sink, size_t len)
{
memcpy(sink->bbs_next->bbs_buffer, sink->bbs_buffer, len);
bbsink_manifest_contents(sink->bbs_next, len);
}
/*
* In case the backup fails, make sure we free any compression context that
* got allocated, so that we don't leak memory.
*/
static void
bbsink_zstd_cleanup(bbsink *sink)
{
bbsink_zstd *mysink = (bbsink_zstd *) sink;
/* Release the context if not already released. */
if (mysink->cctx)
{
ZSTD_freeCCtx(mysink->cctx);
mysink->cctx = NULL;
}
}
#endif