v1-0001-Create-restartpoint-when-replaying-drop-database.patch

From bd98ac1a2e98bc67238103b0f2764cf7fe0edc58 Mon Sep 17 00:00:00 2001
From: Asim R P <apraveen@pivotal.io>
Date: Thu, 12 Sep 2019 17:17:29 +0530
Subject: [PATCH v1 1/2] Create restartpoint when replaying drop database

Drop database replay involves removing the database directory.  We do
not have a mechanism similar to invalid page detection for directories
during WAL replay.  If, due to a crash, WAL replay must resume from a
checkpoint, we should avoid replaying WAL records second time, that
precede the drop database.

Proposed by Paul Guo.
---
 src/backend/access/rmgrdesc/dbasedesc.c |  6 ++-
 src/backend/commands/dbcommands.c       | 84 +++++++++++++++++++++------------
 src/include/access/xlog.h               |  1 -
 src/include/commands/dbcommands_xlog.h  |  3 +-
 4 files changed, 61 insertions(+), 33 deletions(-)

diff --git a/src/backend/access/rmgrdesc/dbasedesc.c b/src/backend/access/rmgrdesc/dbasedesc.c
index c7d60ce10d..b97bac2411 100644
--- a/src/backend/access/rmgrdesc/dbasedesc.c
+++ b/src/backend/access/rmgrdesc/dbasedesc.c
@@ -36,8 +36,10 @@ dbase_desc(StringInfo buf, XLogReaderState *record)
 	{
 		xl_dbase_drop_rec *xlrec = (xl_dbase_drop_rec *) rec;
 
-		appendStringInfo(buf, "dir %u/%u",
-						 xlrec->tablespace_id, xlrec->db_id);
+		int i;
+		for (i = 0; i < xlrec->nspcids; i++)
+			appendStringInfo(buf, "\ndir %u/%u",
+							 xlrec->db_id, xlrec->tablespace_ids[i]);
 	}
 }
 
diff --git a/src/backend/commands/dbcommands.c b/src/backend/commands/dbcommands.c
index 95881a8550..c0c8726698 100644
--- a/src/backend/commands/dbcommands.c
+++ b/src/backend/commands/dbcommands.c
@@ -1400,16 +1400,19 @@ movedb(const char *dbname, const char *tblspcname)
 	 * Record the filesystem change in XLOG
 	 */
 	{
-		xl_dbase_drop_rec xlrec;
+		size_t xlrec_size = sizeof(xl_dbase_drop_rec) + sizeof(Oid);
+		xl_dbase_drop_rec *xlrec = palloc(xlrec_size);
 
-		xlrec.db_id = db_id;
-		xlrec.tablespace_id = src_tblspcoid;
+		xlrec->db_id = db_id;
+		xlrec->nspcids = 1;
+		xlrec->tablespace_ids[0] = src_tblspcoid;
 
 		XLogBeginInsert();
-		XLogRegisterData((char *) &xlrec, sizeof(xl_dbase_drop_rec));
+		XLogRegisterData((char *) &xlrec, xlrec_size);
 
 		(void) XLogInsert(RM_DBASE_ID,
 						  XLOG_DBASE_DROP | XLR_SPECIAL_REL_UPDATE);
+		pfree(xlrec);
 	}
 
 	/* Now it's safe to release the database lock */
@@ -1914,6 +1917,8 @@ remove_dbtablespaces(Oid db_id)
 	Relation	rel;
 	TableScanDesc scan;
 	HeapTuple	tuple;
+	List *dstpaths = NIL;
+	List *spcoids = NIL;
 
 	rel = table_open(TableSpaceRelationId, AccessShareLock);
 	scan = table_beginscan_catalog(rel, 0, NULL);
@@ -1936,31 +1941,39 @@ remove_dbtablespaces(Oid db_id)
 			pfree(dstpath);
 			continue;
 		}
+		dstpaths = lappend(dstpaths, dstpath);
+		spcoids = lappend_oid(spcoids, dsttablespace);
+	}
+
+	table_endscan(scan);
+	table_close(rel, AccessShareLock);
+
+	size_t xlrec_size = (sizeof(xl_dbase_drop_rec) +
+						 sizeof(Oid)*list_length(spcoids));
+	xl_dbase_drop_rec *xlrec = palloc(xlrec_size);
+	xlrec->db_id = db_id;
+	xlrec->nspcids = list_length(spcoids);
 
-		if (!rmtree(dstpath, true))
+	int i=0;
+	const ListCell *cell1, *cell2;
+	forboth(cell1, dstpaths, cell2, spcoids)
+	{
+		char *path = lfirst(cell1);
+		if (!rmtree(path, true))
 			ereport(WARNING,
 					(errmsg("some useless files may be left behind in old database directory \"%s\"",
-							dstpath)));
-
+							path)));
+		pfree(path);
 		/* Record the filesystem change in XLOG */
-		{
-			xl_dbase_drop_rec xlrec;
-
-			xlrec.db_id = db_id;
-			xlrec.tablespace_id = dsttablespace;
-
-			XLogBeginInsert();
-			XLogRegisterData((char *) &xlrec, sizeof(xl_dbase_drop_rec));
-
-			(void) XLogInsert(RM_DBASE_ID,
-							  XLOG_DBASE_DROP | XLR_SPECIAL_REL_UPDATE);
-		}
-
-		pfree(dstpath);
+		xlrec->tablespace_ids[i++] = lfirst_oid(cell2);
 	}
 
-	table_endscan(scan);
-	table_close(rel, AccessShareLock);
+	XLogBeginInsert();
+	XLogRegisterData((char *) &xlrec, xlrec_size);
+	(void) XLogInsert(RM_DBASE_ID,
+					  XLOG_DBASE_DROP | XLR_SPECIAL_REL_UPDATE);
+
+	pfree(xlrec);
 }
 
 /*
@@ -2166,8 +2179,6 @@ dbase_redo(XLogReaderState *record)
 		xl_dbase_drop_rec *xlrec = (xl_dbase_drop_rec *) XLogRecGetData(record);
 		char	   *dst_path;
 
-		dst_path = GetDatabasePath(xlrec->db_id, xlrec->tablespace_id);
-
 		if (InHotStandby)
 		{
 			/*
@@ -2196,11 +2207,26 @@ dbase_redo(XLogReaderState *record)
 		/* Clean out the xlog relcache too */
 		XLogDropDatabase(xlrec->db_id);
 
+		/*
+		 * If we crash after removing directories, we should avoid replaying
+		 * WAL records prior to the current WAL record (drop database).
+		 * Creating a restartpoint ensures that recovery will start from at
+		 * least this point onwards in the event of a crash / immediate
+		 * shutdown.
+		 */
+		RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE |
+						  CHECKPOINT_WAIT);
+
 		/* And remove the physical files */
-		if (!rmtree(dst_path, true))
-			ereport(WARNING,
-					(errmsg("some useless files may be left behind in old database directory \"%s\"",
-							dst_path)));
+		int i;
+		for (i = 0; i < xlrec->nspcids; i++)
+		{
+			dst_path = GetDatabasePath(xlrec->db_id, xlrec->tablespace_ids[i]);
+			if (!rmtree(dst_path, true))
+				ereport(WARNING,
+						(errmsg("some useless files may be left behind in old database directory \"%s\"",
+								dst_path)));
+		}
 
 		if (InHotStandby)
 		{
diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h
index d519252aad..d0582a726b 100644
--- a/src/include/access/xlog.h
+++ b/src/include/access/xlog.h
@@ -221,7 +221,6 @@ extern bool XLOG_DEBUG;
 /* These indicate the cause of a checkpoint request */
 #define CHECKPOINT_CAUSE_XLOG	0x0080	/* XLOG consumption */
 #define CHECKPOINT_CAUSE_TIME	0x0100	/* Elapsed time */
-
 /*
  * Flag bits for the record being inserted, set using XLogSetRecordFlags().
  */
diff --git a/src/include/commands/dbcommands_xlog.h b/src/include/commands/dbcommands_xlog.h
index 46be8a615a..a1654f25e0 100644
--- a/src/include/commands/dbcommands_xlog.h
+++ b/src/include/commands/dbcommands_xlog.h
@@ -34,7 +34,8 @@ typedef struct xl_dbase_drop_rec
 {
 	/* Records dropping of a single subdirectory incl. contents */
 	Oid			db_id;
-	Oid			tablespace_id;
+	uint16		nspcids;
+	Oid			tablespace_ids[0];
 } xl_dbase_drop_rec;
 
 extern void dbase_redo(XLogReaderState *rptr);
-- 
2.14.3 (Apple Git-98)