Fix replay of create database records on standby

Crash recovery on standby may encounter missing directories when
replaying create database WAL records.  Prior to this patch, the standby
would fail to recover in such a case.  However, the directories could be
legitimately missing.  Consider a sequence of WAL records as follows:

    CREATE DATABASE
    DROP DATABASE
    DROP TABLESPACE

If, after replaying the last WAL record and removing the tablespace
directory, the standby crashes and has to replay the create database
record again, the crash recovery must be able to move on.

This patch adds a mechanism similar to invalid-page tracking, to keep a
tally of missing directories during crash recovery.  If all the missing
directory references are matched with corresponding drop records at the
end of crash recovery, the standby can safely continue following the
primary.

Backpatch to 13, at least for now.  The bug is older, but fixing it in
older branches requires more careful study of the interactions with
commit e6d8069, which appeared in 13.

A new TAP test file is added to verify the condition.  However, because
it depends on commit d6d317d, it can only be added to branch
master.  I (Álvaro) manually verified that the code behaves as expected
in branch 14.  It's a bit nervous-making to leave the code uncovered by
tests in older branches, but leaving the bug unfixed is even worse.
Also, the main reason this fix took so long is precisely that we
couldn't agree on a good strategy to approach testing for the bug, so
perhaps this is the best we can do.

Diagnosed-by: Paul Guo <paulguo@gmail.com>
Author: Paul Guo <paulguo@gmail.com>
Author: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Author: Asim R Praveen <apraveen@pivotal.io>
Discussion: https://postgr.es/m/CAEET0ZGx9AvioViLf7nbR_8tH9-=27DN5xWJ2P9-ROH16e4JUA@mail.gmail.com

Author: alvherre

src/backend/access/transam/xlog.c

@@ -8107,6 +8107,12 @@ CheckRecoveryConsistency(void)
 		 */
 		XLogCheckInvalidPages();
 
+		/*
+		 * Check if the XLOG sequence contained any unresolved references to
+		 * missing directories.
+		 */
+		XLogCheckMissingDirs();
+
 		reachedConsistency = true;
 		ereport(LOG,
 				(errmsg("consistent recovery state reached at %X/%X",

src/backend/access/transam/xlogutils.c

@@ -34,6 +34,164 @@
 /* GUC variable */
 bool		ignore_invalid_pages = false;
 
+
+/*
+ * If a create database WAL record is being replayed more than once during
+ * crash recovery on a standby, it is possible that either the tablespace
+ * directory or the template database directory is missing.  This happens when
+ * the directories are removed by replay of subsequent drop records.  Note
+ * that this problem happens only on standby and not on master.  On master, a
+ * checkpoint is created at the end of create database operation. On standby,
+ * however, such a strategy (creating restart points during replay) is not
+ * viable because it will slow down WAL replay.
+ *
+ * The alternative is to track references to each missing directory
+ * encountered when performing crash recovery in the following hash table.
+ * Similar to invalid page table above, the expectation is that each missing
+ * directory entry should be matched with a drop database or drop tablespace
+ * WAL record by the end of crash recovery.
+ */
+typedef struct xl_missing_dir_key
+{
+	Oid			spcNode;
+	Oid			dbNode;
+} xl_missing_dir_key;
+
+typedef struct xl_missing_dir
+{
+	xl_missing_dir_key key;
+	char		path[MAXPGPATH];
+} xl_missing_dir;
+
+static HTAB *missing_dir_tab = NULL;
+
+
+/*
+ * Keep track of a directory that wasn't found while replaying database
+ * creation records.  These should match up with tablespace removal records
+ * later in the WAL stream; we verify that before reaching consistency.
+ */
+void
+XLogRememberMissingDir(Oid spcNode, Oid dbNode, char *path)
+{
+	xl_missing_dir_key key;
+	bool		found;
+	xl_missing_dir *entry;
+
+	/*
+	 * Database OID may be invalid but tablespace OID must be valid.  If
+	 * dbNode is InvalidOid, we are logging a missing tablespace directory,
+	 * otherwise we are logging a missing database directory.
+	 */
+	Assert(OidIsValid(spcNode));
+
+	if (missing_dir_tab == NULL)
+	{
+		/* create hash table when first needed */
+		HASHCTL		ctl;
+
+		memset(&ctl, 0, sizeof(ctl));
+		ctl.keysize = sizeof(xl_missing_dir_key);
+		ctl.entrysize = sizeof(xl_missing_dir);
+
+		missing_dir_tab = hash_create("XLOG missing directory table",
+									  100,
+									  &ctl,
+									  HASH_ELEM | HASH_BLOBS);
+	}
+
+	key.spcNode = spcNode;
+	key.dbNode = dbNode;
+
+	entry = hash_search(missing_dir_tab, &key, HASH_ENTER, &found);
+
+	if (found)
+	{
+		if (dbNode == InvalidOid)
+			elog(DEBUG1, "missing directory %s (tablespace %u) already exists: %s",
+				 path, spcNode, entry->path);
+		else
+			elog(DEBUG1, "missing directory %s (tablespace %u database %u) already exists: %s",
+				 path, spcNode, dbNode, entry->path);
+	}
+	else
+	{
+		strlcpy(entry->path, path, sizeof(entry->path));
+		if (dbNode == InvalidOid)
+			elog(DEBUG1, "logged missing dir %s (tablespace %u)",
+				 path, spcNode);
+		else
+			elog(DEBUG1, "logged missing dir %s (tablespace %u database %u)",
+				 path, spcNode, dbNode);
+	}
+}
+
+/*
+ * Remove an entry from the list of directories not found.  This is to be done
+ * when the matching tablespace removal WAL record is found.
+ */
+void
+XLogForgetMissingDir(Oid spcNode, Oid dbNode)
+{
+	xl_missing_dir_key key;
+
+	key.spcNode = spcNode;
+	key.dbNode = dbNode;
+
+	/* Database OID may be invalid but tablespace OID must be valid. */
+	Assert(OidIsValid(spcNode));
+
+	if (missing_dir_tab == NULL)
+		return;
+
+	if (hash_search(missing_dir_tab, &key, HASH_REMOVE, NULL) != NULL)
+	{
+		if (dbNode == InvalidOid)
+		{
+			elog(DEBUG2, "forgot missing dir (tablespace %u)", spcNode);
+		}
+		else
+		{
+			char	   *path = GetDatabasePath(dbNode, spcNode);
+
+			elog(DEBUG2, "forgot missing dir %s (tablespace %u database %u)",
+				 path, spcNode, dbNode);
+			pfree(path);
+		}
+	}
+}
+
+/*
+ * This is called at the end of crash recovery, before entering archive
+ * recovery on a standby.  PANIC if the hash table is not empty.
+ */
+void
+XLogCheckMissingDirs(void)
+{
+	HASH_SEQ_STATUS status;
+	xl_missing_dir *hentry;
+	bool		foundone = false;
+
+	if (missing_dir_tab == NULL)
+		return;					/* nothing to do */
+
+	hash_seq_init(&status, missing_dir_tab);
+
+	while ((hentry = (xl_missing_dir *) hash_seq_search(&status)) != NULL)
+	{
+		elog(WARNING, "missing directory \"%s\" tablespace %u database %u",
+			 hentry->path, hentry->key.spcNode, hentry->key.dbNode);
+		foundone = true;
+	}
+
+	if (foundone)
+		elog(PANIC, "WAL contains references to missing directories");
+
+	hash_destroy(missing_dir_tab);
+	missing_dir_tab = NULL;
+}
+
+
 /*
  * During XLOG replay, we may see XLOG records for incremental updates of
  * pages that no longer exist, because their relation was later dropped or
@@ -59,7 +217,6 @@ typedef struct xl_invalid_page
 
 static HTAB *invalid_page_tab = NULL;
 
-
 /* Report a reference to an invalid page */
 static void
 report_invalid_page(int elevel, RelFileNode node, ForkNumber forkno,

src/backend/commands/dbcommands.c

@@ -2185,7 +2185,9 @@ dbase_redo(XLogReaderState *record)
 		xl_dbase_create_rec *xlrec = (xl_dbase_create_rec *) XLogRecGetData(record);
 		char	   *src_path;
 		char	   *dst_path;
+		char	   *parent_path;
 		struct stat st;
+		bool		skip = false;
 
 		src_path = GetDatabasePath(xlrec->src_db_id, xlrec->src_tablespace_id);
 		dst_path = GetDatabasePath(xlrec->db_id, xlrec->tablespace_id);
@@ -2203,6 +2205,56 @@ dbase_redo(XLogReaderState *record)
 						(errmsg("some useless files may be left behind in old database directory \"%s\"",
 								dst_path)));
 		}
+		else if (!reachedConsistency)
+		{
+			/*
+			 * It is possible that a drop tablespace record appearing later in
+			 * WAL has already been replayed -- in other words, that we are
+			 * replaying the database creation record a second time with no
+			 * intervening checkpoint.  In that case, the tablespace directory
+			 * has already been removed and the create database operation
+			 * cannot be replayed.  Skip the replay itself, but remember the
+			 * fact that the tablespace directory is missing, to be matched
+			 * with the expected tablespace drop record later.
+			 */
+			parent_path = pstrdup(dst_path);
+			get_parent_directory(parent_path);
+			if (!(stat(parent_path, &st) == 0 && S_ISDIR(st.st_mode)))
+			{
+				XLogRememberMissingDir(xlrec->tablespace_id, InvalidOid, parent_path);
+				skip = true;
+				ereport(WARNING,
+						(errmsg("skipping replay of database creation WAL record"),
+						 errdetail("The target tablespace \"%s\" directory was not found.",
+								   parent_path),
+						 errhint("A future WAL record that removes the directory before reaching consistent mode is expected.")));
+			}
+			pfree(parent_path);
+		}
+
+		/*
+		 * If the source directory is missing, skip the copy and make a note of
+		 * it for later.
+		 *
+		 * One possible reason for this is that the template database used for
+		 * creating this database may have been dropped, as noted above.
+		 * Moving a database from one tablespace may also be a partner in the
+		 * crime.
+		 */
+		if (!(stat(src_path, &st) == 0 && S_ISDIR(st.st_mode)) &&
+			!reachedConsistency)
+		{
+			XLogRememberMissingDir(xlrec->src_tablespace_id, xlrec->src_db_id, src_path);
+			skip = true;
+			ereport(WARNING,
+					(errmsg("skipping replay of database creation WAL record"),
+					 errdetail("The source database directory \"%s\" was not found.",
+							   src_path),
+					 errhint("A future WAL record that removes the directory before reaching consistent mode is expected.")));
+		}
+
+		if (skip)
+			return;
 
 		/*
 		 * Force dirty buffers out to disk, to ensure source database is
@@ -2260,6 +2312,10 @@ dbase_redo(XLogReaderState *record)
 				ereport(WARNING,
 						(errmsg("some useless files may be left behind in old database directory \"%s\"",
 								dst_path)));
+
+			if (!reachedConsistency)
+				XLogForgetMissingDir(xlrec->tablespace_ids[i], xlrec->db_id);
+
 			pfree(dst_path);
 		}
 

src/backend/commands/tablespace.c

@@ -58,6 +58,7 @@
 #include "access/xact.h"
 #include "access/xlog.h"
 #include "access/xloginsert.h"
+#include "access/xlogutils.h"
 #include "catalog/catalog.h"
 #include "catalog/dependency.h"
 #include "catalog/indexing.h"
@@ -1529,6 +1530,22 @@ tblspc_redo(XLogReaderState *record)
 	{
 		xl_tblspc_drop_rec *xlrec = (xl_tblspc_drop_rec *) XLogRecGetData(record);
 
+		if (!reachedConsistency)
+			XLogForgetMissingDir(xlrec->ts_id, InvalidOid);
+
+		/*
+		 * Before we remove the tablespace directory, update minimum recovery
+		 * point to cover this WAL record. Once the tablespace is removed,
+		 * there's no going back.  This manually enforces the WAL-first rule.
+		 * Doing this before the removal means that if the removal fails for
+		 * some reason, the directory is left alone and needs to be manually
+		 * removed.  Alternatively we could update the minimum recovery point
+		 * after removal, but that would leave a small window where the
+		 * WAL-first rule could be violated.
+		 */
+		if (!reachedConsistency)
+			XLogFlush(record->EndRecPtr);
+
 		/*
 		 * If we issued a WAL record for a drop tablespace it implies that
 		 * there were no files in it at all when the DROP was done. That means

src/include/access/xlogutils.h

@@ -23,6 +23,10 @@ extern void XLogDropDatabase(Oid dbid);
 extern void XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
 								 BlockNumber nblocks);
 
+extern void XLogRememberMissingDir(Oid spcNode, Oid dbNode, char *path);
+extern void XLogForgetMissingDir(Oid spcNode, Oid dbNode);
+extern void XLogCheckMissingDirs(void);
+
 /* Result codes for XLogReadBufferForRedo[Extended] */
 typedef enum
 {

src/tools/pgindent/typedefs.list

@@ -3516,6 +3516,8 @@ xl_invalid_page
 xl_invalid_page_key
 xl_invalidations
 xl_logical_message
+xl_missing_dir_key
+xl_missing_dir
 xl_multi_insert_tuple
 xl_multixact_create
 xl_multixact_truncate