pg_dump: Reduce memory usage of dumps with statistics.

yb conflict resolutions:

- src/bin/pg_dump/pg_backup_archiver.c
  >@@ -3849,7 +3890,7 @@ _printTocEntry(ArchiveHandle *AH, TocEntry *te, const char *pfx)
    >-	 * string if any, but we have three special cases:
    >+	 * string if any, but we have four special cases:

  <@@ -3698,12 +3739,50 @@ _printTocEntry(ArchiveHandle *AH, TocEntry *te, const char *pfx)
  >@@ -3862,6 +3903,11 @@ _printTocEntry(ArchiveHandle *AH, TocEntry *te, const char *pfx)
    >+	 * 4. Entries with a defnDumper need to call it to generate the
    <+	 * Entries with a defnDumper need to call it to generate the
    Cause: Missing a45c78e: pg_dump changes to better handle large objects
    Resolution: This improvement does not exist in pg15 so the comment is not relevant.

Right now, pg_dump stores all generated commands for statistics in
memory.  These commands can be quite large and therefore can
significantly increase pg_dump's memory footprint.  To fix, wait
until we are about to write out the commands before generating
them, and be sure to free the commands after writing.  This is
implemented via a new defnDumper callback that works much like the
dataDumper one but is specifically designed for TOC entries.

Custom dumps that include data might write the TOC twice (to update
data offset information), which would ordinarily cause pg_dump to
run the attribute statistics queries twice.  However, as a hack, we
save the length of the written-out entry in the first pass and skip
over it in the second.  While there is no known technical issue
with executing the queries multiple times and rewriting the
results, it's expensive and feels risky, so let's avoid it.

As an exception, we _do_ execute the queries twice for the tar
format.  This format does a second pass through the TOC to generate
the restore.sql file.  pg_restore doesn't use this file, so even if
the second round of queries returns different results than the
first, it won't corrupt the output; the archive and restore.sql
file will just have different content.  A follow-up commit will
teach pg_dump to gather attribute statistics in batches, which our
testing indicates more than makes up for the added expense of
running the queries twice.

Author: Corey Huinker <corey.huinker@gmail.com>
Co-authored-by: Nathan Bossart <nathandbossart@gmail.com>
Reviewed-by: Jeff Davis <pgsql@j-davis.com>
Discussion: https://postgr.es/m/CADkLM%3Dc%2Br05srPy9w%2B-%2BnbmLEo15dKXYQ03Q_xyK%2BriJerigLQ%40mail.gmail.com
(cherry picked from commit 7d5c83b)

Author: nathan-bossart

src/bin/pg_dump/pg_backup.h

@@ -277,6 +277,7 @@ typedef int DumpId;
  * Function pointer prototypes for assorted callback methods.
  */
 
+typedef char *(*DefnDumperPtr) (Archive *AH, const void *userArg);
 typedef int (*DataDumperPtr) (Archive *AH, const void *userArg);
 
 typedef void (*SetupWorkerPtrType) (Archive *AH);

src/bin/pg_dump/pg_backup_archiver.c

@@ -1174,6 +1174,9 @@ ArchiveEntry(Archive *AHX, CatalogId catalogId, DumpId dumpId,
 	newToc->dataDumperArg = opts->dumpArg;
 	newToc->hadDumper = opts->dumpFn ? true : false;
 
+	newToc->defnDumper = opts->defnFn;
+	newToc->defnDumperArg = opts->defnArg;
+
 	newToc->formatData = NULL;
 	newToc->dataLength = 0;
 
@@ -2540,7 +2543,45 @@ WriteToc(ArchiveHandle *AH)
 		WriteStr(AH, te->tag);
 		WriteStr(AH, te->desc);
 		WriteInt(AH, te->section);
-		WriteStr(AH, te->defn);
+
+		if (te->defnLen)
+		{
+			/*
+			 * defnLen should only be set for custom format's second call to
+			 * WriteToc(), which rewrites the TOC in place to update data
+			 * offsets.  Instead of calling the defnDumper a second time
+			 * (which could involve re-executing queries), just skip writing
+			 * the entry.  While regenerating the definition should
+			 * theoretically produce the same result as before, it's expensive
+			 * and feels risky.
+			 *
+			 * The custom format only calls WriteToc() a second time if
+			 * fseeko() is usable (see _CloseArchive() in pg_backup_custom.c),
+			 * so we can safely use it without checking.  For other formats,
+			 * we fail because one of our assumptions must no longer hold
+			 * true.
+			 *
+			 * XXX This is a layering violation, but the alternative is an
+			 * awkward and complicated callback infrastructure for this
+			 * special case.  This might be worth revisiting in the future.
+			 */
+			if (AH->format != archCustom)
+				pg_fatal("unexpected TOC entry in WriteToc(): %d %s %s",
+						 te->dumpId, te->desc, te->tag);
+
+			if (fseeko(AH->FH, te->defnLen, SEEK_CUR != 0))
+				pg_fatal("error during file seek: %m");
+		}
+		else if (te->defnDumper)
+		{
+			char	   *defn = te->defnDumper((Archive *) AH, te->defnDumperArg);
+
+			te->defnLen = WriteStr(AH, defn);
+			pg_free(defn);
+		}
+		else
+			WriteStr(AH, te->defn);
+
 		WriteStr(AH, te->dropStmt);
 		WriteStr(AH, te->copyStmt);
 		WriteStr(AH, te->namespace);
@@ -3697,12 +3738,50 @@ _printTocEntry(ArchiveHandle *AH, TocEntry *te, const char *pfx)
 	 * versions put into CREATE SCHEMA.  Don't mutate the variant for schema
 	 * "public" that is a comment.  We have to do this when --no-owner mode is
 	 * selected.  This is ugly, but I see no other good way ...
+	 *
+	 * Entries with a defnDumper need to call it to generate the
+	 * definition.  This is primarily intended to provide a way to save memory
+	 * for objects that would otherwise need a lot of it (e.g., statistics
+	 * data).
 	 */
 	if (ropt->noOwner &&
 		strcmp(te->desc, "SCHEMA") == 0 && strncmp(te->defn, "--", 2) != 0)
 	{
 		ahprintf(AH, "CREATE SCHEMA %s;\n\n\n", fmtId(te->tag));
 	}
+	else if (te->defnLen && AH->format != archTar)
+	{
+		/*
+		 * If defnLen is set, the defnDumper has already been called for this
+		 * TOC entry.  We don't normally expect a defnDumper to be called for
+		 * a TOC entry a second time in _printTocEntry(), but there's an
+		 * exception.  The tar format first calls WriteToc(), which scans the
+		 * entire TOC, and then it later calls RestoreArchive() to generate
+		 * restore.sql, which scans the TOC again.  There doesn't appear to be
+		 * a good way to prevent a second defnDumper call in this case without
+		 * storing the definition in memory, which defeats the purpose.  This
+		 * second defnDumper invocation should generate the same output as the
+		 * first, but even if it doesn't, the worst-case scenario is that
+		 * restore.sql might have different statistics data than the archive.
+		 *
+		 * In all other cases, encountering a TOC entry a second time in
+		 * _printTocEntry() is unexpected, so we fail because one of our
+		 * assumptions must no longer hold true.
+		 *
+		 * XXX This is a layering violation, but the alternative is an awkward
+		 * and complicated callback infrastructure for this special case. This
+		 * might be worth revisiting in the future.
+		 */
+		pg_fatal("unexpected TOC entry in _printTocEntry(): %d %s %s",
+				 te->dumpId, te->desc, te->tag);
+	}
+	else if (te->defnDumper)
+	{
+		char	   *defn = te->defnDumper((Archive *) AH, te->defnDumperArg);
+
+		te->defnLen = ahprintf(AH, "%s\n\n", defn);
+		pg_free(defn);
+	}
 	else
 	{
 		if (te->defn && strlen(te->defn) > 0)

src/bin/pg_dump/pg_backup_archiver.h

@@ -388,6 +388,10 @@ struct _tocEntry
 	const void *dataDumperArg;	/* Arg for above routine */
 	void	   *formatData;		/* TOC Entry data specific to file format */
 
+	DefnDumperPtr defnDumper;	/* routine to dump definition statement */
+	const void *defnDumperArg;	/* arg for above routine */
+	size_t		defnLen;		/* length of dumped definition */
+
 	/* working state while dumping/restoring */
 	pgoff_t		dataLength;		/* item's data size; 0 if none or unknown */
 	int			reqs;			/* do we need schema and/or data of object
@@ -426,6 +430,8 @@ typedef struct _archiveOpts
 	int			nDeps;
 	DataDumperPtr dumpFn;
 	const void *dumpArg;
+	DefnDumperPtr defnFn;
+	const void *defnArg;
 } ArchiveOpts;
 #define ARCHIVE_OPTS(...) &(ArchiveOpts){__VA_ARGS__}
 /* Called to add a TOC entry */

src/bin/pg_dump/pg_dump.c

@@ -9979,17 +9979,21 @@ appendNamedArgument(PQExpBuffer out, Archive *fout, const char *argname,
 }
 
 /*
- * dumpRelationStats --
+ * dumpRelationStats_dumper --
  *
- * Dump command to import stats into the relation on the new database.
+ * Generate command to import stats into the relation on the new database.
+ * This routine is called by the Archiver when it wants the statistics to be
+ * dumped.
  */
-static void
-dumpRelationStats(Archive *fout, const RelStatsInfo *rsinfo)
+static char *
+dumpRelationStats_dumper(Archive *fout, const void *userArg)
 {
+	const RelStatsInfo *rsinfo = (RelStatsInfo *) userArg;
 	const DumpableObject *dobj = &rsinfo->dobj;
 	PGresult   *res;
 	PQExpBuffer query;
-	PQExpBuffer out;
+	PQExpBufferData out_data;
+	PQExpBuffer out = &out_data;
 	int			i_attname;
 	int			i_inherited;
 	int			i_null_frac;
@@ -10006,10 +10010,6 @@ dumpRelationStats(Archive *fout, const RelStatsInfo *rsinfo)
 	int			i_range_empty_frac;
 	int			i_range_bounds_histogram;
 
-	/* nothing to do if we are not dumping statistics */
-	if (!fout->dopt->dumpStatistics)
-		return;
-
 	query = createPQExpBuffer();
 	if (!fout->is_prepared[PREPQUERY_GETATTRIBUTESTATS])
 	{
@@ -10045,7 +10045,7 @@ dumpRelationStats(Archive *fout, const RelStatsInfo *rsinfo)
 		resetPQExpBuffer(query);
 	}
 
-	out = createPQExpBuffer();
+	initPQExpBuffer(out);
 
 	/* restore relation stats */
 	appendPQExpBufferStr(out, "SELECT * FROM pg_catalog.pg_restore_relation_stats(\n");
@@ -10183,17 +10183,35 @@ dumpRelationStats(Archive *fout, const RelStatsInfo *rsinfo)
 
 	PQclear(res);
 
+	destroyPQExpBuffer(query);
+	return out->data;
+}
+
+/*
+ * dumpRelationStats --
+ *
+ * Make an ArchiveEntry for the relation statistics.  The Archiver will take
+ * care of gathering the statistics and generating the restore commands when
+ * they are needed.
+ */
+static void
+dumpRelationStats(Archive *fout, const RelStatsInfo *rsinfo)
+{
+	const DumpableObject *dobj = &rsinfo->dobj;
+
+	/* nothing to do if we are not dumping statistics */
+	if (!fout->dopt->dumpStatistics)
+		return;
+
 	ArchiveEntry(fout, nilCatalogId, createDumpId(),
 				 ARCHIVE_OPTS(.tag = dobj->name,
 							  .namespace = dobj->namespace->dobj.name,
 							  .description = "STATISTICS DATA",
 							  .section = rsinfo->section,
-							  .createStmt = out->data,
+							  .defnFn = dumpRelationStats_dumper,
+							  .defnArg = rsinfo,
 							  .deps = dobj->dependencies,
 							  .nDeps = dobj->nDeps));
-
-	destroyPQExpBuffer(out);
-	destroyPQExpBuffer(query);
 }
 
 /*