v8-0001-LZ4-compression.patch
application/octet-stream
Filename: v8-0001-LZ4-compression.patch
Type: application/octet-stream
Part: 0
Message:
Re: refactoring basebackup.c
Patch
Same data as JSON:
GET /api/v1/attachments/:id/patch
the parsed metadata as JSON — format, series position, per-file stats; never the diff bytes.
API reference →
Format: format-patch
Series: patch v8-0001
Subject: LZ4 compression
| File | + | − |
|---|---|---|
| doc/src/sgml/ref/pg_basebackup.sgml | 41 | 8 |
| src/backend/replication/basebackup.c | 6 | 1 |
| src/backend/replication/basebackup_lz4.c | 285 | 0 |
| src/backend/replication/Makefile | 1 | 0 |
| src/include/replication/basebackup_sink.h | 1 | 0 |
From ac7f611bd62f81a408cf652f4a0af906a515b3cb Mon Sep 17 00:00:00 2001
From: Jeevan Ladhe <jeevan.ladhe@enterprisedb.com>
Date: Wed, 17 Nov 2021 19:43:20 +0530
Subject: [PATCH] LZ4 compression
---
doc/src/sgml/ref/pg_basebackup.sgml | 49 +++-
src/backend/replication/Makefile | 1 +
src/backend/replication/basebackup.c | 7 +-
src/backend/replication/basebackup_lz4.c | 285 ++++++++++++++++++++++
src/include/replication/basebackup_sink.h | 1 +
5 files changed, 334 insertions(+), 9 deletions(-)
create mode 100644 src/backend/replication/basebackup_lz4.c
diff --git a/doc/src/sgml/ref/pg_basebackup.sgml b/doc/src/sgml/ref/pg_basebackup.sgml
index 9ce8b8d89d..44395a749b 100644
--- a/doc/src/sgml/ref/pg_basebackup.sgml
+++ b/doc/src/sgml/ref/pg_basebackup.sgml
@@ -230,14 +230,7 @@ PostgreSQL documentation
<para>
Allows the tar files generated for each tablespace to be compressed
- on the server, before they are sent to the client. The default value
- is <literal>none</literal>, which performs no compression. If set
- to <literal>gzip</literal>, compression is performed using gzip and
- the suffix <filename>.gz</filename> will automatically be added to
- compressed files. A numeric digit between 1 and 9 can be added to
- specify the compression level; for instance, <literal>gzip9</literal>
- will provide the maximum compression that the <literal>gzip</literal>
- algorithm can provide.
+ on the server, before they are sent to the client.
</para>
<para>
Since the write-ahead logs are fetched via a separate client
@@ -245,7 +238,47 @@ PostgreSQL documentation
the <literal>--gzip</literal> and <literal>--compress</literal>
options.
</para>
+ <para>
+ The following <replaceable>target</replaceable> algorithms for
+ server-compression are supported:
+
+ <variablelist>
+ <varlistentry>
+ <term><literal>none</literal></term>
+ <listitem>
+ <para>
+ Perform no compression. This is the default value.
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>gzip</literal></term>
+ <listitem>
+ <para>
+ Compression is performed using <literal>gzip</literal> and the
+ suffix <filename>.gz </filename> will automatically be added to
+ compressed files. A numeric digit between 1 and 9 can be added to
+ specify the compression level; for instance, <literal>gzip9
+ </literal> will provide the maximum compression that the
+ <literal>gzip</literal> algorithm can provide.
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>lz4</literal></term>
+ <listitem>
+ <para>
+ Compression is performed using <literal>lz4</literal> and the
+ suffix <filename>.lz4</filename> will automatically be added to
+ compressed files.
+ </para>
+ </listitem>
+ </varlistentry>
+ </variablelist>
+ </para>
</listitem>
</varlistentry>
diff --git a/src/backend/replication/Makefile b/src/backend/replication/Makefile
index 8ec60ded76..74043ff331 100644
--- a/src/backend/replication/Makefile
+++ b/src/backend/replication/Makefile
@@ -19,6 +19,7 @@ OBJS = \
basebackup.o \
basebackup_copy.o \
basebackup_gzip.o \
+ basebackup_lz4.o \
basebackup_progress.o \
basebackup_server.o \
basebackup_sink.o \
diff --git a/src/backend/replication/basebackup.c b/src/backend/replication/basebackup.c
index ff26537679..aab6744f73 100644
--- a/src/backend/replication/basebackup.c
+++ b/src/backend/replication/basebackup.c
@@ -64,7 +64,8 @@ typedef enum
typedef enum
{
BACKUP_COMPRESSION_NONE,
- BACKUP_COMPRESSION_GZIP
+ BACKUP_COMPRESSION_GZIP,
+ BACKUP_COMPRESSION_LZ4
} basebackup_compression_type;
typedef struct
@@ -909,6 +910,8 @@ parse_basebackup_options(List *options, basebackup_options *opt)
opt->compression = BACKUP_COMPRESSION_GZIP;
opt->compression_level = optval[4] - '0';
}
+ else if (strcmp(optval, "lz4") == 0)
+ opt->compression = BACKUP_COMPRESSION_LZ4;
else
ereport(ERROR,
(errcode(ERRCODE_SYNTAX_ERROR),
@@ -1013,6 +1016,8 @@ SendBaseBackup(BaseBackupCmd *cmd)
/* Set up server-side compression, if client requested it */
if (opt.compression == BACKUP_COMPRESSION_GZIP)
sink = bbsink_gzip_new(sink, opt.compression_level);
+ else if (opt.compression == BACKUP_COMPRESSION_LZ4)
+ sink = bbsink_lz4_new(sink);
/* Set up progress reporting. */
sink = bbsink_progress_new(sink, opt.progress);
diff --git a/src/backend/replication/basebackup_lz4.c b/src/backend/replication/basebackup_lz4.c
new file mode 100644
index 0000000000..0f49def813
--- /dev/null
+++ b/src/backend/replication/basebackup_lz4.c
@@ -0,0 +1,285 @@
+/*-------------------------------------------------------------------------
+ *
+ * basebackup_lz4.c
+ * Basebackup sink implementing lz4 compression.
+ *
+ * Portions Copyright (c) 2010-2020, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ * src/backend/replication/basebackup_lz4.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#ifdef HAVE_LIBLZ4
+#include <lz4frame.h>
+#endif
+#include <unistd.h>
+
+#include "replication/basebackup_sink.h"
+
+#ifdef HAVE_LIBLZ4
+
+typedef struct bbsink_lz4
+{
+ /* Common information for all types of sink. */
+ bbsink base;
+
+ LZ4F_compressionContext_t ctx;
+ LZ4F_preferences_t prefs;
+
+ /* Number of bytes staged in output buffer. */
+ size_t bytes_written;
+} bbsink_lz4;
+
+static void bbsink_lz4_begin_backup(bbsink *sink);
+static void bbsink_lz4_begin_archive(bbsink *sink, const char *archive_name);
+static void bbsink_lz4_archive_contents(bbsink *sink, size_t avail_in);
+static void bbsink_lz4_manifest_contents(bbsink *sink, size_t len);
+static void bbsink_lz4_end_archive(bbsink *sink);
+static void bbsink_lz4_cleanup(bbsink *sink);
+
+const bbsink_ops bbsink_lz4_ops = {
+ .begin_backup = bbsink_lz4_begin_backup,
+ .begin_archive = bbsink_lz4_begin_archive,
+ .archive_contents = bbsink_lz4_archive_contents,
+ .end_archive = bbsink_lz4_end_archive,
+ .begin_manifest = bbsink_forward_begin_manifest,
+ .manifest_contents = bbsink_lz4_manifest_contents,
+ .end_manifest = bbsink_forward_end_manifest,
+ .end_backup = bbsink_forward_end_backup,
+ .cleanup = bbsink_lz4_cleanup
+};
+#endif
+
+/* Create a new basebackup sink that performs lz4 compression. */
+bbsink *
+bbsink_lz4_new(bbsink *next)
+{
+#ifndef HAVE_LIBLZ4
+ ereport(ERROR,
+ (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+ errmsg("lz4 compression is not supported by this build")));
+#else
+ bbsink_lz4 *sink;
+
+ Assert(next != NULL);
+
+ sink = palloc0(sizeof(bbsink_lz4));
+ *((const bbsink_ops **) &sink->base.bbs_ops) = &bbsink_lz4_ops;
+ sink->base.bbs_next = next;
+
+ return &sink->base;
+#endif
+}
+
+#ifdef HAVE_LIBLZ4
+
+/*
+ * Begin backup.
+ */
+static void
+bbsink_lz4_begin_backup(bbsink *sink)
+{
+ bbsink_lz4 *mysink = (bbsink_lz4 *) sink;
+ size_t output_buffer_bound;
+ LZ4F_preferences_t *prefs = &mysink->prefs;
+
+ /* Initialize compressor object. */
+ memset(prefs, 0, sizeof(LZ4F_preferences_t));
+ prefs->frameInfo.blockSizeID = LZ4F_max256KB;
+
+ /*
+ * 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);
+
+ /*
+ * Since LZ4F_compressUpdate() requires the output buffer of size equal or
+ * greater than that of LZ4F_compressBound(), make sure we have the next
+ * sink's bbs_buffer of length that can accommodate the compressed input
+ * buffer.
+ */
+ output_buffer_bound = LZ4F_compressBound(mysink->base.bbs_buffer_length,
+ &mysink->prefs);
+
+ /*
+ * 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_lz4_begin_archive(bbsink *sink, const char *archive_name)
+{
+ bbsink_lz4 *mysink = (bbsink_lz4 *) sink;
+ char *lz4_archive_name;
+ LZ4F_errorCode_t ctxError;
+ size_t headerSize;
+
+ ctxError = LZ4F_createCompressionContext(&mysink->ctx, LZ4F_VERSION);
+ if (LZ4F_isError(ctxError))
+ elog(ERROR, "could not create lz4 compression context: %s",
+ LZ4F_getErrorName(ctxError));
+
+ /* First of all write the frame header to destination buffer. */
+ headerSize = LZ4F_compressBegin(mysink->ctx,
+ mysink->base.bbs_next->bbs_buffer,
+ mysink->base.bbs_next->bbs_buffer_length,
+ &mysink->prefs);
+
+ if (LZ4F_isError(headerSize))
+ elog(ERROR, "could not write lz4 header: %s",
+ LZ4F_getErrorName(headerSize));
+
+ /*
+ * We need to write the compressed data after the header in the output
+ * buffer. So, make sure to update the notion of bytes written to output
+ * buffer.
+ */
+ mysink->bytes_written += headerSize;
+
+ /* Add ".lz4" to the archive name. */
+ lz4_archive_name = psprintf("%s.lz4", archive_name);
+ Assert(sink->bbs_next != NULL);
+ bbsink_begin_archive(sink->bbs_next, lz4_archive_name);
+ pfree(lz4_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 then 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_lz4_end_archive() is invoked.
+ */
+static void
+bbsink_lz4_archive_contents(bbsink *sink, size_t avail_in)
+{
+ bbsink_lz4 *mysink = (bbsink_lz4 *) sink;
+ size_t compressedSize;
+ size_t avail_in_bound;
+
+ avail_in_bound = LZ4F_compressBound(avail_in, &mysink->prefs);
+
+ /*
+ * If the number of available bytes has fallen below the value computed by
+ * LZ4F_compressBound(), ask the next sink to process the data so that we
+ * can empty the buffer.
+ */
+ if ((mysink->base.bbs_next->bbs_buffer_length - mysink->bytes_written) <=
+ avail_in_bound)
+ {
+ bbsink_archive_contents(sink->bbs_next, mysink->bytes_written);
+ mysink->bytes_written = 0;
+ }
+
+ /*
+ * Compress the input buffer and write it into the output buffer.
+ */
+ compressedSize = LZ4F_compressUpdate(mysink->ctx,
+ mysink->base.bbs_next->bbs_buffer + mysink->bytes_written,
+ mysink->base.bbs_next->bbs_buffer_length - mysink->bytes_written,
+ (uint8 *) mysink->base.bbs_buffer,
+ avail_in,
+ NULL);
+
+ if (LZ4F_isError(compressedSize))
+ elog(ERROR, "could not compress data: %s",
+ LZ4F_getErrorName(compressedSize));
+
+ /*
+ * Update our notion of how many bytes we've written into output buffer.
+ */
+ mysink->bytes_written += compressedSize;
+}
+
+/*
+ * There might be some data inside lz4's internal buffers; we need to get
+ * that flushed out and also finalize the lz4 frame and then get that forwarded
+ * to the successor sink as archive content.
+ *
+ * Then we can end processing for this archive.
+ */
+static void
+bbsink_lz4_end_archive(bbsink *sink)
+{
+ bbsink_lz4 *mysink = (bbsink_lz4 *) sink;
+ size_t compressedSize;
+ size_t lz4_footer_bound;
+
+ lz4_footer_bound = LZ4F_compressBound(0, &mysink->prefs);
+
+ Assert(mysink->base.bbs_next->bbs_buffer_length >= lz4_footer_bound);
+
+ if ((mysink->base.bbs_next->bbs_buffer_length - mysink->bytes_written) <=
+ lz4_footer_bound)
+ {
+ bbsink_archive_contents(sink->bbs_next, mysink->bytes_written);
+ mysink->bytes_written = 0;
+ }
+
+ compressedSize = LZ4F_compressEnd(mysink->ctx,
+ mysink->base.bbs_next->bbs_buffer + mysink->bytes_written,
+ mysink->base.bbs_next->bbs_buffer_length - mysink->bytes_written,
+ NULL);
+
+ if (LZ4F_isError(compressedSize))
+ elog(ERROR, "could not end lz4 compression: %s",
+ LZ4F_getErrorName(compressedSize));
+
+ /* Update our notion of how many bytes we've written. */
+ mysink->bytes_written += compressedSize;
+
+ /* Send whatever accumulated output bytes we have. */
+ bbsink_archive_contents(sink->bbs_next, mysink->bytes_written);
+ mysink->bytes_written = 0;
+
+ /* Release the resources. */
+ LZ4F_freeCompressionContext(mysink->ctx);
+ mysink->ctx = NULL;
+
+ /* Pass on the information that this archive has ended. */
+ bbsink_forward_end_archive(sink);
+}
+
+/*
+ * 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_lz4_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 the compression context by
+ * calling LZ4F_freeCompressionContext() if needed to avoid memory leak.
+ */
+static void
+bbsink_lz4_cleanup(bbsink *sink)
+{
+ bbsink_lz4 *mysink = (bbsink_lz4 *) sink;
+
+ if (mysink->ctx)
+ {
+ LZ4F_freeCompressionContext(mysink->ctx);
+ mysink->ctx = NULL;
+ }
+}
+
+#endif
diff --git a/src/include/replication/basebackup_sink.h b/src/include/replication/basebackup_sink.h
index 6bfea35c22..2558ce5ca2 100644
--- a/src/include/replication/basebackup_sink.h
+++ b/src/include/replication/basebackup_sink.h
@@ -285,6 +285,7 @@ extern void bbsink_forward_cleanup(bbsink *sink);
extern bbsink *bbsink_copystream_new(bool send_to_client);
extern bbsink *bbsink_copytblspc_new(void);
extern bbsink *bbsink_gzip_new(bbsink *next, int compresslevel);
+extern bbsink *bbsink_lz4_new(bbsink *next);
extern bbsink *bbsink_progress_new(bbsink *next, bool estimate_backup_size);
extern bbsink *bbsink_server_new(bbsink *next, char *pathname);
extern bbsink *bbsink_throttle_new(bbsink *next, uint32 maxrate);
--
2.25.1